Document npQx4j0yK1VyNMmbyn0NxDNam

PLAINTIFF'S 5"::":' 111 EXHIBIT Consequences of Synergy between Environmental Carcinogens I M. C. BRIIENRAUM I~c~ptrr.ttnanotf' l ~ x p e r . i t ~ i c nP~t~rtrhl ology, S I. Mtrry 's Hospiltrl Mc~dicirISclrool, London, United Kingdom Received January 31, 1984 As i t is generally impossible to determine close-response relationships for carcinogens at the low concentrations in which they occur in the environment, risk-benefit considerations are by consensus based on the linear, no-threshold model. on the assumption that this represents the worst case. I-lowever, this asstiniption does not take into account the possibility o f synergistic interactions between carcinogens. I t is shown here that, as a result of such interactions, the dose-response curve for added risk due to any individual carcinogen will generally be steeper at lower doses than at higher doses, and consequently the risk at low environmental levels will be higher than would be expected from a linear response. Moreover. this excess risk at low doses is shown to increase as the general level of environmental carcinogens rises and, independently of this effect, it may also increase with the number of carcinogens present. o 198s Academic t'ress. inc. INTRODUCTION The form of the dose-response curve of a carcinogen at low doses (i.e., in the region of interest for most environmental carcinogens) is of considerable public health importance. There has been much argument as to whether such curves have thresholds or not, for the existence of a threshold suggests there may he safe levcls of exposure. However, for most carcinogens, it is almost impossible to determine the shape of the dose-response curve at the low levels which prevail in the environment. Thcre is therefore a consensus that the safest assumption is that curves are generally linear, without thresholds. It is widely believed that this model probably overestimates the effects of carcinogens in the low-dose, lowrate region, that it thus defines an upper limit of risk, and, therefore, that riskbenefit calculations based on this model will err, if at all, on the side of safety (International Commission on Radiological Protection, 1966; Hoe1 et af., 1975; Upton, 1977; Brown, 1977; Schneiderman and Brown, 1978; Pochin, 1978; Committee on Biological Effects of Ionizing Radiations, 1980; Rodricks, 1981). Crump et af. (1976) and Peto (1978) pointed out that, in any case, for carcinogens of environmental concern, the shape of the dose-response curve for an individual carcinogen in isolation is more of academic than of practical importance. They showed that, for a carcinogen added in low levels to an environmenl already containing substantial amounts of other carcinogens, the incidence rate of additional cancers would be very ncarly proportional to the amount of addcd carcinogen, irrespective of the form of its dose-response curve. Further, Gums and Crump (1978) showed that, cvcn whcn a carcinogen has a highly nonlinear 0013-9351/85 $3.00 Copyright n I IYX5 hy Ac:demic Pres.;. Inc. A l l riplit\ ( 1 1 ~ e p r n i l ~ ~ ciln~:~illiyn lorn1 rcwrvcd 3I O doseexperi thc lin ai., ( 1 on ad( The carcin to adc Thus, carcin accepl curve. The do no' to the carcin 1968, koff, 1 about when cinogc dosebe COI Inte ogens criteri from tions produ (Berer with s linear region nonlin case 1 appro indcpt of isol 1953; straigt three) agent5 Figu of E ar the fir! SYNEliCiY BI'I'WEEN ENVIKONMI~N'I'ALCAKC'INOCiliNS 31 I dose-response curve and data are obtained in a perfectly conducted, large-scale experiment with no background effect, statistical tests would probably not enable the linearity hypothesis to be rejected at reasonable significance levels. Guess et d.,(1977) also showed that, at low doses, the upper statistical confidence limits on added risk would almost certainly be linear. These statistical arguments therefore assert that, whatever the true nature of carcinogen dose-response curves at low levels, in practice the extra effect due lo adding a new one to the environnient will be very nearly linear with dose. 'I'hus, the risk-benefit considerations which should decide whether or not a new carcinogen is introduced into the environment and which should determine its acceptable level should be based o n the assumption of a linear dose-response curve. S The purpose of this communication is to show that linear dose-response curves do not in Fact generally indicate a safe upper limit of risk for carcinogens added I to the environment. There is substantial evidence suggesting that environmental L: carcinogens interact synergistically in causing cancers in humans (Selikoff el ai., 1 1968, 1980; Lundin et d.,1969; Kothman and Keller, 1972; Hammond and Selikot'f, 1973; Selikoff and Hammond, 1975; Doll, 1977; Saracci, 1977). Speculation about possible mechanisms of carcinogen interactions is unlikely to be prot'itable I when so much is still obscure about the mechanisms of action of individual carcinogens. However, it is shown here that such interactions affect the shapes of dose-response curves and that, in consequence, risks in thc low-dose region may be considerably greater than indicated by the linear model. 1 the EFFECTS OF INTERACTIONS .iblic Interaction means that the effect of exposure to two or more different carcinrves ogens is not what is expected from their individual dose-response curves. Various Y be criteria have, from time to time, been suggested for determining what to expect sible from combinations of agents that do not interact. The most widespread assump:vail tions are that the effect of a zero-interactive combination should be either the In is product or the sum of the effects of its constituents. It is shown in detail elsewhere this (Berenbaum, 1981, 1985) that the former assumption is correct only for agents low- with simple exponential dose-response curves and the latter only for agents with isk- linear curves. Levels of environmental carcinogens are generally in the low-effect fety region, where it would be difficult or impossible to distinguish between linear and 975; nonlinear curves, so the assumption of linearity is reasonable here, and in this om- case the effect of a combination of noninteracting carcinogens would closely approximate to the sum of their individual effects. However, an approach that is cin- independent of the shapes of dose-effect curves is afforded by the construction r an of isoboles (isoeffect curves or surfaces) (Loewe and Muischnek, 1926; Loewe, Dor- 1953; Berenbaum, 1981, 1985). When agents do not interact, their isoboles are lent waight lines (for combinations of two agents), flat surfaces (for combinations of rale three), and, in general, ( n - I)-dimensional hyperplanes (for combinations of /i ded ugents). less Figures IA, 2A, 3A, and 4A show vaiious types of isoboles for combinations iear of E and X where E is a set of existing environmental carcinogens (considered in the first place as if it were a single agent) and X is a new carcinogen added to 312 M . C. RI<IIENBAUM 1 W u. 0 J w > W J 0 2 4 6 a 10 0 2 4 6 a LEVEL OF X LEVEL OF X FIG. I . ( A ) lsoboles showing zero interaction between existing environmental carcinogen E and an added carcinogen X (levels in pglday). Any particular combination of levels of X and E is represented by a point with these levels as coordinates, and the incidence of cancers per IO6 exposed individuals is indicated by the isobole passing through that point. (6)The excess rate of cancers caused by adding X to a fixed level of E . In the case of zero interaction, the excess rate is independent of the level of E and is equal to the rate caused by X alone. the environment. It is assumed that E and X have linear dose-response curves, as discussed above, and that the risk of cancers per IOh exposed individuals is cqunl to the dose rate in picograms/day (for simplicity E and X are here assumed to have the same dose-response curves, but this is not essential). Now consider the situation in which the level of environmental carcinogens i q fixed, say at 2 pg/day, resulting in two cancers/106 individuals, and that we add to this different levels of the new carcinogen X. In Fig. l A , E and X do not interact, as shown by the straight isoboles. The effect of any combination of E and X is simply the sum of the effects of E and X (as expected from a zerointeracting conibination of agents with linear dose-response curves). The resulting total incidence due to 2 pg/day of E and any particular level of X is indicated by the isobole intersecting the horizontal line representing 2 pg/day of E at the appropriate level of X.Subtracting the two cancers/106 due to E gives the excess rate caused by adding X to the environment. Thus, the added effect due to different levels of X may be depicted as in Fig. 1B. Clearly, the excess due to X is proportional to the added amount of X,as expected from the statistical analyses mentioned above. If this exercise is repeated at different fixed levels of E , we find that the added effect due to X is unchanged, so the straight line function in Fig. 1B is independent of the level of E and, in fact, it is identical to the curve for X alone. Let us now suppose that E and X do interact. This is shown by nonlinear isoboles. When E and X are synergistic the isoboles are concave up, reflecting the fact that a given cancer incidence is produced by lower levels of the carcin- ogens together than would be expected from their individual dose-response F of c (cor excl cur Icv ext d0. the bY 1 bel for is0 CUI sei for car i sc re( do mz les of dU of cu - SYN ERG Y B E`I`W EEN EN V I K O N M I'N'I`A L, CA I<CI N OG E N S 313 / I X 0 c E-0.5 W 3 n c U W U U W n nW n Q . 8 en E and an represented individuals d by adding the level of 0 2 4 6 0 10 0 2 4 6 8 LEVEL OF X L E V E L OF X FIG.2. (A) lsoboles showing synergy between E and X . The production of any particular incidence of cancers requires lower levels of E and X in combination than in the case of zero interaction (compare Fig. IA). ( B ) The excess rate of cancers due to X increases with increasing levels of E and exceeds that in the case of zero interaction (compare Fig. I B ) . e curves, d u a l s is assumed iogens is t we add Y do not lion of E i a zeroThe reof X is &/day of 1 E gives :d effect e excess tat istical levels of function ie curve onlinear :fleeting ! carcinxponse curves (compare Figs. 1A and 2A). If we now calculate the effects of different levels of X added to a fixed level of 2 pg/day of E, it is evident that the curve for extra cancers due to X is no longer linear but is steeper at low levels than the dose-response curve for X alone (Fig. 2B) and that, over the whole dose range, the yield of additional cancers due to X exceeds that which would be produced by X in the absence of environmental carcinogens. Further, in contrast to the case of zero interaction, when there is synergy between E and X the slope of the curve for added risk increases as E increases, for the following reason. This slope depends on the horizontal separation of the isoboles along the line corresponding to the level of E. When the dose-response curves for E and X are linear and the isoboles are concave up, their horizontal separation will tend to decrease a s E increases; therelore the slope of the curve for added risk due to X tends to increase with increasing levels of environmental carcinogens (Fig. 2B). The effects of antagonism between E and X are shown in Figs. 3 and 4. The isoboles are concave down, reflecting the fact that a given incidence of cancer requires more of E and X together than would be expected from their individual dose-response curves. In Fig. 3A, antagonism is moderate and in Fig. 4A it is marked (the difference being that combinations of E and X in Fig. 4A are actually less effective in producing cancer than each of their constituents alone). The effect of moderate antagonism is to make the dose-response curve for added cancers due to X shallower at low doses than the curve for X alone (Fig. 3B). The effect of marked antagonism is to produce a threshold in the curve (Fig. 4B). Although it has been assumed above that E and X have linear dose-response curves, the argument applies also to nonlinear curves. For example, Brown (1977) E '7 W> w A4 !'!020 0 2 4 6 8 10 0 2 4 6 8 LEVEL OF X L E V E L OF X FIG.3. ( A ) Isoboler showing moderate antagonism between E and X . The production of any parlic- ular incidence of cancer5 requires higher levels of E and X in combination than in the case of zero interaction. (13) The excess rate of cancers due 10X decreases with increasing levels of E and is less than that in the case of Lero interaction. and Upton (1977) suggest that dose-response curves for radiation-induced cancer may be linear-quadratic rather than linear, and Hoe1 et a / . (1983) have shown how kinetic processes involved in the metabolism and cellular actions of carcin- ogens may produce "hockey stick" rather than linear curves. In both cases, the dose-response curve is shallower at low doses than at intermediate doses. The effect of synergy on dose- response curves in a linear-quadratic case is illustrated in Fig. 5 . The curves for extra risk due to X are again nonlinear, but this in itself is not of particular significance here as the curve for X on its own is also nonlinear. Of morc importance is the fact that the yield of cancers due to X is again increased over the whole dose range in the presence of environmental carcinogen and that this effect increases with the level of E . (If this exercise is performed with curves that decrease in slope with increasing dose, the reverse is found-the yield of cancers is reduced in the presence of E and the reduction increases with E . Thc biological counterpart would be experiments in the high-dose region in which carcinogenic mechanisms are becoming saturated . Such situations, fortunately. would rarely be relevant to human exposure.) In all the foregoing cases, the curve for extra risk due to X in the presence of nonzero levels of E is nonlinear. However, models can be devised in which in- teractions between X and E give rise to linear curves for extra risk. Suppose, for +example, that risks due to X and E , respectively, are given by cxlX cx2 and PIE + p2 ( a , , p , > 0), and that the combined risk is ( a l X + a 2 ) ( P I E+ pz). This function yields concave-up (synergistic) isoboles. The extra risk due to X is then ( a l X + a2 - ])(PIE+ p?),which is strictly linear in X when a2 = I (otherwise it is affine in X).The extra risk is linear in E when P2 = I . Such artificial models are of dubious biological relevance. Nevertheless, in these cases also, the extra FIG.4 general I1 cancers I a thresh( risk du risk in( A fur treated carcino not sim suppost carcinol be affec Even considei may be there is than twc appears combinz nations of four i two or tl monly v sometim of three __ A '01 / 0';"' S Y N ERGY B EI'W EE N E N VI R O N M EN'I'A L CARCINOGEN S B '01 /E-2 /E=4 315 E=O 8 any partic1st: of zero and is less d cancer e shown f carcinises, the ses. The ustrated in itself inlinear. icreased and that h curves yield of I E . The n which unatel y, jence of hich inlose, for and @ , E 2). This ' is then herwise models le extra 02 4 68 L E V E L OF X ,'LEVEL OF x FIG. 4. (A) lsoboles showing marked antagonism between E and X . Combinations of E and X generally produce fewer cancers than the constituent levels of E or X alone. ( B ) The excess rate of cancers due to X decreases with increasing levels of E and the reduction is large enough to produce a threshold in the dose-response curves of X . risk due to X in the presence of E exceeds that due to X alone, and this extra risk increases with the level of E . EFFECTS OF AGENT MULTIPLICITY A further consequence of these interactions should be pointed out. E has been treated here as if it were a single agent. In fact, the environment contains several carcinogens which may interact, so that the situation that we must consider is not simply that due to possible interactions between two carcinogens, X and a supposedly single E , but that due to adding X to an already interacting set of carcinogens. We must therefore consider how the expression of synergy might be affected by sheer multiplicily of agents. Even for combinations of only two agents, the literature shows evidence of considerable confusion as to the criteria by which interactions between agents may be examined (Berenbaum, 1981, 1985). It is not surprising, therefore, that there is very little useful information on the behaviour of combinations of more than two agents. In fact, adequate information on interactions of this complexity appears to be available only for antimicrobial agents. Here, it has been found that combinations of three antibiotics show higher degrees of synergy than combi- nations of any pair of the three (Berenbaum et d.,1983) and that combinations of four antifungal drugs show greater synergy on average than combinations of two or three (Odds, 1982). Further, whereas combinations of two antibiotics cominonly varied in the type of interactions shown, being sonictimes synergistic and sometimes antagonistic, depending on the ratio of the antibiotics, combinations of three were almost invariably synergistic, irrespective of ratios. Moreover, com- 316 ;a4 M. C. RERENBAIJM viron also I Thc cinog bet wt ther t I am showec 0 24 6a L E V E L OF X L E V E L OF X FIG.5 . ( A ) lsoboles showing synergy between E and X.(B) The dose-response curve for X (indi- cated by E = 0) is given by y = 0.4s + 0 . l . ~w~h.ere y is the number of cancers per lo6 exposed individirals and .Y the dose of X or E i n pglday. I t is assumed that E and x have the same doseresponse curve. The excess rate of cancers caused by adding X to fixed levels of E increases markedly with increasing levels of E. binations of three usually showed synergy even in cases where combinations of any two of the three were almost always antagonistic (Berenbaum et al., 1983). It appears, therefore, in the case of antimicrobial agents, that increasing the number of agents interacting increases the likelihood that the interaction will be synergistic and increases the degree of synergy shown. It cannot be assumed, without further investigation, that other classes of agents will behave similarly, but studies with combinations of two agents show that interactions are very common phenomena, irrespective of the class of agent (these include antibiotics, cancer chemotherapeutic agents, ionising radiations, immunosuppressive agents, and enzyme inhibitors, to name but a few) (Berenbaum, 1981, 1985) and there are no evident differences between these in the sorts of behaviours observed. It is therefore not unreasonable to assume that carcinogens will prove to behave similarly when they are properly investigated, and it appears sensible to assumc this until proved otherwise. CONCLUSION The main conclusions to be drawn from this analysis are that, even when the dose-response curve of a carcinogen is truly linear (or indistinguishable from linear over the relevant range), synergistic interactions between it and carcinogens already present in the environment will generally lead to its dose-response curve for extra risk being nonlinear and steeper at lower than at higher doses, so that extrapolation from a high-dose experiment would underestimate the effect of low doses. These features would be more pronounced at higher levels of en- Berenb Cll Ikre nh TI1 Uei enb COI 13rown lati Conimi EX Crump anc Iloll. R (hies s, Et1 1 Guess, lati Harnmc asst A4 nat Hoel. I: mal Hoel. I: esti Internal tior Loewe, 3, 2 Loewe, Fra Lundin, rad1 Hcc Odds, F azo1 Peto, K Et1 I' Poch i n , 22, Kodrick e 1 8 o r X (indiJ6 exposed ime dosej markedly itions of '., 1983). sing the n will be ssumed, ,imilarly, Are very libiotics, ;agents, nd there :rved. I t I behave I ahsume vhen the Ae from carcino-esponse loses, so le effect Is of en- SY N I:KGY B ETWE 13N E N V IR O N M E N U A I - C` A IiC` IN OG EN S 317 vironmental carcinogens (as long a s these are not at saturating levels). Risk is also likely to increase with the number of carcinogens present. These conclusions are clearly relevant to public policy on environmental carcinogens. They also point to the need for a systematic examination of interactions between carcinogens and of the mechanisms underlying their interactions. Neither topic has hitherto received adequate attention. ACKNOWLEDGMENTS 1 am grateful to the Medical Kesearch Council for support, and to the (unknown) referee who diowed how interactions could generate linear curves for extra risk. REFERENCES lkrenbaum, M . C. (198 I). Criteria for analyzing interactions between biologically active agents. A t h . Coticer Res. 35, 269-335. Uerenbaum, M. C. (1985). The expected effect of a combination of agents- The general solution. J . Theor. Biol., in press. Uerenbaum, M. C., Yu, V. L., and Felegie, T. I? (1983). Synergy with double and triple antibiotic combinations compared. J . Atititnicroh. Chrrtiothrr. 12, 555-563. Brown, J . M. (1977). The shape of the dose- response curve for radiation carcinogenesis: Extrapolation to low doses. Rritlitir. Res. 71, 34-50. Committee on the Biological Effects of Ionizing Radiations (1980). "The Effects on Populations of Exposure to Low Levels of Ionizing Radiation." National Academy Press, Washington, D.C. Crump, K. S., Hoel, 0. G., Langley, C. H . , and Peto, R. (1976). Fundamental carcinogenic processes and their implications for low dose risk assessment. Ctrticor Res. 36, 2973-2979. Doll, R. (1977). Strategy for detection of cancer hazards to man. NLitiiri> (Lotidoti) 265, 589-596. Guess, H. A., and Crump, K . S. (1978). Best-estimate low-dose extrapolation of carcinogenicity data. EtI\virctn. Hecilrli P c ~ . s ~ P (2'2I,. 149-152. Guess, H. A., Crump, K . S., and Peto, R. (1977). UncertLiinty estimates for low-dose-rate extrapo- lations of animal carcinogenicity data. Criticrr R e s . 37, 3475-3483. lianimond, E. C., and Selikofi', I. J. (1973). Relation of cigarette smoking to risk o f death of asbestos- associated disease among insulation workers in the United States. / t i "Biological Effects of Asbestos" (P. Bogovski, J . C. Gilson, V. Timbrell, and J. C. Wagner, Eds.), pp. 312-317. International Agency for Research on Cancer, Lyon. Hoel, I).G.* Gaylor, D.W., Kirchstein, I<. L., S;ifl'iotri. U..iincl Schneiderin;rn, M . A . (1975). Esti- mation of risks or irreversible, delaycd toxicity. J . 7`0.viid. / A i i r o t i . / / r t / l r / i I , 133- IS I. Hoel, I). G., Kaplm, N. L., and Anderson, M . W. (1983). Implication of non-linear kinetics on risk estimation in carcinogenesis. , Y c . i e t i c ~(~Wtishingroti, D . C . ) 219, 1032-1037. Inlernationul Commission on Radiological I'rotection ( IO60). "`l`lrc Evaluation of I<ihk\ I`i.om l<xlia- tion." Pergunon Press, Oxrord. Loewe, S. ( 1953). The problem of synergism and antagonism of combined drugs. ArZtie~it,i.-l;itrsL.h. 3, 285-290. Loewe, S., and Muischnek, H . (1926). Uber Kombin~itionswirktrngen:2. Mitteilung: Hilfsniitlel der Fragestellung, Arc,/i. Evp. Ptirliol. Plrrirt?itikol. 114, 313-326. Lundin, F. E . , Jr., Lloyd, J . W., and Smith, E. M . (1969). Mortality of uranium miners in relation to radiation exposure, hard-rock mining and cigarette smoking-1950 through September 1967. HerilIh P ~ I Y S16. , 571-578. Odds, F, C. (1982). Interactions among amphotericin B, 5-fluorocytosine, ketoconazole, and niiconarole against pathogenic fungi i / i virro. Atiriniicroh. Agrtirs Chetnotlier. 22, 763-770. Rto, K. (1978). Carcinogenic effects of chronic exposure to very low levels of toxic substances. Etiviroti. H r r r l r t i P i w p r c r . 22, 155- 159. Pochin, E. E. (1978). Assumption of linearity in dose-effect relationships. Ltiii/~)tIil.c~trllliI'wspcJct. 22, 103-105. Kodricks, J. V. (1981). Regulation of carcinogens in food. Atiti. N . Y . A(,rrd. Sci. 363, 29-35. 318 M . C. f\liRI<NRAUM Kothman, K., and Keller, A. (1972). The effect ofjoint exposure to alcohol and tobacco on risk of cancer of the mouth and pharynx. J. Ciiroii. Dis. 25, 711-716. Saracci, K. (1977). Asbestos and lung cancer: A n analysis of the epidemiological evidence on the asbestos-smoking interaction. In(. J. Cuncw 20, 323-331. Schneiderman. M. A., and Brown, C. C. (1978). Estimating cancer risks to a population. Emirorr. H d t h Pcrspccr. 22, I IS- 124. Selikoff, I . J., and Hammond, E. C. (1975). Multiple risk factors in environmental cancer. In "Persons at High Kisk of Cancer" (J. E Fraumeni, Jr., Ed.), pp. 467-483. Academic Press, New York. Selikoff, I . J., Hammond, E. C., and Churg, J. (IY68). Asbestos exposure, smoking and neoplasia. .I. A ~ t l ~Mr .ed. ASSOC2. 04, 104- I 10. Seiikoff. I. J.. Seldman, H., and Hammond, E. C. (1980). Mortality effects ofcigaretle smokingamong amositc asbestos factory workers. ./. Ntirl. Criricer Inst. 65, 507-5 13. Upton, A. C. ( 1977). Radiological effects of low doses: Implications for radiological protection. Ntr- dicir. Res. 71, 51-74. KNVIKONMI Met: < j I)cpurtiii I'leur (Metso 11 kit ion were n inhabit; calcific 28.6%' I develor caiise (1 identific dusts ill from 8 respons and bel siniilwi t o the k in the e Diffuse to asbesto: in nature, (Kiviluotc (Navratil : nonoccup: tinrelated t In all of t l ron men taI + sepiolitl sotile in Ti f This stud Scicnces. ES * Ib whom