Predictive Value of Molecular Dosimetry: Individual versus ... › content › cebp › 6 › 8 › 603.full.pdfVoL 6, 603-610, August 1997 Cancer Epidemiology, Biomarkers & Prevention

VoL 6, 603-610, August 1997 Cancer Epidemiology, Biomarkers & Prevention 603

Predictive Value of Molecular Dosimetry: Individual versus Group

Effects of Oltipraz on Aflatoxin-Albumin Adducts and

Risk of Liver Cancer1

Thomas W. Kensler,2 Stephen J. Gange,Patricia A. Egner, Patrick M. Dolan, Alvaro Mu#{241}oz,John D. Groopman, Adrianne E. Rogers, andBill D. Roebuck

Departments of Environmental Health Sciences [T. W. K., P. A. E, P. M. D.,

J. D. G.] and Epidemiology [S. J. G., A. Ml, Johns Hopkins School ofHygiene and Public Health, Baltimore, Maryland 21205; Department of

Pathology, Boston University School of Medicine, Boston, Massachusetts

021 18 [A. E. RI; and Department of Pharmacology and Toxicology,

Dartmouth Medical School, Hanover, New Hampshire 03755 [B. D. R.]

Abstract

Studies in animals and humans have established serumaflatoxin-albumin adducts as biomarkers of exposure toaflatoxin B1 (AFB1), a food-borne hepatocarcinogen. Toassess the utility of measurements of aflatoxin-albuminadducts to predict risk of hepatocellular carcinoma(HCC), 123 male F344 rats were dosed with 20 �g ofAFB1 daily for S weeks after randomization into threegroups: no intervention; delayed-transient (500 ppm ofoltipraz, weeks 2 and 3 relative to AFB1); or persistent(500 ppm oltipraz, weeks -1 to 5). Serial blood sampleswere collected from each animal at weeldy intervalsthroughout aflatoxin B1 exposure and assayed for levelsof aflatoxin-albumin by radioimmune assay. Area underthe curve (AUC) values for aflatoxin-albumin adductsdecreased 20 and 39% in the delayed-transient andpersistent oltipraz intervention groups, respectively, ascompared to no intervention. Similarly, the totalincidence of HCC dropped from 83 to 60% (P = 0.03)and 48% (P < 0.01) in these groups. Tumor multiplicitywas also reduced in the two oltipraz intervention groups,whereas time to HCC was increased. Mononuclear cellleukemia, a common neoplasm in F344 rats, was seen in39% of the control animals, whereas the two oltiprazinterventions reduced incidence to 18% (P 0.05) and13% (P = 0.01), respectively. Overall, a significantassociation was seen between biomarker AUC and risk ofHCC (P = 0.01). However, when the predictive value ofaflatoxin-albumin adducts was assessed within treatment

Received 1/29/97; revised 5/9/97; accepted 5/19/97.

The costs of publication of this article were defrayed in part by the payment ofpage charges. This article must therefore be hereby marked advertisement in

accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

I This work was supported in part by the National Institute of Environmental

Health Sciences Grant ES06052 and Center Grant ES03819 and National Cancer

Institute Grant CA39416.

2 To whom requests for reprints should be addressed, at Department of Environ-

mental Health Sciences, Johns Hopkins School ofHygiene and Public Health, 615North Wolfe Street, Baltimore, MD 21205. Phone: (410) 955-4712; Fax:

(410) 955-01 16; E-mail: [email protected].

groups, there was no association between AUC and riskof HCC (P = 0.56). Thus, aflatoxin-albumin adducts canbe useful for monitoring population-based changesinduced by interventions, such as in chemopreventiontrials, but have limited utility in identifying individualsdestined to develop HCC. As a consequence, the use ofthis biomarker in quantitative risk assessment should bepursued cautiously.

Introduction

Chemical-specific markers have been developed for a numberof environmental carcinogens for use as molecular dosimetersof individual exposures (1). These markers provide the prospectof contributing substantially to the specificity and sensitivity of

epidemiological studies aimed at determining the role of envi-ronmental agents in the etiology of human cancers (2, 3). Some

of these biomarkers may also assist the monitoring of popula-tions for changes in cancer risk, as may be achieved through

exposure avoidance and chemoprevention (4). Biomarkers ofthe biologically effective dose may be particularly useful in thiscontext in that they, in theory, provide a mechanistic linkage

between exposure and disease outcome. The biologically ef-fective dose reflects the amount of carcinogen that has inter-

acted with its critical macromolecular targets and has beenmeasured as either carcinogen-DNA or carcinogen-protein ad-

ducts. However, despite their increasing prominence, there are

few instances where the promise of these biomarkers has beenfulfilled. This reflects both the actual infancy of their develop-ment as well as intrinsic limitations to their application inpopulations (5, 6). To date, no carcinogen-specific biomarkerhas undergone full validation to quantitatively define the extent

of the relationships between external exposure, biomarker 1ev-els, and cancer outcome.

Studies with the aflatoxins have served as a guidepost inthe development of such molecular approaches in the assess-

ment of exposure status and risk of HCC,3 one of the most

common cancers in Asia and sub-Saharan Africa. Over the past30 years, there have been extensive efforts to investigate theassociation between aflatoxin exposure and HCC (7). These

studies have been generally hindered by the lack of adequatedosimetry data on aflatoxin intake, excretion, and metabolismin humans, as well as by the general poor quality of worldwide

cancer morbidity and mortality statistics. However, recent pro-spective studies using biomarkers of the biologically effectivedose of aflatoxin have provided striking confirmation of thepostulated associations derived from the earlier ecological and

3 The abbreviations used are: HCC, hepatocellular carcinoma; CI, confidence

interval; AFB1, aflatoxin B1 ; AUC, area under the curve; MCL, mononuclear cellleukemia.

on June 5, 2021. © 1997 American Association for Cancer Research. cebp.aacrjournals.org Downloaded from

http://cebp.aacrjournals.org/

604 Aflatoxin-Albumin Adducts and Liver Cancer

observational studies. Using urinary aflatoxin biomarkers, Qian

et al. (8) demonstrated a dramatic synergistic interaction be-tween aflatoxin and infection with hepatitis B virus in the riskof HCC in residents of Shanghai, People’s Republic of China.

The presence of urinary aflatoxin biomarkers alone resulted ina significant 3.4-fold increase in risk (95% CI, 1.1-10.0). Un-like measurements of urinary excretion of aflatoxins, which

measure very recent exposures to aflatoxins, the determination

of serum levels of aflatoxin-albumin adducts appears to provide

a more integrated assessment of aflatoxin exposures over a

several-month period. A recent longitudinal study in 120 resi-dents of Daxin Township in Qidong, People’s Republic of

China, demonstrated the consistent presence of aflatoxin-albu-mm adducts in most participants (9). A nested case-control

study in a prospective cohort of individuals in Qidong who areseropositive for hepatitis B surface antigen has indicated that

the relative risk for HCC among individuals also positive foraflatoxin-albumin adducts was 2.4 (95% CI, 1.2-4.7; Ref. 10).

Similarly, in Taiwan, Wang et a!. (11) observed an adjustedodds ratio of 2.8 (95% CI, 0.9-9.1) for detectable compared tonondetectable aflatoxin-albumin adducts in hepatitis B surface

antigen-seropositive men.As a component of the ongoing validation of aflatoxin

biomarkers, the present study has sought to expand these im-portant findings by determining the quantitative relationships

between levels of aflatoxin-albumin adducts and individual riskfor the development of HCC. This study has used a rodentmodel in which one major variable, carcinogen exposure, has

been held constant and risk outcome has been attenuatedthrough the use of persistent and delayed-transient interven-tions with the chemopreventive agent oltipraz. Serial measure-

ments of aflatoxin-albumin adducts throughout the period of

carcinogen exposure have allowed for an analysis of the lon-gitudinal association between biomarker levels and individual

development of liver cancer. Such studies, in turn, facilitate theobjective evaluation of the use and limitations of molecular

biomarkers to predict disease outcomes in individuals at high

risk from exposure to environmental toxicants.

Materials and Methods

Chemicals. Oltipraz was obtained from the ChemopreventionBranch, National Cancer Institute, and was determined to be

>99% pure by high-performance liquid chromatography (12).I was obtained from Sigma Chemical Co. (St. Louis, MO)

as was purified rat albumin and kits for albumin determination.

All other chemicals and reagents were of the highest commer-

cially obtainable quality.

Animals, Diets, and Treatments. Male F344 rats (100 g;

Harlan, Indianapolis, IN) were housed under controlled condi-tions of temperature, humidity, and lighting. Food and water

were available ad libitum. Purified diet of the AIN-76A for-mulation (Teklad, Madison, WI) without the recommended

addition of 0.02% ethoxyquin was used, and fresh diet wasprovided to animals at least every other day. Oltipraz at a final

concentration of 0.05% was mixed into the AIN-76A diet witha V-blender, and diet was stored at 4#{176}C.Rats were acclimated

to the AIN-76A diet for 1 week before beginning the experi-

ment. At this time, 123 rats were randomized to three treatmentgroups: no intervention; delayed-transient intervention with

oltipraz during weeks 9 and 10 of age; and persistent interven-tion with oltipraz during weeks 7-13 of age. All rats received20 �g of AFB1 in 100 �tl of DMSO by gavage each morning

during weeks 8-13. All animals were weighed weekly.

Analysis of Aflatoxin-Albumin Adducts in Serum. Approx-imately 100-150 1.d of blood from each animal were collectedfrom the tail vein in hematocrit tubes at weekly intervals beginning1 week after the first dose of carcinogen and extending to 1 week

after the last dose. Samples were collected 2 h after the daily

administrations of AFB1. The hematocrit tubes were subsequently

centrifuged at 10,000 X g, and the resulting serum was decantedinto Eppendorf tubes and stored at -70#{176}Cuntil assay. Serumsamples were first concentrated by high speed centrifugation using

Amicon Microcon-50 microconcentrators. Briefly, 50 pi of rat

serum were loaded onto a prewashed filter containing 300 �il ofPBS and centrifuged at 8500 rpm for 15 mm. Each serum samplewas then washed with an additional 200 �.d of PBS and recentri-

fuged. The concentrated sample was pipetted from the filter, and

the volume was accurately measured. The amount of serum albu-

mm in each sample was determined by a bromcresol purple dyebinding method using rat albumin as standard (Sigma). Total

protein content was determined by the method of Bradford (Bio-Rad Laboratories, Hercules, CA). Serum proteins were digested

overnight at 37#{176}Cwith nuclease-free Pronase (Calbiochem-Nova-

biochem Co., La Jolla, CA) in a ratio of 1:4 (w/w, enzyme:totalprotein) as described by Shebar et a!. (13). Two volumes ofice-cold acetone were added to the tubes followed by incubation at

4#{176}Cfor 1 h to precipitate the proteins. Following centrifugation at12,000 rpm, the supernatant was decanted into an Eppendorf tube

and concentrated under N2 to remove the acetone. Volumes wereaccurately adjusted with PBS to reflect a final protein concentra-tion of 10 mg/mi. Samples were stored at -70#{176}Cprior to assay.

To minimize potential biases afforded by assay variability

over the course of the analyses, longitudinal sets of serial samplesfrom three animals in each of the treatment groups were assayed

in each assay set. A radlioimmune assay was used to quantitate

aflatoxin-albumin adducts in duplicate rat serum digests based

upon the method described by Wang et a!. (9). Prior to assay, allsamples were adjusted to contain a final concentration of 1 mg ofprotein/mi using digested protein prepared from a normal ratserum pool. Nonspecific inhibition in the assay was determined

from this normal rat serum pool, and the average value wassubtracted from all study samples. Standard curves for the radio-

immune assays were determined using a nonlinear regressionmethod described by Gange et a!. (14).

Analysis of Cancers. All moribund rats were autopsied whenclinical observations indicated that the rat would not survive.The criterion for autopsy was the inability of the rat to rightitself or walk. This ataxia was usually accompanied by a sig-

nificant (>15%) and rapid loss in body weight. All remaining

rats were euthanized and autopsied 20 months after the first

dose of FB1 . All grossly abnormal tissues were taken forhistopathological analysis. This included all spleens weighing

>2 g (i.e., >0.5% body weight). Standardized sections of liverwere cut from the two largest lobes from all livers that appeared

normal; most normal-appearing spleens were also sampled. Alltissues were fixed in formalin, processed by routine methods,and embedded into paraffin, sectioned, and stained with H&E.

Hepatic foci, adenomas, and carcinomas were identified by two

observers blinded to the identity of the treatments.

Statistical Analyses. The relationship of treatment group andoutcome was quantified using both incidence and time-to-event

(“survival analysis”) measures. Separate analyses were con-ducted for the primary outcome ofHCC by considering whetherthe animals had either any or multiple carcinomas separately.Incidence of all hepatic lesions (foci, adenomas, and carcino-

mas) as well as MCL in the three protocols were compared

using Fisher’ s exact test for independence (15). The relation-



0)

I-I(9

>-0000

zw

500

400

300

200

I 00

0

Cancer Epideminlo�, Bionsarkers & Prevention 605

ships between aflatoxin-albumin adduct levels and incidence

was evaluated using standard logistic regression models. Ad-duct levels from each weekly serum collection and each sum-mary measure (initial rate, peak, and AUC) were evaluated in

separate regression models to determine their univariate asso-ciation with cancer risk.

To assess the degree to which treatment modified the risk

of HCC through the biomarker, the bivariate relationship of thetreatments and biomarker levels in predicting disease risk wasinvestigated. Specifically, the strength of association between

treatment and outcome in logistic regression models with treat-ment as the only covariate and with both treatment and biomar-ker as covariates was examined. Notationally, the models were

of the following forms:

logit (Pr(HCC))

logit (Pr(HCC))

= a + f3�I (Delayed-Transient) + f3�I (Persistent)

= a* + f3�’I (Delayed-Transient) + f3�J (Persistent) + yX

where I( ) is an indicator function for intervention group; � and

P2 represent the univariate log-odds of HCC for the delayed-transient and persistent intervention groups relative to the con-trol group; and (3�* and (32* represent the log-odds after ad-justing for the biomarker X. Thus, the reduction of the

magnitude of coefficients � and � �0 (3�� and (32*, respec-tively, reflects the degree to which the biomarker X is a surro-gate for the relationship of cancer risk and intervention. The

two extremes that could occur would be: (a) given the level ofbiomarker, one does not need to know the intervention that was

received (i.e., P1 and (32 significantly different from zero, butPI* and 1�2 � not); and (b) once intervention is known, thelevel of biomarker does not provide any additional informationon risk of disease (i.e., y not significantly different from zero).

Separate survival analyses were conducted to investigate thetime to the occurrence of either any or multiple carcinomas. Thetune from initial aflatoxin exposure to the presence of either of

these events was considered censored for rats who were sacrificedeither prior to the administrative end ofthe study (20 months fromfirst exposure) that did not show the HCC outcome (i.e., any or

multiple tumors) or who did not display ataxia or pronouncedweight loss prior to the end of the study (regardless of subsequent

carcinoma diagnosis). For analyses of multiple HCC lesions, an-imals showing single tumors were treated as tight-censored (mul-tiple tumors could occur after the single tumor event). Each animalwith no carcinoma present at sacrifice was assumed to be at risk for

the occurrence of carcinoma at any time past the censoring time.Because hepatic tumorigenesis is a relatively slow event, theassumption that HCC would occur immediately after the censoring

time may not be appropriate. Thus, analyses using transformedcensoring times were conducted; these times were computed byadding a conservative constant to the censoring time of 180 days

for observations with no lesion or hepatic foci and 30 days forobservations with hepatocellular adenoma. Using these outcomeand censoring features, standard Kaplan-Meier methods (for anyor multiple HCC events; Ref. 16) were used to graphically depict

the time to HCC among the three groups. To evaluate the rela-tionship between intervention groups and biomarker levels both

separately and simultaneously on the time to HCC events, pars-metric log-normal regression models were fit for the time to HCCoutcomes (16). These models have an advantage over standard

Cox proportional hazards models in that they naturally parame-terize the percentiles of the survival distribution, where the pth

0 10 20 30 40 50 60 70 80 90 100

WEEKS OF AGE

Fig. 1. Body weights ofrats treated with 20 �&g ofAFB,/day during weeks 8-13

of age (solid bar) and receiving either no (0), delayed-transient (Li), or persistent

(0) dietary intervention with oltipraz.

percentile is the time at which p percentage develop HCC (hence

50th percentile = median). Additionally, it is easy to compute therelative percentile as a alternative measure to the relative hazardestimated in the Cox model for the difference in time to HCC

between two groups; relative percentiles describe the relative in-crease or decrease in time to HCC between groups (17).

The relationships between intervention groups and bi-

omarker levels were investigated using both simple one-wayANOVA models and regression methods that account for the

correlation among repeated measurements (1 8). The lattermethods were used to estimate and compare the mean of

logarithmically transformed adduct levels among the interven-

tion protocols at each of the six weekly time points. In addition,

summary measures (peak and AUC) of adduct response overthe entire observation period were computed for each animal.

The average of these summary measures was calculated foreach protocol and compared using standard ANOVA.

The repeated-measures regression model described aboveincorporates two parameters that describe the longitudinal

tracking of aflatoxin-albumin adduct levels over time. Specif-ically, the correlation between two adduct measurements on

any single rat taken s weeks apart is assumed to be equal to y’.The parameter y represents the correlation between consecutivemeasurements, and the parameter 0 represents how much the

correlation increases or decreases as the time between meas-urements (lag) increases. Using this model, one can statisticallytest among several forms of the correlation structure, includingwhether the correlation can be summarized using a compoundsymmetry structure (which implies that the correlation is due to

inherent animal effects) or an autoregressive structure (whichimplies that the correlation is due to direct dependence on prior

outcomes). Parameter estimates were obtained using all data

and separately for each intervention protocol.All computations were done using either SAS version 6. 1 1

(SAS, Inc., Cary, NC) or Splus (MathSoft, Seattle, WA).

Results

Growth Rates of Rats. The growth curves for the rats areshown in Fig. 1. All rats were treated with FB1 during weeks

8-13 of age. The overall growth rates of the rats fed oltipraz-supplemented diets in either the delayed-transient or persistentprotocols were similar to that of the rats maintained on thecontrol diet. Moreover, there were no significant differences ingrowth rates in the three experimental groups during the period

of aflatoxin exposure and oltipraz interventions.



00

0

Ca

0

0

0

0

“t.

0

0

0

Fig. 2. Kaplan-Meier estimates forthe proportion free of any (A) ormultiple (B) HCCs in rats treatedwith 20 pg of AFB1/day duringweeks 8-13 of age and receiving

either no (solid line), delayed-iran-sient (dashed line), or persistent

(dotted line) dietary interventionswith oltipraz. Each animal without

the event present at sacrifice orthose not sacrificed by the end of thestudy (20 months) were treated ascensored, where censoring timeswere modified as detailed in “Mate-

rials and Methods.”

70 75 80 85 90 95

0

0

Weeks of age

70 75 80 85 90 95

Weeks of age

606 Aflatoxln-Albwnin Adducts and Liver Cancer

Table I Effect of oltipraz interventio ns on incidence of hepa tic foci, adeno max, and HCCs”

TreatmentIncidence” (%)

No lesions Foci Adenomas HCC

No intervention

Delayed-transient intervention

1/41 (2.4)

1/40 (2.5)

4141 (9.8)

9/40 (22.5)

2/41 (4.9)

6/40 (10)

34/41 (82.9)

24/40’� (60)

Persistent intervention 4/40 (10) 13/40 (32.5) 4/4.0 (10) 19/4()C.d(475)

a Male F344 rats were treated daily with 20 jzg of AFB1. p.o., during weeks 8-12 of age and received either no intervention or delayed-transient or persistent interventions

with 500 ppm oltipraz in the diet as detailed in “Materials and Methods.”

b Only the most advanced hepatic lesion is reported for each rat.

C Significantly different from no intervention (P < 0.05, Fisher’s exact test).d Significantly different from delayed-transient intervention (P < 0.05, Fisher’s exact test).

00I0

C0.ta2

0�

00Ia,0.

0

a,a,

C0

.ea0

0�

0

CsJ

0

Effect of Treatment on Neoplasms. Apart from two earlydeaths due to anesthetic overdose with ether during bloodcollection, the first death in this study occurred during the 26thweek, and the first death with a gross liver tumor was in the

67th week after initiation of aflatoxin dosing. Table 1 presentsthe effects of the oltipraz interventions on the incidence ofAFB1-induced hepatic preneoplastic foci, adenomas, and HCC;however, only the most advanced hepatic lesion is reported foreach rat. Foci were of course present in the livers of rats withadenomas or carcinomas. After 20 months, the 5-week expo-sure to AFB, produced a high incidence of HCC. There was a

significant trend toward more HCCs in untreated animals (83%)relative to the delayed-transient intervention group (60%; Fish-er’s exact test, P 0.028) and persistent intervention group

(48%; Fisher’s exact test, P = 0.001). Almost all of the tumor-bearing animals (32 of 34) in the no intervention group hadmultiple hepatic carcinomas, whereas in the delayed-transient

and persistent intervention groups, significantly lower propor-tions of animals with multiple tumors were observed: 15 of 24

and 10 of 19, respectively. No differences in the histopatho-logical characteristics of the HCCs were observed between the

three groups. However, among the animals with no HCCs, therewere no trends between treatment group and the occurrence ofeither foci or adenomas.

In addition to occurrence, the timing of HCC was signifi-

cantly associated with treatment group. The estimated proportions

free of HCC and free of multiple HCC events over the time fromexposure using the censoring mechanisms described in the “Ma-terials and Methods” section are shown in Fig. 2, A and B,respectively. Comparison of the Kaplan-Meier curves in Fig. 24indicated that the animals in the persistent intervention group had

marginally different HCC-free time than the delayed-transientgroup (P = 0.075)but significantly longer HCC-free time than the

no intervention group (P = 0.002). Log-normal regression modelsestimated that the time to any HCC for the delayed-transient and

persistent groups was 9 and 22% longer than the no interventiongroup, respectively. For the time to multiple HCCs shown in Fig.2B, Kaplan-Meier analyses indicated similar results when com-paring delayed-transient and persistent groups (P = 0.055), butboth the delayed transient and the persistent groups had longertimes free of multiple HCCs than the no intervention group (P <0.001 and P = 0.019, respectively). Log-normal regression models

estimated that the time to multiple HCC lesions for the delayed-transient and persistent groups was 15 and 32% longer than the no

intervention group, respectively.A variety of extrahepatic neoplasms were observed in these

aged rats. The most common extrahepatic neoplasm was MCL.The incidence of MCL was 39% in the no intervention group, and



Cancer Epldemiologjr, Blomarkers & Prevention 607

500

� 450 . 20 �zg afiatoxin B1/day. p.o.

� 400

� .� 350 .!I�� (1�TT’TN�� 100 . 500ppmolipraz \� 50

< 0 � � I I I I I7 8 9 10 11 12 13 14 15

WEEKS OFAGE

Fig. 3. Mean serum levels of serial aflatoxin-albumin adducts in the no (0),

delayed-transient (V), or persistent (D) dietary intervention groups; error bars,SD. One hundred twenty-three male F344 rats were randomized into threetreatment groups as detailed in “Materials and Methods.” Serum samples werecollected at weekly intervals from all rats during the first 6 weeks of theexperiment and analyzed for content of aflatoxin-albumin adducts by radioim-

mune assay. The solid black bar displays the period of aflatoxin exposure,

whereas the striped bars display the periods of oltipraz administration for thedelayed-transient (upper bar) and persistent (lower bar) intervention groups.

this incidence is typical for the F344 rat (19). The incidence ofMCL decreased to 17.5% (P 0.048) with the delayed-transientintervention with oltipraz and dropped even lower to 12.5% (P =0.010) with the persistent intervention. The incidence of otherextrahepatic neoplasms was low in all groups. No metastases wereobserved to the liver from other organs.

Effect of Oltipraz Interventions on Biomarker Levels.Shown in Fig. 3 is the time course for the formation of afla-

toxin-albumin adducts in the serum of rats. Steady-state levels

of approximately 350 pmol of aflatoxin adducts/mg albuminwere reached within 2-3 weeks of daily dosing with 20 p.g ofAFB1. These levels represent 1-2% of the administered dose of

AFB1 . Levels of aflatoxin-albumin adducts declined slightlyduring the remaining period of AFB� exposure; however, theydropped rapidly once AFB, exposure was discontinued, reflect-ing the short half-life of circulating albumin in the rat. Therewere clear differences among treatment groups in the levels ofbiomarkers, as had been seen in earlier intervention studies with

oltipraz (20). At each collection point during the aflatoxinexposure period, the group receiving the persistent intervention

with oltipraz had significantly (P < 0.0001) lower biomarkerlevels than the control, no intervention group, such that steady-

state levels were reduced by 50% to approximately 175 pmol ofaflatoxin adducts/mg albumin. The group receiving the de-layed-transient intervention with oltipraz had biomarker levelsthat began to decline after the addition of oltipraz into the diet,

such that they were intermediate between the control and per-sistent intervention groups after several weeks and converged

with the levels of the persistent intervention group by the endof AFB1 dosing. AUCs for the aflatoxin-albumin adducts overthe entire aflatoxin exposure period decreased by 20 and 39%for the delayed-transient and persistent oltipraz interventions,

respectively, compared to the no intervention group. Therewere linear trends (P < 0.001) among no, delayed-transient,and persistent intervention groups with respect to peak adductlevels (means: 434, 346, and 248 pmol/mg albumin, respec-

tively) and AUC (means: 1321, 1012, and 740, respectively).

Tracking of Biomarkers. The serum samples that were col-lected weekly from the same animals during the period of expo-

sure provided an opportunity to evaluate the degree to whichbiomarker levels tracked over time. An important observation wasthat there was little, if any, decrease in the correlation as the

distance (lag) between the observations increased. This was con-

firmed in a formal analysis with the damped exponential correla-

ton model that indicated that the 0 parameter was not statisticallysignificant from 0.0 when using either data from all three groups

or when using data from each intervention group separately. Thus,

the correlation between observations can be summarized by using

a single intraclass correlation coefficient y, which was estimated tobe 0.296, and the 95% CI (0.205-0.368) indicated this was sig-nificantly greater than zero. There were no statistical differences in

the estimate of y among the three intervention groups. The mag-niwde of this correlation is difficult to interpret absolutely; com-

paratively, higher correlation implies stronger predictability suchthat animals with high (low) adduct levels will tend to remain high

(low).

Relationship of Biomarker Levels to Group and IndividualCancer Outcomes. Aflatoxin-albumin adduct measurementsmade at single time points and summary measures ofadducts overtime were used as covariates in a logistic regression model, pre-

dicting the incidence of HCCs. The variables showed a range inthe strength of the relationship with carcinoma outcome, with

AUC (P = 0.010) and peak adduct level (P = 0.018) showing thestrongest associations. The parametric survival models fit with

biomarker levels predicting the time to any or multiple HCCsshowed similar trends in the direction and magnitude as with thelogistic regression analysis (e.g., any HCC: AUC, P = 0.009,

peak, P = 0.010). Hence, biomarker levels were significantly

associated with both the occurrence and the timing of HCC.Although statistically significant predictors of cancer mci-

dence, neither AUC nor treatment group explained much of thevariability seen in cancer outcomes. Using an R� measure of

explained variability appropriate for binary outcomes (21), it is

estimated that 5.7% of the variability in cancer outcomes wasexplained by AUC, and 6.5 and 13.9% of the variability in cancer

outcomes was explained by the delayed-transient and persistent

groups relative to the no intervention group, respectively.The relationship of biomarker (AUC) and outcome (HCC)

including observations from all three intervention groups is

graphically depicted in Fig. 4A. It shows that the AUC was

higher in animals developing HCCs than those that did not (P =0.01). Fig. 4B shows the data in Fig. 4A stratified by treatment

group. Previously described features of the oltipraz interven-

tions can be seen in this figure; there is a downward trend ofAUC with each of the interventions, and there is a downward

trend of the proportion of HCCs (depicted by the relative

distribution of white and non-white points) among interventiongroups. These relationships fulfill the requirements for the

intervention group to be a confounder for the AUC-HCC rela-

tionship. The distribution of AUC values for rats with eithermultiple or single HCC events substantially overlaps with those

for HCC-free rats once the data are split by intervention group.Thus, there are no differences in the average AUC for those

animals with no HCC versus any HCC when looking at theobservations in each intervention group separately.

The quantitation of the ability of AUC to predict HCC mci-dence and time-to-HCC after adjusting for the relationship of HCC

and intervention group is described in Table 2. The entries in theleft columns of the table relate to predicting incidence of HCC

using a logistic regression model, and the parameter that measures

the strength of the association of HCC incidence and covariates is



000Cl) B

S

S Multiple HCCSingle HCC

0 N0HCC

A

S

S

‘p.

1i-�S

1%

9 S�_��!1- 0 #{149}� BB

� “II-

S� �

S

SB

0

No HCC

HCC

No Delayed- PersistentIntervention Transient

608 Aflatoxin-Alhumin Adducts and Liver Cancer

8..-..L0oc%J

#{149}0#{149}0

Cancer Epidemielogy, Biomarkers & Prevention 609

efficacy of a delayed, transient intervention with oltipraz waspredicted based upon the substantive reduction in the hepatic

burden of GST-P positive foci, a presumptive preneoplasticlesion, reported earlier (23). Interestingly, both interventionprotocols engendered significant delays in the development of

single HCCs or the amplification of single to multiple HCCswhen neoplasms occurred.

A significant decrease in hematopoietic neoplasia was also

observed with oltipraz interventions. MCL is a common neo-plasm in the F344 rat that is not associated with exposure toAFBJ (19). Roebuck et al. (22) initially reported that feeding750 ppm ofoltipraz significantly reduced the incidence of MCL

from 53 to 30%. In the current study, the incidence of MCL wasreduced from 39 to 18 and 13%, respectively, by the delayed-

transient and persistent interventions with 500 ppm of oltipraz.Thus, oltipraz appears to block one or more early processes ofMCL development. Intriguingly, Rao et al. (24) have recentlyreported that oltipraz inhibits the formation of another hema-topoietic neoplasm, thymic lymphosarcoma, in rats treatedwith the heterocycic amine 2-amino-l-methyl-6-phenylimi-

dazo[4,5-bjpyridine (PhIP). With multiple reports of oltiprazpreventing hematopoietic tumors and the paucity of approachesfor the prevention of such tumors, oltipraz may present a unique

opportunity for chemoprevention in this setting.A heretofore neglected opportunity in the development and

use of experimental models of cancer chemoprevention is theability to examine the predictive significance of intermediate bi-

omarkers to the cancer process. Such models afford the opportu-nity to modulate risk of cancer while maintaining carcinogen

exposure constant. A particularly promising class of markers arethose that measure biologically effective dose (4). These biomar-

kers reflect the internal dose at the molecular target (or a reason-able surrogate thereof) and are usually measured as DNA-adducts

or protein-adducts. Numerous epidemiological and experimentalstudies have established dose-response associations between cx-

ternal exposures and adduct levels. Correlations have also beenobserved with chemopreventive regimens in parallel cohorts of

animals between the degree of inhibition of tumorigenesis and theextent of reductions in DNA-adduct and/or protein-adduct levels.

Collectively, these correlative linkages between biomarkers andreduced risk of cancer outcomes suggest that adduct biomarkers

might serve: (a) as short-term end points for assessing the efficacyof chemopreventive agents; and (b) as tools for personalized risk

assessments.Albumin is quantitatively the most abundant target for afla-

toxin, and aflatoxin-lysine has been identified as the major adductin rat and human albumin (25). This adduct is formed by the samemetabolic pathway that leads to aflatoxin-DNA adduct formation.Indeed, a constant relationship between levels of aflatoxin binding

to serum albumin and hepatic DNA have been observed in several

studies in experimental animals (26, 27). Levels of aflatoxin-albumin adducts also qualitatively reflect the relative susceptibility

of different species to aflatoxin-induced hepatocarcinogenesis(28). Moreover, several studies have demonstrated that levels ofaflatoxin-albumin adducts can be modulated by a variety of che-mopreventive interventions with oltipraz and its unsubstituted con-gener, l,2-dithiole-3-thione (20, 29). The repetitive measurements

of aflatoxin-albumin adducts performed during this bioassay re-fleet, quantitatively and qualitatively, the adduct patterns deter-mined earlier from serum samples obtained from animals treatedwith identical aflatoxin/oltipraz regimens, but using daily serialsacrifices of animals for measurements of multiple intermediate

end points (20). Independent analytical methodologies were usedfor the aflatoxin-albumin adduct measurements in these two stud-ies. AUC measurements in both studies revealed approximately 20

and 40% reductions in aflatoxin-albumin adduct burdens through-out the AFB1 exposure period with the delayed-transient and

persistent oltipraz interventions, respectively. Strikingly, compa-rable reductions were seen in the final incidence of HCC with thetwo oltipraz interventions in the present study. As a consequence,there was a strong association between biomarker burden (AUC)and carcinoma outcome (P = 0.01). Another notable feature was

the significant intraclass correlation (tracking) between sequentialbiomarker measurements. These studies, therefore, directly con-firm the usefulness of monitoring levels of the aflatoxin-albuminadduct in population-based approaches to evaluate pharmacody-namic action in interventions with oltipraz and similarly acting

agents. Indeed, this biomarker is currently being used to assess theefficacy of oltipraz in humans at high risk for aflatoxin exposureand development of HCC in rural China (30). Measurements ofalbumin adducts appear to be particularly useful because they use

simple, facile immunoassays that can be applied to large numbers

of samples in field studies.However, the apparent strength of the observed association

with biomarker burden and cancer outcome reflects the propor-

tional distribution of cases and non-cases of cancer within the nointervention and persistent intervention arms. Thus, very strong

associations will be seen when protective efficacy is maximizedthrough judicious selection of dose and duration of carcinogen andanticarcinogen (29, 31). As one example, Hecht et a!. (32) haveshown complete segregation of the distribution of 4-(methylnitro-

samino)-l-(3-pyridyl)-l-butanone-derived hemoglobin adducts

under dosing conditions in which 4-(methylnitrosamino)-l-(3-pyr-idyl)-l-butanone alone induced lung tumors in 70% of the rats,whereas an intervention with phenethylisothiocyanate reduced the

tumor incidence to background. In the present study, the twointervention protocols with oltipraz produced roughly equivalent

distributions of HCC cases and controls within each interventionarm and afforded the opportunity to examine the extent to whichthere exists a prospective link between a molecular biomarker ofthe biologically effective dose of aflatoxin (i.e., aflatoxin albuminadduct) and HCC outcome in individual animals. However, in thisstringent setting, the biomarker provided no assistance in discrim-mating between individual rats destined to develop HCC and those

that did not. Given the complex interactive nature of the carcino-genic process, it is perhaps unreasonable to expect a single, earlymarker to predict cancer outcomes. Production of genetic damage

by hepatocarcinogens is not a sine qua non for cancer. For exam-plc, the genotoxic component of 2-acetylaminofluorene has been

recently suggested to play a minor role in its hepatocarcinogenicity(33). Many other factors, including recurrent cytotoxicity, cellproliferation, and nutritional status, can exert strong postinitiation

effects to either enhance or retard tumorigenesis. Sensitive, non-invasive biomarkers or combinations of markers for these pro-

cesses need to be developed and applied. An additional confound-ing factor relates to the strength of the surrogacy of the aflatoxin-

albumin adduct for genotoxicity. Although there appears to begood correspondence between total aflatoxin DNA and protein-

adduct burden following aflatoxin exposure, reflecting the com-mon metabolic origin of these adducts, the relationship betweenDNA adduct burden and the effective mutational target within thegenome may be less precise. The effective burden of genomic

damage is likely to directly influence tumor incidence and multi-plicity. Such correlative degeneracy may reflect the important rolethat DNA sequence context plays in aflatoxin mutagenesis as well

as the impact of differential rates of repair for damage to CrIticaland noncritical target genes (34). Lastly, it should be noted thatAFB1 exposure levels (and hence, biomarker levels) used in this

rodent experiment were considerably higher than found in humanpopulations. Whether stronger relationships between biomarker



610 Aflatoxin-Albumin Adducts and Liver Cancer

and outcomes might be observed at lower exposure levels remainto be established.

The overall strategy to use chemical-specific biomarkers

as intermediate end points has a wide range of potential, largelyunexplored applications for the refinement of risk quantitation.In the carcinogen testing arena, determinations of these types ofearly markers could accelerate predictions of chronic toxicity

and cancer outcomes well before the traditional 2-3-year timepoints. Furthermore, these mechanistically based biomarkers

might become efficient tools with which to rationally compare

new transgenic animal models with standard animal bioassayspecies. Such applications are needed as the number and diver-sity of chemicals to be tested increase. Similarly, these biomar-kers can be used to assess alterations in risk following chemo-

preventive interventions. In our study, strong concordance isobserved between the reduction in risk of HCC and diminutionof biomarker levels following treatment with oltipraz. Group-wide, population-based changes in biomarker levels are readily

detected as a function of intervention status. However, ourresults also point to strong limitations in the interpretation ofcarcinogen-adduct biomarker measurements. These adducts areunlikely to serve as unilateral measures of personal risk. Al-though changes in relative risk can be readily discerned, these

markers do not appear suitable for absolute prediction of dis-ease outcome. As a consequence, the use of adduct biomarkersin quantitative risk assessment should be pursued cautiously.

References

1. Groopman, J. D., and Kensler, T. W. Molecular biomarkers for human chem-

ical carcinogen exposures. Chem. Res. Toxicol., 6: 764-770, 1993.

2. Perera, F. P., and Weinstein, I. B. Molecular epidemiology and carcinogen-DNA adduct detection: new approaches to studies of human cancer causation. J.

Chronic Dis., 35: 581-600, 1982.

3. Wogan, G. N. Molecular epidemiology in cancer risk assessment and preven-

tion: recent progress and avenues for future research. Environ. Health Perspect.,

98: 167-178, 1992.

4. Kensler, T. W., Groopman, J. D., and Wogan, G. N. Carcinogen-DNA and

protein adduct biomarkers for cohort selection and as modifiable endpoints in

chemoprevention trials. In: B. W. Stewart, D. McGregor, and P. Kleihues (eds.),

IARC Scientific Publication No. 139: Principles of Chemoprevention, pp. 237-

248. Lyon: IARC, 1996.

5. McMichael, A. J. Molecular epidemiology: new pathway or new travelling

companion? Am. J. Epidemiol., 140: 1-11, 1994.

6. Pearce, N., de Sanjose, S., Boffefta, P., Kogevinas, M., Saracci, R., and Savitz,

D. Limitations of biomarkers of exposure in cancer epidemiology. Epidemiology,

6: 190-194, 1995.

7. Eaton, D. L., and Groopman, J. D. (eds.). The Toxicology of Aflatoxins. San

Diego: Academic Press, 1994.

8. Qian, G-S., Ross, R. K., Yu, M. C., Yuan, J-M., Gao, Y-T., Henderson, B. E.,

Wogan, G. N., and Groopman, J. D. A follow-up study of urinary markers of

aflatoxin exposure and liver cancer risk in Shanghai, People’s Republic of China.

Cancer Epidemiol., Biomarkers & Prey., 3: 3-11, 1994.

9. Wang, J-S., Qian, G-S., Zarba, A., He, X., Thu. Y-R., Thang, B-C., Jacobson,

L., Gange, S. J., MuIioz, A., Kensler, T. W., and Groopman, J. D. Temporal

pattems of aflatoxin-albumin adducts in hepatitis B surface antigen positive andnegative residents of Daxin, Qidong County, People’s Republic of China. Cancer

Epidemiol., Biomarkers & Prey., 4: 253-261, 1996.

10. Kuang, S-Y., Fang, X., Lu, P-X., Zhang, Q-N., Wu, Y., Wang, J-B., Zhu,

Y-R., Groopman, J. D., Kensler, T. W., and Qian, G-S. Aflatoxin-albumin adducts

and risk for hepatocellular carcinoma in residents of Qidong, People’s Republic

of China. Proc. Am. Assoc. Cancer Res., 37: 252, 1996.

I 1. Wang, L-Y., Hatch, M., Chen, C-J., Levin, B., You, S-L., Lu, S-N., Wu,

M-H., Wu, W-P., Wang, L-W., Wang, Q., Huang, G-T., Yang, P-M., Lee, H-S.,and Santella, R. M. Aflatoxin exposure and the risk of hepatocellular carcinoma

in Taiwan. Int. J. Cancer, 67: 620-625, 1996.

12. Christensen, R. G., and Malone, W. Determination of oltipraz in serum by

high-performance liquid chromatography with optical absorbance and mass spec-

trometric detection. J. Chromatogr., 584: 207-212, 1992.

13. Shebar, F. L, Groopman, J. D., Qian, G-S., and Wogan, G. N. Quantitative

analysis ofaflatoxin-albumin adducts. Carcinogenesis (Lond.), 14: 1203-1208, 1993.

14. Gange, S. J., Mu#{241}oz,A., Wang, J-S., and Groopman, J. D. Variability of

molecular biomarker measurements from nonlinear calibration curves. Cancer

Epidemiol., Biomarkers & Prey., 5: 57-61, 1996.

15. Agresti, A. Categorical Data Analysis. New York: J. Wiley & Sons, 1990.

16. Kalbfleisch, J. D., and Prentice, R. L. The Statistical Analysis ofFailure Time

Data. New York: J. Wiley & Sons, 1980.

17. Mufioz, A., and Sunyer, J. Comparison of semiparametric and parametric

survival models for the analysis of bronchial responsiveness. Am. J. Respir. Crit.

Care Med., 154: 5234-5239, 1996.

18. Mufioz, A., Carey, V., Schouten, J. P., Segal, M., and Rosner, B. A para-

metric family of correlation structures for the analysis of longitudinal data.

Biometrics, 48: 733-742, 1992.

19. Stefanski, S. A., Elwell, M. R., and Stromberg, P. C. Spleen, lymph nodes,

and thymus. In: G. A. Boorman, S. L. Eustis, M. R. Elwell, C. A. Montgomery,

Jr., and W. F. MacKenzie (edt.), Pathology ofthe Fischer Rat, pp. 369-393. New

York: Academic Press, Inc., 1990.

20. Egner, P. A., Gange, S. J., Dolan, P. M., Groopman, J. D., Mufioz, A., and

Kensler, T. W. Levels of aflatoxin-albumin biomarkers in rat plasma are modu-

lated by both long-term and transient interventions with oltipraz. Carcinogenesis(Lond.), 16: 1769-1773, 1995.

21. Mittlbock, M., and Schemper, M. Explained variation for logistic regression.

Stat. Med., 15: 1987-1997, 1996.

22. Roebuck, B. D., Liu, Y-L., Rogers, A. R., Groopman, J. D., and Kensler,T. W. Protection against aflatoxin B1-induced hepatocarcinogenesis in F344 rats

by 5-(2-pyrazinyl)-4-methyl- 1,2-dithiole-3-thione (oltipraz): predictive role for

short-term molecular dosimetry. Cancer Res., 51: 5501-5506, 1991.

23. Bolton, M. G., Mufloz, A., Jacobson, L P., Groopman, J. D., Maxuitenko, Y. Y.,

Roebuck, B. D., and Kensler, T. W. Transient intervention with oltipraz protects

against aflatoxin-induced hepatic tumorigenesis. Cancer Ret., 53:3499-3504,1993.

24. Rao, C. V., Rivenson, A., Zang, E., Steele, V., Kelloff, G., and Reddy, B. S.

Inhibition of 2-amino-1-methyl-6-phenlimidazo[4,5-b]pyridine-induced lym-

phoma formation by oltipraz. Cancer Res., 56: 3395-3398, 1996.

25. Skipper, P. L., and Tannenbaum, S. R. Protein adducts in the molecular

dosimetry of chemical carcinogens. Carcinogenesis (Lond.), II: 507-510, 1990.

26. Wild, C. P., Garner, R. G., Montesano, R., and Tursi, F. Aflatoxin B1 binding

to plasma albumin and liver DNA upon chronic administration to rats. Carcino-

genesis (Lond.), 7: 853-858, 1986.

27. Groopman, J. D., Hasler, J., Trudel, L. J., Pikul, A., Zhang, L-S., Chen, J-S.,

and Wogan, G. N. Molecular dosimetry in rat urine of aflatoxin-N7-guanine and

other aflatoxin metabolites by multiple monoclonal antibody affinity chromatog-

raphy and HPLC. Cancer Res., 52: 267-274, 1992.

28. Wild, C. P., Hasegawa, R., Barrand, L., Chutimataewin, S., Chapot, B., Ito,

N., and Montesano, R. Aflatoxin-albumin adducts: a basis for comparative

carcinogenesis between animals and humans. Cancer Epidemiol., Biomarkers &

Prey., 5: 170-189, 1996.

29. Groopman, J. D., DeMatos, P., Egner, P. A., Love-Hunt, A., and Kensler,T. W. Molecular dosimetry of urinary aflatoxin-N7-guanine and serum aflatoxin-

albumin adducts predicts chemoprotection by l,2-dithiole-3-thione in rats. Car-

cinogenesis (Lond.), 13: 101-106, 1992.

30. Jacobson, L P., Zhang, B-C., Thu. Y-R., Wang, J-B., Wu, Y., Thang, Q-N., Yu,

L-Y., Qian, G-S., Kuang, S-Y., Li, Y-F., Fang, X., Zarba, A., Chen, B., Enger, C.,Davidson, N. E., Gorman, M. B., Gordon, G. B., ProchaSka, H. J., Egner, P. A.,

Gmopman, J. D., Mufioz, A., Helzlsouer, K. J., and Kensler, T. W. Oltipraz chemo-prevention trial in Qidong, People’s Republic of China: study design and clinical

outcomes. Cancer EpidemioL, Biomarkers & Prey., 6: 257-265, 1997.

31. Kensler, T. W., Groopman, J. D., Eaton, D. L., Curphey, T. J., and Roebuck,

B. D. Potent inhibition of aflatoxin-induced hepatic tumorigenesis by the mono-

functional enzyme inducer 1,2-dithiole-3-thione. Carcinogenesis (Lond.), 13:

95-100, 1992.

32. Hecht, S. S., Trushin, N., Rigotty, J., Carmella, S. G., Borukhova, A.,

Akerkar, S., and Rivenson, A. Complete inhibition of 4-(methylnitrosamino)-l-

(3-pyridyl)-1-butonanone-induced rat lung tumorigenesis and favorable modifi-

cation of biomarkers by phenethyl isothiocyanate. Cancer Epidemiol., Biomark-

ers & Prey., 5: 645-652, 1996.

33. Travis, C. C., Zeng, C., and Nicholas, J. Biological model of EDO1 hepato-

carcinogenesis. Toxicol. Appl. Pharmacol., 140: 19-29, 1996.

34. Loechler, E. L. Mechanisms by which aflatoxins and other bulky carcinogens

induce mutations. In: D. L. Eaton and J. D. Groopman (eds.), The Toxicology of

Aflatoxins, pp. 149-178. San Diego: Academic Press, 1994.



1997;6:603-610. Cancer Epidemiol Biomarkers Prev T W Kensler, S J Gange, P A Egner, et al. of liver cancer.group effects of oltipraz on aflatoxin-albumin adducts and risk Predictive value of molecular dosimetry: individual versus

Updated version

http://cebp.aacrjournals.org/content/6/8/603

Access the most recent version of this article at:

E-mail alerts related to this article or journal.Sign up to receive free email-alerts

Subscriptions

Reprints and

[email protected] at

To order reprints of this article or to subscribe to the journal, contact the AACR Publications

Permissions

Rightslink site. Click on "Request Permissions" which will take you to the Copyright Clearance Center's (CCC)

.http://cebp.aacrjournals.org/content/6/8/603To request permission to re-use all or part of this article, use this link


http://cebp.aacrjournals.org/content/6/8/603http://cebp.aacrjournals.org/cgi/alertsmailto:[email protected]://cebp.aacrjournals.org/content/6/8/603http://cebp.aacrjournals.org/

Documents

Predictive Value of Molecular Dosimetry: Individual versus ... › content › cebp › 6 › 8 › 603.full.pdfVoL 6, 603-610, August 1997 Cancer Epidemiology, Biomarkers & Prevention