Hepatitis B, Aflatoxin B₁, and p53 Codon 249 Mutation in Hepatocellular Carcinomas from Guangxi, People’s Republic of China, and a Meta-analysis of Existing Studies¹

The incidence of HCC³ varies widely worldwide. Among males, the highest incidence rates are found in eastern Asia, particularly in China where HCC is the third most common cause of cancer death (1). Chronic infection with the HBV is the strongest risk factor for HCC worldwide (2, 3, 4). However, populations with similar prevalence of HBV infection have different incidence of HCC (4), suggesting the presence of other important risk factors. Aflatoxins, a group of mycotoxins produced by the common fungi Aspergillus flavus and Aspergillus parasiticus, are established human hepatocarcinogens (5, 6, 7, 8, 9) and are well-known HCC risk factors when present in foodstuffs (3, 10).

The high-risk areas in eastern China include the Guangxi Autonomous Region. Fusui County in southern Guangxi has a standardized rate of primary liver cancer among men of 120/100,000 population/year, a rate 35 times higher than that in the United States (3). HCC accounts for 50% of all of the cancer deaths in men and 25% of all of the cancer deaths in women in this region (11). A cohort study done in this area found a positive linear relationship for AFB₁ levels in foodstuffs and mortality attributable to HCC (3). The levels of AFB₁ were estimated to be as high as 2575 ng/kg/day (3), in contrast to the estimated 3 ng/kg/day exposure in the United States (12). Infection with HBV was significantly associated with HCC mortality, with 91% of HCC deaths occurring in HBV-positive subjects (3).

The tumor suppressor gene p53 is the most commonly mutated gene in human cancers (13). A G to T transversion at the third position of codon 249 of the p53 gene (249^ser) is commonly found in HCC from patients in regions with dietary aflatoxin exposure (14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24). In vitro studies (25, 26, 27) have supported this finding, showing that AFB₁ can induce this mutation. The biological activity of the 249^ser mutant p53 protein remains undetermined. However, experiments (28) conducted with a murine p53 protein carrying the codon 249 homologous mutation (p53ser246) showed that the mutant protein could transform cells in culture and was defective in its transcription activation function. Studies done worldwide (reviewed in Refs. 29, 30) suggest that the frequency of the p53 249^ser mutation in HCC correlates with the exposure level of AFB₁ in the underlying population. However, relatively few studies have been conducted in high aflatoxin exposure areas, and many have been small.

Some epidemiological (10, 24, 31, 32) and animal (33, 34) studies have found evidence for an HBV-aflatoxin interaction in hepatocarcinogenesis (reviewed in Ref. 30). Several mechanisms have been proposed to explain this interaction. The increase in cellular proliferation induced by HBV (35) could increase the probability for clonal expansion of an existing aflatoxin induced-p53 249^ser mutation. An increase in levels of aflatoxin metabolism enzymes (e.g., P450 enzymes) has been described for HBV transgenic mice and has been postulated as a mechanism for interaction (36). The HBVx protein, encoded by HBV, interferes with the nucleotide excision repair pathway (37), which cells use to repair aflatoxin-DNA adducts. Thus, the presence of the HBVx protein could increase the frequency of aflatoxin-induced mutations (37). Lastly, HBV infection could increase oxidative stress, which could lead to an increase in p53 mutations (38).

Epidemiological studies of the interrelationship among aflatoxin exposure, HBV status, and the presence of the p53 249^ser mutation could help clarify whether HBV acts as a confounder or as a synergistic partner with aflatoxin. One study (24) in Taiwan found that, in the presence of aflatoxin, the p53 249^ser mutation was significantly more frequent among HBV-positive than -negative subjects. Other studies found similar but not significant trends (18, 32, 39) or no differences at all (19, 21, 23, 40, 41, 42, 43, 44, 45, 46). The main limitation in investigating possible HBV-aflatoxin interaction is that in high exposure areas HBV infection is generally endemic, and few HBV-negative individuals are available for study.

Our report has two aims. First, we contribute additional data on p53 249^ser mutation and HBV infection in HCC from southern Guangxi. This area has one of the highest aflatoxin exposure rates in the world and has provided valuable data establishing the role of aflatoxin and HBV in HCC (3, 11, 47), but tumors from this region have never been analyzed for 249^ser mutations. Second, we present a meta-analysis using our results along with those from 48 published studies to examine the interrelationships among aflatoxin exposure, HBV infection, and p53 mutations in HCC, focusing on a possible HBV-aflatoxin interaction in producing 249^ser mutations.

We analyzed 64 paraffin-embedded tissue blocks from patients with liver resection for HCC from southern Guangxi, China. All of the patients were permanent residents of Nanning City and surrounding counties.

We used a laser capture microdissection instrument (Arcturus Engineering, Inc.) to remove non-neoplastic tissue from several consecutive 10–15-μm-thick sections of each HCC paraffin-embedded tumor block. We resuspended the microdissected samples in 50–200 μl of digestion buffer containing 10 mm Tris-HCl, 1 mm EDTA, 1% Tween 20, and 0.04% proteinase K (Sigma Chemical Co., St. Louis, MO) and incubated them at 55°C overnight. After proteinase K digestion, samples were boiled for 10 min and centrifuged to pellet tissue debris. The supernatant containing DNA was aliquoted and stored at −20°C until use. Given the old age and long fixation time of the tumor blocks, we were unable to obtain high quality DNA from many samples, which prevented us from performing a reliable systematic analysis of the entire p53 gene. We focused on the analysis of codon 249 by amplifying a region of exon 7 surrounding codon 249 using forward (5′-AACAGTTCCTGCATGGGCGG-3′) and reverse (5′-CCAGTGATGATGGTGAGG-3′) primers. PCR reactions were carried out in a PE 9700 (Perkin-Elmer, Foster City, CA) with an initial denaturation step of 2 min at 94°C followed by 40 cycles of 94°C for 30 s, annealing at 62°C for 40 s, and extension at 72°C for 1 min. The 58-bp long PCR products were purified using exonuclease I and shrimp alkaline phosphatase (Amersham Life Sciences, Cleveland, OH). Direct sequencing was performed using the Amersham Thermosequenase Radiolabeled Terminator Kit (Amersham Life Sciences) with the same primers used for PCR amplification. Mutations were confirmed in both forward and reverse directions.

For p53 protein detection, we deparaffinized tumor sections and stained them using a Ventana NexEs automated immunostainer (Tucson, AZ) using a rabbit anti-p53 primary antibody (NCL-p53-CM1p; Novocastra Laboratories, Newcastle Upon Tyne, United Kingdom). For detection of HBsAg, we deparaffinized tumor sections and quenched them with H₂O₂ for 15 min. We incubated slides first in 5% normal horse serum for 20 min at room temperature and then with a 1:100 dilution of mouse monoclonal primary antibody against HBsAg for 1 h at room temperature (Novocastra Laboratories). After rinsing, slides were incubated with a secondary antibody from the Vector Mouse Elite kit (Vector Laboratories, Burlingame, CA) for 30 min at room temperature, rinsed again, and incubated with labeled antibodies following kit instructions. After a final rinsing, slides were stained with 3,3′-diaminobenzidine and counterstained with hematoxylin for detection. For both proteins, appropriate positive and negative controls were used. A pathologist, blinded to HBV or p53 status, analyzed all of the immunostained tumors. Where possible, non-neoplastic areas surrounding the tumors were also analyzed. For p53 staining, tumors were classified as positive or negative for the presence of nuclear staining above background. For HBV analysis, tumors were scored positive or negative for the presence of cytoplasmic staining.

To examine possible associations between pairs of dichotomous outcomes, we analyzed 2 × 2 tables using Fisher’s exact test (48). Outcomes included the subject’s sex as well as the presence or absence of 249^ser mutation, of p53 protein accumulation, and of detectable HBsAg. We compared mean age at cancer diagnosis among groups defined by combinations of outcome variables using ANOVA methods (48).

We restricted our review to studies published in peer-reviewed journals before March 2000, plus the study presented here. To find studies, we searched Medline using search strings that used combinations of the words: p53 mutation, HCC, liver cancer, aflatoxin, HBV, or codon 249. We also looked at articles cited in those found via Medline.

Our goal was to analyze the proportion of tumors carrying p53 mutations and to see how those proportions changed with the aflatoxin exposure level in the study area and with HBV infection in the patient. To be included in our analysis, a study had to present the total number of tumors analyzed and individual information on p53 mutations obtained by direct DNA analysis. A number of studies, like our own, focused on 249^ser mutations, either by RFLP or by direct sequencing, and did not screen for mutations in the rest of the gene. These studies are used in some analyses. To include a study in our analysis of all of the p53 mutations, we required it to have sequenced at least exons 4 through 8. HBV infection status was assessed in different studies using serum samples, tumors, or both by a variety of methods, including immunostaining, PCR to detect integrated viral genome, ELISA, RIA, or Southern blots. In those studies where several techniques were used, we considered a subject positive if at least one technique showed positive results. To include a study in our analysis of whether HBV infection modified the effect of aflatoxin exposure on mutation frequency, we required cross-classified information on HBV infection and p53 mutation status. Finally, we required each study to provide sufficient information on the geographic origin of the tumor samples so that we could assess aflatoxin exposure. We classified study locations as having high, moderate, or low aflatoxin exposure based on work by Shen and Ong (29). They reviewed various studies and classified each according to either the quantity of aflatoxin in consumed food and/or determination of biomarkers (AFB₁-DNA or protein adducts) in human tissue samples from each region using published information. We performed similar assessments for study locations that they did not classify.

We calculated several response variables for each study. These included the proportion of tumors that had any mutation in p53, the proportion of tumors that had a 249^ser mutation, and the proportion of tumors with a 249^ser mutation among those that had any p53 mutation. We also calculated the proportion of patients that had HBV infections among all of the patients. To examine possible HBV-aflatoxin interactions, we focused on the difference in the proportion of tumors with a given mutation between patients with HBV infection and those without it and looked at whether that difference might change with aflatoxin level. We also looked for interaction on the log odds ratio scale by applying a modified empirical logistic transformation (49) to the proportions and analyzing whether differences in the transformed proportions (i.e., empirical log odds ratios) changed with aflatoxin level. As a measure of interaction for this analysis, we preferred the difference in proportions to the empirical log odds ratios, primarily because with the latter measure observed proportions of zero must be excluded from the analysis, eliminating studies; e.g., if the observed proportions of 249^ser mutations were zero in both HBV-positive and HBV-negative patients in a given study, we included that study in the analysis of differences in proportions (zero minus zero is zero) but not in the analysis on the log odds ratio scale (because zero divided by zero is indeterminate).

Each response variable was analyzed with a random effects model (50) incorporating study as a random effect and using aflatoxin exposure as a categorical covariate. We fitted these models using the MIXED procedure of Statistical Analysis System statistical software (SAS Institute, Inc., Cary, NC) following an approach outlined by Normand (51). This approach takes account of study-to-study heterogeneity in the underlying mean response that might arise, for example, from different populations under study or from different laboratory methods. The resulting estimates of mean response at each exposure level are weighted to take account of both within-study and study-to-study sources of variation and to accommodate the differing sample sizes in each study. When a proportion of zero or one is observed, the usual within-study estimate of the variance for such an observation is zero. Consequently, we examined the sensitivity of our conclusions to minor perturbations in within-study estimated variance. We also checked whether individual studies exerted undue influence on our conclusions by dropping one study at a time and reanalyzing the remainder.

Patients ranged in age from 20 to 72 years. Most were men (77%). In 14 (22%) of the 64 samples, we were not able to obtain PCR products because of poor quality DNA. Of the remaining samples, we found that 18 (36%) had a G to T transversion at the third base of codon 249 of the p53 gene (249^ser), whereas 32 (64%) were wild type. We observed no mutations in other codons of this small region of exon 7. For 22 (44%) of these samples, we performed independent DNA extractions and duplicate analysis to confirm the results and observed 100% concordance between these independent determinations. We found no differences in the frequency of 249^ser mutation between females (33%) and males (37%; P = 1.00). Patients with 249^ser mutation showed no statistically significant difference in mean age from those without the mutation (39 ± 8.3 versus 44 ± 13; P = 0.22).

Thirty samples (50%) showed accumulation of p53 protein in tumor cells. Four samples could not be analyzed. For the 48 samples with both DNA mutation and immunohistochemistry results, we saw concordance between protein accumulation and mutation in 30 samples, whereas 18 were nonconcordant. Among nonconcordant samples, three had 249^ser mutation without p53 protein accumulation, and 15 had wild-type codon 249 but positive immunohistochemistry. We found no statistically significant difference in the proportion of tumors with positive p53 staining between females (43%) and males (52%; P = 0.76). The mean age did not differ between patients positive (42 ± 13) and negative (42 ± 11) for p53 immunohistochemistry (P = 0.99).

For the analysis of HBV infection, we only included the 36 tumors that had non-neoplastic tissue available for staining because previous reports (52, 53) indicated that the majority of liver tumors tend to lose the HBV surface antigen, giving false-negative results. We excluded seven samples because poor tissue quality precluded their analysis. Twenty-eight tumors (78%) were positive for HBsAg, whereas eight (22%) were negative. Among the 28 positive tumors, 23 (82%) expressed the HBV antigen only in the non-neoplastic liver tissue, and 5 (18%) expressed it in both the neoplastic and non-neoplastic tissue. HBsAg status did not differ by gender (P = 1.00). Mean age at cancer diagnosis did not differ by HBsAg status (42 ± 11 versus 48 ± 15; P = 0.20).

Accumulation of p53 protein was detected in 17 (61%) of the 28 HBsAg-positive patients, compared with 2 (22%) of the 9 HBsAg-negative patients (P = 0.06). The 249^ser mutation was found in 9 (36%) of the 25 HBsAg-positive patients, compared with 2 (40%) of the 5 HBsAg-negative patients (P = 1.00). Mean age at diagnosis did not differ among four groups of patients defined by cross-classifying HBsAg status with presence or absence of p53 accumulation in their tumors. Mean age at diagnosis did appear to differ, however, among four groups of patients defined by cross-classifying HBsAg status with codon 249 mutation status of their tumors (P = 0.07). In particular, HBsAg-positive patients whose tumors had the 249^ser mutation had lower mean age at diagnosis than HBsAg-negative patients whose tumors had wild-type p53 (40 ± 6.8 versus 58 ± 1.5 years old, respectively).

In all, 49 studies were available for meta-analysis (Table 1), 48 from the literature plus the current study, although not all of these studies contained data to address each question. We proceeded in two phases: we considered possible associations between p53 mutations and aflatoxin level ignoring HBV infection and, after that, we examined the role of HBV infection.

First, we investigated whether aflatoxin exposure level in a study area was associated with the proportion of tumors carrying any mutation in p53 (Table 2).

The proportion of tumors carrying any mutation in p53 showed a broad range across studies within each aflatoxin level: 45–69%, 13–50%, and 0–35% for high, moderate, and low levels, respectively. Our analysis showed that the mean proportion of tumors with some p53 mutation changed with aflatoxin (P = 0.0001), with the highest weighted average in high aflatoxin areas (57%; Table 2). We saw a significant difference in mean frequency of p53 mutations between high and moderate areas and between high and low areas.

Next, we looked at whether the proportion of tumors with a G to T mutation at the third position of p53 codon 249 (249^ser) varied with aflatoxin exposure (Table 2). We found that these proportions exhibited variations within each aflatoxin exposure level, ranging from 30–83%, 0–30%, and 0–11%, respectively (Fig. 1). The mean proportion of tumors having 249^ser mutations changed with aflatoxin exposure (P = 0.0001) and was significantly larger in high aflatoxin areas (50%). Mean proportions did not differ significantly between the moderate (9%) and low (2%) aflatoxin areas.

The observed decrease in the proportion of 249^ser mutations with decreasing aflatoxin exposure could reflect a general decrease in all of the p53 mutations or a decrease in this specific mutation alone. A comparison of mean proportions suggested that tumors with a 249^ser mutation represented a majority of the tumors with p53 mutations in high aflatoxin areas but a minority in moderate or low aflatoxin areas. We examined this issue directly by analyzing the fraction of all of the tumors carrying p53 mutations that had a mutation in 249^ser. The mean proportion of tumors with a 249^ser mutation among tumors with any p53 mutation was strongly associated with level of aflatoxin exposure (P = 0.0001). At the high exposure level, on average 92% of tumors with p53 mutations were altered at codon 249 (Table 2). The corresponding mean proportions at the moderate and low exposure levels were 27% and 2%, respectively. All of the pairwise comparisons between aflatoxin levels were statistically significant. Thus, at higher aflatoxin levels, tumors with 249^ser mutations represented a sharply increasing fraction of the tumors carrying p53 mutations.

The previous analyses indicate that the increase in p53 mutations with increasing aflatoxin exposure mainly reflects an increase in 249^ser mutations. However, they do not indicate whether the frequency of mutations at p53 sites other than 249^ser also change with aflatoxin exposure. To address this issue, we eliminated all of the tumors with 249^ser mutations from the analysis (both numerator and denominator) and examined whether the proportion of p53 mutations in the remaining tumors changed with aflatoxin level (Table 2). We found evidence that aflatoxin exposure was associated with the proportion of tumors with non-249^ser mutations (P = 0.035). Omission of any one of three studies from high exposure areas that found only 249^ser mutations (i.e., estimated proportion of zero) rendered the association nonsignificant (P > 0.15), whereas omission of a single low exposure study that reported no non-249^ser p53 mutations enhanced statistical significance. Consequently, in our view, how aflatoxin exposure may relate to the frequency of p53 mutations at sites other than codon 249 remains an open question.

The mean proportion of HBV-positive cases among all of the cases was associated with aflatoxin exposure level (P = 0.0001; Table 2). That proportion decreased from 91% in high exposure regions to 76% and 41% in moderate and low exposure regions, respectively.

We asked whether HBV infection modified the effect of aflatoxin exposure on the proportion of tumors with any mutation in the p53 gene (i.e., aflatoxin by HBV interaction). We examined this interaction by looking at whether the difference in the proportion of tumors with any p53 mutation between HBV-positive and HBV-negative tumors changed with aflatoxin exposure. The mean difference in the proportion of tumors that carried any p53 mutation appeared somewhat larger in the moderate exposure areas (18%) than in either the high (−1%) or the low exposure areas (−3%; Table 2; Fig. 2). However, the mean difference in proportions did not change significantly across aflatoxin levels (P = 0.15). Thus, the presence or absence of HBV infection did not appear to modulate the effect of aflatoxin exposure on the prevalence of liver tumors with p53 mutations. Moreover, the mean difference in proportions between HBV-positive and HBV-negative cases, averaged across aflatoxin levels, was only 5% (95% CI, −6–15%; P = 0.39), indicating little or no effect of HBV itself on the proportion of liver tumors carrying p53 mutations.

Lastly, we investigated whether HBV infection modified the effect of aflatoxin exposure on the proportion of tumors carrying a 249^ser mutation. The mean difference in the proportion of tumors with a 249^ser mutation between HBV-positive cases and HBV-negative cases was about the same at all of the aflatoxin levels (P = 0.70; Table 2). The mean difference in proportions between HBV-positive and HBV-negative cases, averaged across aflatoxin levels, was 6% (95% CI, −1–13%; P = 0.11), possibly suggesting a small effect of HBV infection on the proportion of liver tumors with a p53 249^ser mutation.

Accumulation of p53 protein is detectable by immunohistochemistry and is sometimes used as a surrogate for the presence of p53 mutations because the mutant protein has a longer half-life than the wild type. About half of our samples had accumulation of p53 protein, similar to previous reports from Guangxi (43%; Ref. 53) and Qidong, China (55% and 61%; Refs. 17, 54). However, our results do not support an earlier report that HCC patients with tumors that show p53 protein accumulation are younger than those without it (53).

In vitro experiments (55) have shown that the 249^ser mutation may cause loss of tumor suppressor functions in hepatoma cell lines via loss of DNA-binding ability of the p53 protein. A similar finding was described using the mouse p53ser246 homologue mutation (28). Experiments with cells from transgenic mice expressing this mutant, under the control of the albumin promoter, have shown that the mutation promotes the transition of hepatocytes from G₀ to G₁ and/or M to G₁. This result suggests a gain of function mutation (56). Interestingly, these transgenic mice had increased liver tumor development when exposed to AFB₁ (57). Enhanced liver tumor development was reduced in mice that also expressed a wild-type p53, suggesting that loss of the wild-type allele may be required for the mutant to exert its effect (57). In our study from Guangxi, most tumors with the 249^ser mutation did not have a corresponding wild-type allele, suggesting that there had been loss of heterozygosity, resulting in hemizygosity. This observation may support the hypothesis that 249^ser mutations are recessive and do not have the dominant negative function reported for some other p53 mutations.

In vitro studies (27, 58) have shown that AFB₁ can induce the 249^ser mutation. These same studies also showed, however, that the 249^ser mutation is not the only mutation induced in the p53 gene by AFB₁, that this codon is not the preferred site for adduct formation, and that the removal of adducts at this site is neither faster nor more efficient. These findings are not sufficient to explain why a 249^ser mutation is more prevalent in high aflatoxin exposure areas.

Our meta-analysis indicates an association between the p53 249^ser mutation and increasing levels of aflatoxin exposure. This association had been suggested previously by some of the population-based studies carried out in areas with different levels of aflatoxin exposure. However, given the variations in sample sizes, sample origin, and assessment of aflatoxin exposure from study to study, some uncertainty remained. One other meta-analysis (59) of 20 studies has investigated this question but was limited in that it used only two levels of aflatoxin exposure (low and high), included only two studies from high aflatoxin exposure areas, and used statistical methods that did not take into account the sample size of each study. Nonetheless, they also found that areas with high aflatoxin exposure levels were associated with a higher proportion of 249^ser mutations in HCC that had p53 mutations.

Previous reports (18, 24, 32, 39) suggested that 249^ser mutation is more common in HBV-positive tumors than in HBV-negative tumors. In the present study, we did not find strong evidence for such an association. The results of our meta-analysis found no clear evidence for an effect of HBV status on the frequency of any p53 mutation or aflatoxin-HBV interaction on p53 mutation frequency. We also did not find evidence for an aflatoxin-HBV interaction on 249^ser mutation, but we did see some indication, not statistically significant, of a possible effect of HBV on the frequency of this mutation. One of the studies (24) we included in our meta-analysis tested for a possible HBV-aflatoxin interaction using odds ratios of prevalences. In our meta-analysis, we used differences in prevalences instead of odds ratios, as explained in “Materials and Methods.” When we used an empirical log odds approach (49) to reanalyze the studies in our meta-analysis, we found weak evidence for an HBV-aflatoxin interaction on 249^ser mutations (P = 0.10) and significant evidence of an effect of HBV on the frequency of this mutation (P = 0.005). However, these results were strongly influenced by one study. Omitting it, we found no evidence for HBV-aflatoxin interaction (P = 0.57) and weak evidence for an HBV effect (P = 0.07). These results were similar to what we observed using differences in prevalences.

Our meta-analysis has some limitations that are worth discussing. In the majority of the studies, the aflatoxin exposure level was estimated as an ecological measure without using individual measurements of specific biomarkers. This approach may introduce a classification bias because individuals in each study area may vary in their levels of aflatoxin exposure. Regarding HBV exposure, the measurements in all of the studies were done at the individual level. However, two issues should be considered. On the one hand, different methods were used among studies to determine HBV infection status (see “Materials and Methods” for description). These different assays may detect different stages in the infection process that could influence the effect of HBV on p53 mutations. More specifically, some studies used methods that only detect the HBsAg, which can be absent in subjects that do express the HBV core antigen or that have the HBV-X gene integrated in their genome. In our meta-analysis, we did not control for this source of variability. On the other hand, as we showed in our own study in Guangxi, measurements of HBV infection using detection of HBsAg should be carefully done, avoiding the use or pure neoplastic tissue, because that may lead to false-negative results. A few studies included in our meta-analysis assessed HBV status by detecting HBsAg in tumors, and we could not determine whether they had taken this issue into account. Again, this fact might be a source of classification bias.

The results of almost 50 studies of p53 gene mutations in HCC demonstrate evidence of a dose-response relationship between ecological levels of AFB₁ and prevalence of the p53 249^ser mutation in primary HCC. Whether AFB₁ causes these mutations or whether AFB₁ exposure leads to differential promotion of cells that acquired the mutation remains unclear. Although many population-based studies have provided evidence that HBV and aflatoxin can synergistically increase HCC risk, we find little evidence that this interaction also occurs at the molecular level in determining the frequency of p53 mutations.

We thank Julie F. Foley, Patricia S. Stockton, and Cindy R. Moomaw from the Laboratory of Experimental Pathology, National Institute of Environmental Health Sciences, for their assistance in the processing, microdissection, and immunostaining of the samples. We also thank Drs. Robert Maronpot, Abraham Nyska, and Gordon P. Flake, also from the Laboratory of Experimental Pathology, National Institute of Environmental Health Sciences, for their assistance in the histopathological analysis of the tumors, Sue Edelstein from the Arts and Photography department for her assistance with graphics design, and Drs. Ruth M. Lunn and Dale P. Sandler for critical review of the manuscript.