-------
In a study in which p53 deficient mice were intrapleurally injected with UICC crocidolite (200 jig/week),
Marsella et al. (1997) showed that p53 deficient mice exhibited substantially increased susceptibility to
development of mesothelioma. In this study, 12.5% of homozygous mice (p53 deficient) developed
mesothelioma and the rest died of lymphomas or hemangiosarcomas that develop spontaneously in these
mice; 76% of heterozygous mice died of mesothelioma,  and only 32% of wild-type mice (p53 competent)
died of mesothelioma. The authors suggest that p53 deficient mice are susceptible to excess proliferation
induced by crocidolite due to loss of control at the Gl/S check point that is normally mediated by p53.

In further confirmation of this hypothesis, Marsella et a!. (1997) report in the same study that 7.5 jig/cm2
of crocidolite applied to wild murine mesothelial cells in culture induced substantial apoptosis while
p53-deficient cells were resistant to apoptosis. The authors also note that most of the p53-deficient cells
are tetraploid (suggesting a loss of a spindle check point) while the wild type cells are all dipjoid.

Note that, although these studies  suggest that asbestos alone can induce complete transformation of both
lung epithelial cells and mesothelial cells, such studies must be evaluated with caution.  In the case of the
Hei et al. (1997) study, for example, the effects of the (asbestos-independent) mutations required to
initially establish the immortal line of lung epithelial cells are not entirely known. Therefore; the
response to asbestos of epithelial cells in vivo may be substantially different.               :

In the case of the Kravchenko et al. (1998) study, it is clear that mesothelial cells in vitro do not behave in
the same manner as those in vivo; in vitro, they spontaneously transform to tumorigenic cells; This
suggests that one or more growth inhibitory signals exist in vivo (which are absent in vitro) that are
critical to maintaining the health  of the mesothelium

6.3.3   Evidence that Asbestos Acts As a Cancer  Initiator

As indicated in a review by Jaurand  (1997), historically, the status of asbestos as a mutagen was
questioned, due primarily to the failure to produce detectable gene mutations in short-term assays.
However, more recent studies provide clear evidence that asbestos (and other fibrous materials) can
produce mutations. Moreover, fibrous materials may induce mutations by. multiple mechanisms
including, for example:

        •      direct interference with mitosis;

        •      production and release of reactive oxygen species (ROS); or

        •      production and release of reactive nitrogen species (RNS).

Of these, the most consistent, positive evidence that asbestos can act as a cancer initiator (i.el, a genotoxin
or mutagen) has been from assays designed to detect the kinds of genetic damage that result from
interference with mitosis (Jaurand 1997).  Although generation of ROS and RNS plays a cle|r role in
mediating asbestos-induced disease, the direct link to a role of asbestos as an initiator  is somewhat more
tenuous (see Sections 6.3.3.2 and 6.3.3.3).
6.3,3.1
Interference with mitosis
Apparently, the physical presence of asbestos fibers can interfere with proper spindle formation and the
function of other structural units required for mitosis (Jaurand 1997). This typically leads to, aneuploidy
(an incorrect number of copies of the chromosomes contained within a cell), development of micronuclei
(fragments of chromosomes enclosed in a membrane that are isolated from the main nucleus'of the cell),
and has also been shown to lead to clastogenic effects (changes in the organization and structure of the
                                              6.79

-------
chromosomes). Assays for these kinds of genetic alterations have consistently shown asbestos capable of
inducing these effects.

Jaurand (1997) also indicates that:

        •      fibers must be phagocytized by the target cells before they can interfere with mitosis;

        •      once phagocytized, all asbestos types are observed to interfere with mitotic activity; and

        •      samples enriched in long, thin fibers enhance these effects. In contrast, short fibers do
               not appear to contribute to these effects.

Jaurand (1997) notes, however, that results related to fiber size have not been entirely consistent,
primarily because not all studies have rigorously controlled for or even adequately characterized the sizes
of the fibrous structures in test materials. Jaurand (1997) also notes that this mechanism is not dependent
on the formation of reactive oxygen species or any other reactive free radicals.

Among studies that suggest that surface chemistry (or presumably fiber type) may not be an important
factor in determining the degree to which asbestos (or other fibrous materials) interfere with mitosis,
Keane et al. (1999), exposed cultured V79 cells (hamster lung fibroblasts) to untreated and HCl-treated
chrysotile asbestos. The acid treatment substantially reduces the magnesium content on the surface of the
chrysotile. The authors also  noted a "small" effect on fiber size due to treatment (treated fibers showed a
20% excess of short fibers).  Cells were exposed to doses ranging between 0.4 and 12.7 yg/cm2 (each
dose left in place for 24 hours and then rinsed off). Cells were harvested after an additional 24 hours and
evaluated for cytotoxicity and the presence of micronuclei.

Results from the Keane et al. (1999) study indicate that untreated fibers were shown to be slightly more
cytotoxic than treated fibers, but both treated and untreated fibers were observed to increase the
abundance of micronuclei in a similar, dose-dependent fashion. The induction of micronuclei appeared to
saturate at approximately 35/1,000 cells observed at an applied asbestos concentration of 40 ng/cm2. In
contrast, substantial cytotoxicity was only observed at the highest doses employed. The authors thus
concluded that surface chemistry (at least in terms of magnesium content) does not appear to have a major
affect on induction of micronuclei and that the observed genetic damage and cytotoxicity appear to occur
through entirely independent mechanisms.

In other studies, cultured cells were assayed for a variety of genotoxic effects following exposure to a
range of fibrous materials.

For example, Dopp and Shiffmann (1998) dosed human amniotic fluid (HAF) cells or Syrian Hamster
Embryo (SHE) cells with UICC amosite, Rhodesian chrysotile, crocidolite, or ceramic fibers at
concentrations of 0.5, 1.0, 5.0, and  10.0 ug/cm2 and assayed the cultures for formation of micronuclei and
a variety of clastogenic effects.

Based on their study, Dopp and Shiffmann (1998) report that all asbestos types induced formation of
micronuclei in SHE cells in a dose-dependent fashion at rates that were significantly elevated over control
animals. The effect appeared to saturate at doses between 1  and 5 ug/cm2 and rates appeared to peak at
between 48 and 66 hours post-exposure. Ceramic fibers, which were noted to be longer but thicker than
the asbestos fibers tested, also showed significantly increased induction of micronuclei, but at rates less
pronounced than for asbestos. However, it is difficult to judge whether this is due to differences in fiber
type because the data in the paper are not adequate to distinguish effects of fiber size and number from
effects of fiber type. Similar results were obtained with HAF cells, but overall rates were about one third
those observed in SHE cells.

                                               6.80

-------
                                                                                      l
The authors also note observing various disturbance of chromatin structure during interphase.' They
report observing formation of chromatin bridges and chromosome displacements in meta and anaphases
and impaired chromatin separation in mitosis. They also report that cytokinesis was frequently blocked.

In addition to the effects that they observed that are attributable to fiber interference with mitosis, Dopp
and Shiffmann report observing a variety of clastogenic effects.  In all cases where the authors labeled
specific regions  of specific chromosomes, fiber-exposed cells showed significantly greater frequencies of
DNA breaks over controls or gypsum-exposed animals. Different regions of various chromosomes also
showed significantly different frequencies of breakage with patterns that were specific to the different
fiber types.  The authors hypothesize that the observed clastogenic effects may be due to production of
reactive oxygen  species, to formation of some type of clastogenic factor, or to the direct interactions
between fibers and chromosomes during mitosis that are the apparent cause of the disturbances discussed
in the previous paragraphs. However, It was not possible to distinguish among hypothetical causes of the
observed clastogenic effects in this study.                                               •'

Kodamaet al. (1993)  exposed cultures of human bronchiolar epithelial (HBE) cells to asbestos (chrysotile
at 0 to 4 |ig/cm2  and crocidolite at 0 to 300 ug/cm2).  They then examined cells at 24, 48,72, and 96 hours
following exposure for cytotoxic effects and cytogenetic effects. Results indicate that both fiber types
induced concentration-dependent inhibition of cell proliferation and colony-forming ability, but chrysotile
was 100 to 300 times  more toxic. The authors report this translates to a 40-fold increase in toxicity on a
fiber for fiber basis (although the range of sizes  included in this count is not indicated).     ':

Kodama et al. (1993)  also report that, at 72 hours, chrysotile (4 ug/cm2) caused a 2.7-fold increase in
binuclei and a 1.6-fold increase in micronuclei.  Over the same time interval, at 300 ug/cm2, crocidolite
caused a 1.9-fold increase in binuclei, but did not cause micronuclei. They also report that chrysotile
failed to produce significant numerical chromosome changes in HBE cells and increased structural
aberrations only at the 24-hour time point. The  frequency of neither aneuploidy nor polyploidy was
increased at any time  point following exposure to asbestos in this study. The authors indicate that this
contrasts with observations of relatively high incidences of asbestos-induced chromosome changes
observed in some rodent cell cultures and clastogenic effects observed in human mesothelial 'cell lines.
They further speculate that phagocytic cells with high mitogenetic activity are likely most susceptible to
the effects of asbestos, which primarily interferes with mitosis. However, they suggest that epithelial
cells that are exposed  to fibers may undergo terminal differentiation (from Type II to Type I cells) and
thus cease mitosis. This would  effectively prevent adverse genetic effects from asbestos exposure.  Such
pathways are not available to mesothelial cells.                                          !.

Hart et al. (1994) studied the effects of a range of fiber types (with varying size distributions) on Chinese
hamster ovary (CHO) cells. The authors indicate that such cells are  very different from cells in
pulmonary tissues in that they are immortalized, aneuploid, undifferentiated, and preneoplasic.  They also
note that the responses observed in these cells differs from responses observed in cells from pulmonary
tissues.  Nevertheless, the implications concerning fiber size are instructive. Fibers evaluated included
long, medium, short, and UICC crocidolite and chrysotile along with a range of MMVF's, RCF's, and
fibrous glasses.  Exposure concentrations ranged between 10 and 225 ug/cm2.

Results from the Hart  et al.  (1994) study indicate that all of the fibers caused qualitatively similar toxic
effects: concentration-dependent reduction in cell numbers and an increase in the incidence of abnormal
nuclei with little or no loss in viability.  Fiber-induced cell death in CHO cells appears to be minor, even
at relatively high exposure concentrations. Based on mean dimensions (which is problematic), the
diameter dependence on the observed toxic effects, particularly on the formation of aberrant nuclei, was
slight or absent.  However, the effect with length was striking. For lengths up to at least 20 |im, potency
toward both cytotoxicity and the induction of aberrant nuclei increased dramatically  with increasing
length. The authors also note that the lack of an observed fiber composition associated effect on the

                                              6.81

-------
toxicity of CHO cells does not correlate with findings from recent rodent inhalation studies using the
same test fibers. The authors therefore speculate that CHO cells may not represent an appropriate in vitro
model for fiber effects. However, it may also be that effects in vitro occur over time scales that are rapid
relative to those that occur in vivo so that biodurability is not important in vitro.

In a study by Takeuchi et al. (1999) cultured human mesothelioma cells (MSTO211H) and, separately,
cultured human promyelocytic leukemia cells (HL60), which are not phagocytic, were dosed with
crocidolite between 0.6 and 6.6 ug/cm2 (no size range indicated). Studies with latex beads confirmed that
the mesothelioma cells are actively phagocytic, but that the leukemia cells are not. The authors indicate
that dosed mesothelioma cells showed significantly increased numbers of polynucleated cells, tetraploid
cells, and  cells with variable DNA content at the GO/G1  transition in the cell cycle and that the extent of
effects was dose-dependent. Leukemia cells showed no  such effects.

The authors further indicate that, when the mesothelioma cells were sorted by fiber content, those with the
highest fiber content showed the greatest effects.  The authors also indicate that cells are stimulated
neither to  release superoxide nor NO at the concentrations of crocidolite studied.  However, they
hypothesize that intracellular ROS may have been generated because they report finding increased levels
of 8-OH-dGua (and oxidized form of one of the DNA bases) following crocidolite exposure.
Nevertheless, the authors conclude that the mechanisms  by which crocidolite induce cytotoxicity and,
potentially, carcinogenicity is related to phagocytosis. Importantly, the effects described in the Takeuchi
et al. study are entirely consistent with effects attributable to interference with mitosis, despite any
speculation by the authors.

6.3.3,2        Generation of reactive oxygen species (ROS)

ROS have been implicated as mediators in a variety of toxic effects (including cancer initiation)
associated with a broad range of toxins (see, for example, Floyd 1990). Moreover, substantial evidence
indicates that asbestos can induce generation of ROS by several mechanisms  and that asbestos-induced
ROS play a role in several of the toxic effects attributable to asbestos (see below). However, whether
ROS play an important role in asbestos-associated cancer initiation is less clear and needs to be evaluated
carefully. Therefore, evidence that asbestos induces the production of ROS and that ROS contribute to
the adverse health affects attributable to asbestos are reviewed below with particular attention to effects
that may contribute to the initiation of cancer.  Contributions by ROS to other asbestos-related toxic
effects are also evaluated in later sections of this chapter.

Note, that generation of certain reactive nitrogen species (specifically the peroxinitrite ion) is closely
associated with ROS generation so that evidence for the generation of certain reactive nitrogen species
(Section 6.3.3.3) constitutes additional evidence for the generation of ROS.

Asbestos-induced Generation of ROS. Asbestos has been shown capable of generating a variety of
reactive oxygen species (ROS) including hydrogen peroxide (H2O2), superoxide (O2~), and hydroxyl
radical (OH«) via several mechanisms (see, for example, Fubini 1997; Jaurand 1997; Kamp et al. 1992)
including:

        •      catalytic production of superoxide from oxygen in aqueous solution;

        •      catalytic production hydrogen peroxide from oxygen in aqueous solution;

        •      catalytic production of hydroxyl radical by the Fenton reaction (degradation of hydrogen
               peroxide catalyzed by iron on the surface of fibers  or mobilized  from the surface of
               fibers);


                                              6.82

-------
        •      catalytic production of several ROS by redox cycling of iron on the surface of fibers or
               mobilized from the surface of fibers (Haber- Weiss type reactions);         '

        •      catalytic production of ROS by release of heme and heme protein from various cellular
               components;
                                                                                    \t
        •      by inducing release of various ROS species from phagocytes during "frustrated"
               phagocytosis; and
                                                                                    i
        •      by binding to cell receptors or other features of surface membranes that trigger signaling
               cascades mediating the production and release of various ROS (and RNS).  •
                                                                                    i
For a general review of the chemistry involved in these processes, see Floyd (1990).        j;

Fenton, Haber-Weiss, and Related Reactions. Most of the evidence for Fenton and Haberj Weiss
reactions (and related free  radical generating reactions) that take place on the surface of asbestos fibers
comes from experiments in cell-free systems (see below). Therefore, their relevance to the conditions
found in vivo may be limited. Moreover, the size of the fibers (especially in terms of their cumulative
surface area) and the history of their surfaces (in terms of metal contaminants or coatings that might be
present) may substantially alter the effects of such experiments (Fubini 1997).  Unfortunately, however,
few of the available studies report characterization of fiber sizes or surface conditions in sufficient detail
to judge the importance of these effects.

For fibers. Zalma et al. (1987) evaluated a range of fibers (UICC crocidolite, UICC amosite,1 UICC
Canadian and Zimbobwean chrysotile, industrial chrysotile, and magnetite) for their ability to produce
free radicals by the direct reduction of oxygen in aqueous solution.  In some cases, hydrogen:peroxide
was also added to the solution. Results indicate that all of the fibers tested were able to generate hydroxyl
radicals (even in the absence of hydrogen peroxide), but that the efficiency of production was a strong
function of the activation (by grinding) or pacification (by coating with benzene or other agents) of the
fiber surface. Chrysotile was reported to be the most efficient at generating radicals and the authors
assumed that this is due to iron contamination on the surface (since  iron is not a component of the "ideal"
chrysotile fiber).  However, such conclusions are difficult to evaluate in the absence of simultaneous
consideration of fiber size.

Governa et al. (1998) evaluated the ability of wollastonite fibers to generate  ROS in both a cell free
system and a suspension of polymorphonucleocytes (PMN's). The  fibers were observed to produce ROS
in both systems and that ground wollastonite produced substantially more ROS  than unground. The
efficiency of ROS generation in PMN suspension is also reported to be greater for wollastonite than for
either chrysotile or crocidolite (tested in previous studies). However, no size information is given. Based
on additional work with various inhibitors added to the system, the authors also indicate thationly a
fraction of the ROS generated was composed of hydroxyl radicals.                        '
                                               >
Brown et al. (1998) subjected several different fiber types (amosite, silicon carbide whiskers,'RCF-1, and
various fibrous glasses) to  two standard chemical assays for free-radical production (in cell-free systems).
The authors indicate that, of the fiber types tested, only amosite showed free radical activity significantly
above controls in both assays and only RCF-1 additionally showed significantly elevated free radical
activity in one assay. However, there is not enough information provided in this study to determine
whether the observed differences are due to differences in fiber sizes, sample preparation (i.e., surface
condition), or fiber type. In apparent contrast, for example, Gold et al. (1997) report that amosite and
crocidolite produce few free radicals in cell free systems, unless they are ground.
                                              6.83

-------
Weitzman and Graceffa (1984) indicate that chrysotile, crocidolite, and amosite are all capable of
catalyzing the generation of hydroxyl radicals and superoxide from hydrogen peroxide in vitro and that,
based on experiments with various iron chelators, these reactions are iron dependent. The authors further
indicate that hydrogen peroxide is produced in large quantities as a normal bi-product of tissue
metabolism, but that it is effectively scavenged by various enzymes. The authors speculate that, by
physically damaging cell membranes, asbestos may allow release of the precursor hydrogen peroxide
before it can be scavenged.

For particles. Silica, residual fly ash, and ambient air dusts also can create ROS in vitro and the
efficiency of production correlates with ionizable concentrations of various transition metals complexed
on the dust (Martin et al. 1997).  In vivo, such metal containing particles also cause release of ROS from
macrophages (Martin et al. 1997).  Additionally, binding of silica to plasma membranes  of airway lung
cells and phagosomes provokes generation of ROS.

Castronova et al. (1997) further indicate that it is the concentration of contaminating  iron on the surface
of freshly fractured quartz that enhances free radical production in aqueous solution (in cell free systems).
These authors also showed that high iron-containing (430 ppm) quartz dust inhaled by rats (at 20 mg/m3
for 5 hours/day for 10 days) induced 5 times more leukocyte recruitment, 2 times more lavageable red
blood cells, 30-90% increase in macrophage production of ROS, 71% increase in nitric oxide production
by macrophages, and 38% increase in lipid peroxidation of lung tissue than observed in rats exposed to
low iron-containing (56 ppm) quartz.

Although iron is required for the reactions considered here, studies indicate that the iron content of the
fiber itself is not a good indicator of reactivity (Gold et al. 1997).  Studies also indicate that the iron that
participates in these reactions need not originate with the fiber (Fubini 1997; Jaurand 1997) and
biological systems contain abundant sources of iron. Therefore, both iron-containing fibers and iron-free
fibers have been shown to participate in these reactions in vivo.

Release of Items and heme protein. At least one research group (Rahman et al.  1997) indicates that heme
and heme protein cause extensive DNA damage in the presence of asbestos in vitro and, based on
previous studies, that this may involve heme catalyzed production of ROS following  asbestos-induced •
release of heme from cytochrome P-450, from prbstaglandin H synthetase (or perhaps from other heme
containing proteins).  Importantly, the authors indicate that such observations relate to a nuclear pool of
heme, which suggests that ROS generation via this mechanism may occur in the immediate vicinity of
DNA.  The work by this group suggests at least one additional pathway by which asbestos may induce the
production of ROS and by which ROS-mediated damage to DNA might occur.

Frustrated phagocytosis. Numerous studies indicate that long asbestos fibers (longer than somewhere
between 10 and 20 urn) cannot be efficiently phagoycitized by macrophages (see, for example, Sections
4.4 and 6.2) and that macrophages that are damaged by such "frustrated" phagocytosis release ROS (see,
for example, Kamp et al. 1992; Mossman and Marsh 1991). Due to the differences in the size of
macrophages across species (see discussion of Krombach et al. [1997] in Section 4.4). The minimum
length beyond which phagocytosis may become frustrated may differ in different animals. However, it is
clearly longer fibers that contribute to this mechanism for generating ROS. Shorter fibers (<10 (im) are
unlikely to cause frustrated phagocytosis in any of the mammalian species of potential interest.

Lira et al. (1997) showed that alveolar macorphages in culture (after stimulation with
lippopolysaccharide) generated free radicals (ROS) when subsequently exposed to chrysotile, crocidolite,
or amosite (all UICC). They found chrysotile to be the most potent inducer of free radical activity (which
is not surprising given that UICC chrysotile contains the highest fraction of long fibers of the UICC
samples tested (Berman, unpublished).  Based on tests with various inhibitors, the authors indicated that


                                              6.84

-------
the free radicals generated by the alveolar macrophages occurred through a pathway mediated by protein
tyrosine kinase, phosphoISpase C, and protein kinase C and that the effects are dose-related. '

Kostyuk and Potapovich (1998) cultured peritoneal macrophages and showed that treatment with
chrysotile asbestos (1 jj-g, no size data given) resulted in production of frustrated phagocytosis and cell
injury (the latter as evidenced by release of lipid dehydrogenase, LDH, a marker for membrane damage).
By working with various chelators and flavonoids (natural plant products, some of which quench
superoxide and some of which chelate iron), the authors indicate that cell injury was likely induced by
superoxide and that the superoxide was likely produced by an iron-dependent mechanism.  Note that this
contrasts with the above studies that suggests production of radicals by frustrated phagocytosis in culture
is an iron-independent mechanism.

At least for one kind of phagocyte: polymorphonucleocytes (PMN's), a study by Ishizaki et al. (1997)
suggests that crocidolite and erionite may induce production of ROS from PMN's by each of two
mechanisms. The first requires phagocytosis and may represent the traditional, "frustrated" phagocytosis
pathway indicated above.  The second pathway is triggered by an interaction between the fiber and the
cell surface and is mediated by NADPH. The authors also cite evidence that chrysotile may similarly act
through both of these pathways.

Afaq et al. (1998) cultured alveolar macrophages and peripheral red blood cells (RBC's) that were
harvested from rats 30-days following a single 5 mg intratracheal instillation of UICC crocidolite, U1CC
chrysotile, or ultrafine titanium dioxide. The authors indicate that  populations of alveolar macrophages'
were significantly increased (over sham-exposed rats) for all three  particle types and that acid
phosphatase and lipid dehydrogenase (LDH), which are markers of cell membrane damage, were
observed in cell-free lung lavage from animals exposed to all three particle types.  Both alveolar
macrophages taken from asbestos-exposed animals (both types) showed significantly elevated lipid
peroxidation and hydrogen peroxide production over titanium dioxide exposed animals.  However, the
latter also showed elevated peroxidation and peroxide production that were significantly elevated over
sham-exposed animals. Similar results were observed among RBC's from asbestos-exposed|animals, but
not from titanium dioxide-exposed animals. Note, it is possible that ROS production induced by Ti02
occurs by a different mechanism (or set of mechanisms) than that for asbestos (see, for example, Palekar
etal.  1979, Section 6.3.4.4).                                                           ;
                                                                                    !'
In vivo Evidence for Asbestos-generated ROS. Several studies involving whole animals  also indicate
that asbestos exposure induces the generation of ROS. Importantly, in  such studies, evidence for
generation of ROS is generally determined based on observation of the putative effects of ROS, rather
than ROS directly.                                                                  !

Ohio et al. (1998) intratracheally instilled 500 jig of crocidolite (NIEHS) into rats.  This was/observed to
induce a neutrophilic inflammatory response within 24 hours (in contrast to saline-exposed rats). The
authors collected chloroform extracts from exposed lungs and subjected them to electron spin residence
(ESR) spectrometry.  Results indicate the presence of a carbon-centered radical adduct that Has a structure
consistent with products of lipid-peroxidation. The radical signal was only observed in asbestos-exposed
animals and persisted even after one-month following exposure. The authors also report that depletion of
neutrophils did not affect the signal and that dextrin-induced inflammation did not produce the signal.
                         •*
Yamaguchi et al. (1999) studied effects in rat lung tissue at 1,3, 5,7, and 9 days following a^ingle
intratracheal instillation of 2mg of either glass fiber or UICC crocidolite. The authors indicate
significantly increased levels of 8-OH-Guanine (an oxidized form  of the DMA base Guanine) one day
after crocidolite instillation and increasing repair activity for this oxidized form of guanine with time that
became significant at days 7 and 9 following instillation.  Glass fibers (noted to be non-fibrogenic and
                                              6.85

-------
non-tumorigenic) did not produce either increases in 8-OH-Gua or its repair activity. The effects
associated with crocidolite were all noted to be dose related.

Several of these studies also suggest distinctions in ROS generation (either the absolute generation of
ROS or generation of specific ROS species by specific tissues) due to differences in fiber (or particle) size
or type.

Nehls et al. (1997) intratracheally instilled rats either with quartz (2.5 mg) or corundum (2.5 mg).  The
latter mineral is reportedly non-tumorigenic.  Results indicate that lung epithelial cells in quartz exposed
rats exhibited increased 8-oxo-Guanine levels (a DNA adduct generated by reaction with ROS, see above)
that persisted for up to 90 days post-exposure. Elevated levels of the DNA adduct appeared in all  cell
types in all areas of the lung. The authors suggest that the observed persistence of the elevated levels of
8-oxo-Guanine suggests that it was produced at a rate in excess of the lung's capacity for repair. The
authors also report enhanced and persistent inflammation, cell proliferation, and an increase in neutrophil
population in bronchio-alveolar lavage (BAL) fluid and an increase in tumor necrosis factor alpha
(TNF-a) in BAL fluid in the quartz exposed animals.  TNF-a is a cytokine linked to a variety of effects
including the recruitment of inflammatory cells (see Table 6-6).  In contrast, exposure to corundum
produced none of the effects observed with quartz.

Timblin et al. (1998b) report that ROS induced responses by rat lung epithelial cells vary depending on
whether exposure is to crocidolite, hydrogen peroxide, or cadmium chloride (CdCl2). In response  to ROS
generation induced by the first two agents increased levels of cJun protein (a protooncogene, Table 6-6) is
observed.  Further, crocidolite, but not hydrogen peroxide, causes elevation in the levels of manganese-
containing superoxide (MnSOD) dismutase (an enzyme that catalyzes dismutation of superoxide, Table
6-6). Neither of these agents affect levels of either of two common stress proteins (Table 6-6): GRP78 or
HSP72/73, nor do they affect cellular glutathione levels.  In contrast, cadmium chloride does not alter
MnSOD levels, but increases levels of GRP78 and HSP72/73 in addition to cJun protein. Therefore,
ROS-related mechanisms may be complex and may be toxicant-specific. Thus, it may not be correct to
assume that all fibers and particles act through common ROS-related pathways.

In contrast to the results of the above study (which showed no affect on glutathione levels), Golladay et
al. (1997) showed that human lung epithelial cells (cultured A549 cells) exposed to crocidolite (NIEHS
sample) showed substantial reduction in intracellular levels  of glutathione (without increases in  the
oxidized forms of glutathione). Rather, an associated increase in extracellular, reduced glutathione was
observed, suggesting that crocidolite induces release of glutathione from the interior of these cells to the
environment.  The authors also indicate that, given that the half-life for reduced glutathione outside of
cells is on the order of an hour, while extracellular reduced glutathione levels remained elevated for more
than 24 hours following exposure, cells must have been releasing reduced glutathione continuously.
Because no concomitant release of LDH  or labeled adenine  was observed (despite loading of cells with
labeled adenine prior to the experiment), the authors conclude that release of glutathione is not associated
with membrane disruption or apoptosis (which is induced to some degree by exposure to crocidolite).
Also, all of the effects described above were similarly associated with exposure to de-ironized crocidolite.
Thus, the iron content of the fibers does not play a role in this process.

Note that the apparent difference in the reported effect of crocidolite exposure on intracellular glutathione
levels in the above two studies might be due (individually or in combination) to differences in the  cell-
types studied, differences in the size distribution of the crocidolite employed, differences in study  design,
or other factors. Insufficient information is available to distinguish among these possibilities.

Kaiglova et al. (1999) intrapleurally injected rats with 10 mg of long amosite and collected bronchio-
alveolar lavage (BAL) fluid 24 hours following exposure and at later time intervals.  They indicate that
total protein and alkaline phophatase (AP) were both elevated in BAL 24 hours after exposure and that

                                               6.86

-------
AP remained elevated for at least 3 months following exposure.  They also noted increased levels of lipid
peroxides in BAL at 24 hours, but not 3 months following exposure. The authors indicate that
antioxidants were significantly decreased following exposure: glutathione was significantly decreased in
lung tissue at both 24 hours and 3 months following exposure, but was normal in BAL fluid at all time
points; a-tokopherol and retinol were significantly decreased at 3 months in lung tissue; and ascorbic acid
was significantly decreased in both lung tissue and BAL at 24 hours and remained low at 3 months. The
authors indicate that decreases in antioxidants implies a role for ROS (or RNS) in lung tissue injury.  It is
also possible that the varied responses of specific antioxidants may suggest a role for toxin- and injury-
specific ROS/RNS.                                                                  ;

Conclusions Concerning Generation of ROS. Except for frustrated phagocytosis (which is unique to
long fibers), ROS generation by the mechanisms discussed above are considered a common response to
respiration of particles in general (see, for example, Martin et al. 1997 who suggest that ROS "...may be a
global signaling mechanism mediating response to paniculate insult mostly by activation of kinases and
transcription factors common to many response genes." They further indicate that if the load of ROS
generated is too great, or the  airway in which it is generated has been previously impaired, ".Lthese same
mechanisms can result in deleterious respiratory lesions and outright pathology").  However,';not all'non-
fibrous particles are similarly capable of inducing production of ROS. As indicated above, fbr example,
while crystalline silica is a potent inducer of ROS production, carundum is not (Nehls et al. 1997).
Moreover, the spectrum of ROS (set of species) that are induced by particular toxicants are generally
specific to the offending toxicant (Timblin et al. 1998b). Therefore, the generic grouping of ROS
mediated pathways by particles and, especially, by  particles and fibers, does not appear justified. These
mechanisms are more complex and individualized than such generic grouping suggests.    '

ROS can be generated by multiple pathways that are variously dependent on particle size, whether a
particle is a fiber, fiber size, and particle or fiber type (i.e., chemical composition).  The primary
mechanism(s) through which ROS are generated in response to one type of particle or fiber may be very
different than that through which ROS generation is induced by another and the resulting suite of ROS
(set of species) may also differ.  Importantly, the relationship between dose and  response foneach
mechanism may also differ (see, for example, Palekar et al. 1979, Section 6.3.4.4).         !.

Given the above, comparing among the ability of fibrous materials and non-fibrous analogs to induce
generation of ROS requires that such analogs be properly matched before valid conclusions can be drawn.
Thus, for example, the appropriate non-fibrous analog for crocidolite is the massive habit of reibeckite
and the appropriate analog for chrysotile is the massive habit of antigorite or lizardite.  Due to differences
in both chemistry and crystal structure, crystalline silica is not an appropriate non-fibrous analog for any
of the asbestos types. Moreover, because ROS can be generated by different mechanisms, critical
comparison across analogs requires more than the simple confirmation that ROS are generated or even
whether the relative efficiency with which ROS are generated is comparable. It is also necessary to
contrast the specific complexion of ROS (set of species) generated and the specific tissue/cellular
environments (i.e., locations within a cell) in which they are generated in response to each analog.

It is also clear that ROS are generated by both iron-dependent and iron-independent pathways. Even for
the  iron-dependent pathways, however, the source of iron need not derive from the fibers or particles
themselves. Therefore, since iron is abundant in vivo (and the environment), both iron-containing and
iron-free fibers (or particles) can potentially participate in both the iron-independent and the iron-
dependent pathways.
                                                                                   .1
Effects Mediated by ROS. ROS have been implicated as mediators in a variety of toxic effects
(including cancer initiation) associated with a broad range of toxins (see, for example, Floyd 1990;  Martin
et al. 1997). Cellular and tissue effects that have been associated with the effects of ROS include:
                                              6.87

-------
               enhancement of overall uptake of particles by epithelium;
               stimulation of inflammatory responses;
               stimulation of various signaling cascades and production of cytokines;
               inducement of apoptosis;
               cytotoxicity;
               mediation of cell proliferation;
               formation of oxidized macromolecule (including DNA) adducts; and
               induction of DNA strand breaks.

However, only the last two of the above list of ROS effects are potentially relevant to the initiation of
cancer (the topic of this section).  The other effects in the above list likely contribute to the induction of
other asbestos-related diseases and may even promote (but not initiate) asbestos-related cancer. Thus,
they are addressed further in later sections of this chapter (see below).

Although, ROS generation is associated with exposure to various particles and fibers (including all forms
of asbestos), generation of ROS does not necessarily imply carcinogenesis. For example, Zhu et al.
(1998) indicate that ROS generation is induced in response to exposure to asbestos, crystalline silica, and
coal mine dust. However, based on an extensive record of human exposure, the latter (coal mine dust) is
not carcinogenic in humans." Therefore, evidence related to the last two of ROS-associated effects listed
above, are examined  in more detail below.

Several studies indicate that, once generated by exposure to asbestos, ROS can interact with DNA in vitro
and in vivo to produce oxygenated adducts, primarily 8-oxo-Guanine.  Of the studies reviewed above, for
example, Yamaguchi et al. (1999) observed that crocidolite, but not (non-tumorigenic) glass produced
dose-dependent increases in 8-oxo-Guanine in rat lung tissue following intratracheal instillation. Also,
Leanderson et al. (1988), Park and Aust (1998), Keane et al. (1999) indicate that asbestos induces
formation of oxidized DNA adducts in in vitro assays. Brown et al. (1998) indicate that asbestos induces
ROS-mediated DNA strand breaks.

However, not all DNA strand breaks attributable to asbestos occur through pathways involving ROS.
OHikainen et al. (1999) exposed cultures of human mesothelial cells (MeT-5A, transfected with SV-40,
but nontumorigenic)  to hydrogen peroxide (100 uM) or crocidolite (2-4 ug/cm2), either alone or'in
combination with TNF-a and performed assays for DNA strand breaks. The authors note that the
concentrations of asbestos evaluated are well below those that have been associated with cytotoxic
effects.  Crocidolite alone was shown to produce DNA strand breaks at the concentrations tested.  The
authors also note that, at lower concentrations, only reversible effects were observed (presumably
indicating DNA repair). Co-exposure to crocidolite and TNF-a increased the observed incidence of DNA
damage, but the effect was less than additive. The  authors indicate that additional studies with
antioxidants indicate that the DNA damage induced by crocidolite in this study occurs through a
mechanism that does not involve ROS.  In fact, a potentially much more substantial mechanism by which
asbestos may induce  DNA breaks and various clastogenic alterations involves its ability to interfere with
mitosis (Section 6.3.3.1).

There is also evidence that different tissues may respond to asbestos-induced ROS generation differently.
For example, Zhu et  al. (1998) indicate that MnSOD is found in the mitochondria of Type 11 epithelial
cells of rats exposed  to crocidolite. The authors further indicate that, because fibroblasts, alveolar
macrophages, or endothelial cells do not display this protein when stimulated by exposure, this suggests a
difference in the susceptibility of epithelial cells to certain types of asbestos-induced injury.

Importantly, although these studies provide evidence that indicates asbestos is capable of producing
oxidized DNA adducts (or strand breaks) through ROS mediated processes, they do not  address the
question of whether such adducts can lead to heritable mutations in DNA in vivo nor do they indicate

                                               6.88

-------
whether such adducts lead to tumor production. Therefore, such studies should only be construed to
suggest a potential that asbestos can act as a cancer initiator through ROS-mediated pathways;

As previously indicated, not all mechanisms involving the generation or effects of ROS are similarly
fiber-size or fiber-type dependent and those that are do not necessarily depend on these variables in the
same way. At this time, it is not possible to distinguish among the relative importance of the different
mechanisms, so that it is difficult to judge the relative importance of the different effects.  However, the
overall general implication (with the few exceptions noted) is that ROS generation more likely contributes
to other asbestos-related diseases and to cancer promotion than to cancer initiation.         ;!

One final note concerning the effects of ROS that specifically involves the behavior of the hydroxyl
radical is also warranted. The hydroxyl radical is an extremely reactive species. Whether in vitro or in
vivo, this species will react with virtually 100% efficiency with every organic molecule it encounters.
Therefore the effects attributabfe to the hydroxyl radical are limited to those involving reactions in the
immediate vicinity of the location at which it is generated.  Thus, unless asbestos-induced generation of
this radical occurs within the nucleus and in the immediate vicinity of susceptible strands of DNA, it is
unlikely that these radicals are the direct cause  of DNA damage.                          |

Rather, hydroxyl radicals tend to react with cell membranes and other cellular components to'produce
further intermediate radicals (primarily lipid peroxides), which are much more stable than the'hydroxyl
radical and may migrate substantial distances before having an effect.  It is likely that these intermediate
radicals are ultimately responsible for any ROS-mediated DNA damage that may be attributed to
asbestos.                                                                            ••

6.3.3.3     >  Generation of reactive nitrogen species (RNS)                       "•

Nitric oxide (NO) is produced ubiquitously in biological  systems and serves many functions (Zhu et al.
1998). It is highly reactive and, therefore, generally short-lived in vivo.  However, nitric oxide has been
shown to react with superoxide (02~) to form the peroxynitrite ion (ONOO") at near diffusion-limited rates
(see, for example, Zhu et al.  1998). This the peroxynitrite ion (an RNS) may represent the primary
species responsible for the effects attributed to  ROS (at least when the primary ROS formed is
superoxide).                                                                          ;
                                                                                     i
Asbestos-induced Generation of RNS.  Alveolar macrophages, lung endothelial and epithelial cells, and
alveolar epithelium (in both  rats and humans), when stimulated by inflammatory agents, generate both
superoxide and over-produce nitric oxide. These then  combine to produce the perxoynitrite ion (Martin et
al, 1997; Zhu et al.  1998). These cells up-regulate production of NO when stimulated by various
cytokines, lipopolysaccharide, and interferon y. Because the peroxynitrite ion is a strong oxidant and
nitrating agent and is extremely reactive, evidence for its production is generally indicated in most studies
by the presence of nitrotyrosine, the stable product of tyrosine nitration, (Zhu et al.  1998). Evidence for
production of NO is frequently indicated by observation of nitrite.  Numerous studies also provide
evidence of nitric oxide and  peroxynitrite ion production specifically in response to exposurejto asbestos
in various tissues.                                                                     "•          •  .

Both chrysotile and crocidolite up-regulate the production of nitric oxide by alveolar macrophages in the
presence of interferon-y and the interaction between asbestos and interferon is synergistic. Non-
fibrogenic carbonyl iron did  not induce nitric oxide formation (Zhu et al. 1998). These authors also cite a
study in which intratracheal  instillation of silica and coal mine dust caused more inflammation and nitric
oxide formation than TiO2 or carbonyl iron (on an equal particle basis).  This suggests both tHe geometry
and chemical composition of particles determine their ability to up-regulate nitric oxide.    '.
                                               6.89

-------
Inhalation of chrysotile or crocdolite induces secretion of both TNF-a and nitric oxide by pleural
macrophages (Tanaka et al. 1998). Tanaka et al. (1998) studied the effects of RNS in rats exposed by
inhalation to 6 to 8 mg/m3 crocidolite or chrysotile (both NIEHS samples) for 6 hours/day, 5 days/week
for 2 weeks. Rats were sacrificed at 1 and 6 weeks following exposure. The authors indicate that
asbestos induces formation of stable products of nitric oxide in cells obtained by lung lavage 1 week after
asbestos exposure. Nitrotyrosine (a marker for ONOO" formation) was also observed. Also, a greater
number of alveolar macrophages and pleural macrophages were shown to express iNOS protein (the
inducible form of nitric oxide synthase, Table 6-6) than sham exposed animals. Exposed rats showed
significantly elevated immuno-staining for nitrotyrosine in the region of thickened duct bifurcations as
well as within bronchiolar epithelium, alveolar macrophages, and mesothelial cells of both the visceral
and parietal pleura. Nitrotyrosine staining was persistent, being observed at both 1 and 6 weeks following
exposure.

Quinlin et al. (1998) studied the production of nitric oxide in rats exposed to crocidolite or chrysotile
asbestos (both NIESH reference samples). Rats were exposed by inhalation at 6 hours/day, 5 days/week
and lavaged at 3,9, and 20 days. Lavaged macrophages showed significantly increased nitrite/nitrate
(indicating production of nitric oxide) and this was suppressed  with inhibitors to iNOS. Thus, nitric oxide
is produced via an iNOS pathway. The authors also note that nitric oxide production correlated
temporally with neutrophil influx in the lavage fluid. They also indicate that asbestos exposed animals
showed a 3- to 4-fold increase in iNOS positive macrophages in their lungs.

Quinlin et al. (1998) also exposed cultured murine alveolar macrophages (RAW 264.7 cells) to
crocidolite, riebeckite, and crstobalite silica in vitro. These cells showed increased iNOS mRNA
following exposure to asbestos and even greater increases if the cells were also stimulated with
lipopolysaccharide (LPS). Both crocidolite and riebeckite (but not cristobalite silica) stimulated increased
iNOS promoter activity when applied in combination with LPS. Thus, in this case, their appears to be a
mechanism that is sensitive to particle composition, but not size.

Park and Aust (1998) treated cultures of human lung epithelial  (A549) cells with crocidolite and observed
induction of iNOS and  reduction of intracellular glutathione (GSH) levels. Based on studies with
inhibitors, the authors further indicate that iron mobilized from crocidolite was required both for
formation of nitric oxide and to generate 2'-deoxy-7-hydro-8-oxoguanosine, but not for the observed
decrease in intracellular glutathione. The  authors note that approximately 5 times as much chrysotile
(containing approximately 3%  iron) as crocidolite was required to produce the same level of nitric oxide
formation. Importantly, these experiments were conducted in vitro in serum-free medium, so that no
extra-biological source  of iron  was present.

Choe et al. (1998) dosed cultured rat pleural mesothelial cells with either chrysotile or crocidolite (both
NIEHS samples) at concentrations between 1.05 and 8.4 ug/cm2 with or without co-stimulation with 50
ng/ml of interleukin-1 p (IL-lp). The authors report that mRNA for iNOS in asbestos and IL-lp dosed
cells increased progressively from 2 to 12 hours following exposure. Both types of asbestos also
stimulated production of nitric  oxide (measured as nitrite) in IL-lp stimulated cells in a dose- and time-
dependent fashion. Both types of asbestos also induced expression of iNOS protein and formation of
nitrotyrosine (based on nitrate detection) in IL stimulated cells. In contrast, carbonyl iron particles did
not induce any of the effects observed  for asbestos in IL stimulated cells.  Thus, formation of RNS
appears to be either fiber size dependent (not induced by particles) or mineralogy-dependent (or both).

As with the production of ROS, RNS production is apparently a function of multiple, complex
mechanisms. Also, as with production of ROS, the dose-response characteristics of the various
mechanisms differ.  Several of the mechanisms show a strong dependence on fiber size and some
dependence on fiber type. However, there are other mechanisms that are dependent primarily on fiber (or
particle) type, but may  not be dependent on size (or at least not dependent on fiber size).  At this point in

                                              6.90

-------
 time, it is not possible to gauge the relative importance of the various mechanisms by which RNS may be
 generated, except that it is likely that the importance of the various mechanisms likely differ in different
 cells and tissues and likely differ as a function of the specific toxin whose presence is inducing RNS
 production.

 There are also indications that production of RNS may be (animal) species specific.  For example, Jesch
 et al. (1997) report that alveolar macrophages harvested from rats expressed iNOS when stimulated with
 either LPS (Hppopolysaccharide) or interferon-y. In contrast, iNOS expression could not be induced in
 hamster, monkey or human macrophages.

 Effects Mediated by RNS. Based on Zhu et al. (1998), over-production of nitric oxide can:  .

        •      inactivate critical enzymes;                                            '
        •      cause DNA strand breaks that result in activation of poly-ADP-ribosyl transferase
               (PARS);                                                             '
        •      inhibit both DNA and protein synthesis; and                             ;
        •      form peroxynitrite by reaction with superoxide.

 In turn, peroxynitrite ion may:

        •      initiate iron-dependent lipid peroxidation;
        •      oxidize thiols;
        •      damage the mitochondrial electron transport chain; and
        •      nitrate phenolics (including tyrosine).                                   •
                                                                                   ii
 Also, some of the damage to alveolar epithelium and pulmonary surfactant system previously attributed to
 reactive oxygen species may actually be caused through ROS generation of peroxynitrite.  For example,
 Chao et al. (1996) report that crocidolite treatment of human lung epithelial cells (A549 cells) results in
 formation of 8-hydroxy-2'-deoxyguanosine (8-OHdG) in DNA and synthesis of mRNA for iNOS.  An
 iNOS inhibitor reduces intracellular nitrite and eliminates production of 8-OHdG. Addition of
 independent NO donor, recovers production of 8-OhdG.  Thus, production of the oxygenated-DNA
 adduct in this case appears to be generated by reaction with RNS.

 As with ROS, it is primarily the potential for RNS to contribute to DNA damage (e.g., strand breaks or
 generation of oxygenated adducts) that represent the primary pathways by which RNS might participate
 in cancer initiation (as opposed to cancer promotion or other asbestos-related diseases).  It appears that
 RNS-mediated DNA damage is closely associated with ROS generation and mediation of DNA damage
(see  Section 6.3.3.2). Thus, there is little to add here.                                   '
6.3.3.4
Conclusions concerning asbestos as a cancer initiator
The strongest, most consistent evidence that asbestos can act as a cancer initiator relates to the tendency
of asbestos to interfere with mitosis.  Although there is evidence that asbestos may induce production of
DNA adducts and DNA strand breaks (through ROS and RNS mediated pathways), whether such adducts
or breaks ultimately lead to permanent, heritable changes to DNA remain to be demonstrated.1 The
relative importance of the ROS/RNS mediated pathways compared to the pathway involving interference
with mitosis also remains to be determined. As indicated in later sections, however, ROS/RNS mediated
pathways may play substantial roles in cancer promotion and induction of other asbestos-related diseases.

Regarding the primary mechanism by which asbestos may initiate cancer (interference with mitosis), the
pathway is length dependent (short fibers do not appear to contribute to the effect).  Further, although
                                             6.91

-------
there may also be a dependence on fiber type (chemical composition), it is apparent that all types of
asbestos can act through this pathway. The dependence of this pathway on fiber diameter is less clear.
Thus, other than to suggest that fibers must be respirable (and therefore thinner than approximately 0.7
urn, Section 6.1) to have an opportunity to act, whether further diameter constraints are associated
specifically with the mechanism of cancer initiation is not known.

For pathways involving generation of ROS or RNS, some mechanisms (such as those associated with
frustrated phagocytosis) are length dependent, others are not. Although there appears to be some
dependence on fiber type for several mechanisms and a general dependence of several of these
mechanisms on surface chemistry, the relative importance of fiber type to the overall contributions from
these pathways remains to be determined.

Importantly, although crystalline silica may also act to  produce some of the same effects as asbestos
(including potentially induction or promotion of cancer), there is substantial evidence that this material
does not act through the same pathways and that the characteristics of its respective dose-response
relationships may differ.  Thus, for example,  while asbestos likely initiates cancer through a mechanism
that favors long (and potentially thin) fibers,  silica more likely acts through a mechanism that is
dependent on total surface area, with freshly  and finely ground material likely being the most potent.  In
contrast, grinding asbestos fibers tends to lesson carcinogenicity overall.  Due to differences in chemistry
and crystallinity (reinforced by studies indicating a lack of correspondence in behavior) crystalline silica
does not appear to be an appropriate analog for any of the asbestos types. Rather, for example, the
appropriate non-fibrous analog for crocidolite is riebeckite and the appropriate non-fibrous analog for
chrysotile is antigorite or lizardite.

There is also evidence that the relative importance of asbestos as a cancer initiator may differ in differing
tissues. For example, asbestos can only interfere with mitosis in cells that actively phagocytize fibers and
not all cell types actively phagocytize particles (although both mesomelial cells and pulmonary epithelial
cells appear to actively phagocytize fibers, Section 6.3.3.1). However, there is also evidence that
pulmonary epithelial cells (Type II) may undergo terminal differentiation to Type 1 cells and thus escape
potential cancer initiation by asbestos (Section 6.3.3.1). Such a pathway  is  not available to mesotheiial
cells. To the extent that pathways involving  generation of ROS or RNS contribute to cancer initiation, as
indicated throughout this section, the rates of generation and the spectrum of the species generated varies
as a function of cell type.

6.3.4   Evidence that Asbestos Acts As a Cancer Promoter

Primarily, asbestos may promote cancer by facilitating tissue proliferation.  However, additional
mechanisms associated with the observed  interaction between asbestos exposure and smoking (Section
6.3.4.6) also need to  be considered.

Substantial evidence exists indicating that asbestos induces proliferation in target tissues associated with
lung cancer and mesothelioma and this is summarized  below followed by an overview of studies that
suggest the various mechanisms by which asbestos may facilitate such proliferation. Evidence suggesting
the various mechanisms related to the interaction between smoking and asbestos exposure are also briefly
reviewed.

There are numerous mechanisms by which asbestos may facilitate proliferation including:

        •     direct cell signaling to  induce proliferation. This may occur by:
                       — direct interactions between fibers and receptors on the cell surface;
                       — interactions between phagocytized fibers and intracellular components of


                                               6.92

-------
                       signaling cascades; and                                         '
                       - or interactions between cells and intermediate species (e.g., ROS or RNS)
                       whose generation and release has been induced by asbestos; or
                                                                                     i
        •      response to induced cell death in target tissues, which then stimulates stem cells to
               proliferate to replace killed cells. Cell death may be induced either through:
                       - inducing apoptosis (programmed cell death); or
                       - direct cytotoxic effects, which leads to necrosis.                 •'
                                                                                     I-
These pathways are summarized in Table 6-5, which provides a perspective on the complexity of the
interactions between asbestos and the cells and tissues of the body.                        \

Regarding asbestos-induced cell death, asbestos-induced apoptosis (and all of the other effects described
above) typically occurs at exposure concentrations that are much lower than required to induce frank
cytotoxic effects (Sections 6.3.4.3 and 6.3.4.4). Therefore, it is primarily the former that is of potential
interest in terms of implications for asbestos-induced diseases in humans. Due to the high exposure
concentrations typically required, the importance of contributions from frank cytotoxicity to human
disease is unclear (Section 6.3.4.4).                                                     \
6.3.4.1
Asbestos-induced proliferation
Numerous in vivo studies indicate that all types of asbestos induce proliferation in target tissues relevant
to lung cancer and mesothelioma. Such proliferation is also suggested by the animal histopathological
observations previously described (Section 6.2.2). Moreover, many of these studies suggest that the
underlying mechanisms may be fiber size- and fiber-type specific. Responses may also be species-
specific.                                                                             '•:
                                                                                     ('

Importantly, there are some studies (primarily in vitro studies) that suggest asbestos acts to inhibit (rather
than induce) proliferation. Although it is likely that the contrasting observations found in such studies are
due to differences in conditions, timing, dose, or the type of asbestos employed, the underlying reason for
the contrasting observations is not always apparent.

The evidence for proliferation is summarized by tissue below.

In lung epithelium.  Brody et al. (1997) showed that rats and mice exposed for a brief (5 hour) period to
chrysotile asbestos at 1,000 fibers/cm3 (no size indicated) exhibited focal scarring at bronchio-alveolar
duct junctions that are identical  to those seen in asbestos-exposed humans. After 3-consecutive
exposures, the lesions persisted  for 6 months.  In regions where fibers are deposited, macrophages are
observed to  accumulate, epithelium is injured, and proliferation is observed to occur:  In this.study, the
authors also showed by immunohybridization staining that the four genes required to express three
peptide growth factors (TGF-cc,  TGF-P, and the A and B chains of PDGF) and the proteins themselves are
expressed in bronchio-alveolar tissue within 24 hours of exposure. PDGF is expressed almost
immediately and expression remains elevated for 2-weeks post-exposure, but only in regions where fibers
are deposited.

The authors report that PDGF is a potent growth factor for mesenchymal cells, TGF-ec is a potent mitogen
for epithelial cells, and TGF-p inhibits fibroblast proliferation, but stimulates synthesis of extra-cellular
matrix (Table 6-6).  The authors also report additional experiments with knockout mice indicate that
TNF-a is required to induce the  early stages of proliferation. The authors also indicate that Type II cells
produce TGF-pl and TGF-p2 (two of three isoforms of this protein) and that they are stimulated to do so
when co-cultured with macrophages.
                                              6.93

-------
Adamson (1997) intratracheally instilled size-separated (by sedimentation) long and short crocidolite
fibers into rats (0.1 mg in a single dose) and noted that the long fibers damaged the bronchiolar
epithelium and that fibers were incorporated into the resulting connective tissue; granulomas formed with
giant cells containing fibers).  The long fibers also appeared to escape into the interstitium.  Labeled
thymidine uptake (which indicates DNA synthesis and suggests proliferation) following long fiber
exposure was seen in lung epithelial cells, fibroblasts, and pleural mesothelial cells. Such labeling peaked
at 2% in mesothelial cells and 3% in epithelial cells within one week following exposure. Proliferation
appeared to end shortly beyond one week.  Short fibers were observed to have been efficiently
phagocytized by alveolar macrophages and a small increase in macrophage population appeared to have
been induced. Otherwise, none of the other effects attributable to long fibers (described above) were
observed with short fibers. Note that such observations are entirely consistent with those reported for the
range of studies described in Section 6.2.2.

McGavran et al. (1989). exposed both normal (C5+) mice and compliment deficient (C5-) mice to asbestos
(at 4 mg/m3) by inhalation and observed proliferation of bronchio-alveolar epithelium and interstitial cells
at alveolar duct bifurcations (based on incorporation of tritiated thymidine) between 19 and 72 hours after
a single, 5-hour exposure. Sham exposed rats showed fewer than 1% of epithelial and interstitial cells at
alveolar duct bifurcations incorporate labeled thymidine. In contrast, thymidine uptake in asbestos
exposed animals is significantly elevated for the first few days, begins to decrease at 8 days, and returns
to normal by one month following exposure. Both C5+ and C5- mice show similar increases in volume
density of epithelial and interstitial cells at 48 hours post-exposure. However, one month following
exposure C5+ mice developed fibrotic lesions while C5- mice were no different than controls.  The
authors conclude that the depressed macrophage response in C5- mice does not appear to change the early
mitogenic (and proliferative) response to asbestos, but apparently attenuates later fibrogenesis.

Chang et al. (1989) describes the morphometric changes observed in rats following 1 hour inhalation
exposure to chrysotile (at 13 mg/m3). Within 48 hours following exposure: the volume of the epithelium
increased by 78% and the interstitium by 28% at alveolar duct bifurcations relative to sham-exposed
animals. Alveolar macrophages increased 10-fold and interstital macrophages 3-fold. Numbers of Type I
and  Type II epithelial  cells increased by 82% and 29%, respectively.  At 1 month following exposure, the
numbers of Type I and Type II pneumocytes were still elevated, but not significantly. However, the
volume of the interstitium had increased by 67% accompanied by persistently high numbers of interstitial
macrophages, accumulation of myofibroblasts, smooth muscle cells, and an increased volume of
interstitial matrix.

In lung endothelium. In addition to the general evidence for proliferation of epithelial cells provided by
Adamson (1997), McGavran et al. (1989), and Chang et al. (1989), as cited above, a more detailed
description of the nature of asbestos-induced proliferation of endothelial cells is also available.

Proliferation of endothelial cells and smooth muscle cells of arterioles and venules near alveolar duct
bifurcations is induced in rats inhaling chrysotile (no size information given) for 5 hours at 4 mg/m3
(McGavran et al. 1990). This is based on the observed uptake of labeled thymidine (which indicates
DNA synthesis and suggests proliferation) that is significantly increased over controls between 19 and 72
hours following exposure. Twenty-eight percent of vessels near bifurcations exhibited labeled cells 31
hours after exposure.  One month following exposure, the thickness of the smooth muscle layers around
these blood vessels is  significantly increased (doubled). In contrast, labeling of these same endothelial
and  smooth muscle cells in sham exposed rats is zero. The authors indicate that endothelial cells and
smooth muscles associated with pulmonary blood vessels are normally quiescent with turnover rates on
the order of years.
                                               6.94

-------
In mesothelium.  In the second part of the study addressing epithelial proliferation, Adamson (1997)
reports that rats instilled with 0.5 mg of unmodified, UICC crocidolite were sacrificed at 1 week and
6 weeks following exposure and subjected to bronchiolar lavage and pleural cavity lavage. Lavaged
alveolar macrophages were observed to contain fibers, but pleural macrophages did not.  At 1 week,
collected pleural macrophages were shown to induce proliferation of fresh mesothelial cells in culture and
pleural lavage fluid showed an even greater effect. No effects were observed at 6 weeks.  Further work
with anti-bodies to various cytokines indicated that early, transient proliferation of mesothelial cells was
dependent on kertinocyte growth factor (KGF), but not on PDGF, FGF, or TNF-oc (Table 6-6). This
suggests that early, transient proliferation is induced by diffusing cytokines rather than direc^fiber
exposure. Adamson further reports that KGF is a fibroblast-derived cytokine that acts on epithelial cells
so that it's up-regulation likely results from epithelial injury with penetration of asbestos to the
interstitium (where fibroblasts are found, Section 4.4). A similar transient proliferative response in
mesothelial cells was also observed following exposure to chrysotile and in response to crystalline silica
exposure. Thus, it appears that the mesenchymal proliferative response may be mineralogy specific for
particles and is size-specific for fibers.                                                  '  •

Everitt et al. (1997) exposed rats and hamsters by inhalation to RCF-1 (45 mg/m3-650 f/cm3) for 12
weeks and then let them recover for up to an additional 12 weeks prior to sacrifice. The authors indicate
that both rats and hamsters showed qualitatively similar levels of inflammation at time examined (4 weeks
and 12 weeks). They  also indicate the mesothelial cell proliferation was observed in both animal types,
but was more pronounced in hamsters at all time points examined.  The greatest proliferation in both
species was in the parietal pleura lining the diaphragm. The authors also report that fibers (primarily
short and thin) were also observed in the pleural cavities of both species at all time points.   .

Several in vitro studies provide evidence that asbestos either induces or inhibits proliferation' in lung
tissues of interest and that at least some of the mechanisms involved are fiber size and mineralogy
dependent. As previously indicated, the specific reasons for the apparent'contrast between results
observed in vivo (where asbestos consistently promotes proliferation) and the inhibition sometimes
observed in in vitro studies is not always apparent. However, it must be due to the special conditions that
must be created to conduct in vitro studies, which may not support certain mechanisms that are important
in vivo.

Timblin et al. (1998a) completed a study in  which rat pleural mesbthelioma (RPM) cells in culture were
dosed with crocidolite or various cation-substituted erionites.  The  expression of several gene and gene
products were then tracked. Cultures in this study were exposed to 1,5, or 10 ug/cm2 of the various
fibrous materials. Analysis of the fibrous materials indicated that crocidolite contained many more fibers
per gram of material (probably because they are thinner) and that the preparation contains somewhat
longer fibers than the erionites evaluated.  In crocidolite, for example, 88% of the fibers are longer than 5
jim, 68% longer than 10 urn, and 37.5% longer than  20 um. All of the cation substituted erionites  showed
approximately the same size distribution: 50% longer than 5 um, 10-20% longer than 10 jini, and 1-5%
longer than 20 um.

Results indicate that the various cation substituted erionites behave differently and that the Na substituted
erionite shows the largest overall potency, at least for some endpoints, but not others. Fe, Na, and  Ca
substituted erionite all appear to induce c-fos expression in a dose-dependent fashion (increasing regularly
among the 1, 5, and 10 jig applications). 'K-erionite may also show the same pattern, but the'changes
were not indicated as significant over controls (apparently due to greater variability). Only Na-erionite
showed significantly increased expression of c-jun (at 1 and 5 ug applications, but not significantly at 10
jig, apparently due to greater variability. However, the mean result for 10 jig shows a consistent trend
with the lower concentration application results).
                                              6.95

-------
Na-erionite induces c-fos at the same or greater rates as crocidolite asbestos for the same mass application
(but not the same fiber number).  Crocidolite also appears to show a dose-response trend for c-jun
expression, but only the result for the highest application (10 ug) is significantly different from controls.
In contrast, Na-erionite appears to show greater induction of c-jun expression at lower dose than
crocidolite, but the increase with increasing dose is much lower for Na-erionite. Crocidolite also appears
to induce substantial apoptosis (even at the low dose of 5 ng) and that the induction is dose-dependent.  In
contrast, non-fibrous riebeckite does not appear to induce apoptosis.  Comparison between crocidolite
dosed cultures and Na-erionite dosed cultures indicate that crocidolite induces substantial apoptosis at all
time periods following application, but that Na-erionite induces little apoptosis even at higher mass dose
and longer time periods  than crocidolite. The authors also indicate that crocidolite and Na-erionite appear
to stimulate DNA-synthesis, which appears to be a compensating mechanism to fiber cell toxicity. The
authors indicate that chemistry is important in fiber toxicity as Na-erionite was a strong inducer of c-jun,
even at relatively low concentrations, but several of the other cation substituted erionites (including
Fe-erionite) were not.  They further suggest that, given the difference in the fiber lengths of the
crocidolite samples and  Na-erionite samples, that fiber length may be a less important consideration than
fiber surface chemistry.  However, despite the author's assertion, considering that non-fibrous riebeckite
does not induce any of the effects observed for crocidolite, there appears to be a clear size effect. It may
simply require that a defined, minimum length is necessary to induce the effect.

The authors also indicate that balance between proliferation and apoptosis is required to maintain
homeostasis in healthy tissue.  They further indicate that other studies suggest that c-Jun expression is
linked to proliferation and induction of cancer, while c-fos expression is linked to apoptosis.  Thus,
suppression of c-fos may be linked to carcinogenesis by allowing establishment or maintenance of a
transformed cellular phenotype.  This is in fact an early step in carcinogensis. Many environmental
agents stimulate both apoptosis and proliferation and, depending on the degree, may cause imbalances
that lead to disease. Relative stimulation of c-fos and c-jun may reflect some of these pathways.  Since
crocidolite induces both c-fos and c-jun in this study, the implication is that it mediates both apoptosis
and proliferation.

Wylie et al. (1997) dosed hamster tracheal epithelial (HTE) cells and rat pleural mesothelial (RPM) cells
with various asbestos  and talc samples  and-evaluated proliferation based on a colony-forming efficiency
(CFE) assay. The samples were NIEHS crocidolite and chrysotile and three different talcs. Samples were
characterized in the paper by mineralogical composition, surface area, and size distributions.

The authors indicate that both asbestos samples increased colony  formation of HTE cells (suggesting
induction of proliferation), but talc samples did not.  RPM cells, in contrast,  showed only dose-dependent
decreases in colony forming efficiency for all samples, which the authors indicate is a sign of
cytotoxicity.  The authors report that all samples show corresponding effects when concentrations are
expressed as fibers longer than 5 um or by total surface area. They also suggest that the "unique"
proliferative response by HTE cells could not be explained by either fiber dimension or surface area and
suggests a mineralogical effect.

Barchowsky et al. (1997) dosed cultured (low passage) endothelial cells to NIEHS chrysotile, crocidolite,
or RCF-1. After 1 to 3 hours exposure to 5 ug/cm2 (non-lethal concentrations), asbestos (but not RCF-1)
causes changes in cell morphology (cells elongate), increases in cell  motility, and increases in gene
expression.  Further work by the authors indicate that these effects are mediated by interaction between
asbestos and the receptor for urokinase-type plasminogen activator (uPAR).  The authors also suggest that
attachment of asbestos to cell membranes, internalization of asbestos fibers by the cells, and the
morphological changes  induced by asbestos are each mediated by different proteins.

Examples of in vitro studies that indicate asbestos (and other fibrous materials) may inhibit proliferation
in culture include the following.

                                               6.96

-------
In contrast to the above, Levresse et ai. (1997) found that chrysotile and crocidolite act to inhibit
proliferation in cultured rat pleural mesothelioma (RPM) cells. In this study, RPM cells (diplpid, no more
than 25 passages) were dosed with either UICC crocidolite or NIEHS Zimbabwean Chrysotile at
concentrations varying between 0.5 and 20 ug/cm2. The authors also note that the chrysotile sample
contains approximately 4 times the number of fibers as the crocidolite  samples. Cells were then examined
at 4,24, and 48 hours following treatment. In untreated cultures, the number of cells in replicative phase
decrease with time, which indicates that such cells are headed for confluence (completion of a monolayer
on the culture medium). At 48 hours, for example, 10% of untreated control cells were observed to be in
replicative stage. Chrysotile (but not crocidolite) decreased the fraction of cells in replicative phase in a
time- and  dose-dependent manner.  At 48 hours, for example, cells treated with 10 jig/cm2 chrysotile
showed only 1.5% in replicative phase. Further tests confirmed that this was due to blockage of cells at
the Gl/S boundary of the cell cycle. Both chrysotile and crocidolite appears to induce a timerdependent
increase in the number of cells at G2/M in the cell cycle, although this effect was not observed to be dose-
dependent for chrysotile. Even on a fiber-number basis, chrysotile appears to be elicit a greater response
(arrest a greater percentage of cells) than crocidolite.

The authors also indicate that chrysotile caused nuclear-localized, time-dependent increases in p53
concentrations. Crocidolite produced much lower levels that were not detectable in the nucleus.
Chrysotile was  also observed to produce blockage at the GO/G1 transition of the cell cycle, but crocidolite
did not. They also note that p53  is known to mediate arrest at this stage in the cycle so observing that
chrysotile induces arrest at this transition in the cell cycle may be consistent with the observed increased
expression of p53. The authors also note that chrysotile triggers apoptosis in this study and that
crocidolite shows a smaller, but detectable effect.  Spontaneous apoptosis in untreated cultures ran
between 0.5 and 1%  at 24-48 hours whereas chrysotile induced 4% apoptosis, peaking at 72 hours
following exposure to 10 ug/cm2. The authors indicate that the lower level of effects observed with
crocidolite could be due to the substantially smaller number of long fibers in UICC crocidolite compared
to NIEHS chrysotile.                                                                 '..

The previously reviewed (Section 6.3.3.1) study by Hart et al. (1994) also suggests that long,|medium,
short, and UICC crocidolite and chrysotile along with a range of MMVF's show a dose-dependent
inhibition on proliferation of cultured CHO cells and that potency toward the effect is a direct function of
fiber length.                                                                          r

Several of the above-described studies, in addition to providing evidence that asbestos induces
proliferation in  various lung tissues, also suggests certain mechanisms. Asbestos may induce
proliferation, for example, by inducing production of specific cytokine growth factors (Adanison 1997;
Brody et al. 1997), or by inducing certain signaling cascades (Barchowsky et al. 1997, Timblin et al.
1998b). It is also possible that the two effects may be related (i.e., that stimulation of a particular
signaling cascade may result in production of certain growth-stimulating cytokines). Other mechanisms
may also be important (Table 6-5).
6.3.4.2
Asbestos induced cell signaling
Asbestos has been shown to induce a variety of cell signaling cascades in a variety of target cell and
tissue types.  Such signaling may then trigger effects in the stimulated cells that may include:'

        •      proliferation;
        •      morphological changes;
        •      generation and release of various cytokines, enzymes, or extracellular matrix; or
        •      programmed cell death.
                                              6.97

-------
Note, due to the large number of chemical species that need to be considered in this discussion, Table 6-6
provides a summary of the sources of such species (Table 6-6A) and the effects attributable to such
species (Table 6-6B).

In specific cases, asbestos may initiate cell signaling by interacting directly with receptors on the cell
surface, by causing generation and release of intermediate species (e.g., ROS or RNS) that trigger cell
signaling, or (for phagocytized fibers) by interacting with ihtracellular components of a particular
signaling cascade. The specific responses to cell signaling induced by asbestos are frequently cell- or
tissue-type specific. Moreover, depending on the specific mechanism, cell signaling by asbestos may be
dependent on fiber size and/or type.

Barchowsky et al. (1998) showed in a set of studies that long chrysotile and long crocidolite, but not
RCF-1 fibers (at concentrations between 1 and 10 ug/cm2), which is reportedly below levels that typically
induce cytotoxic effects) induced up-regulation of urokinase-type plasminogen activator (uPA) and its
receptor (uPAR) in both lung endothelial cells (vascular cells) and lung epithelial cells. They also
showed that the increased pericellular protolytic activity (requiring cleavage of plasminogen to plasmin)
that is induced by asbestos in these cells is mediated by uPA.

In prior studies, chrysotile  has been shown to cause endothelial cells to elongate and increase expression
of adhesion molecules for phagocytes.  They also show enhanced proteolytic activity and matrix
interactions.  Chrysotile has also been shown to stimulate fibrinolytic activity in lung epithelial cells and
extravasating macrophages.  All such stimulation appears to result from up-regulation of uPA. Thus, this
mechanism may explain the observed asbestos-induced changes in lung endothelial and epithelial cells
including vascular remodeling, development of vascularized granular tissue, increased matrix turnover,
and leukocyte extravasation (which in turn may be caused by cell activation and elaboration of proteases
and adhesion molecules). The authors suggest that asbestos-induced up-regulation of uPA and uPAR
expression may represent a global mechanism for pulmonary toxicity and fibrosis induced by crystalline
fibers. Importantly, chrysotile was shown to induce uPA and uPAR expression in the absence of serum,
so the effect is apparently due to direct binding  of fibers to cell-surface receptors.

Mossman et al. (1997) observed that concentrations of 1.25-5 ug/cm2 of crodidolite (sample from  TIMA)
caused expression of c-jun and AP-1 in both cultured hamster tracheal epithelial (THE) cells and cultured
rat pleural mesothelial (RPM) cells.  Crocidolite was also observed to trigger the EGFR-regulated kinase
(ERK) and mitogen activated protein kinase (MAPK) pathways in RPM cells. The authors indicate that
these pathways are also stimulated by hydrogen peroxide and that NF-K0 induction stimulated by
crocidolite is also stimulated by crystalline silica (although silica stimulation of the MAPK pathway was
not investigated). They also note that the non-fibrous analog to crocidolite, riebeckite does not elicit
these activities.

The authors indicate that induction of the NF-KP cascade was inhibited by excess glutathione, which is
stimulated  by N-acetylcysteine (NAC),  suggesting that this pathway is induced by asbestos-caused
oxidative stress (perhaps through ROS or RNS intermediates).  Application of NAC also diminished
crocidolite induced c-fos and c-jun RNA levels and inhibited activation of the ERK-MAPK cascade.
Further work suggests that asbestos triggers the MAPK pathway by interaction with the Epithelial Growth
Factor Receptor (EGFR), either directly or by phosphorylation of this receptor by ROS. At the
concentrations examined, crocidolite induces substantial apoptosis (apparently through activation of the
ERK-MAPK cascade). In contrast, the  authors note that TNF-a induces the JNK arm of the ERK-MAPK
cascade, which leads to proliferation. Asbestos does not elevate JNK over the time-period of the study.
The authors note that it has been shown in some studies that inhibition of ERK in some cells, also  inhibits
asbestos-induced apoptosis.  Importantly, given that these processes are induced  by crocidolite, but not by
its non-fibrous analog, riebeckite, induction of the ERK-MAPK cascade appears to be a fiber-size
dependent process.

                                              6.98

-------
In a related study to that conducted by Mossman et al. (1997), Zanella et al. (1999) report that the (TIMA)
crocidolite (at concentrations of 2.5-10 ug/cm2, but not its non-fibrous analog, riebeckite, eliminated
binding of EOF to its receptor EGFR. Because EOF does not bind to crocidolite in the absence of
membrane, this is not simply a case of crocidolite tying up ligand. Crocidoiite also induces a greater than
2-fold increase in steady-state message and protein levels of EGFR.                        ;

The authors also note that the tyrphostin, AG-1478 (which specifically inhibits the tyrosine kinase
activity of EGFR), significantly mitigated asbestos-induced increases in mRNA levels of c-fos, but not
c-jun, and that the asbestos action was not blocked by a non-specific tyrphostin, AG-10. Moreover,
pretreatment of RPM cells with AG1478 significantly reduced asbestos-induced apoptosis.  Therefore, the
authors concluded that asbestos-induced binding to EGFR initiates signaling pathways responsible for
increased expression of the protooncogene c-fos and the development of apoptosis.  This apparently
occurs through the EGFR-extracellular signaling regulated kinase (ERK). It is hypothesizedjthat asbestos
may induce dimerization and activation (phosphorylation) of EGFR, which also prevents binding of EGF.
Asbestos apparently serves the same role as EGF in that it promotes aggregation of EGFR, which in turn
promotes binding to the extracellular domain of tyrosine  kinase receptors and the activation of their
intracellular kinases. The authors also indicate that other work suggests that crocidolite fiber exposure
leads to aggregation and accumulation of EGFR at sites of fiber contact and that asbestos also stimulates
biosynthesis of the EGFR and activates ERK in an EGFR-dependent manner.

The authors speculate that asbestos binding may not be ligand-site specific, but may be charge related or
may induce EGFR phosphorylation by local production of ROS, which has previously been demonstrated
to cause EGFR activation. As previously indicated, that this effect is driven by crocidolite exposure, but
not by exposure to riebeckite indicates that this mechanism is fiber-size specific.           ;'

Johnson and Jaramillo (1997) showed that UICC crocidolite, but not JM-100 glass, applied to a culture of
immortalized human Type II epithelial (A549) cells at non-cytotoxic concentrations (for 20 hours) results
in increased expression of p53, Cipl, and GADD153 in a dose- and time-dependent fashion.! Expression
was observed to be maximum at 18 hours. The crocidolite treatment was also  shown to cause an increase
in the number of cells arrested in Stage G2 of the cell cycle (with a persistent decrease in the'number of
cells in Gl). This was considered surprising because both p53 and Cipl  are known to mediate arrest in
Stage Gl. The authors suggest that these findings indicate a strong dependence on both fiber type and
fiber size (JM-100 glass contains substantially  more long fibers than crocidolite).  However, jit is not
possible to separate  the effects of fiber type versus fiber size in this study.

Luster and Simeonova (1998) indicate that at high concentrations, ROS may induce frank cytotoxicity.
At low or moderate  levels, ROS are more likely to induce cell signaling cascades that may, in turn,
contribute to asbestos-related disease. The authors dosed cultures of immortalized human Type II
epithelial (A549) cells, originally derived from a lung carcinoma, and normal human bronchjoepithelial
(NHBE) cells with long (Certain-Teed supplied) crocidolite (reported mean length: 19 um) at
concentrations ranging between 0 and 24 ng/ml. Results indicate that secretion of both Interluken-8
(IL-8) and IL-6 was stimulated by crocidolite exposure in a dose-dependent manner. In contrast,
increases in LDH levels (which indicate cell damage) was only detected at the highest exposure
concentrations tested. Further work indicates that stimulation of IL-6 and IL-8 secretion occurs through
ROS that are generated in an iron-dependent process (that may also include NF-Kp induction).  Note that
the trend of cell signal induction at low and moderate levels of asbestos exposure with evidence for
cytotoxicity observed only at the highest exposure concentrations is common to many of these kinds of
studies.                                                                            '
                                              6.99

-------
Choe et al. (1999) conducted a combined in vivo/in vitro study of the effects of low level exposure to
chrysotile or crocidolite at inducing leukocyte attachment to rat pleural mesothelial cells.  The authors
note that similar populations of rat pleural leukocytes (74% macrophages, 2% neutrophils, 10% mast
cells, and 10% eosinophils) were observed in both asbestos-exposed and unexposed rats.

In the second part of the study, cultured RPM cells were exposed to either crocidolite or chrysotile (both
NIEHS samples) at concentrations ranging between 1.25 and 10 ug/cm2, which was noted to be below
levels at which substantial cytotoxicity is observed.  Attachment of rat pleural leukocytes to RPM cells
was then observed to increase with increasing dose of asbestos to the RPM cells. In contrast, carbonyl
iron (a non-fibrous particle) also induced enhanced attachment, but at much lower levels and the effect
was not dose-dependent.  Further analysis indicated that asbestos-induced adhesion is mediated by up-
regulation of IL-lp (but not dependent on TNF-a or nitric oxide production, although it is noted that
TNF-a independently increases attachment). Asbestos also induces increased expression of vascular cell
adhesion molecule (VCAM-1).  The authors also note that rat pleural leukocytes harvested from asbestos-
exposed rats also showed increased adhesion to in vitro RPM cells over leukocytes harvested from sham
exposed rats. Thus, asbestos appears to trigger alterations in these cells as well.

In a recent paper, Driscoll et al. (1997) reviewed the role of tumor necrosis factor alpha (TNF-a) in
mediating the inflammatory response to lung insult by paniculate matter. The authors indicate that
quartz, coal dust, crocidolite, and chrysotile are all potent inducers of TNF-a production.  Titanium
dioxide (Ti02), corundum (aluminum oxide: A12O3), and latex beads are not. Pulmonary macrophages
have been shown to secrete TNF-a in vivo in response to exposure to some of the dusts listed above  .
(including asbestos). This is also observed among macrophages from asbestosis patients.

The authors indicate that rats immunized against TNF-a show reduced recruitment of neutrophils, which
demonstrates that TNF-a is involved in the recruitment of inflammatory cells.  TNF-a stimulates
macrophages, epithelial cells, endothelial cells,  and  fibroblasts to release chemokines that include
adhesion molecules (Eselectin, ICAM, VCAM). Inflammatory cells then interact with such molecules
and migrate along gradients from vascular structures in the lung to the lung interstitium and even the lung
air spaces. It has also been shown that release of TNF-a by macrophages is apparently dependent on
oxygen stress (i.e., exposure to ROS/RNS) that is induced in pathways that require iron. Production of
TNF-a and other compounds that mediate the inflammatory response are regulated by the oxidant-
sensitive transcription factor NF-Kp. This factor exists as a heterodimer in the cytoplasm in an inactive
form because it is bound to the inhibitory 1-kappaBalpha (1-KBa), which masks a nuclear translocation
signal. Appropriate stimulation of a cell induces phosphorylation of 1-KBa,  which then marks it for
proteolytic degradation. Then, NF-Kp translocates to the nucleus and induces transcription. The process
is stimulated by oxidants and inhibited by antioxidants.

In another review, Finkelstein et al. (1997) indicated that Type II epithelial cells and Clara cells (non-
ciliated bronchiolar epithelial cells) respond to and produce specific cytokines during the inflammatory
process. Early  responses to particle challenge include increases in mRNA and protein for IL-1 p, IL-6,
and TNF-a. These are also accompanied by changes in specific epithelial genes including those for
surfactant protein C and Clara cell secretory protein. The authors further indicate that these responses are
due to direct interaction with particles rather than a result of macrophage-derived mediators and they
suggest a more significant role for epithelial secretions in the overall pulmonary response than previously
suspected. Results also suggest that Type II pneumocyte-derived growth factors may play a significant
role in the pathogenesis of pulmonary fibrosis.

Also in this paper, Finkelstein et al. (1997) report that intratracheal instillation of lipopolysaccharide (a
potent inflammatory agent) caused increases in both lavage fluid and plasma levels of TNF-a and IL-6.
Intrapleural injection induced primarily increases in plasma levels.  The authors indicate that this suggests
that the observed cytokines are produced primarily at the site of injury. The authors further indicate that

                                              6.100

-------
 IL-6 is elevated in lavage fluids following exposure'to Ni2S3, a suspected human carcinogen, but not
 following exposure to Ti02 or NiO. In vitro studies indicate that release of IL-1 p and TNF-« (by Type II
 cells occurred only following exposure to crocidolite or ultrafine Ti02, but not pigment gradeTiO2. The
 authors also indicate that protein C and Clara cell secretory protein were both expressed in their
 respective source cells following exposure only to the fibrogenic of the above particles. They also report
 that crystalline silica has been shown to promote cytokine release and hypertrophy in Type II'cells.
                                                                                      i
 Jagirdar et al. (1997) used immunohistochemistry in a study to show that all three  isoforms of TGF-p
 (1,2, and 3) are expressed in the fibrotic lesions of asbestosis and pleural  fibrosis patients from the
 Quebec mines, primarily by Type II pneumocytes.  The cases examined averaged 38 years of exposure to
 the Quebec  chrysotile. The authors also indicate that the hyperplastic epithelium of silicosis patients also
 show elevated expression of all three isoforms. They further indicate  from previous studies that
 mesothelioma tumor cells frequently express TGF-P2 while the cells in the stroma of such tumors
 frequently express TGF-p 1.  It is also noted in this study that jun and fos  are both transcription factors
 that activate the TGF-pl  promoter.                                                      ;

 Zhang et al. (1993) indicate that macrophages obtained in BAL fluid from idiopathic  pulmonary fibrosis
 (IFF) patients and asbestosis patients show significantly increased secretion of TNF-a and asbestosis
 patients also showed significantly increased secretion of IL-lp. Macrophages and monocytes'obtained
 from both kinds of patients also show elevated expression of mRNA for these cytokines. In an in vitro
 part of this study, Zhang and coworkers, showed that chrysotile, crocidolite, amosite, and crystalline silica
 all stimulated IL-lp and TNF-a release and up-regulated their respective  mRNA in both macrophages and
 monocytes.  The authors also report that these two cytokines have been shown to up-regulate collagen
 Types I and III and fibronectin gene expression in human diploid lung fibroblasts after short-term, serum
 free exposure in vitro.
                                                                                      i
 Holian et al. (1997) exposed cultures of normal human alveolar macrophages (AM) (obtained by lavage)
 to varying concentrations (up to 25 |ig/ml) of short chrysotile, UICC crocidolite, ground silica,
 wollastonite, and titanium dioxide to determine whether these materials cause a phenotypic shift in
 macrophage populations by inducing selective  apoptosis.  The authors indicate that normal lungs contain:
 40-50% RFD1+7+ suppressor AM, and 5-10% RFD1+ immune activator. In this study, the pbrogenic
 subset of the particles tested (not wollastonite or titanium dioxide)  increased the ratio of
 activator/suppressor AM by a factor of 4 within a few hours and the effect was seen to increase with time.
 The authors also note that fibrogenic particles decrease the abundance of RFD7+ AM (phagocytic), but
 the consequences of this phenotypic shift are unclear.   '                                  !•

 The authors indicate that AM taken from fibrotic patients release a variety of proinflammatory mediators
 capable of stimulating fibroblast proliferation and collagen synthesis.  Even in the absence of evidence of
 fibrosis, workers who have been heavily exposed to asbestos yield similarly activated AM.  In contrast,
 they also note that, in vitro studies in which AM are stimulated with fibrogenic particles, while such AM
 are activated to release inflammatory cytokines, such releases are orders of magnitude less than that seen
 from AM derived from fibrotic patients. The authors indicate that the apoptosis-driven phenotypic shift
 in AM that is indicated by this study may explain the apparent discrepancy.

 In a previously described study, Timblin et al. (1998a), see Section 6.3.4.1, showed that crocidolite
 asbestos and several cation substituted erionites all stimulate c-jun and c-fos in rat pleural mesbthelial
 cells, but to  varying degrees  depending on  fiber chemistry. The effect also appears to be size dependent
 as crocidolite, but not its non-fibrous analog riebeckite induces the effect.                   "

 In general, the kinds of signaling cascades that  are potentially stimulated by exposure to asbestos are
 important due to their potential to contribute to the promotion of cancer.  Such pathways, for example,
 may mediate proliferation or may suppress apoptosis. Alternately, they may mediate an inflammatory

                                              6.101
       •      by causing sufficient genetic damage to trigger apoptosis; or    •           !
       •      by triggering a signal cascade that leads to apoptosis.

As previously indicated, asbestos may trigger signaling cascades by interacting directly with receptors on
the cell surface, or by inducing production of intermediate species (such as ROS or RNS) thatjmay in turn
induce cell signaling.

Some of the mechanisms by which asbestos may act to induce apoptosis may be fiber size- or,type-
dependent. Also, responses may vary in different target tissues.

Fibers. As indicated in Section 6.4.3.2, crocidolite (but not the non-fibrous analog riebeckite) induces
apoptosis in hamster tracheal epithelial cells and rat pleural mesothelial cells when applied at non-
cvtotoxic concentrations (Mossman  et al. 1997). Results from the study also indicate that apoptosis is

-------
response that in turn may lead to proliferation or to production and release of other, mutagenic agents
(e.g., ROS or RNS). Pathways that facilitate development of fibrosis may also contribute to cancer
promotion, given the apparent link between fibrosis and the development of lung cancer, which may
relate (among other possibilities) to inhibition of fiber clearance (Section 6.3.4.5).

The ERK-MAPK signaling pathway evaluated in multiple studies by Mossman and coworkers (Mossman
et al. 1997; Timblin et al. 1998a,b; Zanella et al. 1999) in rat pleural mesothelial (RPM) cells is a case in
point (see above). These studies suggest that crocidolite stimulates the ERK-MAPK cascade through
interaction with the EOF receptor. This ultimately leads to transcription of mRNA for c-fos. Crocidolite
has also been shown to induce c-jun (apparently through a separate mechanism) and the balance between
c-jun and c-fos has been implicated in guiding a cell toward either proliferation or apoptosis. Although
the direct connection between c-jun/c-fos and apoptosis has not been established, it is observed that
crocidolite induces substantial apoptosis in RPM cells at the same concentrations at which it induces
substantial expression  of c-fos and c-jun. The  link is also implied because inhibition of the ERK pathway
has been shown in some studies to inhibit asbestos-induced apoptosis. Na-erionite has also been shown to
induce c-fos at levels comparable or higher than crocidolite for comparable exposures (at least on a mass
basis) and induce c-jun at higher levels. However, it is not known whether Na-erionite and crocidolite act
via the same pathways.  Potentially due to the increased, relative expression of c-jun induced by
Na-erioinite, increased apoptosis is not observed in association with exposure to Na-erionite. However,
the link between increased c-jun expression and inhibition of apoptosis has not been demonstrated
explicitly.

Both crocidolite and Na-erionite were also shown by Mossman and coworkers to induce uptake of
bromodeoxyuridine (BrdU) by RPM cells in these same experiments. Uptake of BrdU is an indicator of
DNA synthesis.  Since it has also been shown that crocidolite is capable of damaging DNA via ROS and
other pathways (Sections 6.3.3) and both crocidolite and erionite are known to induce mesotheliomas in
any case, the balance between proliferation and apoptosis in this cell population that is struck by exposure
to these toxins may very well determine whether development of cancers are promoted or prevented.  Of
course, both proliferation and apoptosis may also  be mediated by other pathways independent of the ones
described here.

The problem is that the_range of responses that are induced by asbestos in the lung are varied and
complex (see Table 6-5) so that it has not yet been possible to definitively identify the biochemical
triggers that lead to lung cancer or mesothelioma.  It is even likely, for example, that different
mechanisms (or combinations of mechanisms) predominate under different exposure conditions or in
association with differing fiber types or particle sizes. Still, examination of the dependence of candidate
mechanisms on fiber type and particle size can be instructive, especially to the degree that such
indications are consistent with observations in whole animal studies (see, for example, Section 6.2.2). For
the signaling cascade described above by Mossman and coworkers, for example, the effects attributable to
crocidolite are clearly dependent on fiber size  because the non-fibrous analog to crocidolite, riebeckite
does not induce any of the effects.  It also appears that the chemistry of the fibers is important, given the
observed differences in responses among the various, substituted erionites.

6.3,4.3        Asbestos-induced apoptosis

Apoptosis (programmed cell death) is generally triggered when a cell accumulates certain types of genetic
damage, when cell signaling cascades are triggered by external stimuli that may occur, for example, as
part of the need to maintain tissue homeostasis or to cause a phenotypic shift in response to toxic
challenge (see, for example, Holian et al.  1997, Section 6.3.4.2), or when a cell has completed a pre-
programmed number of divisions.  Asbestos can induce apoptosis in a variety  of cells by several
 mechanisms including primarily:.


                                               6.102
effect). Although non-fibrous particles (such as crystalline silica) may also induce apoptosis, as
previously suggested, this may be through separate mechanisms from those responsible for asbestos-
related effects, even if the same endpoint results.

6.3.4.4        Asbestos-induced cytotoxicity

While there is ample evidence from various in vitro studies that asbestos is cytotoxic, such effects are
observed almost exclusively at the highest concentrations evaluated in an experiment (for example Luster
and Simeonova [1998], Section 6.3.4.2 and Choe et al. [1999], Section 6.3.4.2). Many experiments are
conducted at concentrations below those for which cytotoxicity is important because the other toxic
effects attributable to asbestos occur at substantially lower exposure levels and researchers prefer to study
such effects in the absence of potentially confounding cytotoxicity.  For in vitro studies, for example,
non-cyotoxic effects are typically studied at concentrations less than approximately 10 ue/cm2 (or

-------
The long and short samples were reportedly prepared from the UICC sample by repeated centrifugation.
Goodglick and Kane (1990) report that all three types of crocidolite stimulated release of ROS from
macrophages. At sufficiently high concentrations, all three also caused substantial cytotoxicity, although
apparently due to the longer time required to settle in culture, the foil effects from short fibers take longer
to develop. They suggest that, on a total fiber number or surface area basis, long and short crocidolite
appear to exhibit approximately equal potency toward the cytotoxicity of macrophages. Further work
with various inhibitors indicates that cytotoxicity is mediated by production of ROS and that'ROS are
produced via an iron-dependent pathway. They also indicate that, among the effects of crocidolite
exposure is that macrophage mitochondria! membranes are depolarized.
                                                                                                 %

Goodglick and Kane (1990) also evaluated the effects of long and short crocidolite in vivo. This was
done by evaluating the effects of intraperitoneal injection of the various samples (long, shortj-or mixed
crocidolite or titanium dioxide particles) in C57B1/6 mice.  Results indicate that a single injection of long
crocidolite (480  ug) induced an intense inflammatory response,  leakage of albumin, and fibers observed
scattered across the diaphragm. In contrast, a  single injection of short crocidolite  (600 jig) induced only a
relatively mild inflammatory response and only limited clusters  of fibers observed on the surface of the
diaphragm.

To test whether short fibers would show a greater response, if they were not cleared more readily than
long fibers, Goodglick and Kane (1990)  also subjected mice to 5 consecutive, daily injections of 120 ug
of short crocidolite and noted more substantial aggregations of fiber clusters along the diaphragm as well
as a more pronounced inflammatory response.  Cell injury was also assessed by Trypan blue'staining
(which indicates cell death). All mice singly or multiply injected with mixed or long crocidolite showed
marked Trypan blue  staining. Single injections of short structures showed only limited Trypan blue
staining. However, following 5 daily injections of short fibers, multiple Trypan blue stained; cells were
observed on the diaphragm in the vicinity of the locations were clusters of fibers were also observed. The
authors also indicate  (in contrast) that neither single injections of 160 or 800 ug nor 5 consecutive (160
jig) injections of titanium dioxide produced any Trypan  blue staining.
                                                                                     i
The authors conclude from this study that both short and long crocidolite fibers appear to be cytotoxic to
macrophages while titanium dioxide particles are not (suggesting that not only fiber length, tiut fiber type
is important to cytotoxicity).  They further suggest that, while short fibers tend to be cleared rapidly in
vivo, when such clearance mechanisms are overwhelmed (such as by repeated insult through repeated,
daily injections in this study), then the toxic effects of short structures becomes apparent.  As indicated  in
other studies, however, there may be multiple  mechanisms working to produce similar responses, that
such mechanisms may exhibit varying dose-response characteristics, and that cytotoxicity may not
generally be directly  related to mechanisms that contribute to carcinogenesis.  Moreover, there almost
certainly are at least some longer fibers in the  short fiber preparation and extended analysis to determine
their relative concentration with adequate precision would be helpful to see  if the  relative magnitude of
the observed effects correlate.                                                          :

Palekar et al. (1979) studied the ability of four different samples of commingtonite-grunerite, each also
subjected to varying degrees of grinding, to induce hemolysis of mammalian erythrocytes arid
cytotoxicity to Chinese hamster ovary (CHO)  cells. The samples studied include: UICC amosite
(4.13 m2/g surface area/mass), which is denoted as "asbestiform grunerite; "semi-asbestiform"
commingtonite (3.88 m2/g); acicular commingtonite (ground to three particle sizes: 3.76,2.45, and 0.82
mz/g), and acicular grunerite (2.82 m2/g).

Results from this study indicate that amosite induced the greatest hemolysis of erythrocytes by far
(approximately 50%) while acicular, unground grunerite caused no hemolysis. However,'grinding the
acicular grunerite to increasingly smaller particle  sizes and greater surface area ultimately results in some
hemolysis. Both semi-asbestiform and acicular, unground commingtonite show hemolytic activity

                                              6.105

-------
between amosite and unground,- acicular grunerite and grinding acicular commingtonite also increased its
hemolytic activity.

Similar results were also observed for cytotoxicity.  Amosite was by far the most cytotoxic and the effect
was dose-dependent. A dose of 0.05 mg/ml caused approximately 75% cell death for CHO cells. At 0.2
mg/ml, only 1% of cells survived. Acicular grunerite was nontoxic even at 0.5 mg/ml.  With grinding,
acicular grunerite cytotoxicity increased, albeit only slowly. The most heavily ground sample killed
fewer than 25% of cells at 0.2 mg/ml and killed only 65% at 0.5 mg/ml. The cytotoxicity of semi-
asbestiform commingtonite was substantially less than amosite, but greater even than ground, acicular
grunerite. For this material, 0.2 mg/ml killed approximately 65% of cells and 0.5 mg/ml killed
approximately 90%.  Interestingly, approximately the same dose-response curve for cytotoxicity was
observed for the 3.88 and 1.61 specific surface area samples of this material. The 1.21 samples was
somewhat less cytotoxic.  Acicular commingtonite was somewhat less cytotoxic (for corresponding
doses) than semiasbestiform commingtonite at the highest specific surface area (3.76) and its toxicity
decreased with decreasing surface area.

The authors also indicate that neither surface charges on crystal particles nor Magnesium ion content
appear to correlate with biological activity.  The authors conclude that the degree of "asbestiform"
character of a mineral has a dominant effect on biological activity.  Moreover, although non-fibrous
particles  may also be biologically active (and their activity increases with increasing specific surface
area), the effects of particles and fibers lie along entirely separate dose-response curves. The biological
activity of fibrous materials does not appear to depend directly on specific surface area.

Importantly, the results of the Palekar et al. (1979) study are also consistent with the possibility that
fibrous structures within a specific range of sizes and shapes contribute strongly to biological activity
while largely non-fibrous particles act through a separate mechanism that depends primarily on total
surface area, but that particle-for-particle elicits a substantially lower overall response than the
mechanism by which fibers act.  Such a scenario is supported by several studies.  Jaurand (1997), for
example, indicate that ROS are implicated in the cytotoxicity of long, but not short fibers on tracheal
epithelial cells. Although the evidence for distinct mechanisms for fibers and particles discussed here is
specific to cytotoxic and hemolytic effects, evidence in other studies suggest similar scenarios for other
toxic endpoints (potentially including endpoints that contribute to carcinogenicity).

Comparisons of the rate and extent of effects observed in epidemiology studies, whole animal  dose-
response studies, and in vitro studies suggests that cytotoxicity may not be important to human exposures.
Unfortunately, however, there is currently insufficient information to compare doses and exposures across
these studies in a more quantitative fashion. Therefore, the importance of cytotoxicity to human asbestos
exposure cannot be definitively determined at this time.

6.3.4.5        Association between fibrosis and carcinogenicity

The hypothesis that lung tumor induction is associated with the fibrosis has been examined by several
authors.  There appears to be a debate as to whether fibrosis is a necessary precursor for development of
lung cancer (associated with exposure to fibers), whether the presence of fibrosis is an additional factor
contributing to increased risk for lung cancer, or whether the two diseases are largely unrelated.  This is
an important consideration because the characteristics of the exposure-response relationship between
asbestos  and lung cancer or asbestosis (fibrosis) apparently differ (Sections 6.3.6 and 6.4).

Based on animal studies, Davis and Cowie (1990) found that rats that developed  pulmonary tumors
during inhalation experiments exhibited a significantly greater clinical degree of fibrosis than rats that did
not develop tumors.  Furthermore, Davis and Cowie (1990) reported suggestive evidence that  the


                                              6.106

-------
pulmonary tumors that did develop in the dosed rats tended to develop within portions of the rat's lungs
that were already scarred by fibrosis.
                                                                                     I
As part of a review, Mossman and Churg (1998) indicate that fibrosis of any cause (including diffuse
idiopathic fibrosis) appears to be associated with an increased risk of lung cancer and that this1 is observed
in both human and animal studies.  They also indicate that only those strains of mice and hamsters that
develop fibrotic lesions following exposure to crystalline silica show an increased risk of developing
cancer. They also report that, in parallel to what is reported for asbestos by Davis and Cowie (1990), lung
tumors that develop following exposure to crystalline silica tend to occur primarily (if not exclusively) in
those portions of the lung where fibrotic lesions predominate.
                                                                                     i
In contrast, Case and Dufresne (1997) from their study of lung burdens among Quebec miners and millers
indicate that there is high overlap in the range of concentrations that lead to both lung cancer and
asbestosis and those that lead to lung cancer alone, which the authors suggest show a lack of relationship
between the two diseases (one is not predictive of the other). The authors indicate that, based on
regression of the 111 cases they examined, the only indicator that reasonably tracks lung cancer is
severity of smoking and they indicate that this is true despite the level of fiber content in the lung.
                                                                                     i'
                                                                                     I
Although a definitive determination concerning the relationship between fibrosis (asbestosis) and lung
cancer cannot be developed at this time, it does appear that there is some association between;the two
diseases. Most likely, fibrosis is an additional risk factor for lung cancer and thus represents an additional
set of mechanisms that may contribute to the overall risk of developing lung cancer in association with
exposure to asbestos. However, based on the evidence as a whole, including the evidence that the
character of the dose-response relationship for lung cancer and asbestosis differ, it is not clear that
development of fibrosis is an absolute precursor that is required before asbestos-related lung tumors can
develop. Interestingly, based on the studies reviewed by Mossman and Churg (1998), the relationship
between fibrosis and lung cancer induced by silica may be substantially stronger than that between
fibrosis and lung cancer induced by asbestos.
6.3.4.6
Interaction between asbestos and smoking
Numerous studies have indicated a synergistic relationship between smoking and asbestos exposure
toward the induction of lung cancer (see, for example, Hammond et al. 1979; Kamp et al. 1992,1998;
Mossman et al. 1996).  Smoking is also suspected to facilitate development of asbestos-induced fibrosis
(Kamp et al. 1992). However, a more recent study (Liddell and Armstrong 2002) suggests a more
complicated relationship that is closer to additive than multiplicative (for a more detailed discussion of
this study, see Section 7.2.3).  Therefore, discussion of a more complex interaction (which may not be
specifically synergistic) is addressed below.

Putative mechanisms that may contribute to an interaction between asbestos exposure and smoking
include:

       •      facilitated transport of carcinogenic components of smoke that may be adsorbed on the
               surface of asbestos fibers, which may then serve as vehicles to transport these materials
 ,              through cell membranes to cell interiors and even to locations adjacent to or within the
               nucleus (see, for example, Fubini 1997; Mossman et al. 1996);             !

       •      asbestos-catalyzed production of more highly mutagenic metabolites of the various
               components of smoke, including benzo(a)pyrene (see, for example, Mossman et al.
               1996);
                                             6.107

-------
       •      smoke-product induced inhibition of clearance of asbestos fibers and/or asbestos induced          ^Br
               inhibition of clearance of smoke products (see, for example, Mossman et al. 1996); and

       •      smoke-product induced facilitation of uptake of asbestos by lung epithelium (see, for
               example, Mossman et al. 1996).

Although much progress has been made at elucidating the nature of these mechanisms and other candidate
mechanisms, at this point in time it is possible to indicate definitively neither what mechanisms are
important to any observed interaction between smoking and asbestos exposure nor to indicate the relative
magnitude of the contributions from such mechanisms.  Moreover, a detailed review of such mechanisms
is beyond the scope of this document.

6.3.4.7        Conclusions concerning asbestos as a promoter

There is strong evidence that asbestos acts as a promoter for cancer. While this may primarily involve
mechanisms the contribute to the induction of proliferation, mechanisms associated with the an
interaction between smoking and exposure to asbestos (for the induction of lung cancer; smoking does not
appear to affect mesothelioma) are also important.  There also appears to be an association between the
development of fibrosis and an increased risk for lung cancer.

Evidence indicates that multiple mechanisms may be involved with asbestos-induced cancer promotion
and that such mechanisms may be complex and interacting. The different mechanisms also appear to
exhibit dose-response relationships with differing characteristics. While there are indications that the
most important among these mechanisms may be strong functions of fiber size (with long fibers
contributing most to the induction of disease), mechanisms that depend primarily on surface area or total
fiber (particle) number (for any size range) may also contribute to overall cancer promotion. Importantly,
these latter mechanisms also appear to be strongly associated with the composition of fibers (particles)
and may therefore contribute more substantially to the disease induction of agents that have been shown
to be particularly toxic (such as crystalline silica), as opposed to particles, fibers, or asbestos in general.

At this point in time, the available data may not be sufficient to distinguish among the relative
contributions from the various mechanisms to the overall promotion of cancer, at least in terms of the
mechanistic data itself.  Importantly, however, the mechanistic data should not be considered to be
inconsistent with the results from whole animal studies, where there are clearer indications that fiber size
plays a major role in carcinogenicity and fiber (particle) type  is also important (see Sections 6.2 and 6.4).
Such studies indicate, for asbestos (and other biodurable fibers) that:

        •      short fibers (less than somewhere between 5 and 10 jim) do not appear to contribute to
               disease;

        •      potency likely increases regularly for fibers between 10 urn and a minimum of 20 ^im
               (and, perhaps, continues  to increase up  to lengths of at least 40 p,m); and

        •      fiber type may be important primarily in determining biodurability.

They further indicate that particularly (or uniquely) toxic particles (such as crystalline silica) may act
through a different set of mechanisms that are not dependent on fiber length, but that induce toxic
endpoints paralleling those observed for asbestos.

Importantly, the mechanisms by which asbestos may act as a promoter appear to occur in cell lines that
may contribute both  to the induction of lung cancer and mesothelioma.


                                              6.108

-------
6.3.5   Evidence that Asbestos Induces an Inflammatory Response

There is ample evidence that asbestos induces an inflammatory response in pulmonary tissues and the
pleura (see, for example, Sections 6.2.2 and 6.3). Moreover, there appears to be multiple biochemical
triggers that mediate this response and various mechanisms may be fiber size- and/or fiber type-specific
(Table 6-5). Because the role that inflammation plays in the induction of cancer has been addressed
elsewhere (Sections 6.3.3 and 6.3.4), it is beyond the scope of this document to provide a detailed review
of the mechanisms that lead specifically to inflammation.
                                                                                   !

6.3.6   Evidence that Asbestos Induces Fibrosis

There is ample evidence that asbestos induces fibrosis in pulmonary tissues.(see, for example, Sections
6.2.2 and 6.3). Moreover, there appears to be multiple biochemical triggers that mediate this.response
and various mechanisms may be fiber size- and/or fiber type-specific (Table 6-5). Because the role that
fibrosis plays in the induction of cancer has been addressed elsewhere (Sections 6.3.3 and 6.3.4), it is
beyond the scope of this document to provide a detailed review of the mechanisms that lead specifically
to fibrosis. Such mechanisms have also been the subject of recent reviews (see, for example, Mossman
and Churg 1998; Robledo and Mossman 1999).                                         :;
                                                                                   '!
6.3.7   Evidence that Asbestos Mediates Changes in Epithelial Permeability

As previously indicated (Section 4.4), maintaining the overall integrity of the epithelial surface of the
lung is among the various functions of Type II epithelial cells (Leikauf and Driscoll 1993).  It has been
shown that asbestos induces changes in the morphology of Type II epithelial cells (see, for example,
Ilgren and Chatfield 1998), which has the effect (among others) of increasing the overall permeability of
lung epithelial tissue to various macromolecules and, potentially to asbestos fibers themselves. The
former plays a role in asbestos-induced fibrosis (by allowing cytokines that stimulate fibroblast
proliferation or stimulate fibroblasts to generate extracellular matrix to pass through the epithelium and
reach the underlying fibroblasts, Section 6.3.6).  The latter may be important to facilitating transport of
asbestos from the alveolar lumen  to the interstitium (see, for example, Lippmann 1994).

Changes in epithelial permeability may be triggered by cytokines released from other cells or by the
action of asbestos fibers on epithelial cells directly. Moreover, some of the mechanisms that mediate this
response may be sensitive to fiber size and/or fiber type.  For example, Gross et al. (1994) showed that
monolayers of human bronchial epithelial cells cultured over a porous medium and exposed to
cryogenically ground chrysotile (average length: 1 urn, average aspect ratio:i4 at 15 ug/culture plate)
became permeable to fibrin breakdown products (FBP's).  The cultures were grown over human serum
with labeled fibrinogen. This was based on observed increased concentrations of FBP's (double in 24
hours) in the ablumenal chambers of exposed cells compared to cells in control cultures. Because the
epithelium showed greater permeability to all concentrations, the increased concentrations were not due to
increased breakdown.  The observed FDP flux was not vectoral, not sarurable, and required neither
proteolytic processing  nor active transport. Thus, asbestos increases the paracellular flux of intact FDP
across airway epithelium.

6.3.8   Conclusions Regarding the Biochemical Mechanisms of Asbestos-Related,Diseases

That the specific biochemical triggers for asbestos-related diseases (particularly, the asbestos-related
cancers) have not been definitively delineated as of yet is not surprising.  The detailed interactions
between fibers (and particles) and the cells and tissues of the lung are complex and there are complex,
multiple, interacting mechanisms  by which such interactions may contribute to disease. Despite  great
progress in elucidating candidate mechanisms, the number of candidate mechanisms is large and

                                             6.109

-------
distinguishing among their relative contributions has been difficult. This is because, among other things,
the ability to compare results across studies of different mechanisms is currently limited due to the
inability to reconcile the quantitative effects of dose and response across dissimilar studies.

Nevertheless, a number of important implications can be gleaned from the available literature. First, it
appears that asbestos can function both as a cancer initiator and a promoter. It also appears that both the
initiation and promotion of cancer may occur through more than one mechanism.

Regarding cancer initiation, asbestos likely acts primarily through a mechanism involving interference
with mitosis. By this mechanism, asbestos fibers are phagocytized by target cells, migrate to perinuclear
locations, and interact with the spindle apparatus and other cell assemblages required to complete mitosis.
This tends to result in aneuploidy and may cause various clastogenic effects. This mechanism is driven
by long fibers; short fibers do not appear to contribute to the effect. It also appears that all asbestos fiber
types (and potentially other durable fibers with sufficient dimensions) cause genetic damage via this
mechanisms. If there are effects due to fiber type, they appear only to play a secondary role.

Although there is also" evidence that asbestos may induce production of DNA adducts and DNA strand
breaks (through ROS and RNS mediated pathways), whether such adducts or breaks ultimately lead to
permanent, heritable changes to DNA remain to be  demonstrated. The relative importance of ROS/RNS
mediated pathways for  initiating cancer, compared to the pathway involving interference with mitosis,
also remains to be determined.

There is also some evidence that the relative  importance of asbestos as a cancer initiator may differ in
different tissues. Lung epithelial cells, for example, appear to be relatively resistant to the mechanisms by
which asbestos may initiate cancer. Mesothelial cells are not.  Among several possibilities, this may be
due to the ability of proliferation-competent lung epithelial cells (Type II cells) to undergo terminal
differentiation when challenged with certain  toxins  and this is a pathway not available to mesothelial
cells.

The mechanisms by which asbestos may promote cancer primarily involve mechanisms that contribute to
the induction of proliferation, although mechanisms associated with an interaction between smoking and
exposure to asbestos to induce lung cancer are also  important.  There also appears to be an association
between the development of fibrosis (including asbestosis) and an increased risk of lung cancer.

Evidence indicates that multiple mechanisms may be involved with asbestos-induced cancer promotion
and that such mechanisms may be complex and interacting.  The different mechanisms also appear to
exhibit dose-response relationships with differing characteristics.  While there are indications that the
most important of these may be strong functions of fiber size (with long fibers contributing the most to
carcinogenicity), mechanisms that  depend primarily on surface area or total fiber number (for any size
range) may also contribute to overall cancer promotion. These latter mechanisms also appear to be
strongly associated with the composition of fibers and may therefore contribute more substantially to the
disease induction of agents that have been shown to be particularly toxic, as opposed to particles, fibers,
or asbestos in general.

Although crystalline silica may act to produce some of the same effects as asbestos (including
carcinogenicity), there  is substantial evidence that this family of materials do not act through the same
pathways and that the characteristics of their respective dose-response relationships may differ.  Thus, for
example, while asbestos likely induces cancer through mechanisms that favor long (and potentially thin)
fibers, silica more likely acts through a mechanism that is  dependent on total surface area, with freshly
and finely ground material likely being the most potent. In contrast, grinding asbestos fibers tends to
lesson its carcinogenicity overall.  Due to differences in chemistry and crystallinity (reinforced by studies
indicating a lack of correspondence in behavior), crystalline silica does not appear to be an appropriate

                                              6.110

-------
analog for any of the asbestos fiber types. Rather, for example, the appropriate non-fibrous analog for
crocidolite is riebeckite and the appropriate non-fibrous analog for chrysotile is antigorite or Ijzardite,

6.4     ANIMAL DOSE RESPONSE STUDIES
                                                                                     *,.

Ideally, human epidemiology studies (reviewed in Chapter 7) provide the best data from which to judge
the effects of asbestos in humans and from which to derive exposure-response factors for humans.
However, animal dose-response studies have proven useful for elucidating certain features of the
relationship between asbestos dose and response that cannot be adequately explored in the human studies,
primarily due to limitations in the manner that exposures were characterized in the human studies (see
Chapter 5).

Unlike human epidemiology studies, exposures in animal studies are controlled and better quantified.
Frequently, the characteristics (in terms of fiber size, shape, and type) of such exposures have also been
better quantified and this has allowed exploration of the effects that such characteristics (fibersize, shape,
type) have on disease response. Accordingly, an overview of animal dose-response studies is'provided in
this section. Both injection-implantation studies and inhalation studies are reviewed. Particular attention
is also focused on a "supplemental" animal inhalation study that we conducted with the specific aim of
identifying the characteristics of asbestos that best relate to risk. The strengths and limitations of these
kinds of studies are described in Chapter 5.                                              •'

6.4.1   Injection-Implantation Studies                                             :

Because the fibrous materials in injection and implantation studies are placed immediately against the
target tissue, the effects of processes associated with inhalation, retention, and translocation are avoided.
The only active mechanisms that need  to be considered in these studies are those that occur directly in the
target tissue (including degradation, clearance, and biological responses of the types described in the
previous sections of this chapter). Fibrous materials placed against the tissue surface are subject to
dissolution, phagocytosis by macrophages, and phagocytosis by the cells of the target tissue.  These
mechanisms are described in greater detail in Section 6.2. A  range of biologic responses have also been
observed (described in Section 6.3).                                                    ,

Numerous researchers have performed these types of studies.

The Work of Stan ton and Coworkers. In a series of studies, Stanton and coworkers (1972,' 1977,1981)
implanted fibrous materials and induced mesotheliomas in rats.  In the studies, a pledgette composed of
coarse glass is loaded with hardened gelatin containing sample material and is surgically implanted
immediately against the left pleura of the rats.  Control  studies demonstrate that the coarse glass of the
pledgette does not induce significant tumors in the absence of other tumorigenic agents in the gelatin.

Although the mass dose of material implanted was the same for all experiments (40 mg), the observed
incidence of mesothelioma varied among samples. By characterizing the dimensions of fibrous structures
in the samples using a microscope, the  researchers were able to explore the relationship between fiber size
and the  incidence of mesothelioma.  By studying a wide range of fibrous materials, Stanton and his
coworkers concluded that the induction of mesothelioma is determined primarily by the physical
dimensions of fibers and that mineral composition is secondary. Further, potency appears to  increase
with the length and decrease with the diameter of fibrous structures.  The researchers also concluded that
the incidence of malignant tumors correlates with the degree of fibrosis induced by the presence of the
fibrous materials. This does not necessarily imply, however, that fibrosis is a necessary step in the
induction of asbestos-induced tumors (see Section 6.3.4.5).
                                             6.111

-------
Conclusions from the Stanton et al. (1972,1977,1981) studies indicating that mineralogy is not a factor
in biological response conflicts with evidence provided in Chapter 7 and implications gleaned from
mechanism studies presented in Section 6.3.  However, the studies by Stanton and coworkers have been
shown to suffer from certain methodological limitations (Herman et al. 1995) so that results from these
studies should be considered more qualitative than quantitative.

Due to limitations in the ability to produce samples composed of uniform fibers, quantitative relationships
between size and potency were explored by Stanton and coworkers using a regression analysis.
Structures longer than 8 um with diameters less than 0.25 jim or longer structures with diameters less than
1.5 jim were found to represent the range of sizes that best correlate with carcinogenicity. It was further
stated that such correlations did not eliminate the possibility that other size ranges also contribute to
potency, only that the two size ranges identified appear to correlate best. Samples that varied
significantly from the reported correlations were attributed to errors in the characterization of structure
size distributions in those samples. However, other methodological limitations might also have
contributed to the observed deviations or such "outliers" may also suggest evidence for a mineralogical
effect that is similar to what is reported in other studies (see Section 6.3 and Herman et al. 1995).

The precision of estimates for the ranges of sizes that contribute to biological activity that are derived
from the Stanton and coworkers (1972, 1977, 1981) studies is limited so that such estimates should also
be considered qualitative.  Size distributions were determined by characterizing 200 to 1,000 structures
using TEM and there is no indication that statistically balanced counting rules were employed (Section
4.3).  Under such conditions, counts of structures longer than 8 \im are likely small and subject to large
uncertainties for most of the samples characterized. Confidence intervals are not provided for any of the
exposure values presented in these studies.

Potentially larger errors in the studies by  Stanton and coworkers could have been introduced by the
method employed to relate fiber counts to sample mass.  As indicated in Chapter 5, estimating
contributions to mass by sizing total particles and assuming that this is proportional to total sample mass
is subject to error from the limit to the precision of characterizing structure dimensions (particularly
diameter) and by not accounting for nonasbestos (and possibly nonfibrous) material in the samples.  Thus,
for example, there is no discussion of the precision with which the cut point of 8 um was determined in
these studies.

Re-analysis and Extension of the Stanton Studies. Several researchers have re-evaluated data from the
implantation studies to test additional hypotheses. Using the Stanton and coworkers (1972, 1977, 1981)
data, Bertrand and Pezerat (1980) examined the relationship between mesothelioma incidence and several
characteristics not evaluated by Stanton and coworkers including: average fiber length, average fiber
diameter, average fiber aspect ratio, total fiber surface  area and total fiber volume.  Results from the
regression analysis  indicate that potency varies directly with average length and inversely with average
diameter, but that neither parameter is a good indicator alone. Combining the effects of length and
diameter, average aspect ratio is highly correlated with potency. Biological activity does not correlate
highly with structure count, surface area or volume except when fiber sizes are restricted to the long, thin
structures that Stanton and coworkers defined. Results of this study are not inconsistent with those
originally presented by Stanton and coworkers except that they emphasize a set of characteristics that
relate parametrically to biological activity rather than expressing exposure as a single restricted size range
of structures.

Bertrand and Pezerat (1980) were able to find good correlations between response and specific "average"
characteristics of the samples that are not proportional  to the quantity of the material present in the sample
("intensive" characteristics). Such intensive characteristics as average aspect ratio, average length, or
average diameter are properties that are independent of the mass of material in a sample. Since response


                                               6.112

-------
must be a function of the quantity of sample present, intensive characteristics should have to>'be multiplied
by characteristics that are proportional to the mass of a sample (e.g., fiber number, sample mass, or
sample volume) in order to relate them to response. Properties that vary with the mass of a sample are
termed "extensive" properties.                                                         '

The correlations between intensive properties and response reported by Bertrand and Pezeraf (1980)
likely succeed within the Stanton and coworkers (1972, 1977, 1981) database because a constant sample
mass (40 mg) was employed for all of the implantation experiments. However, to apply dose-response
relationships that are dependent only on intensive characteristics beyond the data presented by Stanton
and coworkers (where mass dose will not be constant), it is necessary to pair intensive characteristics with
extensive characteristics (such as mass or number of fibers per sample). Therefore, it is unclear how the
conclusions from this paper may be generalized to other data sets.
                                                                                     ;»
In a similar study, Bonneau et al. (1986) also examined parametric relationships between  structure
characteristics and mesothelioma induction. The paper examined specifically correlations between
carcinogenicity and dose in terms of two specific relationships: dose expressed as fibers longer than 8 um
that are thinner than 0.25 f±m ("Stanton" fibers) and dose expressed as mean aspect ratio.  The researchers
conclude that mean aspect ratio provides an excellent indication of carcinogenicity for individual fiber
types, but that each fiber type must be treated separately.  Poorer correlations are found for the
relationship between the concentration of "Stanton" fibers and mesothelioma, even when  fiber types are
considered independently.  Although these results appear to be consistent with findings reported from
mechanistic studies (Sections 6.2 and 6.3) in that they posit a role for fiber mineralogy, the relationships
evaluated by Bonneau et al. (1986) also suffer from the limitation of expressing dose only in terms of
intensive quantities,  as discussed above. Direct comparison with other studies is therefore difficult.

Following up on the reported problems of the studies by Stanton and coworkers in characterizing
crocidolite, Wylie et al. (1987) reanalyzed seven crocidolite samples originally studied by Stanton et al.
She and coworkers then used the new  size distributions to reevaluate the "Stanton hypothesis" (that the
concentration of "Stanton" fibers in a sample correlates with carcinogenicity). Wylie and coworkers note
that substantial deviations from the Stanton hypothesis occur for specific samples. They conclude that a
specific structure size range alone is not sufficient to characterize biological activity and that a parametric
relationship with other structure characteristics (potentially including mineral type) may be necessary to
sufficiently describe biological activity.                                                  i

Conclusions from the Wylie et al. (1987) paper must be interpreted carefully because the  researchers
evaluated only the relationship between carcinogenicity and the single specific size range indicated
("Stanton" fibers). Thus, the possibility that improved correlations exist between biological activity and
different size ranges or a combination  of size ranges cannot be ruled out.  Qualitatively, conclusions
presented in this paper  are not inconsistent with the conclusions reported by Stanton and coworkers
regarding the general relationship between response and fiber dimensions.
                                                                                      !
The  Wylie et al. (1987) study appears to suffer from several methodological problems.  These1relate to the
manner in which the sample reanalysis was performed. 'The drop method for preparing electron
microscopy grids (used in this study) is not satisfactory  for preparing grids. In fact, as reported in the
study itself, grids prepared as duplicates by this method were'shown to be non-uniform at the 95%
confidence interval using a chi-square test. In addition, only 100 to 300 fibers were counted for each
sample. Since there is no indication that statistically balanced counting was performed, the uncertainty
associated with counts of "Stanton" fibers may be substantial. Such errors would be further multiplied by
uncertainty introduced during the sizing of total particles to determine the number of fibers per unit mass.

In a later study, Wylie et al. (1993) examined the effect of width on fiber potency. In this latter study,
results from animal injection and implantation studies were pooled and subjected to regression analyses to

                                              6.113

-------
identify correlations between exposure and tumor incidence. The animal studies selected for inclusion in
this analysis were performed on a variety of tremolite samples exhibiting a range of morphological and
dimensional.characteristics.

In their regression analyses, Wylie et al. (1993) evaluated a range of exposure indices that emphasize
different morphological or size characteristics to help elucidate the characteristics of asbestos that induce
a biological response. Because all of the animal studies included in their analyses involved tremolite,
mineralogy was not an issue. Results from this study suggest that fibers longer than 5 |itn and thinner
than 1 um best correlate with tumor incidence among the animal injection and implantation studies
examined.  Further, they suggest that a width limit, rather than a limit on aspect ratio, better reflects the
bounds of the asbestos characteristics that determine biological activity. They also suggest that complex
structures (bundles and clusters) need to be evaluated as part of the determination of exposure because
such structures can breakdown and contribute to the population of thinner fibers.

Although the results of the Wylie et al. study tend to support the general conclusions in this document
related to width, if not to length (see Section 6.5 and Appendix A), as the authors themselves indicate,
such results should be considered qualitative due to the limitations imposed on their study by the
methodology employed. Their study was conducted by:

        •      combining results from multiple studies without careful consideration of variation
               introduced by methodological differences across the studies;

        •      employing asbestos concentrations determined by SEM and without careful consideration
               of differences in the counting methodologies employed by differing'research groups
               across studies; and
       .•      considering injection and implantation studies, which do not account for mechanisms
               related to inhalation and deposition that affect the exposure-response relationship in
               humans.

The limitations imposed by the above constraints are highlighted in Chapters 4 and 5 of this report.

Other Injection Studies. A series of injection studies were conducted by several research groups. In
these studies, fibrous materials were suspended in saline and injected into rats immediately adjacent either
to the pleura or peritoneum. A large number of fibrous materials have now been studied by this process,
as reported by: Bolton et al. (1982, 1984,1986); Davis et al. (1985, 1986a,d, 1987, 1988a); Muhle et al.
(1987); Pott et al. (1974,1976,1978,1982,1987); and Wagner et al.  (1976, 1982,1985).  Newer studies
are also discussed in Section 6.2. Results confirm that it is the fibrous nature of the materials that is the
primary factor leading to the induction of tumors and that potency appears to depend directly on length
and inversely on diameter.

The authors of these studies tend to indicate that, except where fibers are not persistent in vivo, due to
solubility or other degradation processes, the mineralogy of the fibers appears to play only a secondary
role in determining disease  incidence.  Researchers conducting injection experiments also tended to report
a correlation between tumor incidence and the degree of fibrosis induced by the sample.  These
observations are consistent  with the ideas originally articulated by Stanton.

Pott developed Stanton's ideas further by suggesting that carcinogenicity is a continuous function of fiber
dimensions, which decreases rapidly for lengths less than 10 um and also decreases with  increasing
diameter. The possibility was also raised that the apparent inverse dependence on diameter may be an
artifact due to the limited number of thick fibers that can be injected in a sample of fixed mass.
                                              6.114

-------
 Although the published injection studies indicate that potency decreases with decreasing length,
 researchers in these earlier studies were reluctant to identify a length below which contributions to
 carcinogenicity can be considered inconsequential. This may be due in part to the skewed distribution of
 fiber sizes typical of asbestos dusts. Thus, for example, even if structures less than 5 urn are only 1% as
 potent as structures longer than 5 urn, they may be as much as 100 times as plentiful in some asbestos
 dusts, so that the total contribution to potency would be equal for both size fractions.

 Reasonable dose-response curves have been generated using various sample masses of a single material.in
 some of these studies.  This has been demonstrated for UICC crocidolite and UICC chrysotile "A"
 (Bolton  1984). Results indicate that the relationship between tumor incidence and the log of the dose
 may be linear and there is no effective threshold. A consistent difference between the two dusts is
 apparent; the points lie along separate curves and chrysotile appears to be more potent per unit sample
 mass,                                                                                 ;:

 In general, the analytical techniques used for quantifying size distributions in these studies are not fully
 documented.  To the extent that they are, it appears that similar approaches were adopted to those
 described for the implantation studies above.  Consequently, similar limitations apply to the interpretation
 of results.  Briefly, large uncertainties are likely associated with counts of long fibers and estimates of the
 number of fibers per unit sample mass. Counts in several of the studies also suffer from limitations in the
 ability of SEM or PCM to detect thintfibers (Section 4.3); whenever SEM or PCM was employed, the
 thinner fibers were likely under-represented in reported fiber size distributions.
                                                                                      i
 Because samples are placed against mesenchyme in the published implantation and injection studies,
 results of these studies most directly represent processes associated with the induction of mesothelioma.
 Assuming, however, that clearance and degradation processes are similar in the deep lung, once a fiber
 reaches a target tissue, results from the implantation and injection studies may also provide a model of
 biological response in lung tissue and the factors that lead to the induction of pulmonary tumors.  Such a
 model must be considered qualitative at best, however, because it has been shown that the mechanisms of
 tissue response to the presence of asbestos in lung parenchyma and in the mesenchyme differ in detail
 (Section 6.3). The time periods over which the various clearance mechanisms operate in the deep lung
 and the mesenchyme also differ (Section 6.2), although, it is apparent that the general nature of the
 clearance and degradation processes in the two tissue types are generally similar.

 6.4.2   Animal Inhalation Studies

 Animal inhalation studies measure response to exposure in controlled systems that model  most of the
 relevant variables associated with asbestos disease mechanisms in humans (including respirability,
 retention, degradation, clearance, translocation, and tissue-specific response). Thus, the available
 inhalation studies are the best database from which to evaluate the integrated effects that lead to the
 development of asbestos-related disease. Such studies can be used both to identify the characteristics of
 asbestos  that determine biological activity and to qualitatively elucidate the nature of the corresponding
 relationship between exposure via inhalation and the induction of disease.                  '

 In this section, the existing animal inhalation studies are reviewed.  In the following section, a project
 undertaken to overcome the limitations of the existing animal inhalation studies is described (the
 supplemental inhalation study). Because this latter project was specifically designed to support the risk
 protocol  presented in this document, the nature and results of this project are described in  detail.

The existing animal inhalation database consists of approximately 30 studies of which approximately 20
contain dose-response information based on lifetime monitoring of exposed animals, including the work
 by: Bellman et al. (1986, 1995); Bolton et al. (1982); Davis et al. (1978, 1980, 1985, 1986a,d, 1988a,b);
                                              6.115

-------
Goldstein et al. (1983); Le Bouffant et al, (1987); Lee et al. (1981); McConnell et al. (1982); Muhle et al.
(1987); Platek et al. (1985); Smith et al. (1987); and Wagner et al. (1974,1982, 1985, 1987).  The studies
are similar in overall design, although differences in experimental details potentially affect the
comparability of results from separate studies.

In the inhalation studies, plugs formed from bulk samples of fibrous asbestos and related materials are
placed in a dust generator and aerosolized.  The generators (Beckett 1975), usually a modified version of
the apparatus originally designed by Timbrell et al. (1968), consist of a rotating brush that sweeps over an
advancing plug of bulk material and liberates fibers that are entrained in the controlled air flow passing
through the device. The airborne dust  is then passed either into a delivery system for nose-only exposure
or into an exposure chamber where animals are kept for fixed periods of time (usually 7 hours per day) on
a weekly routine (typically 5 days per week). The exposure routine is continued for as long as 2 years in
some of the studies. In some, but not all of the studies, fiber-containing air is passed through a cyclone or
elutriator prior to the exposure chamber so that exposure consists primarily of particle sizes within the
respirable range.

Asbestos concentrations in the animal inhalation experiments are monitored by a combination of
techniques.  The concentration of total dust in the chamber is generally monitored gravimetrically.
Simultaneously, membrane filter samples are collected and fibers counted by PCM.  The quotient of these
two measurements yields the number of (PCM) fibers per unit mass of dust (Section 4.3).  The
distribution of fiber sizes within the dusts introduced into the animal exposure chambers may also be
determined in these studies by any of a variety of methods. As indicated previously (Section 4.3),
however, the utility of such measurements  depends on the precise manner in which they are derived.

To derive fiber size distributions, dust  samples from these studies have generally been collected on
polycarbonate filters for analysis by SEM.  However, such distributions suffer both from the limitations
of SEM (Section 4.3) and from the manner in which they are tied to the inhalation experiments (see
Chapter 5).

Theoretically, the dose of any fiber size fraction can be estimated in a two-step process. The procedure
incorporates consideration of a size fraction termed the PCM-equivalent fraction (PCME), which is the
fraction of structures measured by SEM (or TEM) that correspond to the size range of structures known to
be visible and therefore countable by PCM. First, the concentration of the PCM-equivalent fraction of the
fiber size distribution (measured by SEM)  is normalized by dividing its value by the PCM-measured
concentration per unit dust mass observed  in the inhalation experiment. This ratio is then  multiplied by
the fractional concentration of any specified size range of interest within the distribution (measured by
SEM) to determine the exposure level  for that size fraction.  However, because bivariate (length by
diameter) size distributions have not typically been developed in the available studies and because the
number of total fibers longer than 5 jim observed by  SEM (without adjustment for width)  does not
correspond to the number of total fibers longer than 5 um observed by PCM, it is not possible to derive a
true PCME fraction from the SEM data. Therefore, the theoretical approach described above for
estimating exposure to specific size fractions cannot generally be applied in the existing studies.

Note that SEM analyses are typically conducted on limited dust samples only to provide information on
size distributions.  SEM is not used routinely to monitor daily asbestos concentrations in these
experiments. Therefore a procedure like that described above is required to link the absolute
concentrations to which rats are exposed to the measured and relative size distributions that are
determined by SEM.

As indicated above, the data within the published animal inhalation studies are further constrained by the
limitations of the analytical methods employed to generate the data  (Section 4.3). Comparison of data
between studies is also hindered by the lack of sufficient documentation to indicate the specific methods

                                               6.116

-------
                                                                                            s
and procedures employed in each study. Frequently, for example, it is unclear whether respirable dusts or
total dusts have been monitored. Also, several studies fail to report one or both of two critical pieces of
information: fiber-number-to-mass conversion factors and fiber size distributions. In addition, few
studies indicate the precise counting rules employed for generating size distributions.       j

When structure-number-to-mass conversion factors are provided, unless the conversion factor is derived
by counting fibers in a specific size range in a known mass of sample, and fiber concentrations in other
size ranges are normalized to this count, several types of error may be introduced. For example, if total
sample mass is assumed proportional to calculated mass derived from volume characterizations of the
particles counted, unless isometric particles are sized along with fibers and both asbestos and nonasbestos
particles are included in the count, a bias will be introduced in the conversion factor because total sample
mass will have been under-represented to the extent that such particles are ignored in the estimation of
fiber mass.  Even if such particles are included, significant uncertainty may result from estimating the
volumes of irregular particles and the limited precision associated with the count of the largest particles
(due to their limited number).  The uncertainty in the measurement of a fiber's diameter is squared in
contributing to the uncertainty associated with  a mass estimate.                            !

Among reported variations in study design, differences in the detailed design and operation of the
aerosolization chamber and the frequency and duration of exposure also potentially contribute to variation
in results between studies. Also, use of differing animal strains and species across the various studies
suggest the possibility that physiological differences may contribute to the observed variation in study
results. Such differences are discussed further  in Chapter 5.                               !

A small subset of the asbestos dusts evaluated in the animal inhalation studies have been analyzed by
TEM. However, even the published fiber size distributions from these TEM studies are subject to
variation from differences in procedures used for sample preparation, from differences in counting rules,
and from precision limitations due to the limited number of fibers actually characterized (Section 4.3).
This latter limitation particularly affects the precision with which longer fibers are counted. ',

Although fiber-size distributions are primarily based on SEM analyses rather than TEM analyses in the
existing animal inhalation studies, results generally echo the results of the injection and implantation
studies. Thus, longer fibrous structures are observed to contribute most to asbestos biological activity, at
least qualitatively. For example, dusts containing predominantly long amosite or long chryso'tile fibers
induce far more pulmonary tumors than samples containing predominantly short structures (Davis et a).
1986a,b). However, dusts evaluated in the existing inhalation experiments have not been characterized
sufficiently to distinguish the dependence of biological activity on fiber diameter. Neither are the existing
studies sufficient to evaluate the importance of mineralogy (or other potentially important asbestos
characteristics) in determining risk.

6.4.3   Supplemental Inhalation Study

Given the problems with the existing animal inhalation studies, a project was undertaken to overcome
some of the attendant limitations (Herman et al. 1995). To control for effects from variation in study
design and execution (including choice of animal strain, animal handling procedures, equipment design,
sample handling procedures, dosing regimen, and pathology protocols), the project focused on a set of
studies generated from  a single laboratory (i.e., the studies published by Davis and coworkers).
Ultimately, the results from six studies covering nine different asbestos samples (including fpiir types of
asbestos with samples exhibiting multiple size distributions for two asbestos  types) and a toteil of 13
separate experiments (some samples were studied at multiple exposure levels or in duplicate runs)  were
pooled for analysis. The database of experiments employed in the project is  described in Table 6-7.
                                              6.117

-------












3
HH
11
•g
e
U
•e
s
A
1
Q
>>
J3
"O
S
U
s
e
r )
.52
41
i
it
c
_g
"S
JB
s
»""l
"3
E
'£

1
S
£
i
s
5

t-^
4
H








a
B
£
>
o ? C «
1 ii '
|f ||
^ 2 * -
f-<< ^^ |M

"•« >>
e C „,
« S a a
* 'S ° B
g S S =
"5 fifi *3 £"•
z £.
u _w
lol
a a
Z <

s-s
u 1
a. w

ill?
s S S

H •—
U

8
jjjj

b
<


e
•3
B&
i_
V


U .,
II
rt
OO OO 00
r*" r*" oo
Os CT\ O\

• ^ ^
(S Q Q

— O fl




oo in o





cs oo cs






vo r~ oo




M O O


o o
m —  o ?3
Efl W) EC
> > >
Q Q Q

« o —




r*- i^1 ^5





^D OO ^D






— «•<*•




§ « ON


O O O
f*» NO f"*
« m vo
•-< rM CH
ON OV
O ON Gv







o o o
CO P P


|
1
CO
5

S8§
« P D
U U 4*
000
C/3 VI W
U U U
vg
oo
ON

13
0
m
>
5

O




oo





2






i ^ (^

•3 13 13
4) ^jj "^>
W W trt
> > >
QQQ

O m —




«N — O





O OO O






 cs" p>

0 O O







P J oo







8l?S
5^S

.« .s .-S
"S 'K *J3
<; < <
00 00
r~ p-


rt Q
"S "55
M M
> >
0 13
O O

— o





Q O Q

rs o o




00 00





VO O O






IN O O




O\ O NO
' fN C^ f*^


1


o o o







^. o u







1 II
w
I 1 1
<3 .£> 03
**S ^O OO
OO OO OO
Ch ON ON

13 73 13
4) ^[> "^
tfl OT OT
s s s
Q Q Q

O 0 O




fS fS f>»





- - -






— — —




\O VO "tj"





o o o







u u o







s g s
1 § I
u u u

AJ 
£?
5
f
u
1
u
•s
u
u
3
U
"

2
o
1
«3
g
00
.£
_o
^
^-i
•*?•
_g
**-j
•3
S
S
JS
o
u
o
1/3
a
€
*s
•o
o
y
1
Cu
Sur^
'I
_£>


S "3
>. o.
« T3
81
^•§
!S s
 s
« ?
tl
' MM
„ D,"°
v> O ^ ^
O\ -G m ^
"sli
13-3-S 5
rt 3 «) 'T-*
c§ u^^-
g §51
,'




i,


!







u '
| ^

c
.2 .

V)
^ i
~V i'
>
•a '
'5
u
| .
1 :

s» '
i '
S,
>. '
5
u '
is ,
£ '
E •-,
«s .
•s ,
.s
£ ,
o
E •
U V
S :•
s i-
E '
1'
g
i§ '
c
§
'^
^
1
BO ,
-o
Q
**-« :
"8
irt
3 '
1
1
o>

-------
In the statistical analysis performed in this study, the individual dose-response profiles from each
of the two data sets were fit to a linear dose-response model:
                                                                                 (Eq.6-7)
where:
       "Pi"    is the probability of inducing pulmonary tumors observed in the "ith" study;

       "Q0"   is a parameter that accounts for the background incidence of pulmonary tumors
              (assumed to be the same in all studies);

       "Xy"    is the concentration of the "jth" size fraction of fibers in the "ith" study;

       "aj"    is the coefficient of potency for the "jth" size fraction of fibers; and

       "b"    is a coefficient that represents the absolute potency of asbestos. In some analyses
              this coefficient is allowed to assume different values for different types of
              asbestos (e.g., chrysotile vs. amphibole), or other differences in experimental
              conditions.

The "aj"s in this analysis are constrained to be positive because it is assumed that no fiber
prevents cancer. The "aj"s are also constrained to sum to 1.0, which means that they represent
relative potency rather than absolute potency. Asbestos size fractions evaluated represent
disjoint (mutually exclusive) sets.

The model (Equation 6-7) allows separate potency coefficients to be assigned to individual size
fractions in a dose-response relationship that depends on multiple size fractions.
Simultaneously, the "b" coefficients allows separate potencies to be assigned to different fiber
types or to results from different studies performed under different experimental conditions.

Several investigators (Bertrand and Pezerat 1980; Bonneau et al.  1986; Stanton et al. 1977;
Wylie et al. 1987) have used a logit curve to investigate the dose-response relating various
measures of asbestos exposure to tumor response. The logit formula specifies that the tumor
probabilities satisfy the relation:
                                  log[P/(I-P)]=a + b-logx
(Eq. 6-8)
 where log x is some measure of asbestos exposure, such as log of concentration of fibers in some
 size range. In some instances, the logit model was expanded by replacing b-log x with a term
 representing a linear combination of exposure indices, so that multiple exposure indices could be
 explored simultaneously.  The models were fit using standard linear regression based on normal
 theory.
                                           6.120

-------
 An equivalent form for the logit model is:
                                     P=e"xb/(l+e8xb)
  (Eq.6-9)
 Written in this form, it is clear that this model does not permit a background response (i.e., P=0,
 whenever x=0).  This is not a serious limitation when there are no tumors in control iinimals,
 such as was the case in Stanton et al. (1977). However, the model will not adequately fit data in
 which tumors are found in control animals.  This was one reason for adopting the linear model
 (Equation 6-7) used in the investigation of the animal data reported in the study described here.
                                                                             i
                                                                             I
 There is no evidence from this study that the linear model is inadequate.  For cases in this study
 in which the fit between exposure and response is shown to be inadequate, the lack of fit is
 typically observed to be due to an inconsistent (non-monotonic) dose-response curves so that
 there is no indication that a non-linear model, such as the logit, would provide a better fit.

 The linear model (Equation 6-7) used in this study was fit using a maximum likelihood (Cox and
 Lindley 1974) approach that utilizes the actual underlying binomial probabilities. This is a more
 efficient estimation method than use of regression methods based on normal theory, which was
 the fitting method used in the earlier studies (described above).  In addition, the regression
 procedures applied in the earlier studies indicate only whether the exposure measures'that were
 studied are significantly correlated with tumor response. In contrast, statistical goodness of fit
 tests were applied in this  study to determine whether exposures that are described by a particular
 characteristic (or combination of characteristics) satisfactorily predict the observed turner
 incidence. To illustrate, it is apparent from Text Figure 2 of Stanton et al. (1981) that the
 exposure measure they identify as being most highly correlated with tumor incidence (fibers
 longer than 8 jim and thinner than 0.25 urn) does not provide an acceptable fit to the observed
 tumor incidence.  Similarly, although all of the univariate exposure measures listed in Table 2 of
 Berman et al. (1995) are highly correlated with tumor incidence, none of them adequately
 describe (fit) lung tumor incidence.

 To test for goodness-of-fit in this study, each relationship was subjected to a chi-square
 goodness-of-fit test in which the fit of the model was rejected if the corresponding p-value was
 less than 0.05, indicating that the true model would provide a worse fit only 5% of the time.
 Among models that were not rejected based on a goodness-of-fit test, several hypotheses
 concerning the relative merit of the various models were also examined using the method of
 maximum likelihood (Cox and Lindley 1974).

 An example of an adequate fit to the tumor response data is provided in Figure 6-5 (Figure 3 of
Berman et al. 1995).  Note that Figures 2 and 3 of the original paper were inadvertently switched
during publication; the correct Figure 3 is reproduced here. The exposure index plotted in
Figure 6-5 is the sum:
               Exposures, x^ + a2 x,2 + a3 x^O.OOHC, + 0.853C2 + 0.145C3
(Eq. 6-10)
                                          6.121

-------
s^
e?
••<•
11
ii
•*-" o
•= I
i. -S
*• a
CO (
81
S ^
**
li
 o
 g
2 8P
E^5
 . u
«? J3
^ to
2 g
01
CM
        if!   .•- '„ i   f"  8-  8

-------
where:
       "C,"   (= Xj,) is the concentration of structures between 5 and 40 \im in length that are
              thinner than 0.3 u.m;                                             :
        'C2"   (= Xj2) is the concentration of structures longer than 40 urn that are thinner than
              0.3 [im; and                                                    ;•
        C3"   (= Xj3) is the concentration of structures longer than 40 um that are thicker than 5
This index of exposure represents one of the optimum indices reported in Berman et al. 1995.

As is clear from the figure, when exposure is expressed in the manner described above, the
tumor responses observed in the 13 separate experiments that were evaluated increase
monotonically with increasing exposure. It is also apparent that the data points representing
each study fall reasonably close to the line representing the optimized model for this exposure
index.  This was verified by the fact that the corresponding goodness-of-fit test p-value was
greater than 0.05. Thus, exposure adequately predicts response.

Results obtained from completing more than 200 statistical analyses to determine whether
various measures of asbestos exposure adequately predict lung tumor response (Berman et al.
1995) indicate that:

       •      neither total dust mass nor fiber concentrations determined by PCM adequately
              predict lung tumor incidence;

       •      no univariate  measure  of exposure (i.e., exposure represented by the
              concentration of a single size category of structures as measured by TEM) was
              found to adequately predict lung tumor incidence. Of the univariate measures of
              exposure examined, the concentration of total structures longer than 20 um
              provides the best fit (although still inadequate); and

       •      lung tumor incidence can be adequately predicted with measures of exposure
              representing a weighted sum of size categories in which longer structures are
              assigned greater potency than shorter structures.

The set of analyses completed in support of this work are summarized in Appendix C.
One example of an exposure measure that adequately describes lung tumor incidence is
presented in Figure 6-5. Another exposure index shown to provide an adequate fit is: '•
                                   0.0024Ca + 0.9976Cb
(Eq.6-11)
                                          6.123

-------
where:
       "Ca"   is the concentration of structures between 5 and 40 jam in length that are thinner
              than 0.4 urn; and

       "Cb"   is the concentration of structures longer than 40 |im that are thinner than 0.4 (Am.

The fit of this index is depicted in Figure 6-6.

In addition to the above, a series of hypotheses tests were also conducted to test such questions
as: whether fiber type affects potency or whether the component fibers in complex clusters and
matrices should be counted individually. Questions concerning whether mesothelioma incidence
can be adequately described by the same measure(s) of exposure that describe lung tumor
incidence were also addressed. Taken as a whole, the results presented in Herman et al. (1995)
support the following general conclusions:

       •      structures contributing to lung tumor incidence are thin (<0.5 \im) and long (>5
                  with structures longer than 20 \im being the most potent;
              the best estimate is that short structures (<5 u.m) are non-potent.  There is no
              evidence from this study that these structures contribute anything to risk;

              among long structures, those shorter than 40 urn appear individually to contribute
              no more than a few percent of the potency of the structures longer than 40 urn;

              lung tumor incidence is best predicted by measurements in which the component
              fibers and bundles of complex structures are individually counted;

              at least for lung tumor induction in rats, the best estimate is that chrysotile and the
              amphiboles are equipotent;

              for equivalent size and shape structures, amphiboles are more potent toward the
              induction of mesothelioma than chrysotile; and

              after adjusting for the relative potencies of fiber type, the size categories that
              contribute to lung tumor incidence appear also to adequately describe
              mesothelioma incidence.
                                          6.124

-------
a S
.2 3.
«-*.
8?
« v
8.3
u »
  rf
& s
« a.
If
u  ^
in
CM
CO

-------
A number of supplemental analyses were also conducted, primarily to identify optimal
procedures for performing asbestos analysis and for estimating concentrations.  These analyses
have not yet been published, but they are included in the summaries in Appendix C. The most
important results of the supplemental analyses are that:

       •      tumor incidence can only be adequately fit by data derived from TEM analysis of
              samples prepared by a direct transfer procedure.  Measurements derived from
              indirectly prepared samples could not be fit to lung tumor incidence in any
              coherent fashion; and

       •      it was not possible to identify an exposure measure in which potency is expressed
              in terms of a single, continuous function of structure length.

Regarding the last point, although we were not able to identify a continuous function of length
that provides an adequate fit to the tumor incidence data, the general results from the above
analysis are not inconsistent with the hypothesis that potency is a continuous function of length
(i.e., the Pott hypothesis, Pott 1982). The Pott hypothesis suggests that relative potency is low
for short fibers, rises rapidly over an intermediate range of length, and approaches a constant for
the longest fibers.

6.4.4  Conclusions  Concerning Animal Dose-Response Studies

Results from our evaluation of the animal dose-response data for asbestos (including the existing
injection/implantation studies, the existing inhalation studies, and our supplemental study)
indicate that:

        •      short structures (less than somewhere between 5 and 10 urn in length) do not
              appear to contribute to cancer risk;

        •      beyond a fixed, minimum length, potency increases with increasing length, at
              least up to a length of 20 |am (and possibly up to a length of as much as 40
       •     the majority of structures that contribute to cancer risk are thin with diameters
              less than 0.5 urn and the most potent structures may be even thinner. In fact* it
              appears that the structures that are most potent are substantially thinner than the
              upper limit defined by respirability;

       •     identifiable components (fibers and bundles) of complex structures (clusters and
              matrices) that exhibit the requisite size range may contribute to overall cancer risk
              because such structures likely disaggregate in the lung. Therefore, such structures
              should be individually enumerated when analyzing to determine the concentration
              of asbestos;

       •     for asbestos analyses to adequately represent biological activity, samples need to
              be prepared by a direct-transfer procedure; and
                                           6.126

-------
       •     based on animal dose-response studies alone, fiber type (i.e., fiber mineralogy)
              appears to impart only a modest effect on cancer risk (at least among the various
              asbestos types).

Regarding the last of the above bullets, that only a modest effect of fiber mineralogy ;was
observed in the available animal dose-response studies (when large effects are observed among
human studies, Chapter 7), may be due at least in part to the limited lifetime of the rat relative to
the biodurability of the asbestos fiber types evaluated in these studies, although it is also possible
that different mechanisms drive the effects observed in the animal studies than those .that
dominate for asbestos-induced cancers in humans and that such mechanisms depend more
strongly on mineralogy. Other explanations are also possible.  Issues relating to botK fiber
mineralogy and fiber size are addressed further in Section 6.5 and Chapter 7.

6.5    CONCLUSIONS FROM AN EVALUATION OF SUPPORTING STUDIES

Although gaps in knowledge remain, a review of the literature addressing the health-i-elated
effects of asbestos (and related materials) provides a generally consistent picture of the
relationship between asbestos exposure and the induction of lung cancer and mesothelioma.
Therefore, the general characteristics of asbestos exposure that drive the induction of cancer can
be inferred from the existing studies and  can be applied to define appropriate procedures for
evaluating asbestos-related risk. Furthermore, although it would be helpful to definitively
identify the underlying biochemical triggers and associated mechanisms that drive asbestos-
induced cancer, this is not an absolute prerequisite for the development of a technically sound
protocol for assessing asbestos-related risk.

As previously indicated (Sections 6.3.8),  the biochemical mechanisms that potentially contribute
to the induction of asbestos-induced cancer are complex and varied. Moreover, different
mechanisms appear to exhibit differing dose-response characteristics (i.e., the various
mechanisms do not all show the same kind of dependence on fiber size or fiber type). Some
mechanisms, for example, suggest that fiber length is important and that only structures that are
sufficiently long induce a response.  In contrast, other mechanisms suggest that fibers (and even
non-fibrous particles) may all contribute  to response and that the magnitude of the response is a
function of the total surface area of the offending fibers (or particles). Among these ;!
mechanisms,  additionally, some suggest that fiber type (i.e., mineralogy) is not an important
determinant of potency while other mechanisms indicate that fiber type is an important
determinant of potency.                                                      '

The existing studies are not currently adequate to support definitive identification of the specific
mechanisms that drive the induction of asbestos-related cancer  (versus other mechanisms that
may contribute only modestly or not at all). However, whatever mechanisms in fact contribute
to the induction of disease, they must be consistent with the gross characteristics of exposure that
are observed to predict response in the available whole-animal dose-response studies1 and human
epidemiology studies. Therefore, the implications from these latter studies regarding  the
dependence of asbestos-induced cancers on fiber size and type are reviewed here in some detail.
Further, in Chapter 8, they are used to support development of a protocol for evaluating
asbestos-associated risks.
                                         6.127

-------
Fiber Length.  Fibers less than a minimum length between 5 and 10 \im do not appear to
contribute to risk. This is supported both by the results of our re-analysis of the animal
inhalation studies conducted by Davis and coworkers (Section 6.4.3 and Berman et al. 1995), in
which this hypothesis was tested formally, and by inferences from the broader literature. As
long as fiber size is adequately characterized, the animal inhalation studies (Section 6.4.2) and
injection/implantation (Section 6.4.1) studies consistently indicate lack of ability of short
structures to contribute to the induction of cancer. Furthermore, animal retention studies
(Section 6.2.1) and histopathology studies (Section 6.2.2) provide strong mechanistic evidence
that explains the lack of potency for short structures; they are readily cleared from the respiratory
tract. Even when sequestered in large numbers in macrpphages within the lung, there is little
indication that such structures induce the kinds of tissue damage and related mechanisms that
appear to be closely associated with the induction of cancer.

Although there are mechanism studies that may suggest a role for short fibers in the induction of
asbestos-related disease (see, for example, Goodglick and Kane 1990, Section 6.3.4.4), such
studies dp not track cancer as an endpoint.  Therefore, the relationship between the toxic
endpoint observed and the induction of cancer needs to be adequately addressed before it can be
concluded definitively that short structures can contribute to cancer.  Moreover, such a
conclusion would be surprising given the substantial evidence that exists to the contrary.

Beyond the minimum length below which structures may be non-potent, potency appears to
increase with increasing length, at least up to a length of 20 ^m and potentially up to  a length of
40 um.  The latter limit is suggested by our re-analysis of the Davis et al. studies (Section 6.4.3)
in which it was also found that structures longer than 40 p.m may be as much as 500 times as
potent as those between 5 and 40 u.m in length.  The former  limit is suggested by broader
inferences from the literature that suggest the cutoff in the length of structures that are at least
partially cleared by macrophages from the lung may lie close to 20 ^m and that the efficiency of
clearance likely decreases rapidly for structures between 10  and 20 urn in length (Section 6.2).
Such inferences are further reinforced by measurements of the overall dimensions of
macrophages in various mammals by Krombach et al. (1997), as reported in Section 4.4.

Importantly, the inferences that potency increases for structures longer than 10 urn (up to some
limiting length) from these various studies are strongly reinforcing, even though  the upper limits
to the points at which potency stops increasing do not precisely correspond. Furthermore, that
the longest structures are substantially more potent than shorter structures (and that the shortest
structures are likely non-potent) dictates that asbestos analyses performed in support  of risk
assessment need to provide adequate sensitivity and precision forcounts of the longest
structures.

Fiber Diameter. Because fibers that contribute to the induction of cancer must be respirable,
they must also be thin. The studies reviewed in Section 6.1  indicate that respirable fibers are
thinner than 1.5 \im and the vast majority of such structures are thinner than 0.7 |im.  In fact, the
results of injection, implantation,  and inhalation studies reviewed in Sections 6.4.1 and 6.4.2 and
the results of our supplemental  re-analysis of the Davis et al. studies (Section 6.4.3) indicate that
the fibers that contribute  most to the induction of asbestos-related cancers are substantially
thinner than the limit suggested by respirability alone.
                                           6.128

-------
Importantly, the results of all of the studies cited above indicate that it is a cutoff in absolute
width that defines the bounds of biological activity rather than a cutoff in aspect ratio (the ratio
of length to width) that has been used to define fibrous structures heretofore. That is why the
exposure index recommended based on our review of the studies by Davis and coworkers
incorporates a maximum width as a cutoff, rather than a minimum aspect ratio.      •'

Fiber Complexity. In our supplemental evaluation of the Davis et al. studies (Section 6.4.3), the
tumor incidence data from the animal inhalation studies were best fit (predicted) by exposure
indices in which the component fibers and bundles of complex structures (clusters and matrices)
were separately enumerated and included in the exposure index used to represent concentration
(Section 6.4.3). The appropriateness of such an approach is further supported by theiobservation
that loosely bound structures (including, for example, chrysotile bundles) readily disaggregate
in vivo (Section 6.2). Therefore, it is recommended in this report that those components of
complex structures that individually exhibit the required dimensional criteria be individually
enumerated and included as part of the count during analyses to determine the concentration of
asbestos in support of risk assessment.                                          ,'

Fiber Type (Mineralogy). The magnitude of any effect of mineralogy upon cancer risk in
rodents appears to  be modest at best. On the other hand, mineralogy appears to be an important
determinant for cancer risk in human epidemiology  studies (Chapter 7), with chrysotile
appearing less potent than amphibole for inducing mesothelioma and (with lesser certainty) lung
cancer.  This difference may be due to differences in the life spans of rats and humans compared
to the differential biodurability of the different fiber types. It must also be emphasized that, due
to confounding, the effects of fiber size and fiber mineralogy need to be addressed  '
simultaneously, if one is interested in drawing useful conclusions concerning fiber mineralogy.
                                                                             i: •
Results from some (but not all) of the animal injection, implantation, and inhalation studies
previously reviewed (Sections 6.4.1  and 6.4.2) suggest that mineralogy plays an important role
in determining biological activity. However, the nature of the effects of mineralogy are not
easily separated from size effects, due to the methodological limitations of the studies cited.
Therefore, the evidence from these studies can be considered ambiguous. Formal hypothesis
testing during the re-analysis of rat inhalation studies (Section 6.4.3) indicates that, when size
effects are addressed, chrysotile and the amphiboles exhibit comparable potency toward the
induction of lung cancer. In contrast, amphiboles were estimated to be approximately 3 times
more potent than chrysotile toward the induction of mesothelioma, once fiber size effects are
addressed.

Several of the human pathology studies cited previously (Section 6.2.3) suggest that mineralogy
is an important factor in determining cancer risk, but these studies similarly suffer from
methodological difficulties that introduce ambiguity into the inferences drawn.  However, it is
clear from the human epidemiology data (Chapter 7) that mineralogy plays a substantial role in
the determination of risk for human cancer (primarily, mesothelioma).

The underlying cause(s) for the observed difference in potency between chrysotile and the
amphiboles may relate to differences in fiber durability (Section 6.2), to size/shape related
differences in fibers that are a function of mineralogy and that cause differences in deposition,
retention, or translocation (Sections 6.1  and 6.2), and/or to the dependence on mineralogy of the

                                          6.129

-------
specific mechanisms underlying the biological responses of specific tissues (Section 6.3). The
relative magnitudes of such effects on animal and human pathology also need to be considered,
if the observed differences in potency among animal and human studies, respectively, is to be
reconciled. Such considerations are addressed further in Chapter 8.

Importantly, whether the observed differences in the role of mineralogy toward animal and
human pathology can be reconciled, the effects of mineralogy can be adequately addressed when
assessing asbestos-related cancer risk for humans by incorporating dose-response coefficients
explicitly derived from the human epidemiology data. Because this is the approach proposed in
this document, effects due to mineralogy are properly addressed.

An Appropriate Exposure Index for Risk Assessment. The optimum exposure index defined
based on the re-analysis of the animal inhalation studies conducted by Davis and coworkers
(Section 6.4.3) is a weighted sum of the concentrations of (1) structures between 5 and 40 um in
length that are thinner than 0.4 UJTI, and (2) structures longer than 40 um that are thinner than 0.4
urn (Equation 6-11).

This index was shown to adequately fit (predict) the tumor incidence data across the 13 separate
animal inhalation experiments evaluated (P=0.09).  Whether the exposure index defined in
Equation 6-11 is also optimal for capturing the relevant characteristics of fibers that contribute to
the induction of human cancer is an open question. Because it captures the major characteristics
(concerning length and diameter) identified above that are indicated to be important for human
exposures, it represents a promising candidate.  Unfortunately, however, the data required to
match this index to a set of human-derived exposure-response coefficients does not currently
exist (Section 7.4). Therefore, compromises are required to apply the general conclusions of this
Chapter to the human data. These are addressed further in Section 7.4.1.
                                          6.130

-------
                   7.0.  EPIDEMIOLOGY STUDIES

The existing epidemiology studies provide the most appropriate data from which to determine
the relationship between asbestos exposure and response in humans. As previously indicated,
however, due to a variety of methodological limitations (Section 5.1), the ability to compare and
contrast results across studies needs to be evaluated to determine the confidence with' which risk
may be predicted by extrapolating from the "reference" epidemiology studies to new
environments where risk needs to be assessed. Reliable extrapolation requires both that the
uncertainties contributed by such methodological limitations and that several ancillary issues
(identified in Chapter 2) be adequately addressed.

A detailed discussion of the methodological limitations inherent to the available epidemiology
studies was provided in the Health Effects Assessment Update (U.S. EPA 1986) and additional
perspectives are provided in this document (Section 5.1). The manner in which the uncertainties
associated with these limitations are addressed in this document are described in Appendix A.
As previously indicated (Chapter 2), the ancillary issues that need to be addressed include:

       (1)     whether the models currently employed to assess asbestos-related risk' adequately
              predict the time and exposure dependence of disease;
                                                                            i
       (2)     whether different mineral types exhibit differential  potency (and whether any
              differences in potency relate to the relative in vivo durability of different asbestos
              mineral types);

       (3)     whether the set of minerals included in the current definition for asbestos
              adequately covers the range of minerals that potentially contribute to asbestos-
              related diseases; and

       (4)     whether the analytical techniques and methods used to characterize exposures in
              the available epidemiology studies adequately capture the characteristics of
              exposure that affect biological  activity.

All but the third of the above issues are addressed in this chapter.  Currently, the third issue can
best be addressed by evaluating inferences from the broader literature (see Chapter 6). The
remaining issues are addressed separately for  lung cancer and mesothelioma following a brief
overview of the approach adopted for evaluating the epidemiology literature.

7.1    APPROACH FOR EVALUATING THE EPIDEMIOLOGY LITERATURE

To develop exposure-response relationships (and corresponding exposure-response coefficients)
for use in risk assessment from epidemiological data, two basic types of information are
necessary: information on the disease mortality experienced by each member of the study
population (cohort) and information on the asbestos exposure experienced by each member of
the cohort.  So that disease mortality attributable to  asbestos can be distinguished from other
(background) causes of death, it is also necessary to have knowledge of the rates of mortality
                                          7.1

-------
that would be expected in the study population, absent exposure. Normally, such information
must be determined based on a "reference" or "control" population.

Ideally, one would like to have complete knowledge of exposure at any period of time for each
individual in the cohort and complete access to the data to fit different types of
exposure/response models to the data so that the approach for evaluating the relationship
between exposure and response can be optimized. In most instances, unfortunately, the data
suffer from multiple limitations (see Section 5.1 and Appendix A) and the analysis is further
constrained by less than complete access to the data.

Briefly, the major kinds of limitations that potentially contribute to uncertainty in the available
epidemiology studies (and the effect such limitations likely produce in estimates of exposure-
response coefficients) include:

       •      limitations in the manner that exposure concentrations were estimated
              (contributing to variation across studies);

       •      limitations in the manner that the character of exposure (i.e., the mineralogical
              types of fibers and the range and distribution of fiber dimensions) was delineated
              (contributing to systematic variation between industry types and, potentially,
              between fiber types);

       •      limitations in the accuracy of mortality determinations or incompleteness in the
              extent of tracing of cohort members (contributing to variation across studies);

       •      limitations in the adequacy of the match  between cohort subjects and the selected
              control population (contributing to variation across studies and may have a
              substantial effect on particular studies); and

       •      inadequate characterization of confounding factors, such as smoking histories for
              individual workers (contributing to variation across studies and may have a
              substantial effect on particular studies).

More detailed discussion of the above limitations is provided in Section 5.1. The manner in
which these limitations are being addressed in this evaluation are described briefly below and in
more detail in Appendix A.

The existing asbestos epidemiology database consists of approximately 150 studies of which
approximately 35 contain exposure data sufficient to derive quantitative exposure/response
relationships. A detailed evaluation of 20 of the most recent of these studies, which includes the
most recent follow-up for all of the cohorts evaluated in the 35 studies, based on the
considerations presented in this overview, is provided in Appendix A.

This new analysis of the epidemiology database differs  from the evaluation conducted in the
1986 Health Effects Assessment Update (U.S. EPA 1986) in several ways.  It incorporates new
studies not available in the  1986 update that contain information on exposure settings not
previously evaluated as well as more recently available follow-up for exposure settings

                                            7.2

-------
previously evaluated. It also incorporates new features in the manner in which the analysis was
conducted. These new features include:

       •     estimation of "uncertainty" bounds for the exposure-response coefficients
              (potency factors) derived from each study; and                     ;

       •     for the lung cancer model,  introducing a parameter, a, which accounts for the
              possibility that the background lung cancer mortality rate in the asbestos-exposed
              cohort differs systematically from the rate in the control population.  :

The uncertainty bounds were developed to account for uncertainty contributed by the manner
that exposure was estimated, by the manner that work histories were assigned, by limitations
imposed by the manner in which results were reported in published papers, and by limitations in
the accuracy of follow-up, in addition to accounting for the statistical uncertainty associated with
the observed incidence of disease mortality.  A detailed description of how the uncertainty
bounds were constructed is provided in Appendix A.                             i

Exposure-response coefficients were estimated for each cohort both by requiring that a=l (the
approach followed in the 1986 Health Effects Assessment Update, U.S. EPA 1986) and by
allowing a to vary while fitting the lung cancer model to data. Allowing a to vary addresses
potential problems due to differences in background lung cancer rates between the cohort and the
control population due, for example, to differences in smoking habits in the two populations. As
indicated in Appendix A, this adjustment has a substantial effect on the fit of the U.S. EPA
model to the data for several specific cohorts and a corresponding effect on the estimates of the
lung cancer exposure-response coefficients for those cohorts.                     j

We were also able to obtain the original, raw data for selected cohorts from a limited number of
the more important of the published epidemiology studies.  This allowed us to more formally
evaluate the appropriateness of the existing U.S. EPA models for lung cancer and mesothelioma
(Sections 7.2 and 7.3).

Exposure-response coefficients, and corresponding risk estimates derived therefrom,'must be
based upon an "exposure index" that expresses the relative potency of asbestos fibers of different
dimensions.  For example, the exposure index utilized in the 1986 Health Effects Assessment
Update (U.S. EPA 1986) assigns equal potency to all fibers longer than 5 u.m that exhibit an
aspect ratio >3 and a thickness >0.25 ^m,  regardless of type of asbestos, and assigns zero
potency to shorter, squatter, or thinner  fibers. In this update, we evaluated a range of such
exposure indices, both with respect to agreement with evidence from the literature on the relative
potency of asbestos structures of differing types and dimensions (Section 7.4) and with respect to
overall agreement across the exposure-response coefficients derived from the available
epidemiology studies and adjusted for exposure index. This analysis led to proposed new
exposure indices that better reflect the evidence from the literature on the relative potency of
different structures and provide improved  agreement among exposure-response coefficients
estimated from different environments. Such improvement in agreement across studies
correspondingly increases the confidence with which the exposure-response factors derived from
the existing studies can be applied to new  environments.
                                           7.3

-------
7.2    LUNG CANCER

The 1986 U.S. EPA lung cancer model (U.S. EPA 1986) assumes that the relative risk, (RR), of
mortality from lung cancer at any given age is a linear function of cumulative asbestos exposure
in units of fiber-years/ml (f-y/mf) as measured by PCM, disregarding any exposure in the most
recent ten years. This exposure variable is denoted by CE10, and its use embodies the assumption
that asbestos exposures during the most recent 10 years do not affect current lung cancer
mortality risk. The mathematical expression for this model is:

                                  RR=l+KL*CEi0                          (Eq.7-1)

where the linear slope, KL, is termed the "lung cancer exposure-response coefficient." This
parameter is generally estimated by fitting the model to data from an occupational mortality
cohort  study consisting of observed and expected numbers of cancer deaths categorized by
cumulative exposure, with the expected numbers determined from age- and calendar-year-
specific lung cancer mortality rates from an appropriate control population (e.g., U.S. males).
This model predicts that the mortality rate in an asbestos-exposed population is the product of
the mortality rate in an unexposed, but otherwise comparable, population, and the RR.
Consequently the excess mortality due to asbestos is the product of the background mortality rate
and the excess RR, KL*CEi0.  Since smokers have a higher background mortality from lung
cancer, the model predicts a higher excess asbestos-related lung cancer mortality in smokers than
in non-smokers.

To account for the possibility that an occupational cohort may have a different background
mortality rate of lung cancer than the control population (e.g., due to different smoking habits or
exposures to other lung carcinogens), in the present analysis Equation 7-1 is expanded to the
form,

                               RR = a * (1+ KL * CE,0)                        (Eq. 7-2)

where a is the RR in the absence of asbestos exposure relative to the control population. This
form of the model contains two parameters, the background RR,  a, and the lung cancer
exposure-response coefficient, KL.

7.2.1   The Adequacy of the Current U.S. EPA Model for Lung Cancer

Access to the raw epidemiology data from two key studies allowed us to evaluate the adequacy
of the U.S. EPA model (Equation 7-2) for describing the time and exposure dependence for lung
cancer in asbestos-exposed cohorts. For this analysis, the raw data for the cohort of crocidolite
miners in Wittenoom, Australia was graciously provided by Dr. Nick de Klerk (de Klerk 2001)
and the raw data for the cohort of chrysotile textile workers (described by Dement et al. 1994)
was graciously provided by Terri Schnoor of NIOSH (Schnoor 2001). The Wittenoom cohort
was originally described by Armstrong et a!. (1988), but the data provided by de Klerk includes
additional follow-up through 1999.
                                          7.4

-------
 7.2.1.1
Exposure Dependence
To evaluate the adequacy of the linear exposure response relationship assumed by the U.S. EPA
lung cancer model, the lung cancer model (Equation 7-2) was fit to the raw data from both the
South Carolina and Wittenoom cohorts.  In these analyses, each person-year of follow-up was
categorized by cumulative exposure defined using a lag of 10 years.  The data were then grouped
into a set of cumulative exposure categories and the observed and expected numbers pf lung
cancers were computed for each category. For South Carolina, expected numbers were based on
sex-race-age- and calendar-year-specific U.S. rates. Separate analyses were conducted for white
males, black males, and white females, as well as for the combined group. For Wittenoom
expected rates were based on age- and calendar-year-specific rates for Australian  men. The
categorized data and the resulting fit of the model to the data from Wittenoom and South
Carolina are presented in Tables 7-1 and 7-2, respectively.1                        '

For the Wittenoom data (Table 7-1), the fit with a=l is poor (pO.OOOl), as the model
overpredicts the number of cancers in the highest exposure category and underpredicts at lower
exposures. By contrast the fit of the model with a variable is adequate (p=0.1). This model
predicts a relatively high background of lung cancer in this cohort relative to Australian men in
general («=2.1) and a correspondingly shallow slope, with the RR increasing only from 2.1 at
background to 3.6 in the highest exposure category. A test of the hypothesis that  a=l for this
cohort is rejected (p<0.01) and the model fit to the data (with a variable) predicts  Ki=f0.0047
(f/ml-yr)-1.

The fit of the U.S. EPA model to the South Carolina lung cancer data categorized by cumulative
exposure is shown in Table 7-2. The model with cc=l cannot be rejected both when the model is
applied to white males only (p=0.54) or with all data combined (p=0.92). Since the values « and
KL estimated from white males only are similar to those estimated using the complete,cohort
(black and white males and white females), the fit to the complete data is emphasized in this
analysis. This fit predicts a=1.2 and 1^=0.021 (f-y/ml)'1. A test of the hypothesis that «=1 for
this cohort cannot be rejected (p=0.21), and in this fit KL=0.028 (f-y/miy1
        The fitting of the lung cancer model to the cohort data was carried out by assuming that the observed
numbers of cancers in different exposure categories were independent, with each having a Poisson distribution with
expectation equal to the expected number based on the control population times the relative risk predicted by
Equation 7-2. In the computation of the relative risk for a cumulative exposure category, the person-year-weighted
average cumulative exposure (lagged  10 years) for the category was used to represent the exposure in that category.
With these assumptions, a and KL were estimated by the method of maximum likelihood, confidence intervals were
constructed using the profile likelihood method, and likelihood ratio tests were used to test hypotheses (Cox and
Oakes 1984;  Venson and Moolgavkar 1988).                                                   *>

                                            7.5

-------
 Table 7-1. Fit of EPA Lung Cancer Model to Observed Lung Cancer Mortality Among
   Wittenoom, Australia Miners (Deklerk 2001) Categorized by Cumulative Exposure
                                Lagged 10 Years
Cumulative Exposure Lagged
10 Years (f-y/ml)
Range
0
0-0.4
0.4-1.0
1.0-2.3
2.3-4.5
4.5-8.5
8.5-16
16-28
28-60
60+
Total

Test of H0: a=l
p<0.01
Average
0
0.19
0.69
1.6
3.3
6.2
11.8
21.5
41.1
142.0



Observed
Deaths
5 '
27
11
22
28
38
31
21
25
43
251
Goodness of Fit

Expected
Deaths
4.6
7.9
8.2
11.6
12.9
14.3
13.2
9.2
11.6
11.6
105.1
P-value

Predicted Deaths
by Model
(0=1)
4.6
8.0
8.3
12.1
14.0
16.7
17.4
14.5
24.5
56.5
176.6
<0.0001

(«=2.1)
9.8
17.0
17.6
24.9
27.9
31.4
29.8
21.6
29.6
41.6
251.0
0.1

Estimates of KL (f-y/ml)'1
KL=0.027
90% CI: (0.020,0.035)
      [MLE])
KL=0.0047
90% CI: (0.0017,0.0087)
                                       7.6

-------
  Table 7-2. Fit of EPA Lung Cancer Model to Observed Lung Cancer Mortality Among
   South Carolina Textile Workers (Schnoor 2001) Categorized by Cumulative Exposure
                                    Lagged 10 Years
Cumulative Exposure Lagged
10 Years (f-y/ml)
Range
0
0-0.5
0.5-1.0
1.0-2.5
2.5^4.5
4.5-8.5
8.5-16
16-28
28-60
60-80
80-110
110+
Total
Average
0
0.35
0.75
1.7
3.4
6.2
11.8
21.3
41.5
69.2
93.3
173

Observed
Deaths
0
4
• 7
10
13
11
11
8
14
9
10
25
122
\
Expected
Deaths
0.9
2.6
5.5
9.7
7.9
7/7
7.2
5.3
6.3
3.0
•2.9
4.3
63.4
Predicted Deaths
by Model
(a=l)
0.9
2.7
5.6
10.1
8.6
9.1
9.6
8.4
13.8
9.0
10.8
25.6
114.1
' j[a=1.2)
' ' 1.1
, 3.2'
6.8
12.1
' f ,0.2
.' 10.5 '
10.9
i
, 9.2 •
, 14.3
9.0
1 10.5
24.2
122.0
                              Goodness of Fit P-value
                     0.92
• 0.96
 Test of H0: «=1
 p=0.21

 Estimates of KL (f-y/ml)'1

 (a=l)
 KL=0.028
 90% CI: (0.021,0.037)
  (a=1.2[MLE])
  KL=0.021
.  90% CI: (0.012,0.034)
Appendix A contains fits of the model to published exposure-response data from 18 studies (all,
with the exception of Wittenoom and South Carolina, obtained from the published literature). In
each of these 18 cases the linear model (Equation 7-2) provides an adequate description of the
exposure-response. However, o was estimated as >1.0 in 15 of these cases and statistically
significantly so in six cases, compared to only one case where a was found to be significantly
<1.0. If the true background lung cancer rates in these 15 cohorts with cc>l .0 are equal to that in
the corresponding control population (i.e., so that a fit with a=1.0 is appropriate), the exposure-
response would appear to be supra-linear. At the same time, the data in  11 of the 18 studies
(more than half) can be adequately fit with o=1.0 and the studies for which a=1.0 does not
provide an adequate fit do not appear to be related by mineral type, by industry, or by the size of
the study (Appendix A).  Thus, while we completed our analysis using the current lung cancer
                                          7.7

-------
model (Equation 7-2), further evaluation of the lung cancer exposure-response relationship
appears to be warranted.

7,2 J.2              Time Dependence

The lung cancer model (Equation 7-2) was next evaluated to determine whether it adequately
describes the time-dependence of the lung cancer mortality observed in the Wittenoom and
South Carolina cohorts. Of particular interest is whether the model accurately predicts lung
cancer mortality many years after exposure has ceased. The model predicts that RR increases
linearly with cumulative exposure lagged 10 years until 10 years following the end of exposure,
after which it remains constant from that time forward. However, it has been suggested (e.g.,
Walker 1984) that RRs for lung cancer eventually decline after cessation of exposure. Any
investigation of this issue should control for exposure level, since exposures can also fall with
increasing time due to higher death rates in more heavily exposed subjects. The availability of
the raw data from the South Carolina and Wittenoom cohorts provides an-opportunity to explore
this issue in more depth.

To investigate the assumption inherent in the lung cancer model (Equation 7-2) that the RR
remains constant following the secession of exposure, bivariate tables were constructed from
both the Wittenoom and South Carolina data in which observed and expected numbers of lung
cancers were cross-classified by both cumulative exposure (using the same exposure categories
as in Tables 7-1 and 7-2) and time since last exposure categorized using 5-year intervals. The
lung cancer model (Equation 7-2) was then fit to these bivariate data. A formal statistical test
was conducted of whether the lung cancer RR changed following the end of exposure. This test
consisted of appending the multiplicative factor, exp(-K.TSLE), to the lung cancer model
(Equation 7-2), where TSLE is time since last exposure, and conducting a likelihood ratio test of
whether the estimated parameter, K, is statistically significantly different from zero. For
presentation the bivariate tables were collapsed into univariate tables (Tables 7-3 and 7-4)
categorized only by time since last exposure, by summing observed and expected cancers over
cumulative exposure categories and computing person-year-weighted averages of cumulative
exposure and times since last exposure.

Table 7-3 shows the resulting fit of the U.S. EPA lung cancer model to the Wittenoom lung
cancer data categorized by time since last exposure. The RRs rise to a maximum between 10
and 30 years from last exposure and then decline thereafter.  However, a very similar pattern is
seen with cumulative exposure, which peaks between 10 and 15 years from last exposure, and
then declines due to higher mortality among more heavily exposed workers. The U.S. EPA lung
cancer model (which  does not assume a decrease in RR with time, but does account for any
decrease in exposure with increasing time since last exposure) provides an adequate fit to these
data (p=0.13, « estimated), and there is no apparent tendency for the predicted deaths to fall
below observed at the longest times since last exposure. To the contrary, the model-predicted
number of lung cancer deaths more than 35 years since last exposure (53.3) is very close to, but
slightly larger than, the  observed number (51). Likewise, the parameter K was not significantly
different from zero (p=0.16). Thus, the lung cancer model (Equation 7-2) provides a good
description of the Wittenoom lung cancer data categorized by time since last exposure and there
is no indication of a drop in the RR up to 45 or more years after exposure has ended that cannot
be accounted for by reduced exposures in the longest time categories. It should also be noted

                                           7.8

-------
^
             that the values of a (2.1) and KL (0.0051 f7ml-y) estimated from this bivariate analysis are very
             similar to those estimated from the univariate analysis summarized in Table 7-1.

               Table 7-3. Fit of EPA Lung Cancer Model to Observed Lung Cancer Mortality Among
              Wittenoom, Australia Miners (Deklerk 2001) Categorized by Years Since Last Exposure
Years Since Last Exposure
Range
0-1
1-5
5-10
10-15
15-20
20-25
25-30
30-35
35^0
40-45
45+
Total
Average
0.27
3.0
7.5
12.5
17.5
22.5
27.4
32.3
37.1
43.3
51.2

Average
Exposure
Lagged 10 Years Observed
(f-y/ml) Deaths
1.2
1.2
7.7
24.8
24.1
22.8
22.0
21.7
20.0
16.7
13.0

0
1
8
19
26
42
54
50
31
20
0
251
Expected
Deaths
0.6
1.6
3.9
7.0
11.4
16.4
20.1
20.3
13.9
9.6
0.4
105.1
Predicted Deaths
Relative by Model
Risk (o=2.1)
0
0.6
2.0
2.7
2.3
2.6
2.7
2.5
2.2
2.1
0

1.5
3.9
10.3
19.2
29.1
39.8
'47.2
46.6
31.4
21.1
0.8
250.9
                                                      Goodness of Fit P-value
0.19
             Table 7-4 shows the corresponding fit of the U.S. EPA lung cancer model to the South Carolina
             lung cancer data. This table indicates a marked decrease in RR with increasing time since last
             exposure. However, there is a concomitant decrease in cumulative exposure.  The U.S. EPA
             lung cancer model provides an adequate fit to these data (p=0.31, a estimated), and the estimates
             of a (1.3) and KL (0.20 (f-y/ml)"1) are very similar to those obtained from the univariate analysis
             (Table 7-2). There is no obvious tendency for the model to underestimate risk at the longest
             times since last exposure, and.the value of K estimated for this cohort is not significantly
             different from zero (p=0.12).
                                                      7.9

-------
 Table 7-4.  Fit of EPA Lung Cancer Model to Observed Lung Cancer Mortality Among
          South Carolina Textile Workers (Schnoor 2001) Categorized by Years
                                    Since Last Exposure
                               Average
Years Since Last Exposure     Exposure                                     Predicted Deaths
                           Lagged 10 Years   Observed   Expected  Relative      by Model
                                (f-y/ml)        Deaths     Deaths    Risk        (o=1.3)
    Range
A
Average
0-1
1-5
5-10
10-15
15-20
20-25
• 25-30
30-35
35^0

40-45
45+
Total
0.06
3.0
7.5
12.5
17.5
22.5
27.5
32.5
37.4

43.5
50.4

21.8
13.1
16.5
16.5
11.5
9.9
9.2
8.5
7.8

7.1
31.6

15
7
14
8
3
15
13
11
17

19
0
122
4.1
2.2
3.5
3.8
3.6
5.1
7.1
9.5
10.5
t
13.8
0.0
63.4
3.7
3.2
3.9
2.1
0.8
2.9
1.8
1.2
1.6

1.4
0.0

13.4
7.3
11.5
10.4
7.0
8.8
11.7
15.0
16.2

20.8
0.0
122.0
                                              Goodness of Fit P-value
                                                                   0.31
Raw lung cancer data were not available from any additional studies, so that the corresponding
analysis could not be performed on any other cohort. However, the report by Seidman et al.
(1986) on workers at a factory in Patterson, New Jersey that utilized amosite as a raw material
contains data in a form that permit a similar analysis using time since onset of exposure rather
than time since last exposure. The result of this analysis is shown in Table 7-5, which
corresponds to Tables 7-3 and 7-4 except that Table 7-5 presents mortality by time since first
exposure rather than time since last exposure.2  Since exposures in the Seidman study averaged
       2Table 7-5 was created in the following manner.  Table XIV of Seidman et al. (1986) contains expected
numbers of deaths from all causes cross-classified by cumulative exposure and time since onset of exposure.
Assuming that in each time-since-exposure-onset category the number of deaths from all causes is proportional to
the number of person-years, an average cumulative exposure was estimated for each time-since-exposure-onset
category.  Similarly, using Table V in Seidman et al. (1986), which contains expected numbers of deaths from all
causes cross-classified by cumulative exposure and time since onset of exposure, an average duration of exposure
was estimated for each time-since-exposure-onset category. The ratio of the estimate of cumulative exposure and
duration of exposure provides an estimate of the exposure intensity (f/ml) in each time-since-exposure-onset
category.  Using this estimate of exposure and the midpoint of the duration of exposure range for each cell, an
average cumulative exposure was estimated for each cell in Table X of Siedman et al., in which lung cancer data
were cross-classified using the same scheme as was used in Table V for deaths from all causes. The EPA lung
cancer model (Equation 7-2) was then fit to the lung cancer mortality data in this table. This bivariate table was then
collapsed by summing over cumulative exposure categories to produce Table 7-5, which categorizes the Seidman
et al. (1986) lung cancer data by time-since-exposure-onset.
                                              7.10

-------
only 1.5 years, time since first exposure approximates time since last exposure. Because
Seidman et al. (1986) did not include a lag in their exposure estimates, only data for times since
first exposure >10 years are included in our analysis, since (in this range) lagged and>unlagged
exposures would be very similar.                                              :
  Table 7-5. Fit of EPA Lung Cancer Model to Observed Lung Cancer Mortality Among
   New Jersey Factory Workers (Siedman et al. 1986) Categorized by Years Since First
                                       Exposure
Years Since
First Exposure
10-14
15-19
20-24
25-29
30-34
35+
Total
Observed
Deaths
13
20
17
22
21
16
109
Expected
Deaths
8.1
8.8
9.0
8.6
6.1
4.2
44.7
Relative Risk
1.6
2.3
1.9
2.5
3.5
3.8

Predicted Deaths by Model
(cc=3.4)
16.2
i.
18.8
20.1,
20.8 '
17.8 ',
I
15.3,
109.0.
                     Goodness of Fit P-value
0.55'
The RR of lung cancer increases with time since the beginning of exposure in the Seidman
cohort.  Nevertheless, the fit of the U.S. EPA model to these data with a estimated is very good
(p=0.55) and the numbers of cancer deaths predicted by the model agree closely with the
observed numbers, even up through 35+ years from the beginning of exposure. Time since last
exposure (estimated in each cell as time since first exposure minus average duration'of exposure)
is not a significant predictor of lung cancer mortality (K is not significantly different from zero,
p=0.33). The values of a (3.4) and KL (0.01 (f-y/ml)"1) estimated from this analysis agree closely
with those estimated in Appendix A by a different approach (cf. Table A-13 in Appendix A).
The analysis in Appendix A was also repeated after eliminating workers who worked longer than
2 years (i.e., workers for whom time since first exposure was not a good approximation to time
since  last exposure). Results obtained in this analysis are very similar to those shown in Table
7-5.

These analyses of the relationship between lung cancer mortality and time after exposure ends
are based on cohorts exposed to relatively pure asbestos fiber types: crocidolite (Wittenoom,
Table 7-3), chrysotile (South Carolina factory, Table 7-4) and amosite (Patterson, New Jersey
factory, Table 7-5).  All three of these  were consistent with the assumption inherentjn the U.S.
EPA lung cancer model (Equation 7-2) that RR of lung cancer mortality remains constant after
10 years past the end of exposure. Thus, the U.S. EPA model appears to adequately'describe the
time-dependence of lung cancer mortality in asbestos exposed cohorts.
                                         7.11

-------
7.2,1.3              Smoking-Asbestos Interaction With Respect to Lung Cancer

The existence of an interaction between smoking and asbestos in causing lung cancer has been
known for many years.  For example, Hammond et al. (1979), in a study of U.S. insulation
workers, found a multiplicative relationship between smoking and asbestos exposure in causing
lung cancer. A multiplicative relationship for the interaction between smoking and asbestos is
also the relationship inherent in the U.S. EPA lung cancer model (Equations  7-1 and 7-2), by
virtue of the fact that increased risk from smoking in reflected in the background risk, which is
multiplied by the asbestos-induced RR. More recent work, however, suggests that the
interaction between smoking and asbestos exposure may be more complicated.

As described by Liddell (2001), many reviews of the interaction between smoking and asbestos
exposure have tended to conclude that the interaction is multiplicative due primarily to (1) the
conclusions of Hammond et al. (1979), which was the largest of the early studies and was thus
most heavily weighted, and (2) lack of viable alternatives among the options considered. As
Liddell (2001) explains, commonly, reviewers examined fits of a simple multiplicative model or
a simple additive model and found that (of these two) the multiplicative model tended to provide
a better fit,  although the fits of either model were sometimes relatively poor.  Moreover, only
limited sets of mortality data from asbestos exposed cohorts could be used to evaluate these
model  fits because only limited smoking histories had been collected among such cohorts and,
among the data sets studied, the evaluation was typically limited to distinguishing effects among
dichotomous groups (i.e., non-smokers and smokers, the latter of which were treated as a single,
collective group).

In a more recent analysis of cigarette smoking among the cohort of Quebec miners and millers
exposed to chrysotile (which had been previously described in numerous studies, see Appendix
A), Liddell and Armstrong (2002) found that the interaction between smoking and asbestos
exposure was complex:  certainly less than multiplicative, but not quite linear either.

To get some idea of the possible impact of a non-multiplicative asbestos-smoking interaction
upon our work to reconcile lung cancer exposure-response coefficients calculated from different
environments, consider the generalized model for RR,

                           RR = 6 + ps.S + pA.A + Y«A.S                    (Eq. 7-3)

where  S is a measure of the amount smoked, A is a measure of asbestos exposure, and 6, ps, PA
and Y are parameters. If y=0, the model predicts an additive smoking/asbestos interaction, and if
Y=ps * PA / S> the model becomes

                        RR= 6 * [1 + (Ps/6). S] * [1 + (pA/6), A]                 (Eq. 7-4)

which  is a multiplicative smoking/asbestos interaction. Thus, this model generalizes both
additive and multiplicative interaction, and has been used by a number of researchers, including
Liddell and Armstrong (2002), to study smoking/asbestos interactions.
                                          7.12

-------
Note that the generalized model Equation 7-4 can be written in the form
                                                             ps , S)]         ;,   (Eq. 7-5)
                                                                             !•
Further, note that Equation 7-5 is of the form: const, * (1+constj* [asbestos exposure]), which is
the same form as the U.S. EPA lung cancer model (Equation 7-2), except that in Equation 7-5
const, (=6+ps,S) and const2 (=(pA+Y.S)/(6+ps.S)) depend upon the amount smoked, while in the
U.S. EPA lung cancer model (Equation 7-2) const, (=a) likewise depends upon the amount
smoked, but the constz (=KL) does not.  Thus even with this generalized model, the U.S. EPA
model would still apply, although the KL estimated would depend upon the smoking habits of the
underlying cohort.  That the generalized model Equation 7-3 is at least approximately correct is
supported by the fact that the U.S. EPA model (Equation 7-2), which we have noted is
comparable to Equation 7-3 except that the KL value may depend upon smoking, provided a
valid fit to data from all of the 20 studies to which it was applied.

This at least suggests that, even if the multiplicative interaction assumed by the U.S.,EPA model
(Equation 7-2) is not correct, as long as the smoking habits in different cohorts don't differ
extremely, the KL value from different cohorts will be estimates of nearly the same quantity. In
particular, the  differences in the estimated KL produced by differences in smoking habits in
different cohorts is likely to be a relatively small component of the very large variation in the KL
computed from different studies (see, e.g., Table A-l in Appendix A). In that case, KL computed
from different environments using the U.S. EPA model (Equation 7-2) would still be comparable
and it would be meaningful to search for measures of asbestos exposure ("exposure indices")
that better rationalize KL values (calculated using Equation 7-2) from different studies.3

Although the above argument suggests that efforts to reconcile KL values from different cohorts
is still a valid and useful exercise, even if the smoking/asbestos interaction is not multiplicative,
as assumed by the U.S. EPA model, further evaluation of the interaction between smoking and
asbestos exposure,  including evaluation of different exposure-response models, is clearly
warranted. However,  it must be recognized that the ability to conduct such evaluation will be
limited by the  small number of asbestos-related epidemiology studies in which smoking data are
available, and  possibly further limited by the ability to gain access to the raw data from these
studies. It should also be recognized that smoking data are not available for most of the studies
currently available (see, e.g., Table A-l in Appendix A), and, for most of those studies, the
published data are probably not sufficient to allow fitting of a model that incorporates any
smoking/asbestos interaction other than multiplicative.
       3However, if the interaction is not multiplicative as assumed by the EPA model, it could be problematic to
use a KL value estimated from the model to quantify absolute asbestos risk (e.g., from lifetime exposure) while
taking into account smoking habits.
                                          7.13

-------
7.2.1.4              Conclusions Concerning the Adequacy of the U.S. EPA Lung Cancer
                    Model

Three principle assumptions inherent to the U.S. EPA lung cancer model were considered in this
study to evaluate the overall adequacy of the model for describing the manner in which asbestos
exposure induces lung cancer. These are:

     •       that lung cancer RR is proportional to cumulative exposure (lagged 10 years);

     •       that risk remains constant following 10 years after the cessation of exposure; and

     •       that the interaction between smoking and asbestos exposure is multiplicative.

The findings from this evaluation are summarized briefly below. For convenience, timed-
dependence is addressed first.

In Section 7.2.1.2, the time response predicted by the lung cancer model (Equation 7-2) was
evaluated using the Wittenoom and South Carolina data, along with the data from a New Jersey
amosite factory (Seidman et al. 1986).  Follow-up in the Wittenoom and South Carolina was
sufficient to permit evaluation of the adequacy of the model 40-45 years past the end of
exposure, and follow-up in the New Jersey factory permitted the model to be evaluated up.
through 3 5+years past the end of exposure.                  ^

The data from all three studies were consistent with the assumption inherent in the lung cancer
model that RR of lung cancer mortality remains constant after 10 years past the end of exposure.
That this result holds equally for a cohort exposed to chrysotile (South Carolina) as for cohorts
exposed to crocidolite (Wittenoom) and amosite (New Jersey factory) is particularly noteworthy,
because chrysotile fibers have been reported to be much less persistent than amphibole fibers
in vivo (see Section 6.2.4). The fact that the lung cancer risk in South Carolina remained
elevated 40-45 years following the cessation of exposure, along with the fact that the KL from
the South Carolina cohort is the largest KL value obtained from our review of 20 studies (see
Table A-l in Appendix A) suggests that carcinogenic potency of asbestos is not strongly related
to durability, at least for lung cancer (see Section 6.2).

Based on the results indicated in Section 7.2.1.1, the linear, RR exposure-response model for
lung cancer (Equation 7-2) provides an adequate description of the exposure-response
relationship from 20 studies. There is little evidence that exposure-response is sub-linear or
"threshold-like". However, in a number of these cases the fitted model predicts a background
response that is higher than that in the control population.  If the background rate in the control
population was actually appropriate in  these studies, the exposure-response relationship in these
studies would appear in some cases to be supra-linear. These results further reinforce the
recommendation of the expert panel (Appendix B) that evaluation of a broader range of
exposure-response models (including those that incorporate smoking-asbestos interactions that
are other than multiplicative) for lung cancer is appropriate.
                                          7.14

-------
Although further evaluation of alternate models is warranted, the specific studies for which
«=1.0 does not provide an adequate fit to the data do not appear to be related by mineral type,
industry, or size of study cohort and such studies total fewer than half of the studies|evaluated.
Coupled with the additional evidence that the linear, RR model for lung cancer (Equation 7-2)
provides a good description of the time-dependence of disease mortality during and following
exposure and that it is not likely to grossly underestimate exposure despite limitation in the
manner that smoking is addressed (Section 7.2.1.3), this suggests that use of this model may be
adequate for  estimating exposure-response factors for the existing studies, at least unless and
until a model that provides a superior fit to the data is identified.
                                                                            r
Further support for the U.S. EPA lung cancer model  is also provided by Stayner et a'l. (1997).
Stayner et al. evaluated the relative fit of a variety of additive, multiplicative, and more complex,
empirical models to the same raw data from the South Carolina textile plant that we'evaluated
and conclusions drawn by these researchers generally parallel and reinforce those reported here.

Stayner et al. (1997) found a highly significant exposure-response relationship for lung cancer
and that a linear, RR model (similar to that described by Equation 7-1, except that th'ey initially
assumed a lag of 15 years) provided the best overall fit to the data (among the additive and
multiplicative models evaluated).  Moreover, the fit to the data was not significantly; improved
by adding additional parameters, nor was there any indication of a significant interaction with
any of the covariates evaluated (age, race, sex, or year).  These researchers did find an
interaction with time such that by applying different slopes for different latencies (15-29 years,
30-39 years,  and >40  years), they were able to obtain a significantly better fit.  However,
Stayner et al. (1997) did not evaluate a RR model with a single slope using a lag of 10 years.
Therefore, their analysis cannot be used to evaluate the effect of assuming different slopes for
different latencies with respect to the U.S. EPA model, which assumes a lag of 10 years.

Because we found that mortality among the South Carolina cohort is adequately described with
o=1.0, the Stayner et al. (1997) study does not directly address questions concerning supra
linearity that  have been suggested in our findings. Nevertheless, taken as a whole, the evidence
evaluated here suggests that the model described in Equation 7-2 may be adequate  for evaluating
the relationship between asbestos exposure and lung cancer mortality among the existing
epidemiology studies. Certainly, no clearly superior model has yet been identified. Therefore,
the U.S. EPA lung cancer model was employed in this study for evaluating lung cancer risk
while we also recognize that further evaluation of alternate models is warranted. The primary
obstacle to  more fully evaluating alternate models has been and continues to be lack/of access to
the raw data for a greater number of cohorts.                   ' '   "

7.2.2  Estimating KL values from the Published Epidemiology Studies

The U.S. EPA model for lung cancer (Equation 7-2) was applied to each of the  available
epidemiology data sets to obtain study-specific estimates for the lung cancer exposure-response
coefficient, KL. Based on the results presented in Section 7.2.1.4, while there is some evidence
that models in addition to the U.S. EPA lung cancer model should be explored,  there is little
indication that the current model does not provide an adequate description of lung cancer
mortality that is sufficient to support general risk assessment. The set of KL values derived from
                                          7.15

-------
available epidemiology studies are presented in Table 7-6, which is a reproduction of Table A-l
in Appendix A.

In Table 7-6, Column 1 lists the fiber types for the various.studies, Column 2 lists the exposure
settings (industry type), and Column 3 indicates the specific locations studied.  Column 4
presents the best estimate of each KL value derived for studies, as reported in the original 1986
Health Effects Update and Column 5  presents the reference for each respective study.  Columns
6,7, and 8 present, respectively: the best-estimates for the KL values derived in our evaluation
for all of the studies currently available (including studies corresponding to those in the 1986
Health Effects Update); a statistical confidence interval for each KL value (derived as described
in Appendix A); and an uncertainty interval for each KL value (also derived as described in
Appendix A).  The reference for the respective study from which the data were derived for each
KL estimate is provided in the last column of the table. To assure comparability across studies,
values for all studies (even those that have not been updated since their inclusion in the 1986
Health Effects Update) were re-derived using the modified procedures described in Appendix A.

As explained in Appendix A, the uncertainty intervals for KL values (and corresponding intervals
for KM values, the exposure-response coefficients for mesothelioma) are intended to reflect, in
addition to statistical variation, other  forms of uncertainty that are difficult to quantify, such as
model uncertainty and uncertainty in  exposure estimates.  We interpret these informally as
providing a range of KL values that are reasonable, based on the data available for a given study.
Accordingly, if uncertainty intervals for two KL values do not overlap, these two underlying sets
of data are considered to be incompatible. Potential reasons for such incompatibility include the
possibility that the KL values are based on an exposure measure that does not correlate well with
biological activity of asbestos. One of the goals of this report is to determine an exposure index
that correlates better with biological activity, and consequently brings the KL values into closer
agreement. The degree of overlap of the corresponding uncertainty intervals provides  an
indication of the extent to which a groups of KL values are in agreement.

The KL values derived in this study and the corresponding values derived in the original 1986
Health Effects Update generally agree at least to within a factor of 3; a couple vary by a factor of
4; one varies by a factor of 5; and one varies by a factor of about 15.  However, in every case the
KL from the 1986 update lies within the uncertainty interval for the KL derived in the current
update.

Perhaps the most interesting of the changes between the 1986 KL value estimates and the current
KL value estimates involves the friction products plant in Connecticut (McDonald et al. 1984).
Although a relatively small, positive exposure-response was estimated from this study in the
1986 Health Effects Update, the best current estimate is that this is essentially a negative study
(no excess risk attributable to asbestos). The difference derives primarily  from allowing a to
vary in the current analysis. The exposure groups in this cohort do not exhibit a monotonically
increasing exposure-response relationship and, in fact, the highest response is observed among
the group with the lowest overall exposure. As indicated in Appendix A, lack of a
monotonically increasing exposure-response relationship is a problem observed in several  of the
studies evaluated.
                                           7.16

-------
.1
•3
9
lemiological S
'S.
u
Various
S
o
•e
S
>
S
^_
^
oefficients (1
U
3!
e
o
a
S
w
9
O
a,
m
n
a
y
Ml
o

3
±
R
H


Reference
Uncertainty
Interval*
4>
^ n
ol £
0 S
U


2
•3 *
en ^
e


Reference
^
QO o
^ft C!
•>—' **
(d

O
•g-
u

S
o
• MM
*P»
2
V
D.
0


a
i
tt.
13
•*«*
"o
oo ^*
§OS
0
2-°
C
q
d
o
S


(N
0
d


"rt
3§ i
C wi ^
o — "3
9 -3 .-gg


vn r»»
d d

,S
|1




oo op
c .e
'c HS
is

JU
"o
1
6
Piolatto et al.
1990
McDonald
etal. 1984
O <^
• *°.
S o"

r— r-
d d
S S


V)
o o
d



<-. 2 ^
1 1^
^3 o Q — ;


^^
§ q
d <=>

a 3
.S .0
2 S c •£
ll li




.il
ts "g
•c 2
b CL,




W Kl
t3 t>
4eh
3 <^ .^ os
^ ^H Q ^^
o" 5
^ S

VO "*
d j^-
d — '


«o
^* i
0 ^



13
ts
-t_>
g*
§s
02



oo
CN

1
C3 CQ
S "3,
•c *
? 1 *l
« § g a
?: & &o u




ill
U S H




McDonald
1983a
de Klerk et al.
-1994C
$ i
s- e
^^^
"O" oo
^ d !
4 rf .
s^ 5


r-
^M ^^
0 '
1
i
i
McDonald
etal. 1983a



>/•> i
 o\ e ^ S 13
*> o\ 5 > « **
J — • •< 5^ i< «
^ r
-^. ^ oo
«« •* '(Si
" S' U
S o g
o 2/
^
x-s S^ n
d - K
e 5 ^
S- 0,

1
^ - OS
^H "*) rn
000
.r
!'
!.
1



r

a
!r,l
0> 0 XI C5
•^ o ^ O
CA
S :'


ll .
•3 tn ''
i s :
^is •
..
u
"o
H
Berry and
Newhouse
1983
CO
oj

^-^
oo
d
d^


00
o
d



Berry and
Newhouse
1983


oo
0
d

|
^
•S
m




§1
t> •§
•c £
[i. CU


"S
_x
2

-------
(continued)
en
4)
•3
9
•**
W
"i
f
o

V
3
Q-
w

e^ T3 C
o e 
n
^
O

O)
t
H
t-
<
£
S | -
to « tS
Is f£ IS
S ON 3 ON "3 ON
il^ «—» »JH ^* *% *•*

^Tt* irs ^T^
—> (^ ^^
^ vo ^
^ o M
s s ^



|x-
ON m !>,
 ON
fc S ? iQ

* R
^ d

i-t
I i
illi!
« y 5 ^ ^
O **,» C3
•
JU
'S
{2




~5
tS
2£
U ON
O< ^^

r
1
.S
'i
>,
•C*
.S
1
3
^

^ §§
"rt E (S
go-S>
•sis
i> 3 t
O *^ "N ^^
g ^ 0.
-8 w =>
« O s
imbining 90% con
Terri Schnorr (NI
iw data with folio1
S E e
O ^3
-&*!
T3 03 "R
4> 'g !S
E TJ •§
111
C4 «-,*-,
> flj {rt
C *•* -S
1 1 1
"^ S 8
.s al
533
O je* _f-
O *5 :£3
?FF
CO

-------
Among the KL values derived in the current study, the lowest and highest of the best-estimate
values differ by a factor of 300 (excluding the negative study of the Connecticut friction
products plant, which would make the spread even larger) and several pairs of uncertainty
intervals have no overlap. For example, the KL uncertainty interval for the chrysotile miners in
Quebec lies entirely below the corresponding intervals for chrysotile textile workers' (in either of
the two studies for South Carolina, which are of highly redundant cohorts  in the same plant), for
textile workers in the Pennsylvania plant, and for arriosite insulation manufacturers (in the
Seidman study (1986).

The KL values and the associated uncertainty intervals are plotted in Figure 7-1. Each exposure
environment is plotted along the X-axis of the figure and is labeled with a 4-digit code that
indicates fiber type (chrysotile, mixed,  crocidolite, or tremolite), industry (mining, friction
products, asbestos-cement pipe, textiles, insulation manufacturing, or insulation application); and
a 2-digit code indicating the study from which the data were derived. A key is also provided. In
Figure 7-1, the chrysotile studies are grouped on the left, amphibole studies are grouped on the
right, and mixed studies are in the middle.

For studies conducted at the same facility (generally among highly overlapping cohorts), such as
the Dement et al. (1994) and the McDonald et al. (1983a) studies of the same South|Carolina
textile facility, a single study was selected for presentation in Figure 7-1. Thus, for South
Carolina, the Dement et al. (1994) study is presented because we had access to the r^w data for
this study. It is also a newer study.  Similarly, the Amandus and Wheeler (1987) study was
selected to represent the Libby Vermiculite site over the other study at this facility (McDonald et
al.  1986). The effects of such selection is expected to be small in any case because the KL values
estimated for the individual studies in each pair vary only by a factor of 2.         '

Comparisons of KL values across the available studies are instructive. Within chrysotile studies
alone (and excluding the negative friction products study), lowest and highest KL values vary by
approximately a factor of 70.  Moreover, as previously indicated, the uncertainty intervals for the
lowest (non-zero) value (for Quebec miners) and the highest value (textile workers);have no
overlap. Uncertainty intervals for the negative friction product study and the  other estimates for
chrysotile do overlap, primarily due to the wide confidence interval associated withjthe negative
study.

Among the apparent variations, differences in lung cancer potency observed among'Quebec
miners versus that observed among South Carolina textile workers has been the subject of much
discussion and evaluation, which is worthy of review (Appendix D). The  inability to reconcile
these differences, appears to be among  the biggest obstacles to reliably estimating an overall
chrysotile dose-response coefficient for lung cancer.

As  indicated in Appendix D, the leading hypothesis for the apparent differences in lung cancer
risk per unit of exposure observed between chrysotile mining and textile manufacturing is the
relative distribution of fiber sizes found in dusts in these industries. Evidence from several
studies indicates that textile workers were exposed to dusts containing substantially'greater
concentrations of long structures than dusts to which miners were exposed. Thus, the effects of
fiber size is considered further in Section 7.4.
                                           7.19

-------
                                 Figure 7-1:
Plot of Estimated KL Values and Associated Uncertainty Intervals by Study Environment

     1000-g
      100,
       10-

      0.1 1
     0.01-
     1E-3
                                          T  T
            S
 .
Sfclfclfc
             lOCDh-    COOIOT-
             T-,— ^-    T— T- CJ CM
                                    7.20

-------
                         Key for Figures 7-1 through 7-6 Code
FiberTypes
(First Digit of Code)

A = amosite
C = chrysotile
M = mixed fibers
R = crocidolite
T = tremolite (in vermiculite)
Study Environment
(Second Digit of Code)
A = insulation application
F = friction products manufacturing.
I = insulation manufacturing
M = mining  .
P ='ac pipe manufacturing
T = textile manufacturing
X = misc. products manufacturing :'
Study Cohorts
(Last 2 digits)

1 = Quebec miners (Liddell et al. 1997)
2 = Quebec miners (Liddell et al. 1997, raw data)
3 = Italian miners (Piolatto et al. 1990)                   '.
4 = Connecticut friction product workers (McDonald et al. 1984)
5 = New Orleans ac pipe manufacturers (Hughes et al. 1987)
6 = South Carolina textile manufacturers (Dement et al. 1994, raw data)
7 = British friction product manufacturers (Berry and Newhouse 1983)
8 = Ontario ac pipe manufacturers (Finkelstein 1984)
9 = New Orleans ac pipe manufacturers (Hughes et al. 1987)
10 = Swedish ac pipe manufacturers (Albin et al. 1990)
11 = Belgium ac pipe manufacturers (Laquet et al. 1980)
13 = Retired factory workers (Enterline et al. 1986)
14 = Factory workers (Liddell et al. 1997)                 ;
15 = Insulation appliers (Selikoff and Seidman 1991)
16 = Pennsylvania textile workers (Mcdonald et al. 1983b)  '
17 = British textile workers (Peto et al. 1985)
18 = Australian crocidolite miners (de Klerk, unpublished, raw data)
19 = New Jersey insulation manufacturers (Seidman et al. 1986)
20 = Texas insulation manufacturers (Levin et al. 1998)    !
21 - Libby vermiculite miners (Amandus and Wheeler 1987)
                                         7.21

-------
Ignoring the negative Connecticut friction products study, the range of KL values observed
across chrysotile studies (70) appears to be substantially narrower than the range observed across
all studies (300). However, if the nearly negative study of the Belgium asbestos-cement pipe
manufacturers is also removed, the range observed for chrysotile studies is almost identical to
the range observed across all studies. This is because the KL values for Quebec represents the
low extreme of both ranges and South Carolina represents the high  extreme of both ranges.

Among "pure" amphibole studies, the lowest and highest of the best-estimate KL values vary by
a factor of approximately 9 and the two extremes both derive from within the same industry
(amosite insulation).  The two studies in question are of amosite insulation manufacturing plants
(one in Patterson, New Jersey, and one in Tyler, Texas) that utilized the same equipment
(literally) and apparently had similar sources of asbestos (South Africa). Despite the 9-fold
difference in KL values, the uncertainty intervals for these two estimates have substantial
overlap. If the arithmetic mean of the values for the two amosite insulation studies is used, KL
values estimated, respectively,  for crocidolite mining, amosite insulation manufacture, and
mining of vermiculite contaminated  with tremolite vary by less than a factor of 2.  However, this
is possibly fortuitous, given the magnitude of the associated uncertainty intervals (Figure 7-1).

As indicated in Appendix D, for example, it is possible that mining studies tend to exhibit low
KL values relative to studies of asbestos products industries. This is due to the presence of large
numbers of cleavage fragments in the dusts that may not contribute to biological activity
(because the majority of these may not exhibit the requisite size to be biologically active) but
which, nevertheless, are included in  estimates of asbestos concentrations in the original
epidemiology studies. If the KL estimates for Libby and Wittenoom could be adjusted for this
effect, they might be closer in value  to that obtained from the Seidman study.

It is also instructive to compare variation within and between industries. Within industries
(especially for a single fiber type), the data are limited. The studies from the two chrysotile
mines (Quebec and Italy) show remarkably close agreement, varying by less than a factor of 2.
The three studies involving mining of amphibole (Wittenoom and the two studies of Libby) also
vary by less than a factor of 2.  However, the mean of the amphibole mining group is
approximately 10 times the mean of the chrysotile mining group. Moreover, based on inspection
of their respective uncertainty intervals, the KL values for chrysotile mining in Quebec and
Crocidolite mining in Wittenoom appear to be incompatible.

Across all asbestos types (including  mixed), the asbestos-cement pipe industry shows the
greatest variation, including a nearly negative study (best estimate KL=0.000068) and four more
studies with KL values that range up to 0.0029 producing a variation within this industry of a
factor of 40.  The friction products industry includes one negative and one positive study.  Better
agreement is observed among textiles. The two mixed textile plants show KL values that vary by
no more than a factor of 5  from each other and  from the KL for the South Carolina chrysotile
textile plant.
                                           7.22

-------
7.3   MESOTHELIOMA                                                  ]4

The model proposed in the Airborne Health Assessment Update (U.S. EPA 1986) to describe the
mortality rate from mesothelioma in relation to asbestos exposure assumes that the mortality rate
from asbestos-induced mesothelioma is independent of age at first exposure and increases
according to a power of time from onset of exposure, as described in the following relationship:
              I
              M
= KMf [(T-10)3-(T-10-d)3]

= Kw-f-CT-10)3

= 0
forT>10 + d

forlO + d>T>10

forlO>T
                                                          (Eq.7-6)
     where:
             K
               M
is the mesothelioma mortality rate at T years from onset of exposure to
asbestos for duration d and concentration f;                :

is the proportionality constant between exposure and mesothelioma
response and represents the potency of asbestos;
A more general expression that holds for variable exposure is given by
                        IM »
                                                          (Eq. 7-7)
where f(x) is the concentration of fibers at time x following the beginning of exposure. This
expression reduces to Equation 7-6 when exposure is constant (see Appendix A).

7.3.1 The Adequacy of the Current U.S. EPA Model for Mesothelioma

Access to the raw epidemiology data from a few key studies allowed us to evaluate the adequacy
of the U.S. EPA model (Equation 7-6) for describing the timerdependence for mesothelioma in
asbestos-exposed cohorts. For this analysis, the raw data from a cohort of chrysotile miners in
Quebec was graciously provided by Drs. Douglass Liddell and Corbett McDonald (described in
Liddell et al. 1997), the raw data for the cohort of crocidolite miners in Whittenoom; Australia
was graciously provided by Dr. Nick de Klerk (unpublished) and the raw data for the cohort of
chrysotile textile workers (described by Dement et al. 1994) was graciously provided by Ms.
Terri Schnoor of NIOSH and Dr. John Dement of Duke University. The Whittenoom cohort was
originally described by Armstrong et al. (1988), but the data provided by Dr. de Klerk included
additional follow-up through 1999.

To identify potential effects due to varying statistical procedures, different methods ]for fitting
the U.S. EPA mesothelioma model to epidemiological data were evaluated.  In this evaluation,
three methods were used to fit the U.S. EPA mesothelioma model to data from Wittenoom.
                                         7.23

-------
In the first approach the data were categorized in a manner often available in published form, so
this method mimics the method generally used when raw data are not available. The observed
mesotheliomas and person-years of observation were categorized by time since first exposure,
and the mean exposure level and duration of exposure were calculated for each such category.
The U.S. EPA model was then applied to such data using the approach for the typical situation
(as described in Appendix A) and results for the Wittenoom cohort are presented in Table 7-7.
The KM value estimated for Wittenoom using this approach is 7.15xlO'8 (90% CI: 6.27x10'8,
8.1 IxlO"8).  The fit of this model to data categorized by time since first exposure is good
(p=0.65).

For most of the published epidemiology data sets, the average level and duration of exposure for
individual time-since-first-exposure categories are not provided and have to be estimated from
cruder data representing cohort-wide averages.  Thus for most of the epidemiology data sets, the
calculation of KM is based on cruder information than the calculation presented in Table 7-7.

 Table 7-7. Fit of EPA Mesothelioma Model to Observed Mesothelioma Mortality Among
 Wittenoom, Australia Miners (Deklerk 2001) Categorized  by Years Since First Exposure
Years Since
Range
0-5
5-10
10-15
15-20
20-25
25-30
30-40
40-100
Total
First Exposure
Average
0
0.19
0.69
1.6
3.3 •
6.2
11.8
21.5

Average
Duration
(years)
0.643
0.91
0.958
0.970
0.953
0.957
1.05
1.13

Average
Concentration
(ffml)
30.2
30.2
30.0
29.7
29.5
29.3
29.0
28.1

Observed
Deaths
0
0
1
5
20
25
90
23
164
Predicted Deaths
by Model
0.0
0.0
0.6
6.6
17.6
30.1
78.1
31.1
164.0
                                         Goodness of Fit P-value
0.65
 Estimates of KM
 KM=7.15xlO-s
 90% CI: (6.27xlO-8, 8.1 IxlO'8)

A second approach to fitting the U.S. EPA model to epidemiology data exploits the fact that the
mesothelioma model (Equation 7-7) expresses the mesothelioma mortality rate as the product of
KM and an integral involving the exposure pattern, the time of observation, and the 10-year time
lag, but not any parameters that require estimation.

The value of this integral was calculated for each year of follow-up of each subject.  Person-
years of follow-up and mesothelioma deaths were then categorized according to the values of the
                                          7.24

-------
integral, and the average value of the integral determined for each category. Results are
presented in Table 7-8.                                                      "•.

 Table 7-8. Fit of EPA Mesothelioma Model to Observed Mesothelioma Mortality Among
  Wittenoom, Australia Miners (Deklerk 2001) Categorized by Average Value of Integral
                                    (Equation 6-12)
Average Value of Integral
0
28.4
206.3
732.3
2,038.1
5,153.7
12,385.8
30,639.7
144,801
Total
Goodness of Fit P-value
Observed Deaths
0
0
1
4
10
23
32
27
67
164

. Predicted Deaths
0.0
0.0
1 . 0.3
1.0
2.9
6.8
; 13.8
25.8
113.3
164.0
O.0001
by Model
;
*





;.


•'i
 Estimates of KM
 KM=9.00xlO-8
 90% CI:  (7.89xlO'8, 10.2x10'8)
KM was estimated by maximum likelihood from Table 7-8 assuming that the observed numbers
of cancer deaths in the different categories of the integral were independently Poissori distributed
with a mean equal to the mean value of the integral for that category times KM. The value of Kw
obtained in this fashion was KM=9.00xlO'8 (90% CI: 7.89xlO-8, J0.2xlO'8). This estimate is
about 25% larger than obtained with the data categorized by time since first exposure; although
confidence intervals obtained from the two procedures overlaps The fit of the model to data
categorized by the integral is poor (pO.OOOOl), as the model predicts too few mesotheliomas for
small values of the integral and too many mesotheliomas for large values of the integral.

A third method of estimating KM (termed the "exact method") employs a likelihood that does not
involve any categorization of data.  With this method, the hazard function, h(t)=IM(t) and the
corresponding survival function (probability of surviving to age t without death from
mesothelioma in the absence of competing causes of death), S(t)=exp(-/0' h(s)ds), are'computed
for time at the end of follow-up.  The contribution to the likelihood of a subject who died of
mesothelioma t years after beginning of follow-up is h(t)*S(t), and the contribution of a subject
whose follow-up was not terminated by death from mesothelioma is S(t). The complete
likelihood is the product of such terms over all members of the cohort. The estimate of KM
                                         7.25

-------
obtained by maximizing the logarithm of this likelihood was 7.95x10'8 (90% CI: 7.0x10"8,
9.0x10'8).

We consider the exact method of computing KM to be the most accurate and results from this
method are reported in the summary table for mesothelioma (Table 7-9, which is a reproduction
of Table A-2). The Quebec and South Carolina data sets were thus evaluated using the exact
method. However, it is noteworthy that the two other methods described above, one of which is
often applied to published data, give similar estimates of KM, at least for this data set.

The Quebec cohort was subdivided into three subcohorts, believed to correspond to differing
amounts of amphibole exposure due to tremolite contamination of the ore at different mining
locations, and to use of some imported commercial amphibole at one factory location (Liddell et
al. 1997).  Location I consisted of workers at the mine at Asbestos where the ore reportedly had
less tremolite contamination. Locations 3 and 4 consisted of workers at the large central mine
and at smaller mines, respectively, near Thetford, where the ore was more heavily contaminated
with tremolite. Location 2 consisted of workers at an asbestos products factory at Asbestos,
which processed some commercial amphibole fibers in addition to chrysotile.  The exact method
of calculating KM produced the following estimates: Location 1 (8 cases): KM=1.3xlO"10 (90%
CI: 0.3x10-'°, 4.9X10'10); Location 2 (5 cases): KM=9.2xlO-10, (90% CI: 2.0x10-'°, SSxlO'10);
Locations 3 and 4 (22 cases): KM=2.1xlO-10 (95% CI: 0.65x10'10, 6.5x1O'10).

The relative magnitudes of these estimates track with the relative amounts of amphibole
exposure estimated for these locations (Liddell et al. 1997), which is consistent with the
hypothesis that the mesothelioma risk in this cohort is due, at least in large measure, to exposure
to amphiboles.

There were only two confirmed mesothelioma deaths in the South Carolina cohort and four
additional suspected deaths. These were too few to permit detailed analysis. Based on both
confirmed and suspected mesothelioma deaths, the exact method of analysis gave an estimate of
KM=0.43xlO'8, 90% CI: (0.20xlO'8, 0.79xlO'8). Using only the two confirmed mesotheliomas,
the same analysis yielded KM=0.14xlO-8,90% CI: (0.034xlO-8, OJSxlO'8). Very similar
estimates were obtained by estimating KM from data categorized by time since first exposure and
fitting a linear model to the categorized value of the integral in the definition of the U.S. EPA
model.  Thus, for this cohort comparable KM values are estimated no matter which of the three
methods described above are used for fitting the U.S. EPA mesothelioma model to the
epidemiology data.
                                          7.26

-------



J
•j
5
1
"S
e
.1
4
"E
fct
0
c
'C
>
1^^
|
1
1
r
I
y
^^
JS
.1
u
1
o
U
V
35
a
o


£
tn
i
£
S
1
09
1
cK
I--
—
lO
rot
H




!
\
1
tj
I
a

£.
f%
r i
5 <
2l
3

_
^ g 5
e^ -O C
S «5 "
§£
u

« 2
.s s 2
^!1
=> ^


41
W
D
a

_ o
< ^S
s« S *
M S s
— «

1
3
B
_S
^i
2
V
s

V
. £
u
JB
E
TT
00
f- r- 2 «•> h
ON ON " S 2J
ON ON -j ON ON
•3 T3 " "3 "3
w « 1 « «
? =5 1 1 1
i i i t §
J J S S Q
^-v ;•"••
5T S "- ^
S « c? ^ 2
6 «=> $ "
•s j/^1 o f*i r^i
S I o S S
« o ^ o d
So - -
S » sr
^"" *s] ON
S o ?? ^
0 ^ — 0
oo" _..- o 1 ^-•>
VO 2 _- C?,
S o S e>
II s


i § o S §
00 °











1 s 1
i i i" i 1
rf 13 ~ JJ. §
§ ,0 8 6 . 2
u -^ e i -f s <-•
JD D a 5 c 3 c
< {S 6 2-g. cSl
f
i I „

1 | li 1
e o E g £
S 'c r^ •! &
2. u. U 2 H

V
1
b
JS


S
O1
"S "
ts •=.
McDonald
1983a
de Klerk el

/— V
es
* /— \
""^ 00
<^>
S > "0
<=>. Q
B
«'

^
r*T



so

en
oo
Os
WN
.£
'5
3s
J^
c
£


(N


b
<«3
_o
a


•g
ll
II


X
s


i 1
SiJ ON
^— t-H
"73 •"•
2 *
« «
W5 ^_
•S S
ap 5
x 3

y"*«-,
^-S ^
wri ro
. ;
ON" ®
§ 2
d p
^ S

oo
d
1 *"
o
£
•

f^ ON
° §

I


l
i






New Orleans
plant
Asbestos, Quebec .
N
^
U.
O,
b
. I,
b1

i
!



2
'u
I/I
•M
Seilkoffani
1991


^o
VO
IA"
ts
S

/— V
•<*
rf
v^^


rn



"«
u
*c
^R
S2


<0


U.S. insulation
workers


c .2
1.1
11
£ <







13
S
McDonald
1983b


/— N
NO
NO
t~"
E

^— K
»rj
£
o
N_X


-











Pennsylvania
plant



Textiles




,


If)
oo
ON
"rt
«
a
£


^5
«•;

(2:
C!'
i!'
^* (
£?;:
•|.
u I
§i
j3 •
^3 -
I1'
"J3 %.

Jl

y
bining 90% confiden
iddeir :" ""•
E u
8'S
>>. P
.fftta
*o _s
^ s
1 ^
^*
£ fS
S §
£ E
ti
S 3
8«
§!?





















11
g 
&l
S 8
^*
'^>
E^ -a
CNj
!•>.'

-------
7.3.L1
Time Dependence
The U.S. EPA mesothelioma model (Equation 7-6 or 7-7) was next evaluated to determine
whether it adequately describes the time-dependence of mesothelioma mortality following
cessation of exposure in the Wittenoom and Quebec cohorts. The small number of
mesotheliomas observed among the South Carolina cohort precluded a meaningful evaluation of
this issue for that cohort.   .

For times since first exposure longer than 10 years past the end of exposure the mesothelioma
model (Equation 7-6) can be rewritten as

             IM = 3 * KM  * f * d * (t -10)2 * {1 -  3 *  [d / (t -10)] + [d / (t -10)]2}    (Eq. 7-8)

From this expression we see that, when  time since first exposure lagged 10 years, (t-10), is large
compared to duration of exposure, (d), the model predicts that the mesothelioma mortality rate is
approximately proportional to the product of cumulative exposure (the exposure level, f, (f/ml)
times the duration of exposure, (d) and the square of time since first exposure lagged 10 years.
Thus, the model predicts that the mesothelioma mortality will increase indefinitely with age as
the square of time since first exposure lagged 10 years. The availability of raw data from the
Wittenoom and Quebec cohorts provides an opportunity to evaluate this assumption.

Table 7-10 shows the fit of the U.S. EPA mesothelioma model to Wittenoom data characterized
by time since last exposure, based on the KM estimated from the exact analysis. There is no
indication from this table that the mesothelioma mortality rate declines after the cessation of
exposure, or that the model over-predicts the mesothelioma risk at long times after the cessation
of exposure. In fact, the model under-predicts the number of deaths at the longest times, as it
predicts 76.8 deaths after 30 years from the end of exposure, whereas 96 were observed.

Table 7-11 shows the observed number of mesothelioma deaths in the three separate locations at
Quebec, categorized by time since last exposure, and compared with the predicted numbers
obtained using the KM values obtained using the exact  fitting method. From all three locations
combined the model predicts 10.6 deaths from mesothelioma after more than 30 years following
the cessation of exposure, whereas 10 were observed.  Although the small numbers of
mesotheliomas make it difficult to draw definite conclusions about the adequacy of the model,
there is little evidence that the model under or over-predicts the numbers of mesotheliomas at
long periods after the end of exposure.
7.3.L2
Exposure Dependence
The lack of fit of the mesothelioma model (Equations 7-7 and 7-8) to the Wittenoom data
categorized by the value of the integral in Equation 7-8 (Table 7-8) suggests that the Wittenoom
data may not be consistent with the assumption that the mortality rate is linear in the intensity of
exposure (f/ml). Specifically, the mesothelioma model predicts that for fixed time since first
exposure and duration of exposure the mesothelioma mortality rate varies linearly with f, the
asbestos air concentration. To test this prediction, expected numbers of mesothelioma deaths
were calculated for each of four categories of asbestos air concentrations while controlling for
                                          7.28

-------
both time since first exposure and duration of exposure." Person-years in the first 10 years
following the beginning of exposure were ignored since no mesothelioma deaths occurred in this
time interval, as predicted by the model. The relatively few workers who were employed for
longer than 5 years were also excluded from this analysis (exposure durations were generally
quite short in this cohort, with the average employment duration being <1  year), which means
that all follow-up in this analysis occurred after exposure had ended. Results of this analysis are
shown in Table 7-12.                                         !
     Table 7-10. Fit of EPA Mesothelioma Model to Observed Mesothelioma Mortality
   Among Wittenoom, Australia Miners (Deklerk 2001) Categorized by Years Since Last
                                           Exposure         ;
Years Since
Range
0-1
1-5
5-10
10-20
20-30
30-40
40+
30+
Total
Last Exposure
Average
0.3
3
7.5
14.9
24.7
34
43.6


Average
Value
of Integral
35
72
327
4025
18058
39802
57952


Observed
Deaths
0
0
0
10
58;
79
17 •
96
164.
Predicted
Deaths
by Model
0
0.1
0.6
14.4
52.9
61.3
!5.5
76.8
144.9
Observed/
Predicted
0
0
• o
', °-7
• 1.1
1.3
; 1.1
1.3
;•
                                                        Goodness of Fit P-value
0.21
       4In this analysis the Wittenoom data were categorized by time since first exposure {10-20, 20^-30, 30-40,
and 40+ years), exposure intensity (0-15, 15-30, 30-60, and 60+ f/ml), and duration of exposure (0-1 and 1+ years).
This categorization was facilitated by the facts that in the subcohort being analyzed, exposure had ended prior to the
beginning of follow-up, and in the Wittenoom data base exposure intensity was assumed to be constant throughout
employment. Within each of the eight [time since first exposure] x [duration of exposure] categories tHe total
number of mesothelioma deaths were allocated to the various exposure intensity sub-categories in proportion to the
product of the average exposure intensity times the person-years of observation in each sub-category, thereby
producing the expected number of deaths in each sub-category under the assumption that the response was linear in
exposure intensity within each [time since first exposure] x [duration of exposure] category, as predicted by the
mesothelioma model. These expected deaths and the corresponding numbers of observed deaths were summed
across time-since-first-exposure and duration categories to yield observed and expected deaths categorized only by
exposure intensity                                                                     •'
                                                                                               V
                                              7.29                                  :

-------
 Table 7-11. Fit of EPA Mesothelioma Model to Observed Mesothelioma Mortality
Among Quebec Miners (Liddell 2001) in Each of Three Mining Areas, Categorized by
                         Years Since Last Exposure
Years Average
Since Last Value
Exposure of Integral
Person- Years
Observed
Deaths
Predicted
Deaths
by Model
KM
Location 1
0-10
10-20
20-30
30-40
40-50
50+
30+
Total
94,267.5
119,190
119,971
181,470
278,1 18
339,160


77,008
27,225
22,831 .,
18,321
11,846
6,673


3
2
0
1
2
0
3
8
3.1
1.4
1-2
1.4
1.4
1.0
3.7 .
9.3
1.34x10-'°







Location 2
0-10
10-20
20-30
30-40
40-50
50+
30+
Total
56,377.5
56,445.7
58,307.7
78,685.8
88,541.2
59,590.4


15,065
4,835
3,991
3,129
1,844
795


0
2
0
2
0
1
3
5
2.5
0.8
•0.7
0.7
0.5
0.1
1.4
5.4
9.55x1 0-'°







Locations 3 and 4
0-10
10-20
20-30
30-40
40-50
50+
30+
Total
194,913
262,898
179,052
251,766
348,264
298,743


95,299
23,885
18,777
14,311
8,800
4,451


13
5
0
0
i
3
4
22
12.7
4.3
2.3
2.5
2.1
0.9
5.5
24.8
2.18x10-'°







                                    7.30

-------
  Table 7.12.  Comparison of Wittenoom, Australia (DeKlerk 2001) Mesothelioma Deaths
     to Predicted Deaths Assuming Risk Varies Linearly with Exposure Intensity After
          Controlling for Years Since First Exposure and Duration of Exposure
            Intensity (ffml)
      Range
   Average
                                  Mesothelioma Deaths
Person-Years
Observed
                                        Goodness of Fit P-value
Predicted
0-15
15-30

30-60

>60
Total
9.7
17.0

50.3

100.2

50,736
23,881

18,166

13,353

I 32
51

27
1
40
• 150
;: 19.5
22.4
i
, 41.9
i
,. 66.3
' 150.0
Table 7-12 shows that the assumption that the mesothelioma risk varies linearly with exposure
intensity leads to a 2-fold under-prediction of risks for exposure intensities below 60 f/ml (83
deaths compared to only 41.8 predicted) and a corresponding over-prediction for exposure
intensities above 60 f/ml (only 67 deaths compared to 108.2 predicted). Thus, instead of risk
varying linearly with exposure intensity, Table 7-12 indicates that the exposure response is
supra-linear, with lower fiber intensities being more potent per f/ml.      .         '
7.3.1.3
Discussion of Adequacy of Mesothelioma Model
The mesothelioma model (Equations 7-6 and 7-7) provides adequate fits to each of the three data
sets evaluated (Wittenoom, Quebec and South Carolina) when the data are categorized by time
since first exposure. The value of KM estimated from the cohort of crocidolite miners
(Wittenoom) was largest and was about 60-fold larger than the KM estimated from the South
Carolina chrysotile textile workers (based only on confirmed mesotheliomas in South Carolina)
and more than 100-fold larger than the estimates obtained from the Quebec chrysotile miners
who did not work in the factory that utilized crocidolite. This is consistent with numerous
indications from the literature that crocidolite is more potent than chrysotile in causing
mesothelioma. The relative magnitudes of the KM for the Quebec data estimated from three
locations track with the relative amounts of amphibole exposure estimated for these (locations,
which is also consistent with the hypothesis that the mesothelioma risk is greater from amphibole
exposure than from chrysolite exposure.  Despite the very few mesotheliomas in the South
Carolina cohort, the KM estimated from these data is larger than those from Quebec,' although the
discrepancy is not as large as estimated for lung cancer.

The Wittenoom and Quebec data were evaluated further to see if they were consistent with the
prediction of the mesothelioma model that risk continues to increase indefinitely after exposure
has ceased. By comparing the observed number of mesothelioma deaths to the number predicted
by the mesothelioma model at various times since the cessation of exposure, no evidence was
found that mesothelioma risk dropped off below that predicted by the model, at least up to 40 to
50 years after the cessation of exposure.  To the contrary, in Wittenoom there was some evidence
that the model under-predicted at the longest times since the end of exposure, as past 30 years
                                          7.31

-------
from the cessation of exposure there were 96 observed mesothelioma deaths compared to only
76.8 predicted by the model.

Chrysotile is much more soluble than crocidolite and consequently a chrysotile fiber exhibits a
much shorter residence time in the body than a comparable-sized crocidolite fiber.  Based on
in vitro studies a chrysotile fiber with a diameter of 1 nm will dissolve in body fluid in
approximately 1  year whereas a 1-um crocidolite fiber will take 60 years to dissolve (see Section
6.2.4). It is noteworthy that despite the short residence time of chrysotile fibers, mesotheliomas
deaths have occurred in the Quebec cohort more than 50 years following the cessation of
exposure.  This suggests that either these mesotheliomas are the result of amphibole
contamination of the ore, or else long residence times for inhaled fibers are not necessary for the
production of mesotheliomas.

The mesothelioma model predicts that risk is proportional to the intensity of exposure (Equation
7-6) and, at long times past the end of exposure, to cumulative exposure (Equation  7-8).  Two
analyses of the Wittenoom mesothelioma data suggest that the assumption of a linear exposure-
response may not be valid. First, whereas the model is linear in the value of the integral  in
Equation 7-7, a very poor fit was obtained when the data were categorized according to the value
of the integral (Table 7-8). Second, an analysis that categorized the data by intensity of
exposure, while controlling  for both duration of exposure and time since last exposure, also
provided a poor fit (Table 7-12).  Both of these  analyses exhibit a supra-linear exposure-response
in which less intense lower exposures are more  potent  per f/ml than more intense ones.

In light of these findings, it is interesting that supra-linearity in the exposure-response
relationship for lung cancer has also been suggested by fits of data in several studies (see Section
7.2.1.1). Although the effects for lung cancer are difficult to separate from the confounding
effects from smoking, common suggestions of supra linearity for both disease endpoints
certainly indicate a need to evaluate the nature of the exposure-response models for asbestos in
greater detail. This is consistent with the recommendations of the expert panel (Appendix B)
that evaluation of a broader range of exposure-response models for mesothelioma is appropriate.

This analysis of the Wittenoom data appear to be one of very few that provide information on the
shape of the exposure-response relationship for mesothelioma.  It is possible that systematic
errors in the exposure data for Wittenoom could have resulted in an apparent supra-linear
exposure response.  Like most asbestos-exposed cohorts, the estimated exposures for Wittenoom
are uncertain. If higher estimated intensities are overestimates, a linear exposure response would
appear to be supra-linear, and a linear fit to such data would underestimate the true KM.

In addition, errors in exposure measurement even if unbiased, can tend to make a linear dose-
response appear  supra-linear (Crump 2003). Even  if the supra-linear exposure response in the
Wittenoom data  is real, the exposure-response at lower doses is likely to be linear, but with a
larger KM value than was obtained in the fit to the complete data set. If this is the case, our
analysis (Table 7-12) suggests that the KM for the lower exposures is about a factor of two larger
than the value estimated from the complete cohort. A factor of 2 is not extremely large
compared to the  other sources of uncertainty in  the analysis.  Provisionally, the value of KM
estimated  from the complete Wittnoom cohort will be applied in our analysis. In revisions to this
                                          7.32

-------
document, it will be important to attempt to evaluate the exposure-response for mesothelioma
using data from additional other cohorts.                                        ;
                                                                             i'
Given the importance of these two issues: (1) the relative potencies of chrysotile andjthe
amphiboles and (2) the adequacy of U.S. EPA models for predicting the time and exposure
dependence of disease, limited analysis of raw data from a small number of additional cohorts is
warranted. Recommendations for further research along these lines are discussed in the
conclusions to this Chapter (Section 7.6) and parallel the recommendations of the expert panel
(Appendix B).

7.3.2 Estimating KM Values from Published Epidemiology Studies

At the time that the Health Effects Update was published (U.S.: EPA 1986), four studies were
found to provide suitable quantitative data for estimating a value for KM and six additional
studies provided corroborative support for the mesothelioma model applied. Currently, there are
14 published studies with adequate data for deriving  an estimate of KM (including updates to all
four of the quantitative studies evaluated in 1986).                               ..
                                                                             i
The U.S. EPA mesothelioma model (Equations 7-6 and 7-7) was applied to each of these data
sets to obtain study-specific estimates for the mesothelioma dose-response coefficient, KM.  The
resulting set of KM values are presented in Table 7-9. The format for this table  is identical to that
described in Section 7.2.2 for Table 7-6. As with the KL values in Table 7-6, the KM values for
all studies presented in Table 7-9 (including those studies that have not been updated since their
inclusion in the 1986 Health Effects Update) were re-derived using the modified procedures
described in Appendix A. Uncertainty intervals in Table 7-9 for each estimated KM were derived
using the method described in Appendix A.                 •                    '
                                                                             t
As Table 7-9 indicates, the KM values derived in this study and the corresponding values derived
in the original  1986 Health Effects Update are in close agreement, none vary by more than a
factor of 1.5. Among the KM values derived in the current study, the lowest and highest of the
best-estimate values differ by a factor of approximately 1,400 (excluding the one negative study
of Connecticut friction product manufacturers) and many of the pair-wise sets of uncertainty
intervals do not overlap.  For example, none of the uncertainty intervals for the  KM values
derived for any of the environments involving exposure to chrysotile overlap the uncertainty
intervals associated with the KM values derived for either crocidolite mining or  asbestos-cement
manufacture using mixed fibers at the Ontario plant (Finkelstein 1984). Furthermore^ neither of
the uncertainty intervals for the KM values derived for chrysotile mines in Quebec (Asbestos and
Thetford) overlap the intervals around KM values for  any of the amphibole environments or any
of the mixed environments, except the Quebec factory that is associated with the Asbestos mine
(Liddell et  al. 1997).                                       [

The KM values and the associated uncertainty bounds derived in the current study are plotted in
Figure 7-2. Each exposure environment is plotted along the X-axis of the figure and is labeled
with a 4-digit code that indicates fiber type (chrysotile, mixed, crocidolite, or tremolite), industry
(mining,  friction products, asbestos-cement pipe, textiles, insulation manufacturing, or insulation
application); and a 2-digit numeric code indicating the study from  which the data were derived.
                                          7.33

-------
The key for Figure 7-1 also applies to this figure.  In Figure 7-2, the chrysotile studies are
grouped on the left, amphibole studies are grouped on the right, and mixed studies are in the
middle. As in Figure 7-1, data from the Dement et al. (1994) study are used to represent the
South Carolina textile cohort and data from the Amandus and Wheeler (1987) are used to
represent the Libby mine cohort in Figure 7-2. Also, the estimated KM values for the Quebec
miners (Liddell raw data) from Asbestos and Thetford Mines, respectively, are averaged in this
figure.
                                         Figure 7-2:
          Rot of Estimated K^, Values and Associated Uncertainty Intervals by Study Environment

              1000^
               100.
          00
          o
                0.1,
               0.01 r
               1E-3
                         5    u. 0. I-
                         O    000
00 O)
CL Q.
                                                           10 (o Is-    oo
                                                              55    2
                                           7.34

-------
Figure 7-2 indicates that, within'chrysotile studies alone, lowest and highest KM values vary by
approximately a factor of 15 (excluding the negative friction products study), which is minimal
variation relative to the spread across the values from all of the studies in the table. Also, the
corresponding uncertainty intervals have considerable overlap.                    l:
                                                                             s
The KM values for the two "pure" amphibole studies (crocidolite mining and amosite,
manufacturing) agree to within a factor of 2 and, with the exception of the value from the
Ontario asbestos-cement plant (MP8 in Figure 7-2 and the ninth study listed in Table 7-9) and
the factory plant associated with the Asbestos, Quebec mines (MF14 in Figure 7-2 arid the 11th
study listed in Table 7-9), the mixed exposure KM values lie between the high values;for the
"pure" amphibole exposures and the lower values reported for all of the chrysotile sites.
Although the K.M value for the asbestos-cement plant (MP8) appears high, its corresponding
uncertainty interval overlaps those of the other "pure" amphibole studies as well as those of a
number of the other mixed exposures. The KM value reported for the factory associated with the
Asbestos, Quebec mine (MF14) is the lowest of those reported for mixed exposures, put its
uncertainty interval overlaps those for several of the other mixed exposures, as well as all of
those involving only chrysotile exposures.                  .                    \

As with KL values, the asbestos-cement pipe industry shows the greatest variation in'KM values
across all asbestos types, with a range of more than a factor of 90. Moreover, the uncertainty
intervals for the largest and smallest values in this industry do not overlap. Within the textile
industry, the KM value for the South Carolina chrysotile plant is only an eighth of the values
reported for the two textile plants using mixed fibers, although there is considerable overlap in
their uncertainty intervals. The potential sources of variation across all of the KM  values are
likely attributable to the same sources of variation identified  for the KL values and previously
described in detail (Section 7.22).                          '                   ''.

7.4   EVALUATION OF ASBESTOS EXPOSURE INDICES

As indicated in Chapter 3, for exposure-response coefficients derived from one environment to
be applicable in a different environment requires that both of the following two conditions be
satisfied:                                                ;                   !'
                                                        I

      •      that asbestos be measured in both environments in an identical manner; and

      •      that such measurements reflect (or at least remain proportional to) the
             characteristics of asbestos exposure that determine biological activity:

When these two conditions are met, it is possible to define an "exposure index" that accurately
reflects biological activity, and consequently an exposure-response coefficient based upon such
an exposure index derived in one environment can confidently be applied in a different
environment. Such an index can be defined as a weighted sum of concentrations of categories of
structures of different asbestos types and sizes, where the weights reflect the relative
carcinogenic potencies of the different type and size categories.  For example, as described in
Section 6.4.3, Herman et al. (1995) derived an optimal exposure index from an analysis of rat
inhalation studies involving exposures to different types of asbestos and fibrous structures  of
differing dimensions. The optimum index (defined in Equation 6-11) consists of a weighted  sum

                                          7.35

-------
of the air concentrations of structures between 5 and 40 um in length and >40 urn in length (all
thinner than 0.4 urn).

There is considerable evidence that the manner in which asbestos is quantified in the available
epidemiology studies (i.e., PCM) may not adequately reflect the characteristics that relate to
biological activity (Sections 6.4 and 6.5).  Therefore, the second of the above two criteria may
not be satisfied when exposure-response coefficients (i.e., KL and KM values) derived from these
studies are used to predict risk in new environments.  We therefore investigated the possibility of
adjusting the KL and KM values so as to apply to a measure of exposure that better reflects
biological activity. Such an adjustment requires both (1) data from each studied environment on
fiber size and asbestos type needed to adjust the corresponding KL and KM and (2) evidence
regarding what measure of exposure better reflects biological activity.

7.4.1  Fiber Type and Size Distribution Data Available for Deriving Exposure Indices

Since the range of possible adjustments to the KL and KM values is constrained by the data on
fiber size and type available from each studied environment, we first consider the characteristics
of the data currently available for making such adjustments. The data considered to be pertinent
consisted of TEM analyses of samples conducted in the same environment in which an
epidetniological study was conducted or from an environment involving a similar operation (e.g.,
mining, textile manufacture, etc.). Table 7-13  lists available fiber size distributions obtained
from a search of the literature and categorized  by fiber type and type of operation.  The
epidemiological studies from which KL or KM have been calculated are categorized accordingly.

Assuming that the available TEM size distributions are representative of dust characteristics for
the exposure settings (industries) studied, these were paired with corresponding epidemiology
studies. The TEM size distributions were then used to convert the exposure measurements used
in the epidemiology studies to the new exposure index that potentially  better reflects biological
activity.  Studies were paired as indicated in Table 7-13.

Only a subset of the TEM size distributions listed in Table 7-13 were actually employed in the
effort to normalize KL and KM values. To minimize variability resulting from differences in
TEM analysis methodology employed by different authors,  it was decided to employ
distributions from common studies conducted by common groups of researchers, to the extent
that this could be accomplished without reducing the number of "size-distribution-
epidemiological study" pairs available for inclusion in the analysis. Also, studies containing the
best documented procedures were favored. Ultimately, with one exception the size distributions
selected for use came from only two studies, which were reported in three publications: Dement
and Harris (1979), Gibbs and Hwang (1980), and Hwang and Gibbs (1981). In the case of the
one exception, size distributions for Libby (tremolite asbestos in vermiculite) were derived from
unpublished TEM data recently acquired directly from the site.
                                           7.36

-------
    Table 7-13. Correlation Between Published Quantitative Epidemiology Studies and
                         Available Tern Fiber Size Distributions
Fiber Type  Exposure Setting
                      Distribution Reference
                             Epidemiology Reference
Chrysotile  , Textiles
Chrysotile
and
Crocidolite
Crocidolite

Amosite
            Friction Products
            Mining and Milling
            Asbestos Cement
            Manufacturing
Asbestos Cement
Manufacturing
                      Dement and Harris 1979
                      Cherrieetal. 1979

                      Dement and Harris 1979
                      Marconi etal. 1984
                      Winer and Cossett 1979
                      Roberts and Zumwalde 1982
                      Rood and Scott 1989
                      Gibbs and Hwang 1980
                      Winer and Cossett 1979    '
Dement and Harris 1979

Snyderetal. 1987
Winer and Cossett 1979
Hwang and Gibbs 1981
Mining and Milling      Gibbs and Hwang 1980
                      Hwang and Gibbs 1981
Insulation Manufacturing Dement and Harris 1979
            Insulation Application
            Insulation Clearance

Tremolite    Vermiculite Mining
                      Snyderetal. 1987
                      Cherrieetal. 1979
                      U.S. EPA, unpublished
                             Dement etal. 1994, 1983b
                             McDonald et al. 1983a,b
                             Peto 1980a; Petoetal. 1985
                             Berry and Newhouse 1983
                             McDonald et al. 1984
Liddelletai. 1997
McDonald et al. 1980b
Nicholson et al. 1979
Piolatto et al. 1990
Hughs etal. 1987;'
Finkelstein 1984 ,
               i
               \
Finkelstein 1983 '
Hughs etal. 1987
Weilletal. 1979
Weill etal. 1994
Albin etal. 1990
Armstrong et al. 1988
de Klerk, unpublished data
Levin etal. 1998 ;
Seidmanetal. 1986
Seidman 1984
Selikoffetal. 1979
                             McDonald et al. 1986
                             Amandus and Wheeler 1987
                                          7.37

-------
Table 7-14 presents the resulting bivariate fiber size distributions derived from the published
TEM data that are paired with representative KL and KM values from the corresponding
epidemiology studies. This table shows 17 KL values and 11 KM values matched with a fiber size
distribution from the literature. In this table some length and width categories from some  •
published distributions have been combined, so that, for the most part, only those categories
available for all of the epidemiological studies are presented. The column labeled "PCME"
("PCM-equivalent") provides the relative concentrations of asbestos fibers that would have been
identified by PCM (*5 nm in length and ^0.2 urn in width).
The fiber size distributions in Table 7-14 are not all of equal relevance to the respective
epidemiological studies to which they were paired. As indicated in Table 7-15, some of the
distributions are based on data collected at the same facility, others are based on data collected at
a similar facility, still others are based on a combination of data from similar facilities, etc. The
uncertainty factors listed in Table 7-15 were developed to quantify the relevance of each fiber
size distribution to its paired epidemiological study, where larger factors indicate a less certain
relevance. How these factors were used is explained below. It should also be kept in mind that,
whereas these fiber distributions were based on air samples collected over a fairly narrow time
range, they are used to represent the fiber size distributions throughout the exposure period,
which in most of the epidemiological studies covers many years.

As indicated in Tables 7-14 and 7-15, for the two environments for which multiple studies were
available (i.e., the South Carolina textile plant and the Libby, Montana vermiculite mine), a
single study was selected to represent each environment. For the South Carolina textile plant the
Dement et al. (1994) study was selected and for Libby, the Amandus and Wheeler (1987) study
was selected.

7.4.2 Modification of Existing KL and KM to Conform to a New Exposure Index

The fiber size distribution data in Table 7-14 are used to transform the existing KL and KM values
(which are defined in terms of PCM measurements) so they conform to a different exposure
index based on TEM. To see how this is accomplished, consider a KL value pertaining to a
specific environment, let CPCM be an air concentration from that environment measured by PCM,
let Cns.w be the concentration in the same air measured by a new exposure index using TEM (e.g.,
perhaps defined as a weighted sum of TEM concentrations in various length, width and asbestos
type categories), and let KL* be the adjusted exposure-response coefficient corresponding to the
new exposure index. It is clear that for the U.S. EPA lung cancer model (Equation 7-1 or 7-2) to
estimate the same risk from the given air concentration using either exposure index, it is
necessary that

                               (K. ) (C    ) — (K *) (C   )                        (Eq 7-9)

Using CpcME (the air concentration of PCM-equivalent fibers -  fibers measured by TEM that
would be identified  by PCM) as a replacement for CPCM, we get

                                                '
(Eq.7-10)
                                          7.38

-------
 a

 E

I
 _e


 x

 M
 S

 8
 v.
 S


 I

 S
 S
 "S
TS
 a
A

H
       !


            I
            o

            I
          fs

            I
            e
            S
          rfl
                                    x   ^
                                                        2    S
                       i§  gs
                       §2  §2
                                             o - a i
                   ?   8
                       3
                       3
                  I   I
                            s   S
                           a
S
.s   I   s
3   S   *
I   1   s
I   -g   s
•<   •>   o


!   i   I
d,   d,   S

s   "   £
sis
o   o   ^
                      S.    8
                      o    d

                      a    a
TV    Ml
                      9.   S
                                **i   —
                           o   o

                           2   P
                           p   a
         o
         o"
S   I
5   S
                                        i
                                        i
                                r-   c*1    r»
                                2   2    R
                                    S    3
                                    g    s
                                    I    I
                                    d    d
                                    !   I
                           §    §
                                             §   §
                                             d   d
                                             f*1    NO

                                         S   3    P
                           8
                           d
                                             C%    ^|

                                             P:    S
                                             m    o
                                             1   1
                  £
                  3
                                             O   m   n
                                                      s
                                                      d

                                                      >o
                                                      P
                                                          ^ 3 f § -
                                                                 i
                                                                 d
                                                    S.sf 00 AI 0*
                                                    s ^ s *



                                                I    m   s
                                                                         g
                                                                         d
                                                                                             t '  oo   —
                                                                     §•
                                                                              I   I
                                                          §OK

                                                          «H
                                                                                            fi   I
        I!
                                                                                              ' V °
                                                                                              ;;' s
                                                        £•  I
                                                        r!  3
:,' s
                                                                                             11
                                                        g
                                                        3
                                                        o


                                                        R
                                                        S
                                                        o
                                                                                             s   s
                                                                                             d  ' d
                                                                                            S   8
                                                                                            e  • d
                                                                                            S   g
                                                                                            i  •
                                                                               3
                                                                               d
                                                                      o    d   d
                                                                      ?*•  '  v   v
                                                                               §
                                                                                                a   a
                                                                                                                  0)
                                                                                                                  CO

-------
I
                                 «8
.1
w
 w



I


I
 09
 s

f
.1
 Efl
^
*
 a

I
 8
a

i
 A

I
w

in
                        v

                       1
                       H
1
B
_e

A

1
Q.
X
                               &

                               -S  U
                               "§"§
 e


 i

•o
2
                                             «S
                                             -a
                                                           oo
                                                           ON
                                                           tS
                                                           •Q
                                                                  r~
                                                                  oo
                                                                          T 0
                         &     •&
                                                                   ON

                                                                   C
                                                                                          •§
                                                                             r-

                                                                             ON
                                                                             "S


                                                    8
                                                    1
                                                    OH
                                    5      S
                                    «      -
                                    1
                                    2

                                        I
                                        Q

                                               I
                                               I
                                               CQ
                                                    !
                                                    PU
                                                                                 a.§      >f
                                                                                   1
                                                                                             *e
                                                                                          J 1
                                                                                           o5
                                                                                          Ǥi

 IS
 I

1
 a
•§ ja
a'i
                                             l
                                             1




                                                                                                                   >,


                                                         I
                                                         42
                                                                                CQ
                                                                   2
                                                                   1
                                                                    n
                                                                   o
                                                                   l

-------

g
•+*
QQ



t
gned
*P4
S

K

a
««
r^
       2
Explanation
   •a
   W
     a o
       c
       e
       •s
       S
       o



       I
            !i

            II
            os  S
            •« ^
            3 <8
,£>
(•<•>
00
o\
                    V

                    •o
                         15
                         a,
          i

          •a
          «
          .1

          1
                    to
                    00
                    os


                              "8
          _0

          a,
          ^
          (L>
          a
                              I
                              •O
                              .0    —


                   ao
                   o\


                                         .5 <*5
                    CM
             rN  •— <

             S  2
                                        S


             n>
          w  -
                           i

                «
                            1
                                            -