Skull and limb morphology differentially track population history and environmental factors in the transition to agriculture in Europe

Abstract

The Neolithic transition in Europe was a complex mosaic spatio-temporal process, involving both demic diffusion from the Near East and the cultural adoption of farming practices by indigenous hunter–gatherers. Previous analyses of Mesolithic hunter–gatherers and Early Neolithic farmers suggest that cranial shape variation preserves the population history signature of the Neolithic transition. However, the extent to which these same demographic processes are discernible in the postcranium is poorly understood. Here, for the first time, crania and postcranial elements from the same 11 prehistoric populations are analysed together in an internally consistent theoretical and methodological framework. Results show that while cranial shape reflects the population history differences between Mesolithic and Neolithic lineages, relative limb dimensions exhibit significant congruence with environmental variables such as latitude and temperature, even after controlling for geography and time. Also, overall limb size is found to be consistently larger in hunter–gatherers than farmers, suggesting a reduction in size related to factors other than thermoregulatory adaptation. Therefore, our results suggest that relative limb dimensions are not tracking the same demographic population history as the cranium, and point to the strong influence of climatic, dietary and behavioural factors in determining limb morphology, irrespective of underlying neutral demographic processes.

1. Introduction

The transition from a hunter–gatherer–forager mode of subsistence (Epipalaeolithic and Mesolithic) to one based largely on domesticated animals and plants (Neolithic) occurred approximately 10 000 years ago, and marks a major demographic shift in the population history of Europe, the Near East and North Africa [1–7]. In recent years, it has become increasingly clear that this transition was the result of a mosaic process [8], comprising the movement of farming populations from the Near East and Anatolia (demic diffusion) [9–16] alongside cultural adoption of farming practices by some indigenous hunter–gatherer populations [2,17–20]. Collectively, these processes combined to recalibrate the genetic, morphological and cultural landscape of western Eurasia.

Craniometric data have proved a useful genetic proxy for studying this population history empirically, given that the available cranial data have much higher spatial and chronological coverage than the currently available ancient DNA data from pre- and post-transition individuals [21]. Using skeletal metrics as a means of modelling past population history provides the opportunity to compare directly the populations of interest [16,20–22] rather than relying on modern genetic data from living populations, which may be affected by subsequent demographic events such as empire expansions, diasporas and local extinctions [23]. Global analyses have consistently found that human cranial shape data fit a model of neutral microevolutionary expectation (i.e. resulting from stochastic forces such as mutation, gene flow and genetic drift) [24–31]. Only those populations living in extremely cold climatic conditions appear to have been affected by non-neutral directional climatic adaptation [27,29,32]. However, in the case of the postcranial skeleton, most studies have suggested relatively potent (non-neutral) effects of climatic adaptation [33–38], plasticity related to habitual activity [37,39,40] and the effects of nutritional or environmental stress on growth [41–43]. However, there is evidence to suggest that the morphology of some regions of the postcranium (i.e. pelvic shape) may largely reflect neutral evolutionary processes [44,45], highlighting the need to first control for population history when advocating non-stochastic (i.e. selective) explanations for postcranial variation. Hence, it is important to analyse cranial and postcranial elements together within the same theoretical and methodological framework, in order to develop a more complete understanding of the microevolutionary history of the entire human skeleton.

Given that the transition to agriculture represents a dramatic episode in the microevolutionary history of European, Near Eastern and North African populations, it is important to understand the extent to which this particular demographic event affected the cranium and postcranium. This study represents the first attempt to combine the analysis of cranial shape data with that from the postcranium in an internally consistent methodological and quantitative genetic analytical framework. Although, as reviewed earlier, previous studies have suggested largely non-neutral causes for postcranial variation, the null hypothesis that the transition to agriculture affected cranial and postcranial population affinity patterns in the same manner has never been tested. Therefore, our null hypothesis predicts that population affinity matrices, generated using craniometric and postcranial metric data, will exhibit the same patterns of congruence with explanatory matrices reflecting neutral expectation, archaeological grouping (i.e. whether pre- or post-Neolithic), and environmental factors such as latitude and climate.

2. Material and methods

(a). Materials

A dataset of measurements for crania and postcranial elements was collated for Epipalaeolithic, Mesolithic and Neolithic samples (table 1). The dataset was optimized to maximize the sample sizes for each of the resultant 11 operational taxonomic units (OTUs), while providing sufficient measurements for the cranium, upper and lower limbs. Fifteen cranial measurements describing the overall dimensionality of the vault and face were found to have sufficient sample sizes for each OTU. The postcranial dataset included metric data for the femur, tibia, humerus, radius, ulna and clavicle. As these bones are bilaterally paired, the right side was chosen when both were available or substituted for the left if missing or broken. In order to create postcranial affinity matrices with levels of information comparable with the cranium, data for the femur and tibia were combined into a ‘lower limb’ dataset (10 variables), while data for the clavicle, humerus, ulna and radius were combined into an ‘upper limb’ dataset (eight variables). Whenever possible, OTUs were constructed using specimens from a single archaeological site (e.g. Çatal Höyük, Afalou) or specimens from a specific ‘cultural’ phase in a given geographical region (e.g. Linienbandkeramik; see electronic supplementary material, table S1). Sampling was constrained by uneven spatial, temporal and archaeological representativeness of certain phases, but the resultant dataset (table 1) represents the first attempt to collate cranial and postcranial data representing the same prehistoric ‘populations’ spread over time and space (figure 1).

Table 1.

Collated data for each of the 11 OTUs. Cr, cranium; LL, lower limb; UL, upper limb; min, minimum; max, maximum; temp, temperature; AVK, Alföldi Vonaldíszes Kerámia; LBK, Linienbandkeramik.

OTU name (sites)	archaeological ‘type’	average latitude	average longitude	average date (BC)	sample sizes (Cr/LL/UL)	min temp °C	max temp °C
Afalou	Epipalaeolithic	36.75	5.58	10 000	33/28/22	6.2	24.1
Taforalt	Epipalaeolithic	34.75	−2.25	9500	14/26/28	10.4	25.2
Natufian	Epipalaeolithic	32.44	35.16	11 500	17/28/29	11.6	25.5
Iron Gates	Mesolithic	44.32	22.02	7000	29/34/27	0.8	21.6
western Mesolithic	Mesolithic	47.42	−3.01	5300	30/10/10	6.5	17.3
Nubian	Neolithic	19.60	30.41	2000	60/28/27	19.2	33
AVK/Vinča	Neolithic	46.84	20.72	5300	12/23/23	1.5	20.6
Lengyel	Neolithic	46.44	17.71	4500	45/25/28	1.5	19.7
Italian Neolithic	Neolithic	40.35	14.75	4500	27/12/15	6.5	24
Çatal Höyük	Neolithic	37.10	32.13	6900	24/59/55	0.8	21.1
LBK	Neolithic	48.73	16.65	5200	29/10/10	1	15
				totals	320/283/274

Figure 1.

Geographical distribution of 11 OTUs employed. Circles, Epipalaeolithic; triangles, Mesolithic; squares, Neolithic OTUs. Stars are way-points (Cairo, Istanbul and Venice) used to constrain pairwise geographical distances (km) between the OTUs.

(b). Biological distance matrices

Electronic supplementary material, table S2 details the individual cranial and postcranial measurements taken on each bone. Given the fragmentary nature of many of the archaeological specimens, and the need to ensure a minimum sample size of 10 for each OTU, some missing data were estimated (see electronic supplementary material, materials and methods). Within each of the three main datasets, only individuals with data present for at least 70% of the measurements were included in the analysis [16]. Missing data were estimated in SPSS v. 20 using multiple linear regressions, within sexes where possible and using the specimens with a complete set of measurements within each OTU. Sex was determined on the basis of morphological traits of the pelvis or on the basis of the skull using standard anthropological methods [46]. All three datasets (cranial, lower limb and upper limb) were separately adjusted for isometric scaling by dividing each individual variable by the geometric mean of all variables (within each particular dataset) for that individual [47]. Biological distance matrices were computed for all 11 OTUs separately for the cranial, lower limb and upper limb datasets under varying assumptions of heritability (i.e. h² = 0.3, 0.5, 0.7 and 1.0) using the Relethford–Blangero [48] estimator of population affinities based on quantitative traits (see electronic supplementary material, data matrices). The choice of heritability is unlikely to affect the overall pattern of results as varying assumptions of h² simply scale the biological distance matrices proportionately. However, given that heritability is likely to differ between the cranium and postcranium, testing the effect of using differing assumptions of heritability on the congruence between phenotypic distance and hypothetical test matrices is important. Principal coordinate analyses were employed to visualize the major affinity patterns among the OTUs for each of three biological datasets using the freeware RMET v. 5.0. In addition, overall patterns of between-OTU congruence were visualized using neighbour-joining phenograms [49] in PHYLIP v. 3.66 (J. Felsenstein, http://evolution.genetics.washington.edu/phylip).

(c). Geographical, temporal and environmental matrices

In order to test the predictions of the stated hypothesis, several additional distance matrices were created using the collated data for all 11 OTUs (table 1). A matrix of pairwise great circle [50] geographical distances in kilometres (GeoDist) was created using the average geographical co-ordinates given in table 1. In order to avoid unrealistic distances between OTUs involving large sea crossings, three way-points were used at Cairo (30.05, 31.54), Istanbul (41.03, 28.97) and Venice (45.44, 12.30) (figure 1). Temporal distance (TimeDist) was calculated as the pairwise OTU difference in the average time BC shown in table 1. Average difference in latitude (LatDist) was calculated as a useful proxy for general environmental differences between populations as is reflects both climatic effects (temperature, rainfall) and ecological parameters such as seasonality and average day length. A matrix reflecting ‘archaeological type’ (ArchType) was created as a binary system, whereby Epipalaeolithic/Mesolithic populations were scored as being the same (1) and different (0) from Neolithic OTUs (and vice versa). This matrix is therefore a crude proxy for the affinities we would expect based on whether or not populations have undergone the transition to farming, which in most cases had a major impact on subsistence, growth patterns, activity, behaviour and other aspects of lifestyle (see various contributions in [51]). Finally, two climatic matrices were generated based on average minimum (MinTemp) and maximum (MaxTemp) annual temperatures. Here, the matrices were calculated as the square root of the squared differences between the average temperature values given in table 1.

(d). Analytical methods

The congruence between the biological distances matrices and each of the comparison matrices was quantified using Mantel [52] tests. Partial Mantel tests [53] were used to test for congruence between two matrices while controlling for the effects of one or more potentially confounding factors. Statistical significance (one-tailed) of all full and partial Mantel tests was assessed via a randomization procedure following 10 000 permutations of one matrix. Given the number of tests conducted, and the need to account for the possibility of type I and II errors, all results were scrutinized at both the α = 0.05 and α = 0.01 levels. All Mantel tests were performed using the freeware PASSaGE v. 2.0 (www.passagesoftware.net).

A model of neutral expectation was constructed based on geographical distance, controlling for temporal distance [16,20]. In the absence of neutral genetic data, geographical distance provides a useful proxy for the expected affinities between populations under neutral evolutionary conditions (i.e. in the absence of strong directional selection) [25,54,55] because of the tendency for contiguous populations to exchange genes more frequently and/or because neighbouring populations tend to share more recent common ancestry [56]. In addition, Konigsberg [57,58] provides empirical evidence suggesting that craniometric distance and temporal distance may be correlated within certain archaeological sites, suggesting that two populations may become differentiated through time even under neutral conditions. Here, therefore, we use a partial correlation of GeoDist, controlling by TempDist as the neutral model of evolutionary expectation [16,20].

Thereafter, the following congruence tests are applied to each of three biological distance matrices (cranium, lower limb and upper limb). In each case, the expectation is that the lower and upper limb matrices will exhibit the same congruence patterns as the cranial matrix:

• — all morphological distance matrices are expected to be significantly correlated with archaeological grouping (ArchType), even after controlling for neutral processes (GeoDist/TempDist).

• — all morphological distance matrices are expected not to be significantly correlated with variation in latitude (LatDist), minimum temperature (MinTemp) or maximum temperature (MaxTemp), especially once archaeological grouping (ArchType) and/or neutral processes (GeoDist/TempDist) have been controlled for.

3. Results

(a). Population affinity patterns

Biological distances (under the assumption of complete heritability, h² = 1.0) between OTUs for the cranium, lower limb and upper limb are illustrated in electronic supplementary material, figures S1 and S2, in the form of plots of the first two principal coordinates (PCo) and unrooted neighbour-joining phenograms, respectively. In the case of the cranial data, there is a clear separation on the first PCo between Neolithic OTUs and Epipalaeolithic/Mesolithic OTUs (see electronic supplementary material, figure S1a). The same distinction is evident in the phenogram (see electronic supplementary material, figure S2a). In the case of the lower limb data, PCo1 separates lower-latitude OTUs (Nubian, Afalou, Taforalt, Çatal Höyük) from more northern OTUs, while PCo2 clearly separates the three Epipalaeolithic OTUs from the later Mesolithic and Neolithic OTUs (see electronic supplementary material, figure S1b). The patterns are somewhat similar for the upper limb (see electronic supplementary material, figure S1c) in that PCo1 appears to reflect differences in average temperature, but in this case PCo2 does not exhibit a clear pattern. The postcranial phenograms (see electronic supplementary material, figure S2b,c) show that for both the lower and upper limb data, the Mesolithic OTUs placed among the Neolithic OTUs rather than with the Epipalaeolithic OTUs, as is the case with the craniometric data. In the case of the upper limbs (see electronic supplementary material, figure S2c), the Epipalaeolithic Afalou OTU is linked with the Neolithic Nubian OTU, reflecting geographical and climatic proximity rather than population history.

(b). Correlation of biological distance matrices and alternative explanatory matrices

All results for full and partial Mantel tests (when biological distances are modelled under the assumption of complete heritability) are given in table 2 (analogous results for h² = 0.3, 0.5 and 0.7 can be found in the electronic supplementary material, table S3). It is clear that the congruence patterns are very different for the cranial dataset compared with the upper and lower limb data. The cranial data fulfil the expectations of the neutral model (i.e. expected affinities under the assumptions of geographical and temporal distance), while the limb data do not. The cranial data also fulfil the congruence predictions by reflecting the differences in ArchType, even after controlling for the neutral model (r = 0.356, p = 0.007), and not exhibiting any significant congruence with LatDist or MinTemp/MaxTemp. In contrast, the limb datasets do not reflect ArchType and exhibit strong (r = ∼0.5) and significant (p < 0.01) congruence with LatDist, even when ArchType (r = ∼0.5, p < 0.01) and neutral processes are controlled for (r = ∼0.5, p < 0.05). The upper limb dataset is significantly and strongly (r > 0.5, p < 0.01) correlated with maximum temperature even when controlling for ArchType and the neutral model. The lower limb is also consistently and strongly (r = ∼0.5) correlated with maximum temperature (controlling for ArchType and neutral processes) at the p < 0.05 level. However, overall the pattern of congruence between the limb datasets and minimum temperature differences is weaker (r = ∼0.3–0.4), although still statistically significant at the p < 0.05 level in most cases (see electronic supplementary material, table S3).

Table 2.

Results of the full and partial Mantel tests as r-values (p-values in parentheses). All significant results at α = 0.01 are indicated by double stars (**) and those at α = 0.05 with single stars (*). Results presented here are for biological distance matrices generated under the assumption of complete heritability (h² = 1). Electronic supplementary material, table S3 provides analogous results for all full and partial Mantel tests under the assumptions of h² = 0.3, 0.5 and 0.7.

	cranium	lower limb	upper limb
GeoDist/TempDist (neutral)	0.531 (0.006)**	0.191 (0.196)	0.185 (0.192)
ArchType	0.419 (0.003)**	0.115 (0.118)	0.020 (0.423)
ArchType/neutral	0.356 (0.007)**	0.017 (0.432)	0.034 (0.376)
LatDist	0.254 (0.107)	0.569 (0.007)**	0.517 (0.002)**
LatDist/ArchType	0.317 (0.090)	0.583 (0.005)**	0.521 (0.001)**
LatDist/neutral	−0.070 (0.557)	0.500 (0.025)*	0.571 (0.002)**
MinTemp	0.307 (0.101)	0.469 (0.022)*	0.377 (0.033)*
MinTemp/ArchType	0.331 (0.090)	0.471 (0.032)*	0.377 (0.035)*
MinTemp/neutral	0.151 (0.171)	0.414 (0.045)*	0.423 (0.013)*
MaxTemp	0.145 (0.214)	0.513 (0.020)*	0.565 (0.003)**
MaxTemp/ArchType	0.192 (0.176)	0.526 (0.020)*	0.568 (0.003)**
MaxTemp/neutral	−0.076 (0.558)	0.462 (0.037)*	0.593 (0.002)**

Electronic supplementary material, table S4 shows the results for Pearson correlations between individual limb dimensions and both minimum and maximum temperatures. Not all limb dimensions reflect variation in temperature, and more variables were found to be significantly correlated with maximum temperature but not with minimum temperature. Also, upper limb dimensions were more strongly correlated with temperature differences than lower limb dimensions. In general, higher annual temperature was found to be positively associated with relatively longer and narrower bone dimensions, as might be expected under ecogeographical patterns such as Allen's rule [59]. No consistent relationship was found between relative limb dimensions and archaeological grouping. Moreover, brachial and crural intralimb indices (see electronic supplementary material, materials and methods) were not found to be statistically correlated with temperature (r = 0.543, p = 0.084 brachial; r = −0.190, p = 0.576 crural), nor did they differ significantly between pre- and post-Neolithic OTUs (Student's t-test, p = 0.494 brachial, p = 0.144 crural). However, both lower and upper limb size (geometric mean of all variables) were found to be significantly correlated with ArchType (Mantel tests, r = 0.46, p = 0.010 lower limb; r = 0.246, p = 0.047 upper limb), with Epipalaeolithic and Mesolithic OTUs bigger on average than Neolithic OTUs. Interestingly, in contrast with the climatic results, the lower limb size differences remained significantly correlated with ArchType even after controlling for both LatDist and MaxTemp (r = 0.461, p = 0.010). However, no significant relationships were found between limb size differences and MinTemp or MaxTemp.

4. Discussion

The results clearly demonstrate that cranial shape affinities distinguish between Epipalaeolithic/Mesolithic and Neolithic populations, under the expectations of a neutral model of geographical and temporal distance, as shown also in previous studies [16,20]. By contrast, relative limb dimensions are positively congruent with environmental factors such as latitude and annual temperature. This finding is in line with other studies that also found a relationship between limb morphology and various environmental factors [35,38,44,60,61]. Upper limb dimensions were found to be significantly and strongly correlated with maximum temperature, with OTUs from warmer climates having relatively longer and more slender upper limb bones. This observation fits with theoretical predictions of thermoregulation and limb movement [62]. Lower limb dimensions were more consistent with difference in latitude than annual temperature per se, and fewer individual lower limb dimensions were significantly correlated with temperature. The observed difference between the pattern of upper and lower limb variation may be explained by locomotor constraint and greater canalization of lower limb morphology, as there is evidence that human upper limbs may exhibit between two and three times the level of variation expressed in the lower limb [37]. Despite this constraint, the overall size of the lower limbs was found to be significantly larger among hunter–gatherers compared with farming OTUs, even after controlling for latitude and temperature, suggesting that the transition to agriculture also resulted in a significant decrease in overall body size. This finding is consistent with previous studies noting changes in stature associated with new dietary regimes [43,63], although the greater limb lengths observed among Epipalaeolithic OTUs could provide evidence for a reduction in body size associated with the Pleistocene–Holocene transition rather than the origins of agriculture. In this case, an initial shift towards smaller body size could correspond with climatic warming associated with the onset of the Holocene. Recent research has suggested that limb segment lengths may be influenced by ‘thrifty’ phenotypic mechanisms and exposure to environmental stress [64]. In this context, a decrease in body size associated with the transition to agriculture could be a plastic response to dietary stress. Further research is required to investigate whether the timing of changes in body size are associated with climatic and/or dietary transitions.

The fact that intralimb indices (brachial and crural indices) were not significantly related to either temperature variation or archaeological grouping suggests that classic thermoregulatory (i.e. Allen's) rules alone cannot explain differences in limb dimensions between pre- and post-Neolithic OTUs. While earlier research showed a strong relationship between intralimb indices and temperature based upon the predictions of Allen's rule [33], other analyses have demonstrated that there may have been a considerable time lag between occupation of an environment and the expression of adaptation in limb proportions [34,35]. This ‘lag’ may explain recent evidence that the correlation between intralimb proportions and climate appears to be weaker than expected, and the suggestion that these traits are genetically conservative [38,65]. This evidence for long-term adaptation in limb proportions among past populations, however, sits at odds with evidence [66] demonstrating both rapid changes in limb proportions of migrant populations [67] and mechanisms of developmental plasticity that could underpin the variation we see in intralimb proportions [37,68]. Without a clear understanding of the mechanisms driving variation in the human limbs, the interpretation of limb proportions in the past remains challenging. The results reported here fit with previous studies demonstrating that changes in limb lengths and proportions are exhibited on a long time scale [34,35,65]. While some directional thermoregulatory selection may underpin these changes throughout the transition to agriculture, changes in growth patterns caused by nutritional quality and/or changing biomechanical regimes may also have had a direct effect on the appendicular skeleton of farming populations.

Future work incorporating information from the entire skeleton will allow these factors to be scrutinized in greater detail, as changes in absolute limb dimensions and intralimb indices must be understood in the context of overall changes in stature [63] and robusticity [35,60]. In addition, future studies that overcome the limitations of the current dataset, such as the small numbers of OTUs per archaeological period, the restricted geographical spread of the OTUs and the use of modern climatic data as a proxy for prehistoric climate, will yield important insights into the environmental effects on overall skeletal morphology. Nevertheless, the results of this study demonstrate that, in contrast with cranial shape, which provides a relatively accurate account of population affinities generated via neutral evolutionary processes, limb dimensions record information regarding a combination of non-neutral environmental effects, including thermoregulatory adaptation, changes in nutritional quality and plasticity due to changing activity patterns.

Funding statement

This research was supported by the European Research Council Starting grant (ERC- 2010-StG 263441).