In the pooled Late Pleistocene sample (n = 107), 96 individuals have polar moment of area asymmetries > 5% and hence are sufficiently asymmetrical to indicate side dominance (following Shaw, 2011). Of those 96 paired humeri, 80 (83.3%) have stronger right humeri (hence right-dominance), and 16 (16.7%) are left side dominant. Among the remaining individuals, seven have asymmetries < 5%, and four have contrasting asymmetry directions > 5% between the 35% and the 50% cross-sections.

The distributions of levels of humeral diaphyseal asymmetry (Fig. 1) for the six Late Pleistocene samples range from near zero (symmetry) to > 100% (dominant humerus with at least twice the rigidity of the non-dominant one). In the midshaft, three of the Eurasian samples have high average asymmetry (> 30%; supplementary material, Table S6) and show largely continuous distributions up to ≈ 80% with two high outliers (the La Quina 5 Neandertal and the West Eurasia LUP Arene Candide 2); the small MPMH sample has lower values. Variations in humeral asymmetry among the Neandertal, MPMH, EUP and West Eurasia LUP samples are not statistically significant after multiple comparison corrections (supplementary material, Table S6). In the North African sample, and especially the East Eurasian one, the levels of midshaft asymmetry are generally lower, but the former nonetheless extends up to ≈ 60%. The North African sample is significantly less lateralized than the West Eurasia LUP one (P < 0.05), while the East Eurasia sample is less asymmetric than both the Neandertals (P < 0.05) and the West Eurasia LUP sample (P < 0.01) (supplementary material, Table S6).

In the mid-distal comparison, the pattern is similar, with the higher values present in all but the MPMH and East Eurasia LUP samples (Fig. 1; supplementary material, Table S7). The North African mid-distal humeri have several values ≥ 60%, whereas similarly high values are absent from their midshafts. In this comparison, one Neandertal (Spy 2), three each of the EUP and West Eurasia LUP samples and one North Africa LUP individual (Afalou 27) have asymmetries ≈ 100%. In the mid-distal section, variation in humeral asymmetry among the time periods are non-significant (with multiple comparison correction), except for the East Eurasia sample, which is significantly less asymmetric than the EUP and West Eurasia LUP ones (supplementary material, Table S7).

Variations across the samples by sex are generally non-significant after multiple comparison correction, largely due to small sample sizes (Fig. 4; supplementary material, Tables S6 and S7). Indeed, the only significant comparison at P < 0.05 is between the two largest samples, the midshaft West Eurasia (n = 14) and North Africa (n = 19) LUP males, with the latter substantially less asymmetric on average.

In addition to variation with respect to time and space, plots of asymmetry that reflect humeral diaphyseal size (Fig. 2) and body size (Fig. 3) show an increase in asymmetry with larger dimensions across the Late Pleistocene samples. In the distribution of dominant versus non-dominant polar moments of area (Fig. 2), the mid-distal least squares slope of 1.37 (95% CI: 1.11–1.54) and the midshaft least squares slope of 1.26 (95% CI: 1.11–1.42) are both significantly > 1 (a slope of 1 indicating no change with size) (supplementary material, Table S8). In the comparison of asymmetry to estimated body mass (Fig. 3), the mid-distal and midshaft least squares slopes of 1.90 and 1.36 (95% CIs: 1.34–2.45 and 0.83–1.90, respectively) are significantly > 0 (a slope of 0 indicating no change with size) (supplementary material, Table S8). In a comparison of asymmetry to humeral length, the level of asymmetry increases slightly for the mid-distal level but not for the midshaft (supplementary material, Table S8).

Therefore, as humeral diaphyseal dimensions and body mass (but not especially humeral length) increased across these pooled samples, the level of asymmetry generally augmented. An increase in asymmetry with size, however, is absent or less pronounced in the individual regional samples, in part due to small sample sizes. Yet, these comparisons should be sufficient to open the question as to what extent body size, in addition to other factors (such as different upper limb loading patterns across regions and by sex), might be influencing levels of humeral asymmetry.

The plot of humeral asymmetry by sample and sex (Fig. 4) provides lower average female levels of asymmetry for all except the small East Eurasia LUP sample. Across the pooled Late Pleistocene sample, the male and female samples are highly significantly different (Mann-Whitney U P < 0.0001 for 35% and 50%), although only the West Eurasia and North Africa LUP male-female 35% asymmetry values are significantly different after a multiple comparison correction (the EUP 35% comparison is close to the P = 0.05 level) (supplementary material, Tables S6 and S7).

When the samples are pooled and the sexes are plotted separately, as either dominant versus non-dominant polar moments of area (Fig. 5) or asymmetry values versus body mass (Fig. 6), the least squares slopes variably indicate increases in asymmetry with size (supplementary material, Table S8). In the dominant/non-dominant comparisons, only the female 35% section indicates a within-sex change in asymmetry with size, and the 95% CI (1.04–1.24) extends close to one. In the more appropriate asymmetry/body mass comparison, the females have little asymmetry change with size (35%: P = 0.059; 50%; P = 0.644), but the males (despite considerable scatter) show a significant increase in humeral asymmetry with larger body size at both diaphyseal locations (95% CIs: 35%: 0.70–2.95, P = 0.002; 50%: 0.34–2.35; P = 0.010).

The frequencies for right versus left versus ambiguous side dominance (74.8%, 15.0% and 10.3%, respectively) fall well within expected ranges across extant human samples (Cavanagh et al., 2016 and Raymond and Pontier, 2004). As noted previously (see above), patterns of lateralization documented so far from multiple indicators through the genus Homo are indistinguishable from those of living humans.

Our data, using a larger dataset of Middle and Upper Paleolithic fossil humans, further document that high levels of humeral asymmetry are common in the Late Pleistocene, as shown in previous studies (Churchill, 1994, Churchill et al., 2000, Churchill et al., 1996, Holt et al., 2000, Shackelford, 2007 and Trinkaus, 2015). Asymmetry is especially high among males (except for East Eurasian LUP humans) and generally reaches or surpasses 30% on average, with individual values ranging up to more than 100%. For comparison, recent human individual values generally range up to ≈ 20–40%, and most sample means are < 20% (including recent human foragers); agricultural samples generally show levels of asymmetry around 10%, as do industrial sedentary samples (Ogilvie and Hilton, 2011, Shaw and Stock, 2009, Sládek et al., 2016, Sparacello et al., 2011, Stock et al., 2013, Trinkaus et al., 1994 and Weiss, 2009). Therefore, it is not only the Neandertals that developed marked humeral asymmetry, but individuals in samples spanning the substantial technological and adaptive changes through the Middle and Upper Paleolithic.

Modern athletes with unilateral arm loading frequently reach levels of humeral diaphyseal asymmetry similar to these Late Pleistocene humans (Haapasalo et al., 2000, Ireland et al., 2013, Shaw and Stock, 2009, Trinkaus et al., 1994 and Warden et al., 2009), which has led to suggestions that frequent spear throwing (with or without a spear thrower) influenced the high level of asymmetry in Pleistocene humans (Churchill and Rhodes, 2009 and Churchill et al., 2000). This activity is not only strenuous during hunting but requires continuous training during development for both strength and aim (Cattelain, 1997, Rhodes and Knüsel, 2005 and Whittaker and Kamp, 2006). The high and intermittent loading rates correspond to the pattern that best stimulates osteogenic response (Burr et al., 1996, Burr et al., 2002 and Robling et al., 2002). Patterns of humeral enthesopathies in at least the Upper Paleolithic (Villotte and Knüsel, 2014) further support the importance of throwing among these Late Pleistocene humans. However, other activities may have contributed to the differential hypertrophy of the humeri through the Late Pleistocene, since repetitive domestic activities such as scraping involve asymmetric recruiting of upper limb muscles (Shaw et al., 2012).

Given the increases in humeral asymmetry with humeral diaphyseal and body dimensions, and the ranges of body sizes evident on Fig. 2 and Fig. 3, body size needs to be considered in interpretations of humeral asymmetry. In particular, could the low levels of asymmetry in the East Eurasia LUP sample be due to their generally small body dimensions (Baba and Endo, 1982 and Shackelford and Demeter, 2012)? Could body size account for the moderately lower asymmetry level of much of the North Africa LUP sample compared to the western Eurasian ones, as well as of the MPMH sample (Shackelford, 2007 and Trinkaus et al., 2014)? Furthermore, given the Late Pleistocene sexual dimorphism in body size (Fig. 6), it raises the question as to how much of the perceived sexual differences in humeral asymmetry are secondarily due to larger average body mass in males.

Yet, a significant correlation between body mass and asymmetry is found principally in the pooled male sample (supplementary material, Table S8), and it is present only in the mid-distal section of West Eurasian LUP males when the samples are separated by region/period and sex. Given variation in body size across the Late Pleistocene subsamples and between the sexes, it remains uncertain to what extent regional/temporal, as well as sexual, differences in upper limb differential loading reflect behavioral patterns per se versus a combination of a size effect and behavioral variation.

Given these considerations, the differences between male and female asymmetry values in most of the Late Pleistocene samples, and the differences in the male and female distributions and slopes when humeral asymmetry is compared to body mass, it is likely that some of the variation is due to differential bilateral loading of the humeri between female and male Late Pleistocene humans. In this context, it needs to be kept in mind that, although high levels of asymmetry result from markedly differential loading of the dominant arm, low levels of asymmetry can be the product of either general gracility or bilateral hypertrophy from bi-manual activities (Ogilvie and Hilton, 2011 and Shaw and Stock, 2009). Because the Late Pleistocene pattern occurred in the context of no significant sexual differences in humeral diaphyseal hypertrophy once scaled to body mass and humeral length (SI Section V), it is likely that the average sexual differences were due as much to elevated bilateral loading in individuals with low asymmetry as to differentially increased loading of the dominant arm in those with high asymmetry.

Pleistocene female humeral asymmetry in diaphyseal rigidity has been considered previously to be high (Churchill, 1993), suggesting that Paleolithic females, like males, were habitually engaged in activities (such as throwing) that generated high bending stresses on the dominant limb (Holt et al., 2000). Sexual division of labor in the use of throwing weapons is nearly universal among modern hunter-gatherers (Murdock and Provost, 1973, Testart, 1986, Watanabe, 1968 and Wood and Eagly, 2002). However, among the Agta of the Philippines, women regularly hunted (Estioko-Griffin and Griffin, 1981 and Goodman et al., 1985), and among high-latitude hunter-gatherers, given the importance of large game acquisition, widows and their daughters could become habitual hunters (Briggs, 1970, Jenness, 1922 and Landes, 1938). Nevertheless, the inference that a sexual division of labor might have been less marked in the Pleistocene has been questioned by the male bias of unilateral medial epicondyle enthesopathies suggesting throwing (Villotte and Knüsel, 2014 and Villotte et al., 2010) and by sexual differences in humeral diaphyseal asymmetry in a small European LUP sample (Sládek et al., 2016).

Some of these sexual differences in humeral asymmetry may be secondary to sexual differences in body size. Yet, differences in average asymmetry between male and female tennis players (females: ≈ 40%, males: ≈ 75%; Haapasalo et al., 2000 and Jones et al., 1977) suggest that other physiological factors might play a role in the development of extreme levels of structural adaptations between the sexes (see, for example, Stini, 1969 and Stinson, 1985). However, the contrasts in the male-female distributions relative to body size appear to be substantial enough to suggest the presence of a sexual division of labor affecting the upper limb among most of these Late Pleistocene groups.

It is possible to hypothesize behavioral scenarios to account for these patterns, as with assessments of sex differences in humeral asymmetry in Holocene archeological samples (e.g., Marchi et al., 2006, Ogilvie and Hilton, 2011, Sládek et al., 2016, Sparacello and Marchi, 2008 and Weiss, 2009), recognizing that, despite the average sex differences, there is considerable overlap in the sex-specific distributions. Females could have been differentially engaged in bi-manual processing activities, although currently identified Paleolithic grindstones appear to be uni-manual ones (Revedin et al., 2010). Males could have been more commonly involved in uni-manual processing, such as scraping (Shaw et al., 2012) or throwing weapons (Villotte and Knüsel, 2014 and Villotte et al., 2010). Although it is tempting to explain patterns of asymmetry (and sexual dimorphism in this trait) throughout the Pleistocene to the use of throwing spears (not thrusting spears; Shaw et al., 2012), this inference would have to account for similarly high levels of asymmetry through the Late Pleistocene despite the evolution of hunting weapons from heavy wooden and lithic tipped spears (Middle Paleolithic) to lighter throwing spears (early Upper Paleolithic) to spear throwers and light spears (late Upper Paleolithic).

In this context, it needs to be emphasized that the sexual differences in humeral asymmetry, and hence behavioral implications for females versus males, apply throughout the Late Pleistocene, to both late archaic and early modern humans, to Middle, early Upper and late Upper Paleolithic humans. They are distinctly counter to arguments (e.g., Kuhn and Stiner, 2006) that a sexual division of labor was derived for Upper Paleolithic modern humans.