Testicular Dysgenesis Syndrome and the Estrogen Hypothesis: A Quantitative Meta-Analysis

Background Male reproductive tract abnormalities such as hypospadias and cryptorchidism, and testicular cancer have been proposed to comprise a common syndrome together with impaired spermatogenesis with a common etiology resulting from the disruption of gonadal development during fetal life, the testicular dysgenesis syndrome (TDS). The hypothesis that in utero exposure to estrogenic agents could induce these disorders was first proposed in 1993. The only quantitative summary estimate of the association between prenatal exposure to estrogenic agents and testicular cancer was published over 10 years ago, and other systematic reviews of the association between estrogenic compounds, other than the potent pharmaceutical estrogen diethylstilbestrol (DES), and TDS end points have remained inconclusive. Objectives We conducted a quantitative meta-analysis of the association between the end points related to TDS and prenatal exposure to estrogenic agents. Inclusion in this analysis was based on mechanistic criteria, and the plausibility of an estrogen receptor (ER)-α–mediated mode of action was specifically explored. Results We included in this meta-analysis eight studies investigating the etiology of hypospadias and/or cryptorchidism that had not been identified in previous systematic reviews. Four additional studies of pharmaceutical estrogens yielded a statistically significant updated summary estimate for testicular cancer. Conclusions The doubling of the risk ratios for all three end points investigated after DES exposure is consistent with a shared etiology and the TDS hypothesis but does not constitute evidence of an estrogenic mode of action. Results of the subset analyses point to the existence of unidentified sources of heterogeneity between studies or within the study population.

Impaired spermatogenesis, male reproductive tract abnormalities such as hypospadias and cryptorchidism, and testicular cancer have been proposed to comprise a common underlying syndrome with a common aetiology resulting from the disruption of embryonic programming and gonadal development during fetal life, termed the testicular dysgenesis syndrome (TDS) (Sharpe and Skakkebaek 2003;Skakkebaek et al. 2001). A hormonal etiology most likely underlies this syndrome, although it is believed to have more than one cause, possibly including other than endocrine disruption. Some common causes of endocrine disruption include infection, diet and body weight, lifestyle, genetics, and environmental exposure, but endocrine-disrupting chemicals (EDCs), particularly those with estrogenlike properties, have received the most scientific attention.
The synthetic estrogenic drug diethylstilbestrol (DES) was prescribed to more than 5 million pregnant women from the late 1940s to the early 1970s to prevent abortions and pregnancy-related complications (Palmlund et al. 1993). Evidence later showed that maternal ingestion of DES during early pregnancy increased the risk of vaginal clear cell adenocarcinoma in female offspring (Herbst et al. 1971) and resulted in an increased incidence of malformations of the testes, the development of epididymal cysts, and impaired sperm quality in male offspring . During pregnancy, maternal estrogen levels are significantly elevated. However, more than 90% of maternal endogenous estrogens are effectively sequestered via binding to sex hormone binding globulin (SHBG), and thus the fetus is relatively protected (Joffe 2001;Vidaeff and Sever 2005). On the other hand, DES and ethinylestradiol do not bind well to SHBG, having a higher biopotency if ingested (Sharpe and Skakkebaek 2003;Vidaeff and Sever 2005). Additionally, transgenerational exposure is also possible when lipophilic xenoestrogens are mobilized during pregnancy and lactation (Colborn et al. 1993).
Previous systematic reviews of studies in which pregnant women were exposed to estrogens other than DES have failed to find evidence of an increased risk of urogenital abnormalities in the male offspring (Raman-Wilms et al. 1995;Storgaard et al. 2006;Toppari et al. 1996;Vidaeff and Sever 2005), and have raised the possibility that nonestrogenic or atypical estrogenic effects of DES exposure in utero induce male reproductive abnormalities. However, none of the effects of DES exposure on either male or female offspring of exposed wild-type pregnant mice were induced when administered to ERKO (ER-α knockout) mice (Couse et al. 2001), strongly suggesting an ER-α-mediated mechanism. There is, however, a body of experimental data that is consistent with an effect of antiandrogenic industrial chemicals on male sexual differentiation (Gray et al. 1999(Gray et al. , 2000. Moreover, mechanisms other than endocrine disruption may be involved in testicular toxicity; for example, the nematocide dibromochloropropane, an alkylating agent, is one of the most potent known testicular toxins in adults (Joffe 2001). In this review we focus on the estrogen hypothesis of TDS.
Although several systematic reviews of the literature on the association between estrogenic agents and the disorders thought to belong to the TDS have been published, they are predominantly qualitative and the only quantitative summary estimate of the association between prenatal exposure to estrogenic agents and testicular cancer was published over 10 years ago (Toppari et al. 1996). The primary objective of a quantitative metaanalysis is to combine the results of previous studies examining a specific research question to arrive at a summary conclusion about a body of research. It has been found particularly useful when individual studies are too small to yield a valid conclusion, but it cannot, however, correct for bias and confounding. When applied to observational studies, subset analysis can be a useful tool to explore the reasons for discrepancies among the results of different studies.
The objectives of this research were therefore to carry out a quantitative meta-analysis of the association between three of the end points related to TDS and prenatal exposure to estrogenic agents that would account for both the size and quality of the studies included and yield updated summary estimates in light of Environmental Health Perspectives • VOLUME 116 | NUMBER 2 | February 2008 149 Review the body of research carried out since the formulation of the estrogen hypothesis. Inclusion in this analysis was based on mechanistic criteria, and the plausibility of an ER-α-mediated mode of action was specifically explored. Moreover, subset analysis has been applied to categories of compounds with estrogenic potencies differing by several orders of magnitude in an attempt to detect the existence of any potency-response trend. Most of the studies of sperm quantity or quality have been concerned with time trends rather than etiology, and this end point was not considered further here.

Material and Methods
Identification and selection of literature. A computerized search was conducted using the databases PubMed (National Center for Biotechnology Information 2007) and Web of Science (ISI Web of Knowledge 2007) for the period 1970 to April 2007. The general search keywords were "estrogen," "risk," "dose," and either "hypospadias," "cryptorchidism," or "testicular cancer." A preliminary identification was performed by screening the titles and, if relevant, the abstracts of retrieved literature. The next stage was to check the citations and references of selected studies. This was an iterative process, repeated until no new study could be identified. A set of both inclusion and exclusion criteria was defined, and all relevant literature was then checked for eligibility. The inclusion criteria considered were a) study design, namely, either a case-control, cohort, or clinical trial; b) written in English; c) exposure to one or a mixture of known estrogenic compounds; and d) sufficient data reported to be used in meta-analysis.
The following exclusion criteria were used: • Exposure to a group of compounds (suspected endocrine disruptors) for which mode of action was unspecified, for example, pesticides. • Studies of exposure to phytoestrogens. Some phytoestrogens have been found to have a greater binding affinity for ER-β than for ER-α and can result in agonistic or antagonistic effects (Mueller et al. 2004). • Studies of maternal endogenous hormones. • Studies of the same cohort as this would bias the results towards the particular studies. • Incomplete data.
Data extraction and quality rating. In addition to the number of exposed and nonexposed cases and controls, and risk ratios (RRs) with their confidence intervals (CIs), information regarding the study design, estrogenic agent, geographic location of the study, and year of publication were extracted from the selected literature to allow subset analysis to be carried out. When more than one RR was reported, the following priorities were set for choice: • Adjusted RRs were used, except when the study provided only unadjusted estimates. • When multiple estimates were given, the RR estimator on which the authors had relied for their assessment of causal association was used. • Overall RRs were chosen instead of those derived from further stratifications. If an overall estimate was not provided, the RRs of the maximum duration of exposure or the maximum exposure concentration were chosen. Several aspects of the quality of each study were also recorded according to a rating scheme adapted from those previously described (Altman 1991;Rushton 2000). Every criterion was assessed on a scale of 0 to 2, 0 suggesting that it was not present, 1 when it was unclear, and 2 when that criterion was satisfied. A maximum score of 50 and 52 could be assigned for retrospective (case-control) and prospective (cohort and clinical trials) studies, respectively. This enabled a quality sensitivity analysis to be performed to check the influence of studies with low quality on the pooled estimate.
Data analysis. Graphical representation. The RRs and CIs were plotted against the year of publication to determine whether any positive or negative trends in reporting RRs had occurred over time. Similarly, quality scores were plotted against the year of publication to investigate whether the quality of studies improved over time. To assess publication bias, a funnel plot (SE vs. RR) was produced based on the assumption that smaller studies are less precise in their RRs and thus have less weight and larger SE and should scatter more widely at the lower end of the graph, whereas larger studies will tend to be closer together (Sterne et al. 2001). Forest plots present the RRs against the reference of the study and help check homogeneity visually.
Statistical pooling. Pooled estimates and 95% CIs were calculated using both a fixedeffects model (Mantel-Haenszel method) and a random-effects model (DerSimonian-Laird method), allowing evaluation of the dependence of the conclusions of the analysis on the model assumptions. A summary estimate is considered statistically significant at the 0.05 level if its CI does not include unity.
The Mantel-Haenszel pooled effect estimate was used in a chi-square statistical test of homogeneity to assess the between-study variance. The magnitude of the test statistics depends on the weight of each study. When the number of studies is low or the studies themselves are small, the test statistic Q tends to be small. Tests of heterogeneity in metaanalyses are generally low in their power to reject the null hypothesis of homogeneity. For this reason, the chi-square statistical test of homogeneity was carried out at both 0.05 and 0.1 significance levels. Additionally, pooled estimates calculated using fixed effect and random effect models differ only if there is lack of homogeneity between studies. The estimates obtained by both methods were therefore compared to better assess potential heterogeneity between studies, in which case a single summary estimate of effect may be considered inappropriate.
Subset and sensitivity analyses. To investigate potential sources of heterogeneity between studies, we performed subset analyses for the study design, estrogenic agent, and geographic location.
Some studies exploring the influence of hormonal treatment during pregnancy did not specify the type of hormone. From what is known of the hormonal treatment of common conditions occurring during pregnancy, it was deemed reasonable to assume that they would have been likely to include estrogens, and these studies were included in the analysis. The validity of this assumption was tested by applying stricter criteria and calculating a summary estimate of effect excluding any study in which the hormone used had not been specified. Further sensitivity analysis was performed by excluding low-quality studies and extremes (exclusion of the studies with the largest and smallest RR estimators and exclusion of the studies with the largest and smallest weights) to verify that either the quality of the studies or one particular study did not have an excessive influence on the pooled estimate.

Results
A total of 50 studies were identified for the association between in utero exposure to estrogenic agents and hypospadias and/or cryptorchidism, including 16 that had not been included in previous systematic reviews. Sixteen studies, of which 8 were new studies, were included in the calculation of a summary estimate of effect for either or both end points (Table 1). Studies predating the formulation of the TDS hypothesis often were designed to explore the association of in utero exposure to a range of pharmaceuticals with birth malformations. Other than 2 recent studies for which pesticide exposure was determined by chemical analysis of specific compounds, assessment of exposure to pesticides is generally derived from the occupation of the mother and specific agents are not identified.
Of the 12 studies identified for the association with testicular cancer, only 3 were excluded from the calculation of a summary estimate of effect (Table 2).
Hypospadias. The data from studies included in the meta-analysis for hypospadias are summarized in Table 3. Three extreme values, two greater than and one lower than unity, can be identified visually from the forest plot of the RRs and their CIs ( Figure 1). These extremes correspond to studies with larger SEs, and the shape given to the funnel plot ( Figure 2) by those smaller positive studies would be consistent with publication bias.
These two extreme positive risk ratios were, however, reported after what is commonly referred to as "third-generation exposure" to DES, when the mother herself had been exposed to DES prenatally. It was recognized that the inclusion of such studies in the metaanalysis could have introduced heterogeneity, and the influence of this choice was investigated in the subset analysis. Plots of the quality score and RRs versus year of publication did Testicular dysgenesis syndrome meta-analysis Environmental Health Perspectives • VOLUME 116 | NUMBER 2 | February 2008 Table 1. Studies identified for the association between in utero exposure to estrogenic agent and hypospadias and cryptorchidism.

Reference
End point Comment Previous reviews a Aarskog 1970 Hypospadias Data on progestins treatment only Beard et al. 1984 Cryptorchidism Included R-W, T, S Beral and Colwell 1981 Cryptorchidism Study too small to calculate risk ratio Sx Berkowitz and Lapinski 1996 Cryptorchidism Use of clomiphene before pregnancy recognized Bernstein et al. 1988 Cryptorchidism Maternal endogenous hormones S Bhatia et al. 2005 Cryptorchidism Included Hypospadias Included Bianca et al. 2003 Hypospadias Occupational exposure of fathers to pesticides Burton et al. 1987 Cryptorchidism Maternal endogenous hormone levels S Calzolari et al. 1986 Hypospadias Oral contraceptive use before pregnancy recognized R-Wx, S Cosgrove et al. 1977 Cryptorchidism No control data for documented abnormalities R-Wx, Sx Hypospadias Czeizel et al. 1979 Hypospadias Progesterone treatment R-Wx Czeizel et al. 1999 Cryptorchidism Ecological study design Hypospadias Davies et al. 1986 Cryptorchidism Oral contraceptive use before pregnancy recognized S Depue 1984 Cryptorchidism Same cohort as Depue (1988) Gill et al. 1977 Cryptorchidism No genitourinary abnormalities in exposed infants T Hypospadias Gill et al. 1979 Cryptorchidism Cryptorchidism in men with testicular hypoplasia T, S Harlap et al. 1975 Cryptorchidism All cases exposed to progesterone R-W Hypospadias Harlap and Eldor 1980 Cryptorchidism No cases after oral contraceptive use during pregnancy R-W Hypospadias Harlap et al. 1985 Cryptorchidism Included R-W Hypospadias Included Heinonen et al. 1977 Hypospadias Included Hemminki et al. 1999 Hypospadias No exposed controls Henderson et al. 1976 Cryptorchidism not suggest any significant trends in quality of the studies or estimates of effect over time (not shown).
The pooled estimates of effect by both the Mantel-Haenszel and DerSimonian-Laird methods are very close to unity, and no relationship between in utero exposure to estrogenic agents and hypospadias could be detected (Table 4). None of the chi-square tests allowed the rejection of the null hypothesis of homogeneity between the studies at the 0.05 or 0.1 level of statistical significance. The subsets of studies in which exposure to DES and pharmaceutical estrogens were investigated yielded statistically significant risk ratios with both models, although the modest discrepancy between the fixed-effects and randomeffects estimates suggests heterogeneity. Summary estimates for the latter subset were no longer significant when studies that included DES exposure were excluded. Although these results were based on four studies that all addressed in utero exposure to oral contraceptives, some heterogeneity between studies remained. Excluding the studies of third-generation exposure to DES, values for the summary estimate of effect were found to be 1.33 (95% CI, 0.63-2.83) by the Mantel-Haenszel method and 1.31 (95% CI, 0.52-3.26) by the DerSimonian-Laird method, a very modest and nonsignificant increase in risk. Excluding third-generation exposure from the DES subset yielded estimates of 2.02 (95% CI, 1.12-3.65) by the Mantel-Haenszel method and 2.00 (95% CI, 0.97-4.15) by the DerSimonian-Laird method, on the basis of two studies investigating exposure to any estrogenic drug during the first trimester of pregnancy. The difference between the results obtained by the two models for studies of third-generation exposure to DES was reduced only slightly by excluding the study by Klip et al. (2002); the Mantel-Haenszel method yielded an estimate of 2.46 (95% CI, 0.91-6.67) and the DerSimonian-Laird method of 2.18 (95% CI, 0.64-7.46). The latter study's cohort had been recruited in a fertility clinic, and whether results obtained with subfertile women are generalizable to all women exposed to DES in utero has been questioned (Hernandez-Diaz 2002).
Although the equality of the results obtained by both methods for the environmental estrogens subset suggests those results are robust, the influence of the weight of the study by Vrijheid et al. (2003) cannot be underestimated, as shown by the sensitivity analysis. Exclusion of this study from the analysis yielded a statistically significant Mantel-Haenszel estimate but a lower and not statistically significant DerSimonian-Laird estimate, revealing heterogeneity. A statistically significant estimate was obtained for prospective studies by the Mantel-Haenszel method, but the wide difference with the estimate using the random effect model was suggestive of heterogeneity. Geographic subsets point to a higher risk in Latin America, although the pooled estimates for this location were based on only two studies and did not reach statistical significance. Included Hemminki et al. 1999 No cases Henderson et al. 1979 Included T, S Moss et al. 1986 Included T, S Schottenfeld et al. 1980 Included T, S Strohsnitter et al. 2001 Included S Walcott et al. 2002Phytoestrogens Weir et al. 2000 Included S a The letters T and S stand for Toppari et al. (1994) and Storgaard et al. (2006), respectively. Abbreviations: E, exposed; NE, nonexposed. In addition to the results of the sensitivity analysis presented in Table 4, a pooled estimate of effect was calculated when a stricter inclusion criterion was applied, namely, excluding results from the study by Monteleone-Neto et al. (1981). This had little influence on the overall result, generating summary estimates of 0.97 (95% CI, 0.83-1.13) for the fixed effect model or 0.93 (95% CI, 0.80-1.09) for the random effect model.
Cryptorchidism. Data for the six studies included in the meta-analysis for cryptorchidism can be found in Table 5. The results of only two studies significantly differ from unity, as illustrated by the forest plot (Figure 3). The small number of eligible studies renders analysis of the funnel plot and potential for publication bias difficult (Figure 4). The SEs do, however, illustrate well that the studies were all relatively small. No time trends for the estimate of effect or the quality of studies could be detected (not shown).
As presented in Table 6, the pooled estimates of effect by both the Mantel-Haenszel and DerSimonian-Laird methods are marginally superior to unity, and their relative divergence implies there may be sources of heterogeneity. Chi-square tests did not, however, detect that any of the subsets were significantly heterogeneous. Excluding studies in which DES exposure was examined, either exclusively or along with hormonal therapeutics, yielded summary estimates consistent with no relationship. Statistical pooling of the studies including DES exposure generated a statistically significant estimate by the Mantel-Haenszel method, suggesting a doubling of the risk of cryptorchidism after in utero exposure to DES. The same estimate by the DerSimonian-Laird method did not, however, reach statistical significance and the difference relative to the fixed effect model is indicative of heterogeneity. The heterogeneity introduced by the DES subset of studies can again be observed by comparing the results obtained for all pharmaceutical estrogens with those obtained by pooling the two studies of accidental use of oral contraceptives during pregnancy. Study design also appeared to be a source of heterogeneity. If case-control studies are prone to recall bias, this subset also included the study with the highest estimate, itself a source of heterogeneity, as shown by the sensitivity analysis. Excluding the study by Depue (1988) reduced the difference between estimates by both models, the Mantel-Haenszel estimate then calculated as 1.29 (95% CI, 0.87-1.91) and that by the DerSimonian-Laird method as 1.23 (95% CI, 0.81-1.86). This was also observed for the American subset of studies. When the Depue (1988) study is omitted, the Mantel-Haenszel method yielded a no longer statistically significant estimate of 1.34 (95% CI, 0.84-2.14) and the DerSimonian-Laird method an estimate of 1.27 (95% CI, 0.72-2.23).
Applying a stricter exclusion criterion to studies examining hormonal treatment did not affect which studies were included in the meta-analysis of cryptorchidism. The study with the highest weight appears to lower the overall estimates, whereas increasing quality seems to reduce heterogeneity and lower the estimate of effect toward unity. These variations did not, however, influence the overall conclusion that aside from the DES studies subset, summary estimates did not detect any association between in utero exposure to estrogenic substances and cryptorchidism.
Testicular cancer. Nine studies were included in the meta-analysis of testicular cancer and the data used are summarized in Testicular dysgenesis syndrome meta-analysis Environmental Health Perspectives • VOLUME 116 | NUMBER 2 | February 2008 Table 4. RRs (95% CIs) of the summary estimate of effect, subsets, and sensitivity analyses for the association between hypospadias and prenatal exposure to estrogenic agents.    Abbreviations: E, exposed; NE, nonexposed. Table 7. Of these, 4 had not been included in the summary estimate previously calculated by Toppari et al. (1996). The lack of homogeneity between studies is evident from the forest plot ( Figure 5). Further, the funnel plot ( Figure 6) also illustrates the relatively small size of the included studies. Although a positive trend over time was found for the quality of the included studies (Figure 7), no significant time trend could be detected for the effect size (not shown). Both the fixed and random effect models yield a statistically significant estimate; however, the discrepancy between the two results is suggestive of heterogeneity despite the result from the chi-square test (Table 8). Conversely, the subset analysis was limited by the similarity of the question addressed by the studies included. Eight of the nine studies were interested in hormonal exposure and were conducted in the United States. Despite this, statistically significant heterogeneity between the studies was detected at the 0.1 level. Pooling the two studies examining DES exposure specifically produced a raised but statistically nonsignificant result. Despite the unexplained heterogeneity, all estimates that were calculated point to a doubling of the risk of developing testicular cancer after exposure to estrogenic agents in utero. The work on chlorinated biphenyls (PCBs) by Hardell et al. (2004) was the only study examining environmental estrogens. Its size was relatively small, and it did not detect such an effect.
Applying a stricter exclusion criterion to studies examining hormonal treatment excluded four studies from the meta-analysis; namely, Brown et al. (1986), Gershman and Stolley (1988), Henderson et al. (1979), and Weir et al. (2000). This resulted in a slightly lower Mantel-Haenszel estimate of 1.98 (95% CI, 1.23-3.18) and if the DerSimonian-Laird estimate remained equal to 1.59, because of the wider confidence interval (95% CI, 0.93-2.69), statistical significance was no longer achieved. The sensitivity analysis is consistent with some heterogeneity between the studies, the estimates obtained being relatively sensitive to the exclusion of particular studies varying above and below a risk estimate of 2. The quality of the studies seemed to explain at least some of this heterogeneity.

Discussion
While it is clear that hypospadias, cryptorchidism, and testicular cancer are all positively associated with prenatal exposure to DES, this meta-analysis was unable to produce evidence that such effects were associated with environmental estrogens or even accidental use of oral contraceptives during pregnancy. This is consistent with the results obtained in earlier meta-analyses (Raman-Wilms et al. 1995;Toppari et al. 1996).
The main limitations of meta-analysis are a) the susceptibility of its summary results to publication bias, b) the influence of the quality of studies, c) the possibility of including multiple results from the same study, and finally, d) heterogeneity between studies that could lead to invalid conclusions. The methodology employed in this present review attempts to address these issues. Additionally, the importance of carrying out and reporting a sensitivity analysis was illustrated by the case of hypospadias where the weight attributed to one particularly large study had a nonnegligible influence on the results. In this particular case, the study by Vrijheid et al. (2003) inferred exposure to phthalates from registry data about occupation, and although such an approach can allow the analysis of a great number of cases, assessment of exposure is much more likely to be prone to confounding. The number of studies included in meta-analyses lies typically between 5 and 15, and the results presented here also fall within this range. The size of the homogeneity test statistic depends on both the number and size of individual studies. The funnel plots offer a good visual representation of the precision and size of individual studies, and it is clear that most studies published on the association between estrogenic agents and the probable end points of a TDS were found to be relatively small. The chi-square tests had, therefore, a relatively low power to detect heterogeneity. However, in the absence of statistical Abbreviations: E, exposed; NE, nonexposed.  Brown et al. 1986Depue et al. 1983Gershman and Stolley 1988Hardell et al. 1979Henderson et al. 1979Moss et al. 1986Schottenfield et al. 1980Strohsnitter et al. 2001Weir et al. 2000 heterogeneity, the results of the fixed effect and random effect models should be virtually identical, and the comparison of results obtained by applying both the Mantel-Haenszel and DerSimonian-Laird models enabled the exploration of sources of heterogeneity despite this low statistical power. If the quality of the studies was found to explain some of the heterogeneity observed, particularly in the case of testicular cancer, the remaining heterogeneity could not be explained solely by the fact that environmental, and therefore generally much weaker, estrogens were included in the analysis. The systematic review of published literature yielded relatively few studies examining the association of male urogenital abnormalities or testicular cancer with environmental estrogens specifically; a number of studies concerned with an association with broad categories of putative endocrine disruptor, most often pesticides, were excluded from the meta-analyses. This illustrates the difficulties associated with assessment of exposure, pesticide exposure often being inferred from parental occupation rather than direct measurement. Furthermore, there is increasing evidence that, in accordance with pharmacokinetic theory, the effects of xenobiotics acting via the same mechanism can be predicted fairly accurately by concentration addition (Zhu et al. 2006). Accurately accounting for combined exposure or adjusting for the confounding introduced by environmental exposures will probably require the development of mechanism-specific biomarkers of exposure.
When DES is excluded, there is no conclusive evidence of an effect of pharmaceutical estrogens. Exposure to such estrogens is related mainly to the accidental use of oral contraceptives during pregnancy or hormonal pregnancy tests. Such estrogenic pharmaceuticals often are given in combination with progestagens, and it is legitimate to question whether unopposed estrogens would have the same effects as opposed estrogens. This also highlights another difficulty associated with exposure assessment, that of critically sensitive periods of development and the ascertainment of whether exposure took place during a "window" of susceptibility to hormone disruption. Nonetheless, studies in which maternal levels of hormones were measured in the first and third trimester of pregnancy do not support an association with elevated estrogen levels but rather indicate that a lower estrogen/androgen ratio and/or higher levels of α-fetoproteins may be beneficial (McGlynn et al. 2005;Zhang et al. 2005). If in animals both estrogenic and antiandrogenic compounds have been associated with end points consistent with those of human TDS (Fisher 2004;Veeramachaneni 2000), epidemiologic evidence remains elusive. Alternatively, the doubling of the risk estimates of all three effects associated with DES exposure would be consistent with a shared etiology and the TDS hypothesis. It does not constitute conclusive evidence of an estrogenic mode of action, however, as common etiologic factors could be related to the underlying condition for which DES was prescribed. Furthermore hypospadias, cryptorchidism, and testicular cancer have all been found to be associated with low birth weight, suggesting a potential association with an underlying placental defect.
The understanding of the importance of endogenous estrogens in normal adult testicular function is becoming clearer. Their roles during fetal life, however, remain relatively unclear, but those mediated by the ER-α or ER-β have been shown to differ (Habert et al. 2006). Interestingly, DES has been found to have similar affinity for both receptors, whereas estradiol has only a slightly stronger affinity for ER-α compared with ER-β (Mueller et al. 2004). ER-α has been detected in undifferentiated gonads as early as 10 days postconception in the mouse and found to be localized in the Leydig cells of fetal testis in rodents (Habert et al. 2006). Studies of the expression of ER-α and ER-β in human and nonhuman primates have so far yielded inconsistent results. Gaskell et al. (2003) reported that ER-α could not be detected in human fetal testes between weeks 12-19 of gestation, whereas Shapiro et al. (2005) found that ER-α was apparent by week 12, its levels peaked at 16 weeks before diminishing, and it was localized in Leydig cells. Current research focus has shifted to the role played by testosterone, anti-Müllerian hormone and insulin-like factor 3 produced by the fetal testes during masculinization. In the male rat, exposure to high levels of estrogens has been shown not only to suppress testosterone production but also to downregulate the expression of the androgen receptor protein in reproductive target tissues including the testes, Wolffian duct, and prostate (Sharpe 2006). Further research in this area may help shed light on possible mechanisms of injury or relevance of the rodent model.
The subset analyses did not generate many clues to explain the heterogeneity of the collected data. This is, however, consistent with the wide geographic variability in the incidence of the conditions of interest (Boisen et al. 2004;Richiardi et al. 2004). Interactions between genetic susceptibility and the environment have been the focus of research in this area (Martin et al. 2007), and advances in genomics have allowed the identification of polymorphisms associated with hypospadias, cryptorchidism, and testicular cancer (Beleza-Meireles et al. 2006;Kurahashi et al. 2005;Starr et al. 2005;Yoshida et al. 2005). Such discoveries may, Testicular dysgenesis syndrome meta-analysis Environmental Health Perspectives • VOLUME 116 | NUMBER 2 | February 2008 Table 8. RRs and 95% CIs of the summary estimates, subsets and sensitivity analyses for the association between testicular cancer and prenatal exposure to estrogenic agents.  Year of publication however, give rise to as many questions as they offer to answer. This is well illustrated by the recent identification of the association of a variant of the gene for the ER-α with hypospadias and cryptorchidism in Japanese cohorts (Watanabe et al. 2007;Yoshida et al. 2005) that has now been found to be associated with a decreased incidence of hypospadias in a European cohort (Galan et al. 2007).

Conclusion
The modest increase in risk for all three end points associated with DES exposure is consistent with a shared etiology and the TDS hypothesis, whereas the results of the subset analyses suggest the existence of yet unidentified sources of heterogeneity between studies or within the study populations. Although 10 years of further research on the potential effects of endocrine disruptors on male reproductive health have provided some clues regarding the etiology and mechanism of conditions such as hypospadias, cryptorchidism, and testicular cancer, there is still no conclusive evidence of the role played by environmental estrogens.