The Weakest Link: Exposure assessment in childhood cancer epidemiology studies

Most epidemiologic studies of pediatric cancer rely on questionnaires to assess exposure and, thus, recall bias can be a problem. In particular, parents of children with cancer tend to think more about specific exposures that they perceive as being harmful, and may report events that have no etiologic significance. While some recent studies have incorporated environmental (e.g., electromagnetic field monitoring) and/or biological measures (e.g., VDJ recombinase events) to improve exposure assessment, there is a need for further methodological improvements. For example, several studies have reported statistically significant positive associations between parental pesticide exposure and childhood cancer. Most contemporary-use pesticides such as carbamates, organophosphates, pyrethroids, etc, however, have very short half-lives-- usually a few days: also, they do not accumulate in tissues. In contrast, organochlorine-containing pesticides (including DDT, polychlorinated biphenyls (PCBs) and others) are degraded very slowly and are readily taken up in human adipose tissue, with half-lives estimated to be between 5-30 years. Due to this persistence, except for rare uses, most organochlorines have been banned in the United States. Therefore, biological assessment of pesticide exposure in epidemiological studies is constrained by either the short-half life of contemporary-use pesticides (in which case there would be no measurable blood levels unless there was very recent exposure) or the long-half life of the mostly-banned organochlorines (in which case it is difficult to determine the source of exposure). 

One recent pilot study explored the feasibility of direct measurement of pesticide levels prenatally [Cooper et al, Am J Ind Med 2001; 40:578-585]. Nine pregnant farmworkers were recruited from a city at the Texas-Mexico border.  At a prenatal visit close to the time of delivery, a brief questionnaire was administered to the women. The women consented to provide a urine specimen at delivery, and to have cord blood collected along with a sample of the placenta. Biological samples were screened for 51 analytes. All women had detectable urine levels of 2-4-D, a pesticide that is used on cotton, fruits and vegetables, as well as golf courses and residential lawns. The estimated half-life of 2-4-D is about 10-20 hours. All nine umbilical cord blood samples demonstrated detectable levels of DDE, a metabolite of DDT. While DDT is banned in the U.S., it continues to be used in Mexico. Eight of nine cord samples had detectable levels of volatile organic compounds including toluene, dichlorobenzene and/or trimethylbenzenes; few other compounds were detected. The authors discuss the difficulty of measuring pesticides with short half-lives. 

COMMENT: The Cooper et al study illustrates the difficulty with direct measurement of agents such as pesticides that have such a short-half life in vivo. However, this type of study is useful and there are other relationships that could be explored in a study such as this in pregnant women. While the authors administered a brief questionnaire, it appears that they did not perform any validation of the questionnaire with the types of pesticides the women may have come into contact with. It would be of interest to do a similar study over a period of time with multiple urine or blood samples collected during pregnancy to attempt to correlate reported exposures with biological measures. It is also important to note that there are several biological markers of effect associated with exposure to pesticides (as well as other agents) that are more persistent (e.g., DNA adducts, HPRT mutations, glycophorin A assay). These markers of effect have been utilized in epidemiologic studies to determine potential correlations with reported exposures. These types of methodological studies are necessary in order to determine the level of validity to questionnaires that assess specific exposures. 

When does Wilms’ tumor begin?

As noted previously [see C3 Vol 11, No 5; Vol 11, No 3; Vol 11; No 2; Vol 4, No 3; Vol 2, No 5 ], genomic imprinting is the differential expression of a gene depending on the parent of origin. Several genes, including insulin-like growth factor-2 (IGF2), are imprinted in humans. For IGF2, only the paternal allele is expressed in most normal tissues. Loss of imprinting (LOI) of IGF2 has been noted in Wilms’ tumor, as well as other embryonal tumors of childhood including rhabdomyosarcoma, Ewings’ sarcoma, germ cell tumors, and hepatoblastoma. In this study by Ravenel et al [JNCI 93:1698-703, 2001], Wilms’ tumor specimens (n=60) were examined for LOI of IGF2. Additionally, the authors examined imprinting status by two distinct Wilms’ tumor subgroups: intralobar nephrogenic rest (ILNR)-like Wilms’ tumors, which often contain smooth muscle, cartilage and adipocytes, and are thought to arise within the interior of the renal lobe early in fetal development; and perilobar nephrogenic rest (PLNR)-like Wilms’ tumors, which lack these elements, and are thought to arise in the peripheral area of the renal lobe later in fetal development. Of the 60 tumors, 36 were informative for the IGF2 ApaI polymorphism and were examined further. Twenty-one tumors displayed normal imprinting, while 15 had LOI at IGF2. In an independent pathological evaluation, 25 of the 36 tumors were classified as either ILNR-like (n=15) or PLNR like (n=10). Interestingly, 9/10 (90%) of the PLNR-like tumors demonstrated LOI at IGF2, while only 1/15 (7%) of the ILNR-like tumors did (p < 0.001). LOI was not related to stage. Using real-time PCR, the authors also reported that there was a 2.2 fold increase in IGF2 expression in tumors with LOI compared to tumors with normal imprinting at IGF2. Finally, the patient’s age at diagnosis was associated with LOI; patients whose tumors showed LOI had a median age at diagnosis of 65 months, which was more than twice the median age at diagnosis for patients whose tumors showed normal imprinting. Additional analyses explored common Wilms’ tumor mutations (including WT1 mutations as well as decreased expression) and loss of heterozygosity (LOH) at chromosome 11p. While ILNR-like tumors frequently (9/15) demonstrated WT1 mutations and LOH at 11p; none of the PLNR-like tumors demonstrated either. The authors propose a molecular model to explain their results: LOI at IGF2 in mature nephroblasts leads to increased IGF2 expression and the development of PLNR-like tumors in the periphery of the renal lobe. In contrast, LOH at 11p (along with other common Wilms’ tumor mutations) interferes with cell maturation and leads to ILNR-like tumors. 

COMMENT: The findings from this study are very provocative, as they provide molecular evidence that may help pinpoint the timing of Wilms’ tumor development. It may be of interest to explore whether epidemiologic risk factors vary across these clearly distinct molecular subgroups. Exposures early in fetal development may be more etiologically-relevant in ILNR-like WT, while exposures later in fetal development may be more etiologically-relevant in PLNR-like WT. 

Folate deficiency and childhood ALL

There is growing interest in the protective role of folate in human carcinogenesis. Folate supplementation has been associated with a decreased risk of colon cancer, and to some extent, brain tumors in children. In this case-control study from Australia [Thompson et al; Lancet 358:1935-40, 2001] mothers of 83 children diagnosed with common ALL up to age 14 years and 166 control children were interviewed in person during the period 1984 to 1992. 
Mothers were asked about demographic factors, reproductive and medical history, tobacco and alcohol use, occupational exposures, etc. The interview contained approximately 180 questions and lasted about 2.5 hours. Fathers were mailed a questionnaire, which consisted of 50 questions that assessed occupational history, x-ray history, etc. Several associations were statistically significant and strongly positive, including child exposed to pesticides (OR=3.08; 95% CI=1.51-6.29), mother exposed to agricultural chemicals in pregnancy (OR=2.91; 95% CI=1.40-5.63); and father’s report of work-related paint or pigment exposure prior to the child’s birth (OR=3.29; 95%CI=1.46-7.45). Unexpectedly, the authors found a statistically significant inverse association with mother’s report of ‘folate or iron’ supplementation during pregnancy and risk of childhood ALL (OR=0.37; 95%CI=0.21-0.65). In an attempt to determine whether the observed association was due to folate or iron, the authors stratified the data on women who reported intake of only one of the supplements. While only one mother reported taking folate without iron, 56 mothers (17 cases, 39 controls) reported using iron without folate (OR=0.75; 95% CI=0.37-1.51). Given the observed attenuation of the OR with iron alone, the authors suggest that the significant inverse association observed with ‘folate or iron’ is probably due to folate. This relationship remained even after adjustment for several other potential confounders. The authors suggest that folate supplementation reduces the risk of common ALL in children. 

COMMENT: While the finding is striking, there are several limitations to interpreting these analyses. The sample size was very small, and there is little discussion of age-specific results. While the authors describe that polymorphisms in folate metabolizing genes including methylenetetrahydrofolate reductase (MTHFR) have been associated with childhood leukemia, they did not perform any molecular analyses in this study. As noted in a recent C3 [Vol 12, No 2], homozygosity for the C677T polymorphism in MTHFR has been linked with a decreased risk of MLL infant leukemia; no association was reported with the more common type of childhood leukemia, TEL-AML1. It is unclear how folate supplementation during pregnancy could protect against development of leukemia in older children. Further, it is difficult to determine from the article how the questions regarding folate use were phrased. The authors note that ‘details of iron and folate use in pregnancy were elicited by the last in a series of questions about use of medications during pregnancy’. Most pregnant women are prescribed a multivitamin pill that contains recommended levels of folate and iron, so it is unclear whether this supplementation was above and beyond a multivitamin pill, or, was used in lieu of a multivitamin pill. Based on the small sample size, the number of comparisons, lack of sufficient detail regarding exposure assessment, and limited analyses by age-specific groups, the study received more press attention than perhaps warranted. Nevertheless, the finding is intriguing, and needs to be explored further. 

C3 Quarterly Newsletter
Children's Cancer Research Fund
Epidemiology Research Unit
Division of Pediatric Epidemiology
Clinical Research
University of Minnesota
420 Delaware St. SE, Box 422
Minneapolis, MN 55455
pedsepi@umn.edu

Editors: 
Stella M. Davies, MD, PhD, and Julie A. Ross, PhD