Int Arch Occup Environ Health ( 1996) 68:484-494

Daniel M Byrd · M Luann RoegnerJames C Griffiths · Steven H LammKaren S Grumski · Richard Wilson · Shenghan Lai

Carcinogenic risks of inorganic arsenic in perspective

Abstract Induction of cancer by inorganic arsenic occursinconsistently between species and between routes of ex-posure, and it exhibits different dose-response relation-ships between different target organs Inhaled or ingestedarsenic causes cancer in humans but not in other species.Inhaled arsenic primarily induces lung cancer, whereas in-gested arsenic induces cancer at multiple sites, includingthe skin and various other organs Cancer potency appearsto vary by route of exposure (ingestion or inhalation) andby organ site, and increases markedly at higher exposuresin some instances To understand what might explain

these inconsistencies, we reviewed several hypothesesabout the mechanism of cancer induction by arsenic Ar-senic disposition does not provide satisfactory explana-tions Induction of cell proliferation by arsenic is a mech-anism of carcinogenesis that is biologically plausible andcompatible with differential effects for species or differ-ential dose rates for organ sites The presence of other car-cinogens, or risk modifiers, at levels that correlate with ar-senic in drinking water supplies, may be a factor in allthree inconsistencies: interspecies specificity, organ sensi-tivity to route of administration, and organ sensitivity todose rate.

Work presented at the " 23rd Congress on Occupational and Envi-ronmental Health in the Chemical Industry" (Medichem 1995)"The Chemical Industry as a Global Citizen Balancing Risksand Benefits", 19-22 September 1995, Massachusetts Institute ofTechnology, Cambridge Massachusetts

D M Byrd, R Wilson, S Lai and S H Lamm are membersof the Inner Mongolia Cooperative Arsenic Project

D M Byrd () M L RoegnerConsultants in Toxicology,Risk Assessment and Product Safety, Suite 1150,1225 New York Ave , NW, Washington, DC 20005, USAFax ( 202) 484-6019

J C GriffithsInternational Specialty Products, 1361 Alps Road,Wayne, NJ 07470, USAFax ( 201) 628-4180

S H Lamm · K S GrumskiConsultants in Epidemiology and Occupational Health, Inc ,2428 Wisconsin Ave , NW, Washington, DC 20007, USAFax ( 202) 333-2239

R WilsonLyman Laboratory of Physics ( 231), Harvard University,11 Oxford Street, Cambridge, MA 02138, USAFax ( 617) 495-0416

S LaiBiostatistical Working Group, University of Miami,Room 2001 (D-91), 1800 NW 10th Avenue, Miami,FL 33136, USAFax ( 305) 279-7815

Introduction

At present, the scientific consensus is that arsenic inges-tion causes human skin cancer lIl and that arsenic inhala-tion causes human lung cancer l 2 l, but neither route of ex-posure causes cancer in other species l 3, 4 l Inhaled ar-senic primarily induces lung cancer, whereas ingested ar-senic notably induces skin cancer but also is associatedwith leukemias and cancers of the bladder, lung, kidney,gastrointestinal tract, liver, and other organs l 5 l Becauseinhaled arsenic is absorbed into the general circulation,distributed, metabolized, and eliminated through urinel 6 l, the relative absence of leukemias or cancers at sitessuch as skin, bladder, and kidney after inhalation appearsto contradict the ingestion data.

Studies of drinking water supplies with high arseniclevels in India, Japan, Mexico, Inner Mongolia, and SouthAmerica have extended the general observations of an as-sociation between ingestion and skin cancer l 7, 8, 9, 10,11 l Many studies of arsenic ingestion lack quantitativeexposure data, however, so that quantitative evaluationsof ingested arsenic as a skin carcinogen usually depend ona single report by W P Tseng and coworkers l 12 l.

Tseng studied an area in Southwestern Taiwan withwell water as the primary source of drinking water Manyof these wells had high concentrations of arsenic Resi-dents had a high incidence of several diseases, most no-

Springer-Verlag 1996

485

tably blackfoot disease, a vascular occlusion of the ex-tremities, and dermatological conditions, including skincancer Prevalence correlated with dose rate of arsenic andwith age, which was a surrogate measure of duration ofexposure, because of lifetime exposure l 12 l ProfessorChien-Jen Chen and his coworkers later augmentedTseng's observations by expanding the cohort size and bystudying other diseases, including goiter and hepatitis B,in the same cohort l 13, 14, 15 l Through detailed studiesof death certificates, they found that the mortalities fromliver, kidney, and bladder, and even lung cancer, also cor-related with arsenic dose rate and duration of exposurel 16 l.

The lung cancer potency of inhaled arsenic appears toincrease with dose rate l 2 l In this paper we similarly an-alyze the dose rate dependence for skin cancer potency ofingested arsenic, using the Tseng data, and review poten-tial explanations for apparent inconsistencies in inter-species sensitivity to arsenic, organ sensitivity to route ofexposure, and organ sensitivity to dose rate.

Methods of analysis

We identified all data from the published scientific literature thatquantitatively relate skin cancer prevalence to arsenic ingestion l 9,11, 12, 17, 18, 19, 20, 21 l and applied several modeling ap-proaches The terms in the models are defined as follows:

y = prevalence, f = some function, D = dose rate, Dg = a fixed doserate (> 0), and q O, ql, q 2, qn, = constants (unknown parameters).

Linear models

y = q O + q 11D (q O> 0, q 0)

A linear process reflects direct proportionality; i e the increase inprevalence for an incremental increase in dose is constant over alldose rates.

Segmented linear models

y = q + q D, if O < D Dg, andy = q O + q 1 D + q 2(D Dg), if D > D,where Dg is a fixed dose, Dg> O and q O 2 0, q, 2 0, and q 2

> 0.

A segmented linear model describes a combination of two directproportionality processes, at least one of which has no threshold.These models can easily be extended to more than two processes.This model describes a biological mechanism in which different,independent, concurrent processes act to increase prevalence, but

some of these processes do not contribute until some level of ex-posure is reached.

Truncated polynomial fit models

Another possibility is to expand the parameter y into a power se-ries in dose These models sometimes are described as truncatedpolynomial fit models A modification must be made to accountfor the fact that no one dies more than once The formula becomes

y = 1 e (qO + q lD

+q 2 D 2 + qn Dn)

where N is constrained to one less than the observed number of doses,because of the limited degrees of freedom We used MSTAGE, amodified power law expansion model created by Dr EdmundCrouch to calculate power law expansion models and the variousmultistage models described below The error bars in the figuresdescribe the statistical uncertainty that is attributable to the numberof persons with cancer (or cancer fatalities).

Multistage models

Multistage models are biological models, often attributed to Ar-mitage and Doll l 22 l, although many other persons contributed totheir earlier development Multistage models arise by assumingthat there are a number of sequential biological steps, each requir-ing time, that must take place before a cancer develops Severalvariants of the multistage model are identical to truncated polyno-mial fit models, as above, but with positive coefficients for eachterm in the power series This constraint leads to a response that in-creases in a strictly monotonic way as a function of increasingdose.

Linearized multistage models

The form of the linearized multistage preferred by the U S Envi-ronmental Protection Agency (EPA) uses a truncated polynomialfit model with an upper limit in some statistical sense on the coef-ficient of the linear slope (dose term ql) with the other coefficientsheld constant l 23, 24 l We follow this procedure with the MSTAGEprogram but allow the other coefficients to vary to obtain the bestfit overall.

Results

The eight studies of arsenic ingestion we found in the lit-erature (Table 1) yield a dose-response relationship forskin cancer induction in which prevalence increases withdose rate (Table 2, Fig 1) The risk of skin cancer appearslinearly related to arsenic dose rate above a discontinuityor threshold in the range of approximately 100 ppb With-in the overall data, the original data from Tseng l 12 l are

Table 1 Studies on ingestedarsenic and skin cancer Author Year Place Exposure Dose Pop Cancers

levels

Fierz 1965 Germany Fowler's Solution ? 262 21Tseng 1968 Taiwan Well Water 4 47,921 428Goldsmith 1972 California Well Water 1 92 0Zalvidar 1974 Chile Water 1 ? ?Tay 1975 Singapore Herbal Medicines ? 74 6Harrington 1978 Alaska Well Water 1 211 OCebrian 1983 Mexico Water 2 614 4Southwick 1979 Utah Well Water 1 250 0

486

Table 2 Dose-response data from(Fig 1 plots these data)

studies of arsenic ingestion

Author Year Dose Popu Cancers Preva-rate lation lence (%)(ppb)

Tseng 1968 785 8251 185 2 20Tseng 1968 473 5413 60 1 10Cebrian 1983 410 296 4 1 40Harrington 1978 224 211 0 0 00Southwick 1979 200 250 0 0 00Tseng 1968 171 9526 21 0 20Goldsmith 1972 120 92 0 0 00Tseng 1968 5 7500 0 0 00Cebrian 1983 5 318 0 0 00

Table 3 Sample weighted means for exposure intervals in thestudy of Tseng et al l 12 l

D (ppb) Mid-point Wtd mean

Low dose 0-300 150 171Medium dose 300-600 450 473High dose > 600 1200 785

consistent with the overall data and have the largest popu-lation size These exposure data are ecological averagesrather than individual exposures (Table 3).

The assignment of dose (or dose rate) is a crucial stepin modeling In this paper the concentration of arsenic inwell water serves as a surrogate for dose rate The pre-ferred approach would use population weighted means Inthe absence of data that allow the calculation of popula-tion weighted means, we used sample weighted meanswith mid-ranges for closed exposure intervals and an ap-

Fig 1 Skin cancer prevalencerates as a function of arsenicconcentration in drinkingwater for all literature Theline represents a best fit toall data using the MSTAGEprogram with a threshold of103 ppb ly = 3 23 E-05 (D)

3 33 E-03 for D > 103 ;P value = 0 80 l The squaresymbols illustrate the datafrom Tseng l 12 l; the trian-gles all other data (seeTable 2)

2.5 %

2.0 %-

a0Cco

a0

CUQc_NW

.5 %-

1.0 % -

0.5 %

0.0 %0 100

proximation for the first quartile, which is in an open ex-posure interval, because its lower bound is not zero but isunknown Table 3 describes the application of this strat-egy to the arsenic levels in well water and skin cancerprevalence in Tseng's study l 12 l.

Because of the discontinuity in the dose-response rela-tionship, any linear model will fit Tseng's data poorly(Fig 2) The error bars in Fig 2 represent one standard de-viation Alternatively, it is possible to fit Tseng's data witha single straight line and a threshold Assuming a thresh-old at 100 ppb, a slope of 3 14 x 10 - 5 x D provides a bestfit to the data with a correlation coefficient of approxi-mately 0 9 A segmented linear model of the data im-proves on this fit We obtained a segmented linear modelwith two successive slopes of 9 2 x 106 x D and 2 2 x10-5 x (D-143), in units of case prevalence rate per ppb ar-senic in drinking water (ppb- 1).

We initially hoped to link each of these two slopes tounderlying biological processes but could not find ade-quate experimental data to justify such a procedure Thereis a problem in extracting two linear relationships from adata set with only four data points Because of limitationson the available degrees of freedom, it is not possible totell if, for example, three independent process might beinvolved Our efforts to separate these slopes using a sta-tistical justification, through regression on a segmentedlinear model, failed Slightly different, but reasonable, ap-proaches to estimate the first slope, generated a variety ofthresholds between 100 to 200 ppb Lacking a biologicaljustification at present for two independent processes, wereserved investigation of segmented linear models forlater research.

Unlike the linear or segmented linear models, whichonly utilize dose rate as the explanatory variable, the mul-

200 300 400 500

Arsenic concentration in ppb (g/I)600 700 800

A

/.

/II

/E

I

II

II

I

II

I

I

I,

1

_ ie

Fig 2 Multistage model ofoverall skin cancer prevalenceas a function of arsenic doserate in the study by Tseng andcoworkers l 12 l best fit,

95 % limit on ql

Table 4 Drinking water arsenic levels and skin cancer prevalencein the study by Tseng et al l 12 l

Exposure level As dose Popu Cases Preva-rate (ppb) lation lence (%)

Background 5 7,500 0 0 00

Low exposure 171 9,526 21 0 22

Medium exposure 473 5,413 60 1 11

High exposure 785 8,251 185 2 24

Table 5 Time distribution ofskin cancer prevalence in Tai-wanese cohort

a = hypothetical; b = calculated

Fig 3 Multistage model ofmale skin cancer mortality asa function of arsenic dose ratein the study by Wu and co-workers l 16 l -best fit,

95 % limit on ql

tistage model permits consideration of age or duration ofexposure, because model coefficients represent the timesfor events to occur Since lifetime exposure to well wateroccurred within Tseng's cohort, age is a measure of dura-tion of exposure Tseng's paper provides age data both asage intervals, assuming a maximum age of 100 years, andas lifetime exposures, assuming a 76 year mean lifetime.We recalculated age-specific skin cancer prevalence ratesfrom these data, as shown in Table 4 We assume that the

Age 0-19 20-39 40-59 60 + Lifetime Average

t (years) 10 30 50 70 76 a 40 b

In (t) 2 3 3 4 3 9 4 25 4 33 3 7Cases 0 31 197 208 436

Persons at risk 22,813 9,527 6,146 2,020 40,506

y = prevalence < 0 00013 0 00325 0 032 0 103 0 1 5 7 b 0 0108

ooo

do

0

0'.4

487

8V

488

age dependence of prevalence and the time dependence ofprevalence are approximately separable.

The time distribution of skin cancer prevalence is shownin Table 5 The rate of change in prevalence with age inthis information describes a process in which prevalenceincreases with the fourth power of age From these datawe also estimated a cumulative lifetime risk ( 0 157), whichresembles the risk for the elderly ( 0 103), an order ofmagnitude greater than the hypothetical average risk( 0.0108) These data demonstrate that the skin cancerprevalence risk in Tseng's cohort was markedly age (orduration) dependent This nonlinear relationship is bestapproximated as a carcinogenic process with four stages.Fig 2 illustrates the overall prevalence of skin cancer as afunction of dose rate.

We used the subsequent data on mortality l 16 l and ar-senic levels in well water measured during 1964-1966l 12 l, to analyze internal organ cancers Arsenic levelsranged from 10 to 1,750 ppb with two clusters at 50 to

Fig 4 Multistage model offemale fatal skin cancers as afunction of arsenic dose rate inthe study by Wu and cowork-ers l 16 l best fit, 95 %limit on ql

000

-4o

43

250 ppb and 450 to 650 ppb Chen and coworkers classi-fied villages by median arsenic levels of well water andcategorized exposures into three groups, corresponding tothose used by Tseng: less than 300 ppb, 300 to 590 ppb,and above 600 ppb.

To calculate multistage models, we took the number ofcancers observed as the outcome numerator, R, and thenumber of cases expected, N, as the denominator, per unitof time, such that R/N was the rate Figures 3 and 4 dis-play the data and models for fatal skin cancer cases inmales and females respectively Because mortality wasapproximately 10 % of prevalence for skin cancer, thenumber of cases in these figures are lower than in Fig 2.Although statistical accuracy decreased, there is generalagreement with the nonlinearity of skin cancer prevalencewith arsenic dose rate Arsenic apparently has lower po-tency in inducing skin cancer at low dose rates than athigh dose rates Figures 5 and 6 show the data and modelsfor fatal kidney cancer cases in males and females respec-

arsenic level (ppb)

Fig 5 Multistage model ofmale fatal kidney cancers as afunction of arsenic dose rate inthe study by Wu and cowork-ers l 16 l best fit, 95 %limit on q I

best fit: 95 % limit on q Iq O= 6 602 e-6 q O= 5 824 e-6ql= 3 821 e-7 ql < 5 287 e-7

100 200 300 4110 500 800 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

1 D o 200 30 40 500 60 D 700 800arsenic level (ppb)

000o

o 200-

; 150-

0

,, 50-l 1

U

900 1000-

-

I

Fig 6 Multistage model offemale fatal kidney cancers asa function of arsenic dose ratein the study by Wu and co-workers l 16 l best fit,

95 % limit on ql

Fig 7 Multistage model ofmale fatal bladder cancers asa function of arsenic dose ratein the study by Wu and co-workers l 16 l best fit,

95 % limit on ql

Fig 8 Multistage model offemale fatal bladder cancers asa function of arsenic dose ratein the study by Wu and co-workers l 16 l best fit,

95 % limit on ql

arsenic level (ppb)

000

d0'-4

S.

.

Id

489

00(o

0o0 o

'd,-

00a

d0o'-4

.,.J

1 I

a

Pl

490

tively, and Figs 7 and 8 show the data and models for fa-tal bladder cancer cases in males and females respectively.Despite the small numbers of cases, only small differ-ences were found between the linearized and best fittingmultistage models, with a best fit and with the upper 95thpercentile for the linear term In contrast to the findingsfor skin cancer, the internal cancer cases exhibit a directlyproportional relationship with arsenic dose rate, except forfemale kidney cancers, and even these have less curvaturethan the female skin cancer cases.

Discussion

At least five hypotheses could explain different dose-re-sponse relationships for cancers of skin versus internal or-gans in the same cohort of persons exposed to arsenic Insummary, (a) nonarsenical substances associated with ex-posure to arsenic, or various metabolites of arsenic, mightfunction as proximal carcinogens for different organ sites.(b) Some level of exposure might saturate a limited hu-man capacity to metabolize arsenic, so measures of doserate poorly reflect the dose of arsenic delivered to tissues.(c) Arsenic might accelerate a later stage of the carcino-genic process in skin than in internal organs (d) Measuresof dose rate might reflect tissue insult differently in dif-ferent tumors because of nonlinearities in the biologicalmechanism that relates delivered dose of arsenic to the ex-pression of cancer (e) Irregularities might exist in the Tai-wanese data.

Different proximal carcinogens for different organs

F.J Lu has pointed out that the well water consumed bythe Taiwanese cohort contained many substances besidesarsenic, and some of them correlate with arsenic concen-tration l 25 l The water supply in Inner Mongolia has sim-ilar characteristics yet there are differences in health out-comes For example, no blackfoot disease is seen in Mon-golia although other symptoms of chronic arsenicism arefound l 10 l High levels of organic materials, includinghumic acids occurred in the Taiwanese well water In thepresence of arsenic, or other transition metals capable ofreadily changing valence, which serve as catalysts, humicacids undergo a well-described reaction and generate flu-orescent and/or highly mutagenic substances Thus, ar-senic levels will correlate with the levels of the mutagenichumic acid byproducts.

Lu l 25 l initially investigated the possibility that thesehumic acid byproducts caused blackfoot disease How-ever, his preliminary results suggest that the humic acidbyproducts are potent bladder carcinogens If inorganicarsenic only induces human skin and lung cancers, then itbecomes easier to understand why bladder cancer appar-ently occurs only after ingestion of drinking water, sinceinhalation studies involve arsenic exposures in the ab-sence of humic acids.

Apparently, no epidemiology studies of arsenic indrinking water have systematically characterized the pres-

ence of humic acids or humic acid byproducts, or exam-ined whether these substances account for the associationwith bladder cancer (or other cancers) more strongly thanarsenic Further, we see no reason to limit the search forconfounding exposures to humic acid byproducts For ex-ample, selenium and fluoride levels generally correlatewith arsenic levels in groundwater, and these substancescan modify the effects of arsenic l 26 l The presence ofother carcinogenic substances, or modifiers of carcino-genesis, at levels that correlate with arsenic in drinkingwater, might provide satisfactory explanations for the in-terspecies and interorgan inconsistencies in arsenic car-cinogenesis In theory, substances in drinking water couldsuppress the lung cancer response observed after arsenicinhalation, while other substances could induce cancers ofskin and internal organs.

Saturation of metabolism

Humans are directly exposed to arsenite (As+ 3), the re-duced form of inorganic arsenic, as well as arsenate(As+ 5), the oxidized form, depending on the oxidation/re-duction state of their water supply, whereas most bioas-says tested only arsenate Species differences might re-flect differential exposure to arsenite Indeed, the U S.National Toxicology Program has a bioassay of arsenite inprogress.

Arsenate is less acutely toxic than arsenite Arsenateabsorbs better than arsenite, possibly because arsenate re-acts less with gastro-intestinal tract membranes l 27 l Inbiological systems, however, arsenate and arsenite inter-convert freely, depending on the oxidation-reduction statein the gastrointestinal tract and body Higher arsenic lev-els in drinking water will directly alter the reducing ca-pacity of the water Mammals also excrete arsenic intobile with greater appearance after arsenite than arsenateadministration In either case, the gastrointestinal tractrapidly and efficiently reabsorbs arsenic from bile l 28 l.

At a whole body level, the distribution of absorbed ar-senate depends on clearance from blood Man, dog, mouse,and rabbit clear absorbed arsenate rapidly, with 90 % ofthe administered arsenic disappearing with a half-life ofone to two hours, followed by a second phase of clearancewith a half-life of approximately thirty hours and a thirdphase clearance with a half-life of approximately twohundred hours l 28 l In contrast, rats accumulate arsenic inblood, through binding to red blood cells, and exhibit ahalf-life of sixty to ninety days l 28 l Absorbed arsenite re-acts directly with thiol groups in plasma Humans at au-topsy, or rabbits and mice exposed to arsenite or arsenate,have elevated levels of arsenic in liver, kidney, lung, andintestinal mucosa l 29 l.

In mice, uneliminated arsenic accumulates in bone,kidney cortex, intestinal mucosa, and hair follicles l 30 l.Arsenate is isosteric and isoelectronic with phosphate, re-sulting in arsenate substitution for phosphate One conse-quence of this exchange is the distribution of arsenic intobone matrix, specifically into apatite crystals l 31 l Fur-

thermore, arsenate can function as a substrate for manyenzymes in place of phosphate l 32 l.

When dose rate is modeled, dietary intake may not in-fluence parameters, because drinking water levels maygenerally correlate with dietary levels Often, drinking wa-ter also supplies local agriculture, resulting in a feedbackloop Thus, drinking water dose rate could serve as an in-dicator of overall arsenic exposure but still not constitutethe major source of arsenic Clearly, exposure throughdrinking water ingestion can have dramatic effects on therisk of skin cancer, although only approximately one-thirdof total arsenic intake usually arises from drinking water.

At a cellular level, arsenate rapidly accumulates withincells, via a phosphate pump l 33 l All animals rapidly re-duce intracellular arsenate to arsenite; glutathione appearsresponsible for almost all such reduction l 34 l Intracellu-lar inorganic arsenic either enters the phosphate pool asarsenate or conjugates with glutathione as arsenite Ap-proximately one-third of absorbed arsenic binds to thiolgroups of structural proteins in the form of arsenite, andreducing agents, such as mercaptoethanol, can release itinto solution l 35 l The overall effect is to trap arsenicwithin cells in the vicinity of the site of absorption, but thespecificity of uptake is influenced by tissues rich in thiolgroups, i e , keratin-rich tissues, such as skin, hair, or cu-ticle.

The major metabolites of inorganic arsenic are mono-methyl arsenate (MMA) and dimethyl arsenate (DMA;cacodylic acid) The biochemistry of arsenic methylationis poorly understood, but it appears to involve a gluta-thione-arsenite complex as substrate l 35 l All three arsenicspecies (inorganic, MMA, and DMA) are excreted in urineby humans Most evidence supports the hypothesis thatthe process of arsenic methylation is primarily a detoxify-ing process l 4 l, however there is evidence that DMA maybe a proximal toxicant for some endpoints l 34 l.

DMA apparently induces tetraploidy in cell culturesl 36 l DMA also enhances kidney carcinogenesis l 37 l andbladder carcinogenesis l 38 l in animal bioassays Yama-naka and coworkers l 39 l speculate that tumor promotionby DMA may result from DNA strand scission by oxygenradicals arising from the reaction of DMA with molecularoxygen.

The extent of methylation in intact animals depends onthe valence of the arsenic administered Arsenite yields agreater degree of methylation in rats and mice than arsen-ate l 28 l Since arsenite appeared before DMA in the urineof rabbits given intravenous arsenate, arsenate must be re-duced to arsenite prior to methylation l 40 l Methylationalso depends on the species l 28, 29 l Marmoset monkeysare the only species so far identified that show virtually nomethylation of inorganic arsenic Mice and rabbits alsomethylate inorganic arsenic In both species almost all ar-senic is excreted as DMA.

While it is easy to understand how different arsenicmetabolites could serve as proximal carcinogens at differ-ent organ sites, and while it is easy to understand that dif-ferential localization of arsenic after different routes ofexposure could lead to different patterns of cancer induc-

491

tion, it remains difficult to explain the absence of evi-dence that skin, bladder, or kidney cancers occur after ar-senic inhalation, solely on the basis of metabolism In-gested arsenic will undergo first pass metabolism throughthe liver before entering the general circulation, whereasapproximately one-third of the arsenic absorbed after in-halation passes through the liver, but this difference cannot account for the inconsistencies in organ-site patternbetween inhalation versus ingestion epidemiology studies.

The hypothesis has been advanced that humans havelimited metabolic capacity for arsenic methylation and,thus, that the nonlinearity in the dose-response relation-ship for skin cancer reflects saturation of detoxificationl 41 l The idea is that human liver methylates most arsenicbelow some dose rate, and, when methylation capacity issaturated, inorganic arsenic is delivered to other tissues.The misconception that saturation of detoxification wouldlead to a threshold, below which arsenic would not inducecancer, probably has generated more attention for this hy-pothesis than it deserves Saturation of detoxification ac-tually implies a shift in potency at lower doses, but not athreshold.

The data analyzed in this paper suggest that kidney andbladder cancers apparently have linear dose-response re-lationships, but skin cancer prevalence is a highly nonlin-ear function of arsenic dose rate It is difficult to under-stand how arsenic metabolism could saturate among somemembers of a cohort who develop skin cancer, but not sat-urate for persons in the same cohort, who develop internalorgan cancers No good evidence exists that only methyl-ated forms of arsenic circulate in humans, but this is atestable hypothesis In addition, the hypothesis of sat-urable human methylation capacity is at variance with theidea that arsenic metabolism occurs on a more regionalbasis, depending on route of administration, for example.

Oral administration of arsenate to humans, mice, anddogs resulted in humans eliminating 68 % of the dose inone week and mice and dogs eliminating 99 % of the dosein a week l 28 l Humans are the only species to excreteMMA in addition to DMA l 29 l, and this evidence is citedin support of saturable human methylation capacity Thereis, however, no good evidence that the urinary excretionof inorganic arsenic, MMA, or DMA reflects metaboliccapacity instead of renal handling.

If high arsenic levels do saturate human methylationcapacity, not renal handling of metabolites, methylation ofDMA should plateau at higher arsenic exposures, and theratio of MMA to DMA should dramatically increase Nogood quality evidence demonstrates such an excretionpattern Excretion patterns of arsenic metabolites also aredifficult to interpret, because methylation capacity reflectsdietary status A high dietary arsenic load could depletemetabolic reserves of glutathione, shifting the arsenate toarsenite level, and diminishing the precursor for methyla-tion Thus, higher arsenic exposures might well lead tohigher arsenite exposures.

Valentine and coworkers l 42 l measured total arseniclevels in human blood, urine, and hair in five differentcommunities, each with different and variable arsenic lev-

492

els in drinking water They found increases of arsenic inurine and hair with increasing drinking water concentra-tions over a range from 6 to 393 ppb, but no increase inblood concentration until drinking water levels reachedapproximately 100 ppb The correlations were establishedusing group averages, instead of data for each individualin the cohort No correction was made for the contributionof arsenic from food Individual arsenic metabolites werenot measured In contrast, Smith and coworkers l 6 l didmeasure levels of individual arsenic metabolites in theurine of workers exposed after inhalation of arsenic triox-ide They found no plateau in DMA excretion with in-creasing urinary output, and no notable change in MMAto DMA ratio occurred.

If rodents had much greater methylation capacitiesthan humans, higher arsenic levels in bioassays would notsaturate animal metabolism, and detoxification could ex-plain the inconsistency in species specificity of arsenic asa carcinogen Unfortunately for this hypothesis, rats ex-crete only 4 % of the administered inorganic arsenic asmethylated forms of arsenic l 28 l, yet rats do not developtumors in response to arsenic exposure Several labs aredeveloping physiological pharmacokinetic models to de-scribe the absorption, distribution, metabolism and excre-tion of arsenate, arsenite, MMA, and DMA Because somearsenic metabolites bind covalently to structural macro-molecules and/or undergo intracellular sequestration, con-struction of adequate deposition models for each humanorgan likely will prove difficult Eventually, better qualitydata and good models may resolve the questions about therole of arsenic disposition in carcinogenesis.

Overall, neither hypotheses that invoke saturation ofhuman methylation capacity nor inconsistencies in theTaiwanese data explain the inconsistencies in the induc-tion of cancer by arsenic The disposition of arsenic in-evitably plays a role in the induction of cancer and meritsmore study, especially the idea of regional metabolismand distribution It is worth recalling that most inhaled ar-senic is in particulate form, which may help explain someof the inconsistencies of effects after different routes ofexposure.

Late stage carcinogenesis

Application of the multistage model to lung cancer datafrom a cohort of copper smelter workers suggested thatarsenic acted on a late stage l 43 l The analyses of skin,but not bladder or kidney, cancer data in this paper areconsistent with this hypothesis, because late stage car-cinogens exhibit highly nonlinear dose-response relation-ships Thus, arsenic might accelerate a late stage of thecarcinogenic process in skin but an early stage in kidneyor bladder If the late stage is irrelevant to the carcino-genic process in rodents, or if rodents lack initiated cells,late stage carcinogenesis also might explain the speciesspecificity of arsenic carcinogenesis Late stage carcino-genesis does not, however, explain the exposure depen-dent pattern of organ sites in humans.

Nonlinear dependence of mechanism on delivered dose

Measures of dose rate may poorly reflect carcinogenic po-tency, because arsenic acts on some enzymatic or recep-tor-mediated process which has a nonlinear dose-responserelationship The leading candidate for such a process isstimulation of cell proliferation Arsenic is not mutagenicin the traditional sense of either generating DNA adductsor inducing revertants at specific loci l 44 l Arsenic doesinduce both cell proliferation and clastogenic events inSyrian hamster embryo cells l 45 l Arsenic does not, how-ever, appear to induce cancer in Syrian hamsters.

The mechanism of arsenic-induced cell proliferation isnot yet known Following exposure of intact rodents orcells in culture to arsenic, induction of heat shock proteinsand metallothionine occurs l 46 l These effects precedemore obvious manifestations of chronic cell injury It maybe that arsenic modifies the release of some factor thatcontrols cell proliferation or changes the activity of thisfactor once bound to cells.

Cell proliferation alone can provide a sufficient expla-nation of carcinogenesis l 47 l Such a mechanism has im-portant consequences for the dose-response relationship.If arsenic promotes the transformation of previously initi-ated cells through the induction of cell proliferation, anonlinear dose-response relationship will result, similar tothat observed in Fig 2 If cell proliferation alone inducescancer, by increasing the number of cells at risk of spon-taneous mutation, a true threshold should occur, but anymechanism involving a simultaneous increase in the ini-tial mutation rate by arsenic will not generate a threshold.

Aberrant cell proliferation can lead to abnormal mito-sis, resulting in chromosomal abnormalities Clastogene-sis alone can induce cancer The likely mechanism is theloss of suppressor genes after inactivation during chromo-somal rearrangement Thus, hypotheses about arsenic-in-duced cancer that depend on late stage carcinogenesis orcell proliferation are not exclusive of each other.

Keratoses, a definitive sign of aberrant cell prolifera-tion, were noted at earlier times and in a higher proportionof the arsenic exposed persons in the Taiwanese cohortl 12 l Further, skin cancer incidence did not occur inde-pendently of the incidence of keratoses Like skin cancer,the prevalence of keratoses also correlated with age Per-sons with keratotic lesions were at much greater risk ofskin cancer The keratoses in the cohort were neitherquantitatively attributed to specific individuals, who ei-ther did or did not get skin cancer, nor were shown to pre-cede skin cancers at the same anatomical sites, so a pre-cursor-successor relationship was not established Clasto-genic effects also have been noted in humans exposed toarsenic l 48 l A mechanism of arsenic-induced human can-cer that involves cell proliferation would explain bothspecies specificity and different dose-response relation-ships for different organ sites, but it is difficult to see whydifferent routes of exposure would generate different pat-terns of cancer, based on this mechanism.

493

Data irregularities

Some anomalies have surfaced in the Chen data l 49, 50 l.Wide variations in arsenic levels occurred at several vil-lages and excessively high mortality rates attributed to ar-senic were observed in some villages with only low ar-senic levels in the drinking water Thus, some distortionof the dose-response relationship is likely, and this mis-classification probably applies to the Tseng study of skincancer as well, since it used the same exposure data Be-cause the arsenic levels in the water supply varied acrosswells in the same village and temporally in the same well,even individual exposure specifications represent aver-ages The death certificate data for the Chen cohort appearsound, but similar reservations apply to them as for all ret-rospective studies based on death certificates For this rea-son, it seems unwise to place much confidence in the ex-act parameters of any model fit to similar epidemiologydata.

The effect of exposure reclassification would be tomove some individuals from one exposure category to an-other If this occurred at random, our conclusion, that thedose-response relationship for skin cancer differed fromthe dose-response relationships for internal organ cancers,would not change Only if, for example, deaths related toskin cancer in the highest exposure category were selec-tively misclassified from the lowest exposure category,would the relative differences, and our conclusions,change This seems unlikely Quantitatively, the relation-ship between prevalence and exposure for Tseng's studyis consistent with all other prevalence studies (Fig 1) TheTaiwanese drinking water supply was remediated in themid-1950 's, and it may now be impossible to establishprecise quantitative exposures While we have no prob-lem with efforts to evaluate the adequacy of the data (or torefine them) retrospectively, for purposes of risk assess-ment, our response to recent criticism of the Chen andTseng studies is that they neither undertook nor publisheda risk assessment Scientifically, their work appearssound, because independent studies have replicated theirmain conclusions: prevalence correlates with dose rateand with duration of exposure The exposure data appear,however, too imprecise for reliable risk estimation.

We are continuing our evaluation of dose-response re-lationships for internal cancers in the Taiwanese cohort,using the more detailed information compiled by C-J.Chen The Taiwanese cohort no longer is exposed, andthus little biological sampling is possible In addition, in-dependent replication of the studies with an unrelated co-hort is always desirable For these reasons, we hope tostudy a new population in Inner Mongolia l 10 l.

Arsenic-associated cancer remains a significant world-wide public health problem There are reasons to believethat arsenic still contributes significantly to overall U S.cancer risk, despite the high quality of U S drinking wa-ter This belief has two different bases One is that ar-senic-induced skin cancer is biologically different andmore lethal than sunlight-induced skin cancer, but thecases of arsenic-induced skin cancer get lost in the much

higher prevalence of sunlight-induced skin cancer Theother belief is that arsenic may account for a substantialportion of all U S bladder cancer Based on extrapolationof data from ingestion studies and a U S case controlstudy, Alan Smith and coworkers l 51 l have suggested thatthe low levels of arsenic present in U S drinking watermight cause a major portion of the bladder cancer cases.

Acknowledgement We thank Barbara Beck, Kenneth Brown, D.Warner North, and Randy N Roth for their comments on an earlierversion of this manuscript.

References1.Chen C-J, Wang CJ ( 1990) Ecological correction between ar-

senic level in well water and age-adjusted mortality from ma-lignant neoplasms Cancer Res 550:5470-5474

2 Lamm SH, Lederer WH ( 1985) Inorganic arsenic: The impor-tance of accurate exposure characterization in risk assessment.In: Risk analysis in the private sector Whipple C, Covello VT(eds) Plenum Press, New York, pp 149-163

3.International Agency For Research on Cancer, IARC mono-graphs on the evaluation of the carcinogenic risk of chemicalsto man, Vol 23, Some metals and metallic compounds WorldHealth Organization, Lyon, France ( 1980)

4 Petito CT, Beck BD ( 1991) Evaluation of evidence of nonlin-earities in the dose-response curve for arsenic carcinogenesis.Trace Subst Environ Health 24: 143-176

5.Bates MN, et al ( 1992) Arsenic ingestion and internal cancers:a review Am J Epidemiol 135:462-476

6 Smith TJ, et al ( 1977) Airborne arsenic exposure and excretionof methylated arsenic compounds Environ Health Persp 19:89-93

7 Chakraborty AK, Saha KC ( 1987) Arsenical dermatoses fromtubewell water in west Bengal Ind J Med Res 85:326-334

8.Tsuda T, et al ( 1995) Ingested arsenic and internal cancer: ahistorical cohort followed for 33 years Am J Epidemiol 141:198-209

9.Cebrian ME, et al ( 1983) Chronic arsenic poisoning in thenorth of Mexico Human Toxicol 2: 121-133

10 Lou Z-D, et al ( 1995) Chronic Arsenicism and Cancer in InnerMongolia -Consequences of Arsenic in Deep Wells J EnvironGeochem Health (in press)

11 Zalvidar R ( 1974) Arsenic contamination of drinking water andfoodstuffs causing endemic chronic poisoning Bitrage Patho-logica 151:384-400

12.Tseng WP, et al ( 1968) Prevalence of skin cancer in an en-demic area of chronic arsenicism in Taiwan J Nat Cancer Inst40:453-463

13.Chen C-J, et al ( 1988) Atherogenicity and carcinogenicity ofhigh arsenic artesian well water: multiple risk factors and re-lated malignant neoplasms of blackfoot disease Arteriosclero-sis 8:452-460

14.Chang T-C, et al ( 1991) Higher prevalence of goiter in en-demic area of blackfoot disease of Taiwan J Formosan MedAssoc 90:941-946

15 Lu J-N, Chen C-J ( 1991) Prevalence of hepatitis B surface anti-gen carrier status among residents in the endemic area ofchronic arsenicism in Taiwan Anticancer Res 11: 229-234

16.Wu M-M, et al ( 1989) Dose-response relation between arsenicconcentration in well water and mortality from cancers andvascular diseases Am J Epidemiol 130: 1123-1132

17.Fierz VU ( 1965) Katamnestische Untersuchungen iber dieNebenwirkungen der Therapie mit anorganischem Arsen beiHautkrankheiten Dermatologica 131 :41-58

18 Goldsmith JR ( 1972) Evaluation of health implications of ele-vated arsenic in well water Water Res 6:1133-1136

19 Harrington JM, et al ( 1978) A survey of a population exposedto high concentrations of arsenic in well water in Fairbanks,Alaska Am J Epidemiol 108: 377-385

494

20 Southwick JW, et al ( 1983) An epidemiological study of ar-senic in drinking water in Millard County, Utah In: LedererWH, Fensterheim RJ (eds) Arsenic: industrial, biomedical, en-vironmental perspectives Van Nostrand Reinhold, New York,pp 210-225

21.Tay C-H, Seah C-S ( 1975) Arsenic poisoning from anti-asth-matic herbal preparations Med J Austria 2:424-428

22 Armitage P, Doll R ( 1961) Stochastic models for carcinogene-sis In: Neyman J (ed) Proceedings of the Fourth Berkeley Sym-posium on mathematical statistics and probability 4: 19-38

23 Crump KS, et al ( 1976) Fundamental carcinogenic processesand their implications for low dose risk assessment Cancer Res36:2973-2979

24.Crump KS, et al ( 1977) Confidence intervals and tests of hy-pothesis concerning dose response relations inferred from ani-mal carcinogenicity data Biometrics 33:437-451

25.Lu FJ ( 1990) Blackfoot disease: arsenic or humic acid? Lancet336:115-116

26.Lin SM, Yang MH ( 1988) Arsenic, selenium, and zinc in pa-tients with blackfoot disease Biol Trace Elem Res 15:213-221

27.Pontius FW, et al ( 1994) Health implications of arsenic indrinking water J Am Water Works Assoc, September: pp 52-63

28 Vahter M ( 1983) Metabolism of arsenic, In: Fowler BA (ed)Biological and Environmental Effects of Arsenic Elsevier Sci-ence Press, New York

29 Vahter M, Marafante E ( 1985) Reduction and binding of arse-nate in Marmoset monkeys Arch Toxicol 57:119-124

30 Deak ST, Csaky KC, Waddell WJ ( 1976) Localization and his-tochemical correlation of 73As by whole-body autoradiographyin mice J Toxicol Environ Health 1:981-984

31.Lindgren A, Vahter M, Dencker L ( 1982) Autoradiographicstudies on the distribution of arsenic in mice and hamsters ad-ministered 7 4 As-Arsenite or Arsenate Acta Pharmacol Toxi-colog 51: 253-265

32.Jung C, Rothstein A, ( 1965) Arsenate uptake and release in re-lation to the inhibition of transport and glycolysis in yeast.Biochem Pharmacol 14:1093-1112

33 McKinney JD, ( 1992) Metabolism and disposition of inorganicarsenic in laboratory animals and humans Environ GeochemHealth 14:43-48

34.Thomas DJ ( 1994) Arsenic toxicity in humans: research prob-lems and prospects Environ Geochem Health 16: 107-111

35 Thompson DJ ( 1993) A chemical hypothesis for arsenic methy-lation in mammals Chem Biol Interactions 88: 89-114

36 Endo G, et al ( 1992) Dimethylarsenic acid induces tetraploidsin Chinese hamster cells Bull Environ Contamin Toxicol 48:131-137

37 Shirachi DY, et al ( 1983) Tumorigenic effect of sodium arsen-ite in rat kidney Proceeding of the Western Pharmacology So-ciety 26:413-415

38 Yamamoto S, et al ( 1995) Cancer induction by an organic ar-senic compound, dimethylarsenic acid (cacodylic acid), in F 344/Du Crj rats after pretreatment with five carcinogens CancerRes 55:1271-1276

39 Yamanaka K, et al ( 1991) Cellular response to oxidative dam-age in lung induced by the administration of dimethylarsenicacid, a major metabolite of inorganic arsenics, in mice ToxicolAppl Pharmacol 108:205-213

40 Vahter M, Envall J ( 1983) In vivo reduction of arsenate in miceand rabbits Environ Res 32: 14-24

41.Marcus WL, Rispin AS ( 1988) Threshold carcinogenicity usingarsenic as an example In: Cothern CR, Mehlman MA, MarcusWL (eds) Risk Assessment and Risk Management of Industrialand Environmental Chemicals Princeton Scientific, N J , pp133-158

42 Valentine JL, Kang HK, Spivey G ( 1979) Arsenic levels in hu-man blood, urine, and hair in response to exposure via drinkingwater Environ Res 20:24-31

43.Brown CC, Chu KC ( 1983) Implications of the multistage the-ory of carcinogenesis applied to occupational arsenic exposure.J Nat Cancer Ins 70:455-463

44 Jacobson-Kram D, Montalbano D ( 1985) The reproductive ef-fects assessment group's report on the mutagenicity of inor-ganic arsenic Environ Mutagen 7: 787-804

45.Lee T-C, et al ( 1985) Comparison of arsenic-induced celltransformation, cytotoxicity, mutation, and cytogenic effects inSyrian hamster embryo cells in culture Carcinogenesis 6: 1421-1426

46 Bauman JW, Liu J, Klaassen CD ( 1993) Production of metal-lothionine and heat shock proteins by metals Fundam ApplToxicol 21:15-22

47.Moolgavkar SH ( 1991) Carcinogenesis models: an overview.Bas Life Sci 58: 387-96

48.Vig BK, et al ( 1984) Chromosome studies in human subjectschronically exposed to arsenic in drinking water Am J IndMed 6:325-338

49.Brown KG, Chen C-J ( 1994) On the observed dose-responsefor internal cancers and arsenic in drinking water in the Black-foot disease endemic region of Taiwan In: Arsenic: exposureand health Chappell WR, Abernathy CO, Cothern CR, (eds)First International Conference on Arsenic Exposure and HealthEffects Sci Technol Lett, Northwood, England, pp 153-169

50.Brown KG, Chen C-J ( 1995) Significance of exposure assess-ment to analysis of cancer risk from inorganic arsenic in drink-ing water in Taiwan Risk Analysis Vol 15:475-484

51 Smith AH, et al ( 1992) Cancer risks from arsenic in drinkingwater Environ Health Persp 97:259-267