Document qdpD2xgLLmLoBZg2n5erw50VR

Analysis. Vol IS. No. 2. 1995 The Sizes, Shapes, and Mineralogy of Asbestos Structures that Induce Lung Tumors or Mesothelioma in AF/HAN Rats Following Inhalation1 D. Wayne Berman,2 Kenny S. Crump,3 Eric J. Ghatfield,4 John M.G. Davis,5 and Alan D. Jones5 i i i i Data from inhalation studies in which AF/HAN rats were exposed to nine different types of asbestos dusts (in 13 separate experiments) are employed in a statistical analysis to determine if a measure of asbestos exposure (expressed as concentrations of structures with defined sizes, shapes and min eralogy) can be identified that satisfactorily predicts the observed lung tumor or mesothelioma inci dence in the experiments. Due to limitations in the characterization of asbestos structures in the original studies, new exposure measures were developed from samples of the original dusts that were re-generated and analyzed by transmission electron microscopy using a direct transfer technique. This analysis provided detailed information on the mineralogy (i.e., chrysolite, amosite, crocidolite or tremolite), type (i.e., fiber, bundle, cluster, or matrix), size (length and width) and complexity (i.e., number of identifiable components of a cluster or matrix) of each individual structure. No univariate measure of exposure was found to provide an adequate description of the lung tumor responses observed.among the inhalation studies, although the measure most highly correlated with tumor incidence is the concentration of structures >20 pm in length. Multivariate measures of ex posure were identified that do adequately describe the lung tumor responses. Structures contributing to lung tumor risk appear to be long G>5 pm) thin (0.4 pm) fibers and bundles, with a possible . contribution by long and very thick >5 pm) complex clusters and matrices. Potency appears to increase with increasing length, with structures longer than 40 pm being about 500 times more potent than structures between 5 and 40 pm in length. Structures <5 pm in length, do not appear to make any contribution to lung tumor risk. This analysis did not find a difference in the potency of chrysolite and amphibole toward the induction of lung tumors. However, mineralogy appears to be important in the induction of mesothelioma with chrysotile being less potent than amphibole.' KEY WORDS: Asbestos; tumorigenicity. lung rumors: mesothelioma; inhalation; dose/response; AF/HAN rats. INTRODUCTION human exposure settings; more than 50 positive epide miology studies have now been published.1 li However, Inhalation of asbestos dust has been clearly linked to lung cancer and mesothelioma in a broad range of a quantitative dose-response relationship that applies across exposure environments has not been established for either disease; potency estimates derived from dif- ' This research was completed under contract to the US. Environ mental Protection Agency. Such support, however, does not signify ? 1CF Kaiser Engineers, Oakland, California. that the contents of this paper necessarily reflect the views and poi- 3ICF Kaiser Engineers, Ruston, Louisiana. jdlpcs of the US. Environmental Protection Agency, nor does the J Chatfield Technical Consulting Limited, Mississauga, Ontario, Can J^^Hiention of trade or commercial products constitute endorsement or ada., .- recommendation for use. \ 3 Institute of Occupational Medicine, Edinburgh, United Kingdom.' 181 \ sLU EXHIBIT 0272-4jJ2/95/0*0U-0l8l$07.5(yi 0 )995 Socwiy for Risk Analysis i\ t\1 V} HWBUI0010287 ferent epidemiological studies differ by up to 600-fold for lung cancer (Table 6-10 of an HEI-AR report*'*) and potency estimates reported for mesothelioma induction are also quite variable. Although there are many features of epidemiolog ical studies that may contribute to the variability ob served in dose-response relationships, a major compo nent may be the inadequate characterization of the asbestos dusts to which individuals are exposed in the epidemiological studies. In most epidemiological stud ies, asbestos dust concentrations were measured (if at all) by phase contrast microscopy (PCM), or by midget impinger.*2* The impingers produce a count of total dust particles that are frequently converted to PCM-equivaIent counts when comparison data are available, al though the correlations are generally poor.1** PCM measurements may not be useful for distinguishing among exposure environments that differ in their poten tial to induce disease because (1) PCM is not capable of. distinguishing asbestos from non-asbestos structures15* and (2) existing animal studies suggest that asbestos structures outside the range of sizes visible by PCM may contribute to risk.*'* Because of the limitations in the characterization of asbestos exposures in epidemiological studies, the best information regarding the effects of size and mineralogy on the relative potency of asbestos structures has come from animal studies. Injection and implantation experi ments (in which asbestos or other fibrous material is ei ther injected or implanted into the pleura or peritonea .of rats) generally indicate that long, thin fibers exhibit the greatest tendency to induce mesothelioma.1*-51* However, injection and implantation experiments bypass the proc esses associated with inhalation, retention in the lungs, and transport from the lungs, which may be important in modulating the effects of airborne exposure. Thus, results obtained from animal inhalation studies are likeiy to be more relevant for evaluating human risk than in jection and implantation experiments. A number of inhalation studies have been con ducted in which animals (generally rats) have been ex posed to varying concentrations of asbestos dusts of various types and the incidence of tumors observed in the animals recorded.*7-'524-28-52-54* In addition to verifying that different types of asbestos can cause lung cancer and mesothelioma when inhaled by animals, these stud ies generally indicate that longer fibers tend to be more carcinogenic than shorter ones. However, no measure of asbestos exposure that satisfactorily predicts tumor in cidence is identified in these studies. In this study, data on tumor incidence in AF/HAN rats from 13 inhalation experiments reported in a series of studies conducted by Davis et al.17-'5* are combined in a statistical analysis to determine if a measure of ex posure can be identified that satisfactorily predicts lung tumor or mesothelioma incidence. Because of limitations in the characterization of asbestos exposures in the orig inal Davis et' al. studies (use of scanning electron mi croscopy precluded observation of structures thinner than 0.2 pm; only SEM visible fibers and bundles were, included in the characterized size distributions while clusters and matrices may also contribute to tumorigenicity in a unique way; and lack of bivariate character ization of the size distributions precluded evaluation of the combined effects of structure length and width), ar chived samples'of the original stock, samples were used to regenerate asbestos dust clouds that were collected on filters and characterized in detail by transmission elec tron microscopy (TEMJ. TEM is capable of detecting and identifying even the thinnest asbestos structures. DATABASE AND METHODS The Animal Inhalation Database The series of animal studies by Davis et al.*7'"-'5* all employed a common protocol, utilized the same strain of rat, and were conducted in the same laboratory by the same group of investigators. In these studies, groups of 40 male AF/HAN rats, aged 3 months at the beginning of the experiment, were exposed by inhalation for 7 Hours per day, 5 days per week for 224 days over one year and then observed for a minimum of an addi tional'year. These studies involved UICC crocidolite, Korean tremolite, four types of efurysotile and three types of amosite (Table 1). Several of the samples were also studied at two doses or On multiple occasions. Details of the experimental procedures used in these studies (along with the' sources of the asbestos samples em ployed) are reported in the studies cited in Table 1. Due to the small number of mesotheliomas observed, the present evaluation was limited primarily to lung tumors. Benign and malignant lung tumors were pooled for this evaluation. Also, control groups from the various studies were combined into a single group. Regeneration and Analysis of Dusts from the Animal Studies To obtain more definitive characterization of the as bestos dusts used in the Davis et al. studies, dusts' were RrTT'rtSu HWBUI0010288 Asbestos Structures that Induce Tumors in Rats 183 Table 1. Summary Data for Animal inhalation Experiments Conducted by Davis and Coworkers' Fiber type Description Abbre viations Mass concen tration (mg/m-5) PCM 17ml Number of animals Number of benign pulmonary tumors Number of malignant -pulmonary tumors Total number of pulmonary tumors Mesothe liomas Reference Chrysolite Chrysotile Chrysotile Chrysotile Chrysotile Chrysotile Chrysotile Amosite'. Amosite . 'Amosite'' UICC-A . UICC-A Long Short UICC-A UICC-A (Discharged/* . WDC Yam" UICC Long ; ' Short uc uc LC SC UC DC wc 2 10 10 10 9.9 9.9 . 3.6 'UA - 10 . LA 10 SA . 10 390 1,950 . 5.510 1,170 2,560 2,670 679 . 42 40 40 40 36 39 - 41 - 550 2,060 70 43 40 42 6 .7 8. 1 6 4 5 2 3 .0;. 2 8 12 6 8 6 13 0 8 0 8 15 20 7 14 10 18 2 11 0 i 17) 0 CO 3 (13) 1 (13) 0 (12) I (12) 0 . (11) 0 (7) 3 (10) I ' cio) Crocidblite Ccoctdolite ...... :uicc UICC UR 4.9 UR 10 . 430 43 860 30 2 I 0 .2 0 "1 1 (7) . 0 (7) Tremolite Korean KT 10 1,600 39 2 15 18 2 (9) None None None None None Control Control Control Control Control C0 c0 c 0 c0 c .0 20 0 36 0 61 1 64 1 47 I 00 00 12 12 1, 2 0 (7) 0 (9) 0 (10) 0 (11) 0 (13) Exposure occurred for 7 hours a day. 5 days a week for I year. ft UJCC-A Chrysotile in this experiment was treated with mixed polarity air (produced with a source of beta radiation) following generation to reduce the surface charge on individual particles within the dust. e Chrysotile samples used for dust generation in this experiment were obtained from material treated by a commercial wet dispersion process. regenerated from samples archived from the original studies using the same equipment, procedures, and per sonnel as in the original studies. For each asbestos sam ple type, three or four sets of filters at three different, particle loadings were collected at regular time intervals over approximately one hour during which the rate of dust generation was kept constant. The set of optimallyloaded filters (one from each time interval) was 'then prepared and analyzed and the results from the individ ual filters were combined. A separate set of filters was also collected for PCM analysis at the same time as the filters collected for analysis by TEM. A detailed discus sion of the preparation and analysis of samples for this study and a characterization of the samples based on this analysts is currently in preparation (Berman et al.. Jour nal article in preparation). The regenerated dusts were analyzed by TEM using the counting criteria from the Interim Superfund Method for the Determination of Asbestos in Air,15'* These cri teria provide that, in addition to examination of a portion of each filter in which all asbestos structures present are identified, a separate examination of a different portion of each filter is performed at lower magnification, during which only structures > 5 |im in length are evaluated. The stopping rules in the Superfund method were mod ified for this study to assure that a minimum of 200 structures derived from total structure examination and, separately, 200 structures derived from examination for long structures (S 5 pm in length) would be evaluated for each sample. By using a separate examination for long structure's, the present study achieves a level of pre cision for the determination of long structures that could otherwise requirc the counting of thousands of. structures when all sizes are evaluated simultaneously.(1,) In the data base developed from the regenerated dusts, fibers, bundles, clusters and matrices (as defined in the literature*57>) are characterized separately along with measurements of the length and width of each such structure. Up to five fibers and bundles that are com: ponfents of clusters or matrices are also characterized for those complex structures (i.e., clusters and matrices) where individual fibers or bundles can be characterized.,5B) In the analyses reported herein, two methods of utilizing information on complex structures are applied. In one set of analyses, only the primary structures enter the analysis. In a second set of analyses, whenever com- Berman.et al. poneht" fibers or bundles in a cluster or matrix are characterize'd; these components are included as if they are independent structures and the parent cluster or matrix is riot included. . Estisnarion of.Exposure Concentrations " .,;Cbrieritrations of asbestos structures in each regen erated dust that exhibit specific characteristics .based on ^ze or'type'werecalculated by multiplying the, number. <q.CStnictureS liri'a particular sample displaying the char\^etetasti'c(s^pr-ihter'est by the- total area of the filters %po.sed.dtmri^ air'-sample collection and. dividing by the. product of the area of the portion of the filters examined during the analysis and the volume of air passed through _ the filters' during air sample collection. The concentra. tions of structures in the original dusts to which the an imals were exposed were estimated by multiplying the corresponding concentration in the regenerated dust by the ratio of the concentration measured by PCM in the original dust to the PCM concentration in the regener ated dust. Statistical Methods Once the. re-generated dusts were characterized, a statistical analysis was performed to test for relationships between the size, shape, and mineral type of an asbestos structure and its relative potency for inducing lung tu mors. In the statistical methods employed, the probabil ity of a lung tumor (benign or malignant) response in an animal is assumed to be of the form: P = l-exp{--a--P*WQ (I) where a specifies the response probability in unexposed animals (i.e., the probability of-a response in unexposed animals is 2 - exp[-aj), 3 represents the absolute potency of a dust for producing a tumor response, and WC is a weighted sum of the concentrations of structures in dif ferent structure categories. A structure category is de fined by restrictions on structure sizes {e.g., a defined range of lengths, widths, and/or aspect ratios) and struc ture types (e.g., fibers, bundles, clusters, and/or matri ces). The weights in the weighted sum are estimates of the relative potencies of structures in the different struc ture categories. Thus, WC = ql*xl+q2*x2+....+qJ[*xl = Eq^ (2) here Xj is the concentration of airborne asbestos struc tures in the jth structure category and qj represents the potency of structures in the jth structure category, relative to the potency of structures in other categories. The qjS are constrained to be non-negative (q^ > 0 for j = l, .... k) and to sum to one (q, + ... + qj = 1). The latter constraint makes each q; a measure of relative potency rather than.absoiute potency. The non-negativity constraint implies that no category , of structures is capable'of reducing the probability of a'tumor response. In a few of the analyses the model is expanded to permit the probability of response to depend upon the chemical composition (mineralogy) of the structures : (e^g.i chrysotile bramphibole). This expanded.version of the-model is of the form: . P; = .l.-exp(-oc-Pi*WQ (3) where P; is the probability of tumor response in animals exposed exclusively to material of type i, and {3; a meas ure of the absolute potency of this material. Thus, in this more general model, a different absolute potency is as sumed for materials of differing mineralogy but the rel ative potencies of structures in different structure categories are assumed to be independent of mineralogy. In still other analyses, the tumorigenic potential of a structure is assumed to be embodied in a single quan titative measure (e.g., the surface area of the structure). In these analyses the probability of a tumor response in an animal is assumed to be of the form: . P = l-exp(-a-3*SM) (4) where SM is the sum of the quantitative measures over the structures contained in a specified volume of air. For example, if the quantitative measure is surface area, then SM is the total surface area of structures per milliliter of air. The parameters of these models (a, p and q}) are estimated using maximum likelihood methods and like lihood ratio tests are used for testing hypotheses.'39* Con fidence intervals for individual parameters are con structed using the "profile likelihood method."'39* The goodness-of-fit of each model to the data is assessed by applying a chi-square distribution to the deviance statis tic.'401 The p-value of the chi-square statistic is approx imated using the chi-square distribution with degrees of freedom equal to the [number of dose groups] -- (num ber of qj's estimated as being non-zero] -- l. The non-negativity constraints imposed on these models make this approach very different from ordinary (unconstrained) regression. In ordinary regression, a per fect fit of a model is guaranteed whenever the number of parameters equals or exceeds the number of data points. However, in the constrained model described above, typically all but a few of the q^'s are estimated Asbestos Structures that Induce Tumors in Rats 185 Mass Concen&aSon (mgftrP) Fig. 1. Fit of model. Tumor incidence versus mass concentration from animal study. to be zero so that the estimation results are the same as if these q^s had not been included in the estimation pro cedure. Consequently these, models may legitimately in clude more categories of asbestos structures than there are animal exposure groups. . Evaluation of Mesothelioma Incidence oma following exposure; P,, is the probability of devel oping lung tumors following exposure; and c is tbe constant of porpotionality between the probability of de veloping a mesothelioma and the probability of devel oping a lung tumor. This test was applied to the total data set and to the chrysotile and amphibole studies separately. Although there were too few mesotheliomas ob served in the animal inhalation studies to perform an extensive analysis similar to that conducted for lung tu mors, a test was .conducted of whether the risk of mes othelioma was proportional to the risk development of lung tumors^ If this test cannot be rejected, it.suggests that any fiber size distribution that describes die lung tumor responses also describes the mesothelioma re sponses. The specific test conducted was a likelihood ratio test of whether the probability of mesothelioma could be expressed as P1 moo = cPli where Pmi=u, is the probability of developing mesotheli RESULTS Univariate Measures of Exposure Figure I is a plot of the percentage of animals with lung tumors versus the mass concentrations (mg/m1) of total dust reported in the original Davis et al. studies. The curve in the figure represents the maximum likeli hood fit of the dose-response model represented by Equation 4, where SM is the measured dust mass for each experiment reported in Table 1. It is clear from Figure 1 that mass concentration of total dust does not provide a consistent dose-response !6 Berman et a). Waited Afrfeoma Ccncrtrstert Length Categories: 5pjm-40^mp * 40/xin Wkhh Categories: < 0.3^m and *5pfn Fig. 2. Fit of model. Tumor incidence versus PCM concentration from animal srudy. for these studies. Several of the Davis et al. studies em ployed a target mass concentration of 10 mg/mJ (Table 1) and results of these studies appear grouped together on the right side of the figure. Despite having been ex posed to similar dust mass concentrations, responses in these animal groups are quite variable, ranging from 0 pet (0/42 animals) for short amosite to 50 get (20/40 animals) for long chrysolite. Moreover, even though the experiment , involving wet dispersed chrysotile involved exposure to a total dust mass concentration of only 3.5 the tumor response was 44 pet (18/41), which is considerably greater than that of several of the experi ments employing a total mass concentration of 10 mg/m-\ Figure 2 is similar to Figure 1. except that the con centration of fibers measured by PCM (f/cc) in the orig inal study replaces mass concentration as the measure of exposure. This figure likewise exhibits no consistent dose-response pattern. Thus, neither of the two measures of exposure reported in the original Davis et al. studies (total dust concentration and PCM fiber concentration) relate to lung tumor risk in a consistent manner. Table 2 is a summary of the tests of goodness-offit for the exposure-response models presented in Fig ures 1 and 2 and similar models based on other univariate measures of exposure that are derived from TEM measurements of the re-generated dusts. For each univariate measure of exposure, the table contains the correlation coefficient (R2) for the relationship between the airborne asbestos concentration (defined by the uni variate measure) arid the negative Logarithm of the prob ability of remaining tumor-free (-Ln(l-P)]. The table also contains the deviance of the maximum likelihood fit of the univariate dose-response model {defined using Equation 4) and the p-value for the goodness-of-fit test associated with that deviance. Large deviances and, cor respondingly, small p-values (i.e., p < 0.05) indicate a poor description of the lung tumor response by a partic ular univariate measure of exposure. Both total dust mass and PCM concentrations give very large deviances and corresponding, highly signifi cant lack of fit, 138.0 (p < 0.0001) and 56.1 (p S 0.0001), respectively (Table 2), which is consistent with the visual impressions from Figures 1 and 2. Likewise, the concentrations of total asbestos structures measured by TEM, of TEM structures longer than 5 pm, and of TEM structures longer than 20 pm all give poor fils to the data, although the deviance statistic decreases as. the Asbestos Structures that Induce Tumors in Rats 187 Table II. Summary of Fils of Univariate Concentrations 10 Lung Tumor Data' Dose measure R Deviance if />-value* Total dust mass concentration 0.05 116.0 12 <0.0001 PCM structure concentration Total TEM structure concentration* 0.50 0.32 56.1 1)2.1 12 <0.0001 12 <0.0001 Concentration of structure longer than 5 pm* 0.39 87.7 12 <0.0001 Concentration of structure longer than )0 pm* . 0.50 51.4 !2 <0.0001 Concentration of structure longer than 20 pm* 0.72 31.4 12 0:0017 Concentration of structure longer than 30 pm* 0.7) 41.7 12 <0.0001 Concentration of structure longer than 20 pm and thinner than 0.4 pm* 0.60 37.4! 12 0.0002 Concentration of structure longer than 20 pm and thinner than 0.2 pm* 0.29 . 92.9 12 <0.000! Concentration of structure longer than 20 pm and thicker than 0.4 pm* 0.71 38.0 12 0.0002 Total structure surface area per air volume*' 0.65 38.4 12 0.0001 Total structure volume per air volume 0.54 50.2 12 <0.0001 Sum of aspect ratios per air volume* 0.32 )08.8 12- <0.0001 Sum of (aspect ratio)1-* per air volume* .. .. .. ._ 0.35 .100.3- . 12 <0.000! Stanton Index** 0.38 . 79.85 - 12 <0.0001 ' P-values <0.05 indicate a significant lack of lit of model to lung rumor data, based on chi-square distribution for the defiance with 12 degrees of freedom (14 dara sets and two parameters). * Concenlralions derived counting components of less complex structures as individual structures and ignoring the parent structure. 4 Concentrations derived counting only primary structures. J The index proposed by Slanton-er is the concentration of all fibers 2 8 pm in length and < 0.25 pm in width. structures are restricted to increasing lengths in this se ously suggested by Wylie et al.,131* however, the fact that ries, which indicates an improving fit. However, restrictr none of the exposure measures listed in Table 2 provide ing the analysis to even longer structures (> -30 pm), an adequate description of the database suggests that the restricting the widths of structures longer than 20 pm to features of asbestos that relate to risk are' too complex to thinner structures (< 0.4 pm or < 0.2 pm) or to thicker be represented by a single univariate exposure measure structures (> 0.4 pm) all make the fit worse. and that multivariate measures may be required to ade Several other measures evaluated in Table 2 are quately describe lung tumor response to asbestos. similar to exposure measures found by other investiga tors to be significantly correlated with tumor incidence: structures longer than 8 pm and thinner than 0.25 pm,1-4* Multivariate Measures of Exposure total surface area of the asbestos structures per unit vol ume of air (implied by Lippman,'41* assuming a relation To gain an understanding of the combinations of ship between tumor induction and fibrosis as described structurec.ateg'ories.that relate best to potency, more than by Davis and Cowie*421); total volume of asbestos per 100 statistical analyses were conducted in which relative unit volume of air (which is proportional to the total potencies were estimated for various combinations of mass concentration of asbestos'43*), the concentration of length and width categories. The results of these ex aspect ratios (sum of aspect ratios of structures per unit ploratory analyses are summarized by an analysis that volume of air*2**); and the concentration of aspect ratios incorporates a matrix of five length categories (< 5 pm, raised to the 1.8 power.130* 5-10 pm, 10-20 pm, 20-40 pm, and > 40 pm in length) All measures reported in Table 2 are significantly in combination with five categories of width (< 0,15 correlated (p < 0.05, based on the slope of the regression pm, 0.15-0.3 pm, 0.3-1.0 pm, 1.0-5.0 pm, and > 5.0 ; line) with lung tumor incidence. Nevertheless, alf of these measures provide a poor fit to the lung tumor data pm in width), for a total of 25 categories of structures that potentially contribute independently to potency. In (p < 0.0017, based on the deviance statistic) despite a this analysis, the relationship between exposure and re significant correlation coefficient. sponse was modeled using Equation 1 with WC con The results in Table 2 suggest that the tumorigenicity sisting of a 25-term function that is the sum of the of the asbestos dusts studied by Davis et al. are more product of each size category multiplied by a relative closely related to the concentration of longer structures potency for that size category (Equation 2). Table 3 in than to the concentration of shorter structures. As previ- ' dicates the maximum likelihood estimates of the relative \ \ HWBUI0010293 Berman et a). Table ])I. Relative Potencies for Inducing Lung Cancer" Width (|lm) <5 5-10. Length (pm) 10-20 20-40 > 40 Total Analysis based on .primary structures -V: " . ** -. " 11" -*/: *" ' Less complejC'clusters. and matrices )laced by components^' .. . ' < 0.15 0.15-0.3 0.3-1.0 '. f.O-5.0" ' > 5.0. .X Total 0 Deviance = 13.33 (7 df) ' p'*f- 0.06 0.0055 0.42 0.040 ' 0.043 0.0055 .. .0.50 - ' . . ; * * . " <-o.is ; ' .. ' : .0.15-0.3; '" . 0.3-I.0 1.0-5.0 > 5.0 X Total 0 Deviance = 12.03 (7 df) p = 6.10 0.0018 6.0018 . 0.033 0.033 X 0.074 0.074 0.10 -' 0.02 0.12 x 0.42 0.42 0.42 . 0.040 -. . 6.. , P\, 0.54: . 1.0' .- . X , 0.72 0.13 0.85 -- Q.I3 0.72. 0 . 0 . -. 0.15 1.6 juries represent an estimate of the relative potency assigned to structures in that size category by the model defined by Equation (1). An "X" tdicates a category that contains no structures. A blank indicates zero potency estimated for that category. P-value is for fit of model to hmg >mor data in Table I, based on chi-square distribution for the deviance (small p-values indicate poor fits). ies (i.e., the q^'s from the mode!) of the structures each of the 25 size categories. Table 3A presents results in which only primary .natures are included, whereas Table 3B presents relts in which parent clusters and matrices (for which mponent structures' were characterized) are replaced their components. The statistical analysis in which mplex structures are resolved into components (Table )) provides the best fit to the data and the resulting ode! provides an overall adequate description of the ta based on goodness-of-fit (deviance = 12.03 [7 df], = 0.10). The fit is also marginally acceptable from the alysis (Table 3A) that includes only primary structures eviance = 13.33 [7 df],p = 0.06). These two analyses hibit several common features. Both of the analyses xibute zero potency to structures shorter than 5 pm. isitive potencies are confined to structures that are ei-. ;r thin (< 0.3 pm) or very thick (> 5 pm). For both .n and thick structures, potency tends to increase with creasing length, although the observed relationships ; not entirely monotone. Resolution of complex structures into component ers and bundles significantly increases the precision th which fibers and bundles can be categorized, since jghly 50 pet of the fibers and bundles characterized re-generated dusts are components of complex res. Consequently, from this point on, only, anal yses that resolve clusters and matrices into components are considered. .- Although the statistical analysis reported in Table 3B adequately describes the data, the relative potencies assigned to the two narrowest categories of structures seems somewhat unrealistic; one expects dose-response relationships to vary smoothly with size. Positive poten cies are assigned, to structures thinner than 0.15 pm in : combination with one long and one short length category (5-10 pm and > 40 pm) and to.structures between 0.15 and 0,3 pm in width in combination with two interme diate length categories (10-20 pm and 20-40 pm). This suggests that the potencies of structures in these two width categories cannot be distinguished using this da tabase. To test this hypothesis, these two categories were . combined into a single category (width < 0.3 pm). The results of this analysis are presented in Table 4A. Com bining these width categories only slightly increases the deviance (from 12.03 to 12.15), so that the hypothesis that structures in the two width categories are equally potent cannot be rejected. Moreover, the fit of the model actually improves somewhat (the p-value increases from 6.10 to 0.14), due to a reduction in the number of de grees of freedom-. Because the pattern of potencies observed among the length categories of the model presented in Table 4A Asl A: I B: < -Er. ini tui als* plil the wa stn bin she cat rie: Str ren in fit "o det sli| the 0.2 wa inti poi All opt HWBUI0010294 .sbestos Structures that Induce Tumors in Rats 189; Table IV. Additional Analyses Performed to Obtain an "Optimum Exposure Index"' W^Th (pm) Length (pm) < 5 5-10 10-20 20-40 > 40 Total A: Intermediate analysis < 0.) . 0.3-1.0 1.0-5.0 > 5.0 X Total 0 Deviance = 12.15(840 p = 0.14 0.0012 0.0054 0.0012 0.0054 0.012 0.012 0.84 0.14 0.98 0.847 0 0 0.152 10 B: Optimum exposure index < 0.30 > 5.0 Total Deviance = 12.66 (10 dj) p = 0.24 0.0017 0.0017 0.853 0.145 0.998 0.855 0.145 1.0 Entries represent an estimate of the relative potency assigned to structures in that size category by the model defined by Equation (1). An "X" indicates a category that contains no structures. A blank indicates zero potency estimated for that category. A-value is for fit of model to lung mrnor data in Table I, based on chi-squared distribution for the deviance. (Small p-values indicate poor fits). also appear somewhat unrealistic, this model was sim plified by reducing the number of length categories in the same manner that the number of width categories was reduced for ihe model described above. Therefore, structures between 5 pm and 40 pm in length were com bined into a single length category. Also, structures shorter than 5 pm were removed from the analysis be cause, even with the reduced number of length catego ries, the short structures were assigned zero potency. Structures between 0.3 and 5.0 pm in width were also removed because these too were assigned zero potency in the reduced model. The model resulting from the maximum likelihood fit of the remaining size categories (referred to as the "optimum exposure index") is shown in Table 4B. The deviance of 12.66 obtained from this analysis is only slightly larger than that of the full model (12.03) andthe goodness-of-fit p-value"from the reduced: model is 0.24 (10 df), which indicates a better statistical fit than was obtained using the original model (Table 3B) or the intermediate model (Table 4A). The "optimum exposure index" assigns a relative potency of: 0.0017 for structures < 0.3 pm in width and be tween 5 and 40 pm in length; 0.853 for structures < 0.3 pm in width and > 40 pm in length; and 0.145 for structures > 5.0 pm in width and > 40 pm in length. All other structures are assigned a potency of zero in the optimum exposure index. Inlereslingly, whereas the lat ter size category is composed completely of complex structures, the two categories of thin structures are com posed almost totally of fibers and bundles (including, many that are components of complex structures). The fit to the experimental data obtained by the "optimum exposure index" is presented in Figure 3. The x-axis of this plot is exposure expressed as the weighted airborne concentration formed by the sum. of the relative potencies from this analysis times the con centrations of structures in the corresponding categories (WC in Equation 2). Unlike Figures 1 and 2, this graph indicates a consistent- dose-response relationship. Table 5 contains 90 pet confidence intervals for the three structure categories that are assigned positive po tency by the optimum exposure index. The fact that none of these three confidence intervals contains zero potency implies that none of these three structure categories can. be removed from the model without significantly de grading the fit (i.e. the hypothesis that one of these cat egories has zero potency can be rejected). Thus, to. obtain adequate fits to the lung tumor incidence data within the framework of this analysis, it is necessary to assign positive potencies to all three of the size cate gories of structures that contribute to the optimum ex posure index. Table 5 also contains 90 pet confidence intervals for two categories of structures assigned zero potency in the optimum exposure index: structures < 5 pm in length and structures 5--40 pm in length and > 5 |im in thickness. The upper 95 pet confidence bound for the relative potency of structures shorter than 5 pm is esti mated as 0.00008. This upper bound is 0.00008/0.0017 10 73 - Berman et a). o c Wrtfcsl bws Indcatft 90% confidsoeg friftwals. AS&revistteffT* ter data are sptetnad TeM 1. >00 2009 3000 4000 PCM GortesftfrafJort itfcc) Fig. 3. Fit of model. Tumor incidence versus structure concentration by TEM. 5000 Table V. Relative Potency Estimates Derived from the 4`Optimum Exposure Index** with 90% Confidence Intervals Category (units in Um) Relative potency estimate 90% confidence interval length < S 5 5 length < 40 width S 0.3 5 5 length < 40 width > 5.0 length k 40 width < 0.3 length & 40 width 2: 5.0 0 0.0017 0. 0.8S 0.1 S' ? (0, 0.00008][0.0010. 0.0032] (0, 0.030]- (0.67.0:95] (0.04, 0.32} " The size category is assigned 2ero potency in the model derived from the maximum likelihood fit and therefore does not contribute to the "optimum exposure index.'* = 1/20 as large as the estimated relative potency of structures between 5 and 40 pm in length and thinner than 0.3 pm, and only 0.00008/0.85 = 1/10,000 as large Kthe estimated relative potency of structures longer an 40 pm and thinner than 0.3 pm. Results of various additional hypothesis tests are presented in Table 6. All of the analyses presented in dicate a zero potency for all categories of structures shorter than 5 pm. Since this is the outcome most fa vorable to the hypothesis that such structures are nonpotent, p = 1.0 is the p-value associated with a test of this hypothesis. Although some analyses reported herein assign a positive potency to. structures 5-10 pm in length, the hypothesis that structures shorter than 10 mi crons in length are non-potent cannot be rejected either (p = 0.09). Results of hypothesis tests for a difference between the potency of chrysotile and amphibole, conducted within the framework of the optimum exposure index, are also presented in Table 6. This test was performed by estimating different absolute potencies (i.e., different values for (3; in the model) for chrysotile and amphibole, but assuming that the relative potencies of structures in different size categories (i.e., the qj's) are the same for chrysotile and amphibole (Equation 3). The resulting pvalue for this test is 0.72 (non-significant), indicating no detectable difference between the potency of chrysotile and amphibole. It is also not possible to reject the com- Asbestos Structures that Induce Tumors in Rats 191 pound hypothesis that chrysotile and amphibole have the same relative potencies and that the relative contribu tions to potency from different size fractions are the same for chrysotile and amphibole. The optimum exposure index assigns positive po tency to thin (< 0.3 pm) and very thick (> 5 pm) struc. . hires but not to structures of intermediate thickness. Although a potency for very thick structures could be due to thin structures imbedded within complex struc tures (see discussion), this result could also be an artifact ' caused by limitations in the data base and the specific ~outpoints, used to create the optimum exposure index. Additional analyses taken to further explore' this issueidentified a slightly different index.that also provides an adequate fit to the data (p = 0.09) and which does not result in a positive potency for very thick structures. This index assigns a relative potency of: 0.0024 for structures < 0.4 pm in width and be tween 5 and 40 pm in length; and 0.9976 for structures < 0.4 pm in. width and > 40 pm in length. Statistical Analyses of Mesothelioma Data Results in Table 7 indicate that the hypothesis that mesothelioma incidence is proportional to lung tumor incidence for all sample types combined can be rejected. However, the hypotheses that mesothelioma incidence is proportional to lung tumor incidence either for chrysotile sample types or amphibole sample types (considered separately) cannot be rejected. The best indication (i.e., maximum likelihood) is that the probability of a chrysotile (amphibole) exposure inducing mesothelioma is 0.058 (0.20) times the prob ability that the same exposure induces lung tumors. These two values are significantly different (p = 0.032). Since the analyses of the lung tumor data indicate that there is no difference between chrysotile and amphibole in their ability to induce lung tumors, these results sug gest that amphibole is 0.20/0.058 = 3.4 times more po tent than chrysotile for causing mesothelioma in rats (assuming that the relative potencies of structures in dif ferent size categories are approximately the same for mesothelioma and lung tumors. DISCUSSION Results from this study indicate that the induction of lung tumors in rats following inhalation of asbestos , Table VI. Hypothesis Test Results for Lung TumorTncidence Decrease in deviance lvalue of Hypo thesis test* Structures shorter than 5 pm are nonpotent Structures shorter than 10 pm are non-potent* Structures shorter than 40 |im are non-poienF' Structures longer than 40 pm are non-potenH Structures thicker than S pm are nonpotent*' Structures thinner than 0.3 pm are non-potent Amphibole and chrysotile structures.-. arc equally potent^ Fibers and bundles are equally po tent* 0 I.7& 74.6 73.1 10.0 54.4 0.12 0.78 i (NST 0.09 (NS) y<0.0001 is <0.0001 (S) 0.007 <S) <0.0001 . (S) 0.72 (NS) 0.68 (NS) * A P-value less than 0.05 indicates that the hypothesis can be rejected. `NS - not significant (hypothesis cannot be rejected). ' Based on analysis of data in Table 5A. d Based on analysis of data in Table SB 'S - significant (hypothesis can be rejected). - /Due to a limitation in the computer program used to perform these calculations, the value of the background parameter, a, was fixed at the value that provides a perfect fit to the control data. is a function of the size and shape of asbestos structures within the inhaled dusts. Structures contributing to lung tumor risk appear to be long (5 5 pm) thin (< 0.3 or 0.4 pm) fibers and bundles, with a possible contribution by long and very thick (> 5 pm) complex clusters and matrices. Potency appears tojncrease with increasing length, with structures longer than 40 pm being about 500 times more potent than structures between 5 and 40 pm in length. Fibers and bundles of similar dimensions appear to contribute equally to potency. For all types of asbestos structures, results of this study suggest that a minimum length (between 5 and 10. pm) exists such that the best estimate of the potency of structures shorter than this minimum is zero. This anal ysis also indicates that, as a worst case, a 95% upper limit to the potency of individual structures shorter 5 pm is no more than about 1/20 of the potency estimated for individual fibers or bundles between 5 and 40 pm in length and no more than about 1/10,000 of that esti mated for individual fibers or bundles >40 pm in length and <0.3 pm in width. In contrast, the evidence sug gesting a range of widths within which structures can be considered non-potent is less compelling. HWBUI0010297 Berman et al. Table VII. Hypothesis Test Results for Comparing Mesothelioma and Lung Tumor Incidence Hypothesis tested . Increase in P-value of deviance hypothesistest* The relative probability of inducing mesotheliomas and lung tumors are the same for all dusts tested. IP.... = . :V$' 20.9, Therelative probability of inducing mesotheliomas and lung tumors are the same for all chrysolite dusts tested. . v; . : TTif trelauive-probability of inducing mesotheliomas and Sung tumots. are the same for ail antphibole dusts ;"Aesied. ? c.?,,>* . .fheproportionality constant relating mesothelioma-incidence to lung tumors incidence is the same for chrysolite . .^'di.th'e^phiBoies. (c;. = c,)- .- ....... .. / 9.97 . 6.30 . 4.62 0.028 0.20 0.28 0.032 than 0.05 indicates,that P is significam and the hypothesis can be rejected.. . .. , ATfie- tjSsi estimate for the constant of proportionality between mesothelioma and lung rumor incidence is c = 0.094 when all dusts arc considered. .-'Theibest estimate for the constant of proportionality betwen, mesothelioma and lung minor incidence is c, = 0.058 when chrysotile. dusts are '''canstderedseparately. . :` . ' 'The best estimate for the constant of proportionality between.mesothelioma and lung' tumor incidence is c. .= 0.20 when antphibole dusts are . considered separately. . '. . . For mesothelioma, it appears that mineralogy is an important determinant of the relative potency of inhaled dusts, with chrysotile being less potent toward the induc lion of mesotheliomas than the amphtboles in comparison their relative potency for inducing lung tumors. . * Despite the special emphasis placed-on the charac terization of long structures, the data base employed in this,- study contains only a limited number of very long structures (e.g.-, >40 pm in length). This contributes un certainty to the specific quantitative estimates- of relative potency for long structures, the most potent'structures identified. Our statistical analysis does not address this type of uncertainty; fiber concentrations were assumed to be known with certainty. Additional uncertainties in the exposure estimates result from the reliance.on regenerated dusts in this study rather than the actual dusts to which the. experimental animals were exposed. This problem was mitigated by having the dusts regenerated by the same personnel using the same stock material,, equipment, and procedures that were employed in the original studies. Despite the above considerations, we know of no other set of experimental data that is more suitable for quantitative determination of relative potencies of as bestos structures of different types and dimensions that are apt- to be relevant to conditions of human exposure. Our study involves inhalation, which is more relevant to human exposures than injection or implantation studies. Our data base contains data on a variety of different types of asbestos including crocidolite, tremolite, chrys otile, and amosite. It includes four different samples of chrysotile and three different samples of amosite, which were chosen to represent a range of asbestos structure size dimensions. It also involves a more detailed char acterization of long structures and complex structures than those available from other experimental settings. The only other studies in which the relationship be tween structure size and shape and tumor response has been examined formally (using some type of statistical analysis) are the.studies by Stanton et al.123-*'' and other statistical studies employing the database reported by Stanton et al.1"-?0* or employing a re-analysis of the sam ples studied, by Stanton and coworkers.(3l> However, these studies all incorporate a statistical procedure that differs radically from that employed in the present study. While the earlier studies only identify exposure meas ures that are' significantly correlated with tumor inci dence, the present study identifies an exposure measure that satisfactorily describes the tumor incidence (i.e., provides, an acceptable fit to all of the data).6 In fact, it 6The methodology utilized in this study avoids several methodological problems associated with the analysis by Stanton et al. and the others who re-evaluated their data and conclusions.' Stanton et al. computed a correlation between the logarithm of the concentration of fibers in specific dimensional ranges and the logit of the probability of a tu mor. A number of exposure categories for individual fiber types in vestigated by Stanton et al. contained no fibers, which implies that the logarithm of the fiber concentration for these categories is un defined. Stanton et al. assigned a zero in place of the logarithm of concentration in these cases. However, this decision is entirely ar bitrary and the value selected could have had a large impact upon the correlation coefficient. Similarly, no tumors were detected in sev eral experimental groups investigated by Stanton et al. and the logit is undefined whenever no tumors occur. Although it is not clear how Stanton et al. handled this problem, it seems likely that such groups were omitted from their analysis. Note that, because of these prob lems, the correlation coefficients presented in Table 2 (which are calculated based on a linear relationship that is not undefined at zero) may not be directly comparable to those calculated by Stanton et al. Asbestos Structures that Induce Tumors in Rats 193 is apparent from Text Figure 2 of Stanton et aI.|2Jl that the exposure measure they identify as being most highly correlated with tumor incidence (fibers longer.than 8 [im and thinner than 0.25 jim) does not provide an accept able fit to the observed tumor incidence. Similarly, al though all of the univariate measures considered in the present study (Table 2)'are highly correlated with tumor incidence, none of them adequately describe lung tumor incidence. In contrast, when asbestos exposure is expressed as the weighted sum of the concentrations of structures in the three size categories of structures defined in this study by the optimum exposure index, this exposure in dex provides a statistically adequate fit to the lung tumor response data reported in the studies by Davis et al. (Ta ble 1). Consequently, the hypothesis that the model de termined by the optimum exposure index completely characterizes the potency of asbestos structures in the induction of lung tumors in AF/HAN rats cannot be re jected. This implies that this model can potentially pro vide adequate descriptions of risk in a broader range of exposure settings. Stanton et found that fibers longer than 8 pm appear to correlate best with mesothelioma incidence, which parallels the findings in this study that structures longer than some minimum length (between 5 and 10 pm) tend to contribute to the induction of lung tumors fo!fowing inhalation and that potency tends to increase with increasing length. However, our analysis indicates that the potency of structures increases with length up to lengths of at least 40 pm,, whereas Stanton et al. did not explicitly consider the contribution to potency by such long structures. The maximum thickness for potent fibers and bun dles found in this study (< 0.3 or 0.4 pm) is comparable to the maximum thickness reported by Stanton et of 0.25 pm. Stanton et al. did not report information on complex structures in their data base and, indeed, the meaning of complex structures, in a gel suspension (which they used for implantation) is not clear. Conse quently, their study provided no information on the ef fects of thick clusters. Complex asbestos structures do occur as isolated species in the air, however, and results in this study indicate that dusters longer than 40 pm and thicker than 5 pm may contribute to the potency of an asbestos dust. That long, thick clusters (structure for structure) may contribute about one-sixth.as much to total potency as long, thin fibers or bundles (Table 5B) suggests a mechanism in which some of these may break down and liberate long fibers or bundles that may then contribute to the induction of tumors. The structures longer than 40 pm and thicker than 5 pm that are evaluated in this study are open structures with settling velocities that are likely comparable to their component fibers or bundles. Thus, it is likely that they are respirable and can pene trate the deep lung. Once in the deep lung, however, when such structures impact a wail, some may liberate component fibers or bundles and the longest and thinnest of those may then contribute to the induction of tumors. Although neither Stanton et al.'2'-J' nor Bertrand and PezeraP29' concluded that mesothelioma induction following injection or implantation is a strong function of fiber mineralogy, such a conclusion was reached by Bonneau et al.'30' In our study, there is no evidence that mineralogy is a determinant of potency toward the in duction of lung-tumors. However, assuming the size range that induces lung tumors and mesothelioma- are similar, results in this study suggest that for inhaled dusts that exhibit comparable potency toward lung tumor induction, amphibole dusts are approximately three times as likely to induce mesothelioma as chrysotile dusts. Implications for Human Exposure For human exposures, potency estimates obtained from different epidemiological studies, which are based upon PCM measurements, have been found to differ by large factors.''' This study likewise demonstrates that PCM measurements do not provide a consistent dose response in the animal inhalation studies either. How ever, more complex exposure measures based upon TEM measurements are shown to provide a consistent dose-response relationship for the animal data. This sug gests that the observed lack of a consistent dose response across epidemiological studies is due at least partially to the fact that the features of asbestos important to deter mining potency are not adequately represented by PCM measurements. Regarding mineralogy, human epidemiology stud ies (taken as a whole) suggest that mineralogy is im portant at least for determining potency toward the induction of mesothelioma."' Our results likewise indi cate that amphibole dusts are more likely (by a factor of about three) to induce mesothelioma than chrysotile dusts whenever their potential to induce lung tumors are comparable. The importance of mineralogy in determining the relative potency of a dust toward the induction of lung tumors in human exposure is less clear."' Results from this study suggest that, when the relative size distribution of structures in a dust is taken into account, the miner 94 . Berman et al. alogy of a structure does not contribute to the determi nation of potency toward the induction of lung tumors. However, this finding may not be applicable to humans because chrysotile degrades more rapidly than amphibole in vivo. Since humans live much longer than the animal species typically used in experiments, the relative persistence of chrysotile and- amphibole in wWIJ may have a much larger impact on the induction of lung tu mors in humans while remaining unimportant in ani mals. gression analyses; and Robyn York and Lynn-Williams for typing the revisions to this manuscript. The work could not have been completed without each of their contributions. Finally, we wish to remember Richard Howe who developed the software we used to complete the regression analyses reported in this study. REFERENCES Implications for Asbestos Measurement Results from this study indicate strongly that meth ods for the measurement of asbestos should be modified to include better characterization of longer structures and that such characterization should be performed using TEM (as opposed to PCM), due to the need to include thin structures (and the need to distinguish asbestos structures from non-asbestos structures). Better charac terization of longer structures can be achieved, just as in our reanalysis of the data from the animal experiments, by examination of grid specimens at lower magnifica tions in which only structures exceeding some minimum length are recorded. At a minimum, we recommend that separate examinations be made for structures > 5 (im in length. However, because of the indication in this study that very long structures (> 40 pm in length) are highly potent relative to shorter structures, we also recommend that a separate examination of still longer structures (e.g., structures > 20 Jim in length) be included in the routine analysis of asbestos. Similarly, to obtain expo sure measures that relate more closely to. risk without unduly increasing the cost of the analysis, characteriza tion of structures shorter than 5 pm can be deempbasized. ACKNOWLEDGMENTS We wish to thank Kent Kitchingman and Steven Bayard for their advice and support. We also acknowl edge the National Asbestos Task Force of the U.S. En vironmental Protection Agency, chaired by Hans Crump, for funding this work. Recognition is extended to Mike Whittington, who generated the dust clouds and col lected the samples; Greg Lewis and Alice Liebert, who performed the many hours of TEM analysis required; Jane Chatfield who completed the quality assurance (work to confirm data quality; Tammie Lambert and Shipei Weng, who performed the data manipulation and re- 1. HEl-AR (Health Effects Institute-Asbestos Research), "Asbestos irr Public and Commercial Buildings.' A Literature Review and Synthesis of Current Knowledge" (HEl-AR, I4i Portland St., Suite 7100, Cambridge. Massachusetts, 1991). 2.' USEPA (Unite'd States Environmental. Protection Agency), Airhome Asbestos Health Assessment Update (Office of Health Ef fects Assessment, Washington, D.C., Report 600/8-84-003F, 1986). 3. WJt. Walton, `"The Nature, Hazards and Assessment of Occu pational Exposure to Airborne Dust: A Review." Annals of Oc cupational Hygiene 26, 115--247. 4. R.E. Bolton, J.M.G. Davis, K. Donaldson, and A. Wright, "Var iation in the Carcinogenicity of Mineral Fibres," Annals of Oc cupational Hygiene, 26, 569-582 (1982). 5. R.E. Bolton, J.M.G. Davis, B.G. Miller, K. Donaldson, and A. Wright, "The Effect of Dose of Asbestos on Mesothelioma Pro duction in the Laboratory Rat," Proceedings of the Vlth Inter national Pneumoconiosis Conference^ 2,-1028-1035 (1984). 6. R.E. Bolton, J. Addison, J.M.G. Davis, JC. Donaldson, A.D- Jones, B.G. Miller, and A. Wright, "Effects of the Inhalation of Dusts from Calcium Silicate Insulation Materials in Laboratory Rats," Environmental Research, 39, 26-43 (1986). 7. J.M.G. Davis, S.T. Beckett, R.E. Bolton, P. Cotlings, and A.P. Middleton, "Mass and Number of Fibres in the Pathogenesis of Asbestos-Related Lung Disease in Rats," British Journal ofCan cer. 37. 673-688 (1978). 8. J.M.G. Davis, S.T. Beckett, R.E. Bolton, and K. Donaldson, "A Comparison of the Pathological Effects in Rats of the UICC Ref erence Samples of Amosite and Chrysotile with Those of Amosite and Chrysotile Collected from the Factory Environment" (J.C. Wagner, ed.. Biological Effects ofMineral Fibres, (ARC Scientific Publications, 288-292; 1980). 9. J.M.G. Davis, J. Addison, RJE. Bolton, K. Donaldson, AJ3. Jones, and R.G. Miller, "Inhalation Studies on the Effects of Tremolite and Brucitc Dust in Rats,'' Carcinogenesis, 6, 667-674 (1985). !9. J.M.G. Davis, J. Addison, R.E, Bolton, K. Donaldson, A.D. Jones, and T. Smith, "The Pathogenicity of Long Versus Short Fibre Samples of Amosite Asbestos Administered to Rats by Inhalation and Inrraperitoneal Injection," British Journal of Experimental Pathology. 67,415-430 (1986). 11. J.M.G. Davis. J. Addison. R.E. Bolton, K. Donaldson, and A.D. Jones, "Inhalation and Injection Studies in Rats Using Dust Sam ples from Chrysotile Asbestos Prepared by a Wet Dispersion Pro cess," British Journal of Pathology, 67, 113-129 (1986). 12. J.M.G. Davis, R.E. Bolton, A.N. Douglas, A.D. Jones, and T. Smith, "Effects of Electrostatic Charge on the Pathogenicity of Chrysolite Asbestos," British Journal of Industrial Medicine, 45, 292-309 (1988). !3. J.M.G. Davis, and A.D. Jones, "Comparisons of the Pathogenicity ofLong and Short Fibres of Chrysotile Asbestos in Rats." British Journal of Exposure Pathology, 69, 171-738 (1988). 14. E.J. King, J.W. Clegg, V.M. Rae, "The Effects of Asbestos, and of Asbestos and Aluminum, on the Lungs of Rabbits," Thorax. ' 1, 188-197 (1946). . HWBUI0010300 Asbestos Structures that Induce Tumors in Rats 195 15. H. Muhle, B. Bellmano, S. Takenala, and U. Ziem, "Inhalation and Injection in Rats to Test the Carcinogenicity of MMMF." Annals Occupational Hygiene, 31, 755-754 (1987). 16. F. Poll, F. Huth, and K.H. Friedrichs, "Tumorigcnic Effect of Fibrous Dust in Experimental Animals," Environmental Health Perspectives, 9, 313-315 (1974). 17. F. Pott, F. Huth, and K.H. Friedrichs, "Results of Animal Car cinogenesis Studies After Application of Fibrous Class and Their Implications Regarding Human Exposure" (Occupational'Expo sure to Fibrous Glass, U.S. HEW Publication No. 7fr-l5l. pp. 183-191, 1976). 18. F. Pott. "'Some Aspects on the Dosimetry of the Carcinogenic Potency of Asbestos and Other Fibrous Dusts." Staub-Reinhalt. 38. 486-490 (1978). 19. F. Pott and H. Muhle, "Animal Experiments with Mineral Fibres" Proceedings from a symposium. Short and Thin Mineral Fibres: . Identification, Exposure, and. Health Effects, National Board of Occ. Safery and Health, Solnic, Sweden pp. 133-161, 1982). 20. F. Pott, U. Ziem, and If. Mohr, "Lung Carcinomas and Mesothe liomas Following Intratracheal Instillation of Glass Fibres and As bestos" (Proceedings of the 6th Pneumoconiosis Conference, pp. 747-756. 1983). 21. F. Pott, U. Ziem, R.J. Reiffer, F. Huth. H. Ernst, and U. Mohr, "Carcinogenicity Studies on Fibres, Metal Compounds, and Some Other Dusts in Rats," Experimental Pathology, 32, 129-152 (1987). 22. M. Stanton, and C. Wrench. "Mechanisms of Mesothelioma In duction with Asbestos and Fibrous Glass," Journal of the Na tional Cancer Institute, 48, 797-821 (1972). 23. M- Stanton, M. Layard. A. Tegeris, E. Miller, M. May, and E. Kent, "Carcinogenicity of Fibrous Glass: Pleural Response in the Rat in Relation to Fiber Dimension," Journal of the National Cancer Institute, 58. 587-597 (1977). 24. M. Stanton, M. Layard, A. Tegeris, E. Miller, M. May, and E. Morgan, "Relation of Particle Dimension to Carcinogenicity in Amphibole Asbestos and Other Fibrous Minerals," Journal of the National Cancer Institute, 67, 965-975 (1981). 25..J-.C. Wagner, G. Berry, and J.W. Skidmore, "Studies of the Car cinogenic Effects of Fiber Glass of Different Diameters Following Innapleural Inoculation in Experimental Animals" (NIOSH 76151, pp. 193-197, 1976). 26. J.C. Wagner, G. Berry, R.J. Hilt, D. Munday, and J.W. Skidmore, "Animal Experiments with MMM(V)F-Effects of Inhalation and Intrapleural Inoculation in Rats. Biological Effects of Man-made Fibres" (Proceedings ofa WHO/LARC Conference, Copenhagen, pp. 209-233, 1982). 27. M.M.F. Wagner. J.C. Wagner, R. Davies, and D.M. Griffith. "Sil ica-Induced Malignant Histiocytic Lymphoma: Incidence Linked with Strain of Rat and Type of Silica," British Journal of Cancer, 41, 908-917 (1980). 28. J.C. Wagner, J.W. Skidmore. R.J. Hill, and D.M. Griffith. "Erionite Exposure and Mesotheliomas in Rats," British Journal of Cancer. 51. 727-730 (1985). 29. R. Bertrand and H. Pezerat, "Fibrous Glass: Carcinogenicity and Dimensional Characteristics. Biological Effects of Mineral Fi bres" (J.C. Wagner, ed_, /ARC Scientific Publications. 901-911, 1980). 30. L. Bonneau. C. Malard, and H. Pezerat, "Studies on Surface Prop erties of Asbestos," Environmental Research 41, 268-275 (1986). 31. A.G. Wylie, R.L, Virta, and J.M. Segrctti, "Characterization of Mineral Population by Index Particle: Implications for the Stanton Hypothesis," Environmental Research, 43, 427-439 (1987). 32. K.P. Lee, C.E. Banns, R.S. Griffith. R.S. Waritz, and C.A. Lapin. "Comparative'Pulmonary Responses to Inhaled Inorganic Fibers with Asbestos and Fiberglass." Environmental Research, 24, 167191 (1981). 33. E. McConnell, J.C. Wagner. J.W. Skidmore, and J. Moore, '"A Comparative Study of the Fibrogenic and Carcinogenic Effects of UiCC Canadian Chrysotile 8 Asbestos and Glass Microfibre (JM 100)" (Biological Effects of Man-made Fibres. Proceedings of a WHO/IARC Conference, pp. 234-252, 1982). 34. D.M. Smith, L.W. Oritiz, R.F. Archuleta, and N.F. Johnson,. "'Long-Term Health Effects in Hamsters and Rats Exposed Chron ically to Man-Made Vitreous Fibres," Annals of Occupational Hygiene, 34, 731-754 (1987). 35. J.W. Wagner, G. Deny, J.W. Skidmore, and V. Timbreil, "The Effects of the Inhalation of Asbestos in Rats," British Journal of Cancer, 29, 252-269 (1974). 36. J.C. Wagner, D.M. Griffiths, and D.E. Munday, `'Experimental Studies with Palygorskite Dusts," British Journal of Industrial Medicine, 44, 749-763 (1987). 37. E.J. Chatfield, and D.W. Berman, "Interim Superfund Methodfor the Determination ofAsbestos in Ambient Environments. Part t: Method" (USEPA publication: 540/2-90/005a, May 1990). 38. D.W. Berman and E.J. Chal field, "Interim Superfund Methodfar the Determination ofAsbestos in Ambient Environments, Part 2: Technical Background Document" (USEPA publication: 540/290/005b, May, 1990). 39. D. Cox and D. Lindley, Theoretical Statistics. (Chapman and Hall, London, 1974). 40. National Research Council. Health Effects of Exposure to Low Levels ofIonizing Radiation. BEJR V. Committee on the Biological Effects of Ionizing Radiations (National Academy Press, Wash ington. D.C., 1990). 41. M. Lippman, "Review: Asbestos Exposure Indices." Environ mental Research, 46, 86-106 (19B8), 42. J.M.G. Davis, and H. Cowie, "The Relationship Between Fibrosis and Cancer in Experimental Animals Exposed to Asbestos and Other Fibers," Environmental Health Perspectives, 88, 305-309 (1990). 43. W.J. Nicholson, Airborne Levels of Mineral Fibres in the NonOccupational Environment (Division of Environmental and Oc cupational Medicine, ML Sinai School of Medicine, City University of New York, 1988).