Minireview: Organizational Hypothesis: Instances of the Fingerpost

Abstract

There is now compelling evidence that the ratio of the length of the second digit divided by the length of the fourth digit (2D:4D) is affected by prenatal androgens in humans. This ratio is greater in females than males from fetal life through adulthood, correlates with polymorphism in the androgen receptor gene in men, is feminine in XY androgen insensitivity syndrome, and masculinized in congenital adrenal hyperplasia. Using 2D:4D as a correlate, researchers have found evidence that prenatal androgens affect many sexually differentiated human behaviors, including sexual orientation in women (but not in men), attention deficit disorder, autism, eating disorders, aggression, and risk-taking. In each case, lower 2D:4D, indicative of greater prenatal androgen stimulation, is associated with behavior more commonly displayed by males than females. The correlation between 2D:4D and prenatal androgen stimulation is too imperfect to accurately predict the phenotype of a particular individual, even in terms of sex. However, digit ratio is the best available retrospective marker of average differences in prenatal androgen stimulation between groups of people, and/or correlations of prenatal androgen stimulation with particular behaviors and characteristics within a group. Thus digit ratios offer a valid test of the organizational hypothesis that androgens act early in life to masculinize various human behaviors.

The comparative lengths of adult human fingers may provide an imperfect but readily accessible correlate of prenatal androgen stimulation, allowing tests of the organizational hypothesis that prenatal androgens affect human behaviors.

“When in the investigation of any nature the understanding is so balanced as to be uncertain to which of two or more natures the cause of the nature in question should be assigned … instances of the fingerpost show the union of one of the natures with the nature in question to be sure and indissoluble, of the other to be varied and separable; and thus the question is decided, and the former nature is admitted as the cause, while the latter is dismissed and rejected.”

—Francis Bacon, Novum Organum (1620)

For 50 yr the dominant theme in behavioral endocrinology has been the “organizational hypothesis” that exposure to androgens such as testosterone (T) early in life permanently masculinizes the brain and behavior (1). A persistent caveat has been whether this dictum, which summarizes thousands of findings in nonhuman animals, applies to our own species at all. The biggest stumbling block to resolving that caveat is ethical: we cannot manipulate androgens in male or female fetuses to see how that affects their later behavior. Thus tests of the organizational hypothesis in humans must rely on correlations between prenatal androgen and adult behavior. Even this approach is hampered by the slow course of human development. Assuming we could safely measure circulating androgens in fetuses throughout gestation, it would be years before we could assay many behaviors of interest, such as sexual orientation, to see whether prenatal T correlates with adult behaviors. Thus any means of retrospectively assessing prenatal androgen stimulation in adults, whose behavior can be assessed immediately, could test the application of the organizational hypothesis to human behaviors.

Just such a retrospective marker was suggested by the 1998 report from John T. Manning and colleagues (2) of a sex difference in the ratio of the length of the index finger (2D) to that of the ring finger (4D), because this 2D:4D is greater in females than in males, even in 2-yr-old children. The literature about mammalian sexual differentiation offered only four hypotheses about how girls might develop a larger 2D:4D than boys: 1) the sex difference in perinatal androgen exposure, which accounts for most sex differences in body morphology, affects 2D:4D; 2) secretion of some other gonad-specific hormone, such as anti-Mullerian hormone from the testes, affects digit ratio; 3) the sex chromosomes affect digit development directly through nonhormonal mechanisms; 4) social factors encouraging sex differences in the behavior of children affect digit growth.

Do Prenatal Androgens Affect Digit Ratios?

Subsequent findings confirmed the first hypothesis and decisively ruled out the others. The strongest evidence that androgens affect digit ratios is the report that normal polymorphism in the androgen receptor (AR) gene correlates with digit ratios in men. The greater the number of CAG repeats in exon 1 of AR (coding for a longer polyglutamine stretch), the less effective AR protein is produced (3,4), in extreme cases causing reduced sperm counts and gynecomastia (5). In a nonclinical sample of 50 men there was a significant positive correlation between the number of CAG repeats and 2D:4D (6), as would be predicted if less effective AR produced less masculine (i.e. larger) digit ratios. The study estimates that variation in AR’s CAG repeat domain alone accounts for about 9% of the variance in men’s digit ratios. Any other AR polymorphism would contribute additional variance to 2D:4D, as would individual differences in prenatal androgen secretion. Mutations severely disrupting AR function, carried by XY women with androgen insensitivity syndrome (AIS), result in feminine digit ratios (7), further implicating androgen’s role. The variability in digit ratios among the AIS women in this study presumably reflected variability in AR alleles, although these were not examined. Similar variability is seen in other androgen sensitive characters in women with AIS (e.g. Ref. 8).

The effect of these various AR polymorphisms on digit ratio highlights an advantage of such biomarkers: they reflect androgen sensitivity as well as hormonal levels. Thus digit ratios reflect total androgenic stimulation, proportional to both the level of circulating androgens and the individual’s sensitivity to those hormones. For testing the organizational hypothesis, total androgen stimulation is a more appropriate measure than prenatal androgen levels alone, because any androgenic stimulation of the brain affecting behavior would also be proportional to both androgen sensitivity and circulating hormone levels. While it may be convenient to speak loosely of digit ratios as though they reflect only prenatal levels of androgen, it is important to keep in mind that in fact they reflect total androgen stimulation.

While both of these finding concerning the AR gene strongly implicate androgens in affecting digit ratios, neither addresses when androgens might affect them. Reports that the sex difference in digit ratio is present in utero (9,10) make it clear that if androgens are responsible for the sex difference, they must act before birth. Confirming that inference, digit ratios are also masculinized in females with congenital adrenal hyperplasia (CAH) (11,12), a condition that exposes fetuses to elevated levels of androgens before birth. Likewise, males with CAH are also exposed to elevated androgens before birth, primarily in the form of several-fold higher levels of androstenedione than in control males (13,14), and they have hypermasculinized digit ratios (11,12). As excess androgen production in cases of CAH is typically controlled by medication at birth, these findings demonstrate that increased prenatal androgens masculinize digit ratios in both sexes. Because the prenatal studies measured bone lengths and human bone growth plates contain AR (15), androgens may act directly on developing bone to affect digit growth, but the cellular site of androgen action on 2D:4D has not been determined.

Not only do these findings strongly confirm the idea that prenatal androgens affect digit ratios, they also refute all three remaining alternatives. If direct, nonhormonally mediated effects of sex chromosomes or nonandrogenic hormones from testes were responsible for masculinizing digit ratios, then 2D:4D should be masculine rather than feminine in AIS women, who possess an XY karyotype and fetal testes that suppress Mullerian duct development. Likewise, the sex difference in fetal 2D:4D rules out the possibility that socially-induced sex differences in behavior cause the sex difference in ratios. Sex differences in behavior in utero may contribute to the sex difference in digit ratios, but the other findings indicate that such putative sex differences from behavior would have to be driven by androgens. To use Francis Bacon’s analogy, the fingerposts from these many observations all point to an effect of prenatal androgens on human digit ratio (Fig. 1A).

Figure 1.

Instances of the fingerposts. A, Many different experimental findings concerning 2D:4D are consistent with the notion that prenatal androgens affect digit ratios in humans, while several point away from alternative hypotheses, such as a direct role of sex chromosomes, or social influences. B, The correlations of digit ratio with various behaviors, each of which is normally displayed more by one sex than the other, all point in a particular direction—the direction indicating that prenatal androgen stimulation contributes to the masculinization of diverse human behaviors, despite the obvious importance of social factors and learning. The particular behavior-digit ratio relationships displayed are among those that have been replicated at least once (see text).

What Have Digit Ratios Told Us about Human Behavior?

Thus, a deluge of reports have related digit ratios to a host of sexually differentiated human behaviors. These studies do not correlate digit ratios and behaviors across both sexes, as that would not test any hypothesis: such a correlation is inevitable if there’s known to be a sex difference in both the behavior and 2D:4D. Rather, the test of the organizational hypothesis is to detect an effect of prenatal androgen stimulation, as reflected by digit ratio, within one sex or the other. Using the logic of strong inference, if the sex difference in prenatal androgens is responsible for the difference in behavior between sexes, then individual variation in prenatal androgen stimulation within a sex should affect the behavior as well. In that case, you would predict a relationship between digit ratios and the behavior in one or both sexes, and that relationship should be in a particular direction, namely more masculine behavior should be associated with more masculine, smaller 2D:4D. Obtaining that result would confirm the prediction and support the hypothesis. If you see no relationship between digit ratios and the behavior, then the hypothesis is not supported (assuming you measured accurately, sampled randomly, have a reasonable sample size, etc.) If the relationship between digit ratios and the behavior is significant in the direction opposite to that predicted, then the hypothesis is refuted, and you must replace or adjust your hypothesis to account for the 2D:4D data.

For example, our group reported that mean 2D:4D is more masculine in lesbians than heterosexual women (16), a finding that has been replicated many times (17,18,19,20,21,22,23,24,25) [but was not seen by Lippa (26)]. These findings indicate that females exposed to greater prenatal androgen stimulation are more likely to be sexually attracted to women in adulthood, as would be predicted if the organizational hypothesis applies to human sexual orientation (i.e. if prenatal androgen in males promotes sexual attraction to females). A recent meta-analysis of all published and several unpublished datasets concluded that the difference in digit ratios between lesbians and straight women is significant, and that “58 additional studies with null effects would be required to produce a statistically nonsignificant difference” (27).

Interestingly, we saw no difference between gay and straight men, and this negative finding has also been confirmed repeatedly (23) and in the meta-analysis (27). This absence of a difference is also informative, requiring modification of the hypothesis, as it suggests that either prenatal androgens do not affect sexual orientation in males, or that there is a “ceiling effect” such that all males are exposed to sufficient prenatal androgen stimulation to receive the maximum effect of androgens on orientation. A similar situation applies to male reproductive behavior in adult rodents: while eliminating T entirely through castration abolishes male copulatory behavior in many species, among gonadally-intact males there is no correlation between circulating T and sexual vigor, and a fraction of normal levels of androgen can maintain full copulatory behavior in rats (28). If an analogous situation applies to sexual orientation in human males, such that all or nearly all males are exposed to enough prenatal androgen to maximize its effects on sexual orientation, then variation in sexual orientation among males would be due to other factors, including, but not limited to, variation in whatever genes androgens might act upon to affect orientation. Alternatively, if males exposed to either low (26) or high levels (21) of prenatal androgen are more likely to be gay in adulthood, the mean digit ratio might be equivalent in homosexual and heterosexual men.

There have been many other reports of correlations between digit ratios and human behaviors that are more common in one sex than the other. For example, if greater prenatal androgen stimulation contributes to the higher incidence of autism in males than in females, as the organizational hypothesis would predict, then individuals with autism should have more masculine digit ratios, on average, than same-sex control subjects, and symptoms of autism spectrum disorder should negatively correlate with digit ratios, both of which have been reported (29,30,31). Likewise, if the organizational hypothesis explains why boys are more likely than girls to display attention deficit disorder (ADD), then ADD symptoms should be negatively correlated with digit ratios within a sex, as has been seen at least three times (32,33,34). Conversely, if low levels of prenatal androgen contribute to females’ greater susceptibility to eating disorders, then risk for eating disorders should correlate, in the predicted positive direction, with digit ratios, which has been reported in both females (35) and males (36).

Curiously, sex differences, AR polymorphisms, and clinical conditions all seem to have a greater effect on 2D:4D of the right hand than the left, suggesting this hand is more responsive to androgenic effects, for reasons that remain unknown. For example, two studies of females with CAH found no significant difference in the ratios on the left hand (11,37). Tellingly, correlations between human behaviors and digit ratios are also typically stronger on the right hand than the left, which is also consonant with the notion that the behavior is affected by prenatal androgens.

Many other sexually differentiated behaviors have been correlated with digit ratios and replicated, including aggression (38,39) and risk taking (40,41). A PubMed search [ “digit ratio(s)” or “2D:4D”] at the close of 2009 yielded 226 studies with only Manning et al.’s report (2) predating 2000. Not all of these papers are about human behavior, and not all of those report significant correlations between 2D:4D and behavior, but it is striking how many do report significant correlations. Importantly, the direction of the correlation is as would be predicted by the organizational hypothesis, despite the diversity of behaviors examined. One is left with the impression that androgens are affecting not just a few but myriad aspects of human behavior that differ between the sexes (Fig. 1B).

However, these studies also suggest that prenatal androgens’ effect on human behavior is subtle, as one might expect in a socially complex, learning-dependent species. For example, in studies of female sexual orientation and 2D:4D, there are many lesbians with very feminine ratios. Presumably prenatal androgen had no effect on their sexual orientation, as the plasticity of sexual orientation in women (42)would also suggest. For every behavior, the correlations between digit ratio and behavior are small or medium in size, indicating that androgens cannot account for the majority of the variance in behavior. On the other hand, because digit ratio is an imperfect indicator of prenatal androgens, our topic for the remainder of this review, correlations between behavior and 2D:4D are necessarily conservative estimates of the effect of prenatal androgen. If digit ratios reflected prenatal androgen stimulation perfectly, such correlations might be higher.

How Accurately Do Digit Ratios Reflect Prenatal Androgen?

The best benchmark is the basic sex difference, because virtually all males are exposed to higher levels of prenatal androgen than any females. Yet the sex difference in 2D:4D is only about one half of a sd [d = 0.5(7)], so there is considerable overlap in the distributions of digit ratio in men and women. This overlap between groups engenders misunderstanding about digit ratio research. The recent report that 2D:4D is feminine in AIS, strongly confirming the role of androgens on digit ratios and contradicting all the alternative hypotheses, nevertheless concluded, “… digit ratio is not a good marker of individual differences in prenatal androgen exposure” (7). The researchers justify this view because discriminant analysis did not accurately classify individuals as male, female, or XY female with AIS, from digit ratio alone. But there is an important distinction between using digit ratios to discriminate and predict the phenotype of an individual, vs. using them to compare differences between groups, or correlations within a group.

Cohen’s (43) discussion of effect sizes is useful for considering this distinction. He regards ds of 0.5, such as the sex difference in 2D:4D (7), as medium-sized effects, equivalent to the difference in height between 14-yr-old and 18-yr-old girls. Because many individual 14-yr-old girls are taller than many individual 18-yr-olds, a discriminant analysis would make many mistakes trying to predict each girl’s age from height alone. Still the mean difference between them represents an indisputable biological mechanism of aging: girls tend to grow. Therefore, a group of randomly selected 14-yr-old girls will reliably have a shorter mean height than a group of randomly selected 18-yr-old girls. How often their means will differ is directly proportional to the size of d. Likewise, we could correlate their height with particular behaviors to see whether age is a factor affecting the behavior. If the behavior really does change as girls age, then the correlation between behavior and height would not be as strong as between the behavior and age per se. But with a large enough sample size, we would detect a significant correlation of behavior with height, reflecting the change in behavior with age.

Similarly, one cannot make accurate predictions about an individual’s phenotype, not even their sex (44), based on 2D:4D alone. For example, despite the repeatedly replicated difference between lesbians and straight women in 2D:4D, one cannot use digit ratios to classify individual women as straight or gay, an important point that is often absent from popular media reports on the issue. In any random sample, one could correctly predict each woman’s sexual orientation 95% of the time simply by guessing “heterosexual” for everyone, no matter what their digit ratios. So indeed “digit ratio is not a good marker of individual differences in prenatal androgen” (7) if the goal is discriminant analysis to classify individuals by sexual orientation. But if the goal is to test the organizational hypothesis for human behaviors by comparing two groups, or looking for correlations within a group, then digit ratios are quite useful markers of individual differences in prenatal androgen stimulation. Now that it is clear that prenatal androgens affect digit ratios, the ease of reliably measuring them in a large sample of people offers the opportunity to test the organizational hypothesis for any behavior that can be measured in adults. From that perspective, the assertion that digit ratios are not a good marker of individual differences in prenatal androgen rather begs the question. If they don’t offer a way of comparing prenatal androgen stimulation across individuals, then how can one explain the data? How could mean digit ratio reliably differ between groups of men and women, male and female fetuses, men and AIS women, and people with CAH vs. same-sex controls, and how could digit ratio correlate with polymorphisms in AR in men if it is not a good marker of individual differences in prenatal androgen stimulation? Likewise, why does digit ratio consistently correlate with so many sexually differentiated human behaviors (e.g. sexual orientation, ADD, and eating disorder risk in females; aggression, risk-taking, ADD, and eating disorder risk in males), if it is not a good marker of individual differences in androgen stimulation?

A flip side to this criticism of the use of digit ratios as “markers of individual differences” is the assertion that if other factors also affect digit ratios, then they cannot be used to gauge prenatal androgen. For example, Wallen (45) writes, “human digit ratios, although influenced by androgens, are also the product of some additional factor or factors other than androgens,” and concludes, “The data presently available allow rejection of 2D:4D digit ratios as a reliable proxy for prenatal androgen exposure.” The most obvious additional factor(s) affecting digit ratio are genes, and several studies report that 2D:4D is indeed heritable (46). But this is expected because the classic mode of androgen action is to regulate gene expression. We’ve already seen that two different polymorphisms in the AR gene itself affect digit ratios. Presumably polymorphisms in any of the downstream genes mediating androgen’s effects on digit ratios would also have an influence. So the effect of genes on digit ratio, far from contradicting the influence of prenatal androgen, is an inescapable consequence of androgenic influence on digit ratios. It is because of these other influences that digit ratios only imperfectly reflect prenatal androgen, and the reason we use parametric statistics to compare digit ratios is precisely to overcome those additional sources of variance. Most life scientists use statistics in all our work because we never work with any characteristic that responds to only a single variable. We could not expect digit ratios to be the sole exception.

In fact, the established heritability of digit ratios can be exploited to experimenters’ advantage. By comparing relatives, thus controlling in part for genetic influences affecting digit ratio, it should be possible to detect differences in prenatal androgen with smaller samples. That reasoning is validated in one CAH study where a subset of the subjects included six pairs of males with CAH and their unaffected male relatives. Despite the small sample size, repeated measures analysis not only confirmed the differences in digit ratio on the right hand, but also detected a difference on the less sensitive, left hand, which was not significant in the larger overall samples using an independent t test (11). Another example, this one relating to behavior, is the difference in 2D:4D between lesbians and straight women replicated by Hall and Love (20) using a mere seven pairs of monozygotic twins discordant for sexual orientation. Future researchers may benefit by such comparison of relatives to control for the genetic contributions, allowing a more sensitive assay of the contribution of androgens to digit ratio.

Those Who Cannot Remember the Past Are Condemned to Repeat It

Ironically, this failure to distinguish between using 2D:4D to classify individuals vs. comparing groups dogged the first 50 yr of digit ratio research. The earliest known report of the sex difference in digit ratios, by Ecker in 1875(47), offered anecdotal reports, “that the relatively greater length of the index finger is more often found among the female sex than among the male sex.” Ecker never claimed that all women displayed this trait [“… the variations are very large, so that it is in no way possible to (deduce) a certain law”], only that they were more likely to have this trait. Yet a rather pointless debate ensued as researchers disputed Ecker’s claim by reporting individual women and men who did not fit the rule. By 1930, a large, systematic sample finally made it clear that while there is overlap between the sexes, the index finger is indeed longer than the ring finger more often in women than in men (48). This settling of the “debate” meant that digit ratios became forgotten until Manning’s 1998 report.

Manning and other researchers acknowledge the imperfect relationship between digit ratios and prenatal androgen when they use statistics to look for group differences in mean digit ratios or correlations between digit ratios and various characteristics. Even when speaking of using digit ratios to “predict” athletic ability or autism characteristics, researchers use the same probabilistic language used in any other linear analysis, and acknowledge the imperfection of the “prediction.” Hopefully, this resurrected conflation of the usefulness of digit ratios for analyzing groups (to test hypotheses) vs. classifying individuals (as a parlor game? as a shibboleth to detect someone’s secrets?) can be resolved in fewer than 50 yr. The resolution may be as simple as this: digit ratios imperfectly reflect prenatal androgen stimulation, so you must gather a sufficiently large sample size to relate that stimulation to human behavior statistically.

What Does the Future Hold?

The conjecture that the sex difference in human digit ratios might reflect prenatal androgen has been confirmed: the sex difference is present before birth, correlates with polymorphism in AR sensitivity, is feminine in XY individuals with AIS, and is masculinized in both females and males with CAH. One obvious test remains: to correlate 2D:4D with prenatal measures of androgens from amniocentesis. A drawback of any correlation with amniocentesis samples is that they offer a glimpse, only a single time point in the 8 wk of fetal development when males normally secrete androgens (49), which is presumably when body and brain are sensitive to their masculinizing influence. So there is always a chance of a false negative result. On the other hand, a positive correlation, despite only a single measure of prenatal steroids, would be compelling. The only published attempt to date had a total sample of 29 children, which was too small to detect the sex difference in digit ratios (50), so they found no significant correlation of T and 2D:4D even across the two sexes (which, as noted above, would have to be present if there is a sex difference in both measures), much less within either sex. Surely larger samples will be gathered eventually.

In the meantime, we can expect to see more papers relating digit ratios to various human behaviors. This is partly because the ratios are so easy to gather and objectively measure, and partly because of long-standing interest about whether sex differences in human behavior are affected by biological factors such as prenatal androgens. It will be important to see which findings are replicated, as false positive results are possible with any experimental measure. Replications are especially important in digit ratio research because the ease of measuring them might lead a researcher to correlate them to many behaviors, increasing the odds of a false positive result. While some might discourage publication of digit ratio research for fear of false positives, such a policy would provide no chance to learn which behaviors genuinely correlate with prenatal androgens. Finally, as in any correlational work in humans, it is important to gather sufficiently large random samples and to minimize sources of experimental error, including using “blind” observers to measure ratios and behaviors. Given the results so far, it seems likely that we will hear of more cases where behavior that is more commonly seen in men is negatively correlated with 2D:4D within one or both sexes.

The many reports so far make it clear that there is some relationship between digit ratios and sexually differentiated human behavior. Those relationships can most parsimoniously be explained by a masculinizing effect of prenatal androgen on both digits and behavior. Thus Francis Bacon’s advice on how one should regard such “instances of the fingerpost” indicates that the organizational hypothesis applies to many sexually differentiated behaviors, even in a socially complex learning-dependent species such as our own.