Digit ratio (2D:4D) and maternal testosterone-to-estradiol ratio measured in early pregnancy

The ratio of index to ring finger (2D:4D) has been hypothesised to indicate prenatal androgen exposure, yet evidence for its validity is lacking. We report the first pre-registered study to investigate mothers’ early pregnancy sex hormone concentrations in relation to their children’s digit ratios measured at 18–22-month follow-up. Although the testosterone (T) to estradiol (E) ratio correlated negatively with right hand digit ratio (R2D:4D) and directional asymmetry (right-minus-left) in digit ratio (D[R−L]), neither effect remained statistically significant once demographic and obstetric covariates were controlled for. Nevertheless, the multivariate level of analysis did reveal that T correlated positively with left hand digit ratio (L2D:4D) and negatively with D[R−L]. However, the first of these effects is in the opposite direction to that predicted by theory. Taken together, the results of our study suggest research with larger samples is required to determine whether digit ratios are valid proxies for maternal sex hormone exposure.

www.nature.com/scientificreports/ We pre-registered our analysis plan on the Open Science Framework (https:// osf. io/ bmp6h) and predicted that neither maternal T nor E assayed from late first trimester/early second trimester maternal circulation would be a significant predictor of digit ratio when measured in offspring at 18-22-month follow-up. However, we also predicted that there would be negative correlations between the T:E ratio and R2D:4D and L2D:4D (but not D [R−L] ). The reason we predicted no effect for D [R−L] is that evidence for the validity of this measure is particularly weak. Although a small study (n = 26 mother-offspring dyads) reported that maternal urinary testosteroneto-estrone conjugate levels in pregnant Titi monkeys correlated negatively with D [R−L] in their offspring, the effect disappeared after covariates were controlled for 61 . Human studies have reported that D [R−L] does not differ between people with and without CAH 10 , and that it does not correlate with sex hormones measured from amniotic fluid 40,41 , umbilical cord blood 48,50 and second trimester maternal plasma 40 .
Intraclass correlation coefficients (ICC) (two-way mixed, single measures with absolute agreement definition) were computed to determine inter-rater reliability (please note that in our pre-registration it was stated that only one set of direct measurements was obtained: this was incorrect [two sets of direct measurements were available for a subsample of participants]). These revealed that inter-rater reliability was low for the direct measures: Descriptive statistics for digit ratio and maternal hormone variables are presented in Table 1. The hormonal measures did not differ in regard to fetal sex. L2D:4D measured from photocopies was lower in males than females, though all other comparisons were non-significant. Compared with direct measures, photocopies yielded Although not in our pre-registration plan, we report bivariate associations between maternal hormones and children's digit ratios to facilitate comparison with studies that have not controlled for covariates (Table 2). T:E correlated negatively with R2D:4D and D [R−L] but there was no association with L2D:4D (Fig. 1). None of the other hormone-digit ratio correlations were statistically significant. In our pre-registration, we specified that we would conduct bootstrapped (10,000 resamples) hierarchical multiple regression analyses (Step 1: enter covariates; Step 2: enter T, E, and the T × sex and E × sex interaction terms; Step 3: enter T:E ratio and the T:E × sex interaction term). However, in the interests of parsimony, we decided instead to run separate models for each predictor (along with its respective interaction term with sex). We included the following covariates: infant sex Table 1. Descriptive statistics for digit ratio (2D:4D) and hormone variables. Sample sizes differ for 2D:4D variables because measurements were only taken directly or indirectly from some participants; additionally, in some cases it was only possible to collect data for the right or left hand, e.g., because the second and/or fourth fingertips were missing from the photocopied images. Equal variances were not assumed for each of the independent samples t-tests reported here. Effect displayed in bold is statistically significant (p < 0.05, two-tailed). a Estradiol was measured in nmol but the values reported here are divided by 10,000 for ease of interpretation. b T:E was calculated as T (nmol)/E (nmol) but the values reported here are multiplied by 1000 for ease of interpretation. www.nature.com/scientificreports/ (0 = female, 1 = male), maternal PCOS status (1 = absent, 2 = present), maternal hirsutism (1 = no areas affected, 2 = one area affected, 3 = more than one area affected), infant's birth weight (grams), infant's age at follow-up corrected for gestational age (days) (we did not include child's birth length as specified in our pre-registration because this variable was not made available). Since the covariates were included for the sole purpose of controlling for factors that may be associated with hormonal profiles during pregnancy and infant growth factors that could affect 2D:4D, we report in Table 2

Discussion
The current study aimed to investigate whether maternal circulatory T:E ratio in early pregnancy predicts digit ratios in infancy. Our main prediction was that a high T:E ratio would correlate with low (male-typical) R2D:4D and L2D:4D, and that there would be no association with D [R−L] . Although we detected the predicted association with R2D:4D, T:E did not correlate with L2D:4D, and there was a negative correlation with D [R−L] . However, once covariates (including sex, and the relevant hormone × sex interaction term) were controlled for, T:E did not correlate significantly with any digit ratio variable, although T correlated positively with L2D:4D and negatively with D [R−L] . It is unclear why the results changed in this manner with the inclusion of covariates. However, we do not attempt to provide specific explanations for this, as they would necessarily be speculative in nature. A negative correlation between maternal T:E ratio and offspring R2D:4D would be consistent with the theory that differential prenatal exposure to sex hormones affects development of digit ratios 1,2,4,6,7 , and in line with results from studies of human infants 3 and experimental animal research 5 . However, this effect was not significant after controlling for covariates, and other research has reported no correlation between R2D:4D and the T:E ratio measured from amniotic fluid 41 or between R2D:4D and the androgen-to-estrogen ratio measured from perinatal umbilical cord blood 50 . Additionally, the only other study to investigate offspring digit ratio in relation to maternal sex hormones in early pregnancy 60 did not examine the T:E ratio. Further research using larger samples will be required to determine whether these variables are meaningfully related.
Contrary to our pre-registered prediction, a significant negative correlation between maternal T:E ratio and D [R−L] was observed in bivariate association. However, this did not remain significant after controlling for demographic and obstetric covariates. This may therefore be considered consistent with observations that D [R−L] does not correlate with sex hormones measured from amniotic fluid 40,41 , umbilical cord blood 48,50 and second trimester maternal plasma 40 , and that it does not differ between people with and without CAH 10 . Although Baxter et al. 61 recently reported a significant negative correlation between the urinary testosterone-to-estrone conjugate ratio of pregnant Titi monkeys and the D [R−L] of their offspring, the effect reported there also did not retain statistical significance once covariates had been controlled for. It further remains unclear to what degree, if at all, maternal urinary sex hormone concentrations relate to those of the developing fetus.
The current study adds to a literature replete with inconsistent findings and replication failures 11 . In particular, there appears to be a concerning pattern in which smaller studies attempting to test the 2D:4D validity question report positive findings and larger ones do not. For example, small studies of CAH 1,62 , which have been heavily cited in the literature, have reported significant effects, whereas larger ones have not [63][64][65] . The same pattern is observed for twin research, with some early small studies observing significant effects 30,35 but the largest in the Table 2. Associations between maternal hormone concentrations and offspring digit ratios. BCa 95% CI = bias corrected and accelerated 95% confidence intervals. Bivariate analyses are boostrapped (10,000 resamples) Pearson's correlations; multivariate analyses are bootstrapped (10,000 resamples) multiple linear regression. ES = effect size (Pearson's r for bivariate analyses; β for multivariate analyses). The following variables were included as covariates in the multivariate analyses: child's sex (0 = female, 1 = male), maternal polycystic ovary syndrome (PCOS) status (1 = absent, 2 = present), maternal hirsutism score (1 = no areas affected, 2 = one area affected, 3 = more than one area affected), child's birth weight (grams), child's age at follow-up corrected for gestational age (days). Statistically significant effects (i.e., those for which the BCa 95% CIs do not include 0) are presented in bold.  36 . Similarly, the negative correlation between amniotic T:E ratio and R2D:4D reported by Lutchmaya et al. 3 was not replicated in a larger cohort 41 . The current study may also fit this general pattern: although some statistically significant effects were observed, the larger study by Barrett et al. 60 found no correlation between early pregnancy maternal sex hormone concentrations and the 2D:4D of their children. These issues present a serious challenge to the credibility of digit ratio research, particularly when considered in conjunction with the ease with which data can be collected and the considerable researcher degrees of www.nature.com/scientificreports/ freedom afforded at the analysis stage 66,67 . For instance, it has been noted that researchers often examine several digit ratio predictor variables in the same study (e.g., R2D:4D, L2D:4D, D [R−L] , and the average of R2D:4D and L2D:4D [M2D:4D]) and also stratify their analyses by sex 10,14,68 . Unless effective controls for alpha inflation are in place, this necessarily increases the chances of observing statistically significant effects, and, hence, making Type 1 errors. This problem is further compounded when researchers measure multiple outcome variables 68 , and particularly so if not all of those outcomes are reported 67,69 . A further issue is that one significant effect in the predicted direction may be taken as evidence in favour of rejecting the null hypothesis despite the greater weight of evidence being in favour of its acceptance 10 . To take a conservative example: if researchers were to examine R2D:4D and L2D:4D, separately in males and females, in relation to a single outcome variable, one statistically significant effect (e.g., for R2D:4D in males) might be emphasised over three concurrent null results for the same hypothesis (i.e., null effects for L2D:4D in males and for R2D:4D and L2D:4D in females). Such practices do not only lead to biased interpretations of individual datasets, but they also make it difficult to detect publication bias 10 , a problem which may be prevalent within the digit ratio literature 10,14,70 . This is because if there is a bias for publishing positive findings, it may be irrelevant in this field whether a statistically significant effect is found in relation to the right hand or left hand or in males or in females etc. An obvious way to address this issue going forwards is to pre-register studies with specific a priori hypotheses, predictions, and analysis plans, and to note in publications where analyses have deviated from these 14,71,72 . R2D:4D and L2D:4D values in the current study were lower when measured from photocopies than when measured directly from participants' hands. This corroborates findings from studies of older populations [73][74][75] , and implies that the process of photocopying may distort the soft tissue in a way that acts differentially across the second and fourth fingers. What was more surprising was that the direct and photocopy measures were uncorrelated in our sample, as previous research has identified both techniques to be reliable 76,77 . As (after removal of outliers) 2D:4D measurements taken directly (self-measured) from young adults have been reported to correlate moderately (R2D:4D, r = 0.518; L2D:4D, r = 0.409) with those taken from photocopies (researcher-measured) 78 , the lack of intercorrelation observed within the current study may simply reflect the difficulty associated with obtaining accurate measurements directly from the hands of toddlers. Correlations between direct and photocopy measures of digit ratio in young children may generally be fairly low (e.g., R2D:4D: r = 0.421; L2D:4D: r = 0.373 [Constantinescu, 2009, p. 32 65 ]) but it remains unclear why we observed no association at all. We suggest that, despite the slight distortion caused, photocopies/scans may be the most effective method of obtaining reliable data in this population. Although one might still make the case for using direct measurements, we examined those derived from photocopies because they are more likely to be reliable. This is indicated by the higher ICCs and lower SDs observed in the current study, and by previous research reporting computer-assisted measurements to yield the most reliable results 76,77 . Additionally, we used these measures because it would be possible to check the data against the original photocopies whereas this is not so for the direct measurements. Analysis of both sets of measurements would have increased the number of null hypothesis significance tests used, and, therefore, the chances of making Type 1 errors.
The current study has strength in that it not only reported on two hormones in isolation and their interactions with fetal sex, but also examined the T:E ratio, a variable posited to play a key role in the determination of digit ratios [3][4][5][6] . It is additionally important that sex hormones were assayed in close proximity to the time at which they are hypothesised to exert their greatest influence on digit ratio development 59 . However, some limitations should also be considered. Though comparable with previous research 3,41,48,50 , our sample is only modest in size; it therefore, lacks the statistical power required to detect small effects. Second, although inter-rater reliability for digit ratios measured from photocopies was high, it is notable that these measures did not correlate with those taken directly from participants' hands. Third, fetal steroids were not measured directly, and it remains unclear how representative maternal serum samples may be of the fetal circulation. Previous research has shown estradiol measured from second trimester amniotic fluid to be positively correlated with that present in maternal serum sampled in the second and third trimesters 42 , and human chorionic gonadotropin (hCG) levels measured from amniotic fluid and maternal serum are positively correlated during the second (but not third) trimester 79 . Conversely, although there are exceptions (e.g. 80,81 ), most studies have found maternal serum testosterone to be unrelated to fetal sex 42,44,[82][83][84][85] , and testosterone concentrations in the maternal and fetal plasma have been reported to be uncorrelated 43 . Considering that digit ratios have been hypothesised to relate primarily to sex steroid concentrations present within the fetal rather than maternal circulation, the findings of the present study should therefore be interpreted cautiously.

Conclusions
The current study contributes the first pre-registered analysis of 2D:4D in relation to maternal sex hormone concentrations, as well as the first empirical test of whether the right-left difference in digit ratios (D [R−L] ) is correlated with hormonal concentrations measured during early pregnancy. We attempted to test the hypothesis that mothers' T:E ratio is a predictor of their offspring's digit ratios. Although we observed statistically significant effects linking high T:E with low R2D:4D and D [R−L] at 18-22-month follow-up, neither effect remained statistically significant once covariates had been controlled for. Furthermore, multivariate analyses revealed that T correlated positively with L2D:4D and negatively with D [R−L] , the first of these effects being inconsistent with well established theory. Taken together, the results of this study suggest that further research with larger sample sizes will be required to determine whether digit ratios are valid proxy measures of maternal sex hormone exposure.  (1) little/no consumption of alcohol during pregnancy, (2) no smoking or recreational drug use during pregnancy, (3) a singleton fetus whose measurements did not indicate intrauterine growth restriction or largefor-gestational age, (4) absence of any major fetal anomalies, and (5) birth of a clinically healthy baby. At time of scan mothers were asked to complete a pregnancy history questionnaire to self-report metabolic, reproductive, and diagnosed conditions. Detailed description of this sample has already been reported elsewhere 46,86 . Hormone assays. Serum samples were collected by a specialist phlebotomist at the Rosie Hospital and stored at − 80 °C, as part of a national screening programme at the end of first trimester/start of second trimester (M = 12.7 [SD = 0.7] weeks gestation 46 ) for biomarkers of Down's Syndrome and other conditions. Samples from CUSP participants (n = 122) were thawed and transferred to separate vials (1 ml aliquots per sample), which were anonymised and sent for analysis at the Core Biochemical Assays Laboratory (CBAL) at Addenbrookes Hospital, Cambridge.
Concentrations of testosterone (T), estradiol (E), dehydroepiandrosterone sulphate (DHEAS), progesterone (P), and sex hormone-binding globulin (SHBG) were measured. Samples were analysed on a DiaSorin Liaison ® XL automated immunoassay analyser using a one-step competitive chemiluminescence immunoassay for each hormone and two monoclonal antibodies for each peptide. All reagents, standards and consumables are those supplied by DiaSorin (DiaSorin S.p.A, 13040 Saluggia [VC], Italy). Although SHBG was assayed, allowing for estimation of the free testosterone index and the free estradiol index, we instead include the total hormone levels in our statistical models. This is because SHBG does not easily cross the placenta 87 , and so it is unclear whether SHBG in the maternal serum is reflective of fetal bioactivity.
Digit ratio (2D:4D) measurements. Parents and infants were invited for an in-person follow-up visit to measure 2D:4D and physical growth (infants' M age = 19.87, SD = 0.86, range = 18.20-21.95 months). Direct measures of finger length were taken by Research Assistants using a standard tape measure. A Canon LiDE 300 flatbed scanner was used to scan infants' left and right hands, and colour images were made at a resolution of 2400 × 4800 dpi. 2D:4D ratios were calculated from these by two researchers (GR and EA) using AutoMetric 2.2 for Windows 88 .

Statistical analysis.
We computed the averages for R2D:4D, L2D:4D, and D [R−L] across the two sets of measurements (separately for direct and photocopy measures). We checked for correlation (Pearson's tests) and differences (paired samples t-tests) between the direct and photocopy measures, and used independent samples t-tests to examine for sex differences. We then used bootstrapped (10,000 resamples) Pearson's correlations and multiple linear regression analyses to determine whether the maternal hormones (T, E, and T:E ratio) were associated with digit ratio (R2D:4D, L2D:4D, and D [R−L] ). We included the following covariates: child's sex (0 = female, 1 = male), maternal polycystic ovary syndrome (PCOS) status (1 = absent, 2 = present), maternal hirsutism score (1 = no areas affected, 2 = one area affected, 3 = more than one area affected), child's birth weight (grams), child's age at follow-up corrected for gestational age (days). We included covariates to control for factors related to the maternal hormonal environment (PCOS and hirsuitism are associated with elevated androgen concentrations 89,90 ) and infant factors related to growth trajectories that could affect 2D:4D (2D:4D may fluctuate considerably during early postnatal life 91,92 ; it may also correlate with birth weight, although empirical findings are mixed [93][94][95][96]. We also included infant sex as a covariate because 2D:4D exhibits marked sex differences 9 and because associations with prenatal hormonal variables could differ between males and females 39 . When utilising a bootstrapping approach, a specified number of resamples (in this case, 10,000) the size of the original is drawn with replacement from the available data. The chosen statistic is then computed for each resample. These resamples are considered equivalent to samples derived in the usual way from an infinitely large population with similar characteristics to those of the observed data. The variation among resamples indicates what would be expected from sampling variation under such circumstances (see Loehlin et al., p. 300-301 97 ). We used bootstrapping because it does not assume a normal distribution of the error term 98 , and may be advantageous when examining variables that exhibit marked deviations from the normal distribution as well as presence of datapoints that would be considered outliers in the context of a normal distribution 41,97,99,100 .

Data availability
The datasets generated and/or analysed during the current study are not publicly available due to limited ethics approval for the wider clinical study (CUSP) by CUH and to the specific consent provided by the participants. They may be available from the corresponding author on reasonable request and pending approval of any future analyses by CUH. The R script used to run the analysis is available on the Open Science Framework: https:// osf. io/ uj3mv/. www.nature.com/scientificreports/