The strongest evidence for androgens playing a direct role in the development of digital formula is its association with congenital adrenal hyperplasia (CAH). CAH is a condition resulting in elevated androgen production, which is usually treated soon after birth. Prior to treatment, CAH often results in the masculinization of the external genitalia of newborn girls as well as other aspects of phenotype, including psychology (Hines 2004; Meyer-Bahlburg et al. 2004). Two studies have reported that both males and females with CAH have smaller, that is, more masculine, 2D:4Ds (Brown et al. 2002; Ökten et al. 2002). Although another study later failed to replicate this finding using 2D:4D measures obtained from left-hand radiographs, they did not employ a case–control design with age matching and their subjects ranged in age from 1 to 20 years old (Buck et al. 2003). The means obtained in that study were in the expected directions but not significant. In retrospect, their failure to replicate the earlier results can be understood partly in light of evidence that the 2D:4D increases during childhood (McIntyre et al. in press), potentially confounding comparisons between CAH-affected people and controls. Not having employed a case–control design might also have introduced other confounding factors, such as ethnic differences (Manning et al. 2000; McIntyre et al. in press).

As further evidence of a direct relationship between androgen action and digit ratio, other recent research has shown a relationship between a low 2D:4D and fewer CAH elements in the transactivational domain of the androgen receptor (Manning et al. 2003). The CAH elements encode a polyglutamine tract. The length of the tract is determined by the number of CAH elements in the allele, which is highly polymorphic in healthy people. The length of the tract has been inversely associated with both in vitro androgen receptor transactivation (Callewaert et al. 2003), and phenotypic masculinization of tissues (Ding et al. 2004). The association between this androgen receptor polymorphism and the 2D:4D suggests that not only do androgens directly influence the development of the digits, but they do so at least partly through the androgen receptor.

Sex steroids are known to play an important role in bone growth and skeletal maturation. Most steroid effects on growth plates in long bones have been shown to operate through oestrogen receptors alpha and beta (Cutler 1997; Kusec et al. 1998; Nilsson et al. 1999; Weise et al. 2001), with the effects of testosterone mediated through its local aromatization to estradiol (Öz et al. 2001). However, Abu et al. (1997) have reported that the androgen receptor is also expressed in the growth plates of long bones, though its direct physiological effects are unclear. Perhaps more relevant in the case of prenatal effects on bone growth is the finding by Ben-Hur et al. (1997) that both androgen and oestrogen receptors are expressed in foetal cartilaginous tissue, leaving open the possibility that androgens influence the development of the digital anlagen. Differences in the effect of androgens on the growth of different bones or digital rays could then be understood as resulting from differences in steroid sensitivity, enzyme activity or coactivational environments among tissues.

Garn et al. (1974) long ago noted sex differences in the foetal development of the fingers, with males having more advanced onset of ossification for a given crown–rump length (an unexpected pattern given the general female advancement at other ages and in other respects). In the following year, Garn et al. (1975) also showed that phalangeal length ratios typical of adulthood are attained early in gestation, a period of high testosterone production in males (Forest 1990). If, as an alternative to direct effects on particular bones, androgens are envisioned as having a global effect on digital development, such as advancing ossification, then sex differences could arise from global effects in one period affecting bones that develop at different rates. More research is needed to establish the physiological-developmental pathways mediating sex and relative digit lengths.

However, whatever the mediating physiological pathway anatomical sex differences in young children are most likely to result from perinatal androgen action, rather than, for example, from sex-linked genetic interactions or from the effects of other steroids, such as oestrogens. While the sex-determining region Y (RSY) certainly determines gonadal sex and, along with many other genes, regulates differentiation of internal reproductive organs, neither SRY nor other regions of the Y chromosome have been found to influence secondary sex differences in other tissues. Although X-linked genes often have different effects in males and females, such as the many well-known X-linked recessive conditions, X-linkage per se has not been found to play an important role in secondary sex differences. The process of X-inactivation seems to limit the potential role of X-linked genes in determining sex through, for example, reliable differences in gene product dosage between the sexes (though it fails to ameliorate the susceptibility of XY tissues to recessive allelic variants). Genes located on the X chromosome would normally be expressed in both males and females unless regulated by other sex-different factors, that is, hormones. Rather than genetic differences, sex steroids secreted from the differentiated gonads play the pivotal role in secondary sex differentiation in vertebrates. Furthermore, as discussed earlier, sex differences arising prior to puberty develop largely under the influence of testosterone alone, which is secreted by testes but not by the still-quiescent ovaries. If oestrogen was to play a role, it would likely be to promote masculinization and not feminization, as has been shown in other mammals. This pattern is attributable to the higher perinatal exposure of males, rather than of females, to oestrogens, due to the peripheral aromatization of testosterone. However, this pathway for oestrogen action is restricted by the action of alpha-fetoprotein in binding and inactivating oestrogens, probably to prevent foetal masculinization arising from placental oestrogens.

That said, in humans, the evidence thus far suggests that oestrogens normally play no role in masculinization prior to puberty, perhaps partly because of the action of alpha-fetoprotein which binds and inactivates oestrogens in foetal circulation (in contrast to the potentially substantial maculinizing effects of synthetic oestrogens, especially diethylstilbestrol, which are not bound by alpha-fetoprotein). This claim is most strongly evinced by the combined observations that (i) XY males with complete oestrogen insensitivity or aromatase deficiency do not present with signs of hypomasculinity and (ii) XY females with complete androgen insensitivity, who were nevertheless exposed to male-typical or even further elevated levels of oestrogens, while lacking female internal reproductive organs, have thus far not been observed as being masculine in other respects, including psychologically (Wilson 2001). Therefore, clear demonstration that sex differences in digital formula arise prior to puberty provides evidence for the involvement of perinatal androgens.

This paper, therefore, focuses on the most important question about digit ratio validity, namely, ‘To what extent do digit ratios or other measures from the fingers approximate sex differences arising prior to puberty?’ Identifying the age at which sex differences in digit ratios arise only partly answers this question. It is important to understand the developmental processes producing variance in adult digit ratios and, specifically, in sex differences. An association with childhood sex differences that have disappeared by adulthood would augment the utility of digit ratios as a marker of childhood or prenatal sex differences. A strong association between digit ratios and important growth processes which are not different between the sexes would warn us to interpret digit ratios carefully and to expect many spurious results.

The criteria for the selection of subjects for this study were (i) having at least one measurable radiograph from the first year of life or (ii) having radiographs taken at or within two years of ages 1, 5, 9, 13, and 17 years. Analyses were conducted on the two, largely overlapping groups of radiographs corresponding to the two selection criteria, hereafter infant and serial. All available and measurable infant radiographs were measured (varying from one to five for each subject; median of three radiographs each). Exactly five serial radiographs, one corresponding to each target age, were measured from subjects included in the serial group. The infant sample contains 399 radiographs from 124 subjects (56 females and 68 males). The serial sample contains 555 radiographs from 111 subjects (52 females and 59 males). Of these, 107 subjects (49 female, 58 male) were included in both datasets. From these subjects, 99 radiographs were also included in both datasets, including all radiographs taken at exactly one year old.

In most cases, serial radiographs were taken within the target birth month. When target-age radiographs were unavailable or not measurable, the nearest available age was selected. When two radiographs at equally younger and older ages were available, the older age was selected.

Radiographs were digitized to a resolution of 0.85μm using a desktop scanner with backlight. The method of measurement and software used follow a published protocol (McIntyre et al. in press), yielding nine segment lengths: a proximal, medial and distal segment from each of the second (index), third (middle) and fourth (ring) fingers.

The observer making the original measurements also remeasured all 111 subjects in the serial group at ages one and five years to assess repeatability. At age one, the reliabilities (α) of the segment lengths range between 0.961 and 0.988; were 0.988 for the second digit (2D), 0.982 for the third digit (3D), and 0.985 for the fourth digit (4D); and were 0.970 for 2D:4D and 0.947 for 3D:4D. By age five, reliabilities were higher for all measures: between 0.987 and 0.994 for all segment lengths, 0.998 for 2D, 3D and 4D, 0.985 for 2D:4D and 0.969 for 3D:4D. None of these values would render measures unreliable, though the measures from younger ages are universally less reliable.

The analytical methods employed in this study involved more than simply measuring digit ratios at different ages and noting when sex differences arise. We had two goals in selecting analytical methods. First, we wanted to understand the development of digit ratios well enough to identify possible sources of bias in interpretation, especially any strong association with important growth processes that are not sex dimorphic, and to quantify the effect of any sources of bias. Second, we wanted to quantify the amount of information that digit ratios contain about pre-pubertal sex differences, which is the information of interest to most researchers.

Beyond further substantiating previous claims about digit ratios, this study also extends our understanding of adult digit ratios by relating them to early childhood growth processes, including the development of sex differences which include, but are not limited to, sex differences in the digit ratios. Our most striking finding was that digit ratios in the most mature hands reflect childhood sex differences in the fingers, expressed as a continuous variable, much more strongly than might be expected on the basis of the small group sex differences observed among adults. While binary sex (namely, being male or female) accounts for less than 5% of the variance in mature digit ratios, patterns of sex difference in childhood might account for as much as 20% of the variance in mature digit ratios. The high serial reliability both of digital formula measures (including digit ratios and principal component scores) and of sex-discriminant scores contributes to the reported pattern.

It is important to note that these measures were taken from radiographs of the left hand. Most research employing digit ratios involves measures taken on the skin surface, often from the right hand. Therefore, direct, quantitative comparisons are problematic, even if most conclusions can be applied generally.

Differences in method might explain why we found adult 3D:4Ds to be a more valid descriptor of childhood sex differences. Using measures taken on the skin surface, both right and left hand 2D:4Ds are more sexually dimorphic than 3D:4Ds. However, digit ratios other than 2D:4Ds have not been widely reported for children. Therefore, contrary findings using radiographs from children might at least argue for the continued investigation of digit ratios other than 2D:4Ds. In particular, the size of adult sex differences ought not to be taken as primary evidence for the validity of 2D:4Ds. Rather, the associations between digit ratios and relevant developmental variables (such as independent proxies of pre-pubertal androgen production or pre-pubertal sex differences) are crucial.