Results and Discussion

We used densely tiled oligonucleotide probes (9) and a recently described protocol (11) to capture the protein-coding exons from 17,367 genes in a ∼49,000-y-old (12) (uncalibrated radiocarbon date) Neandertal from Spain (SD1253, El Sidrón Cave) and a ∼44,000-y-old (uncalibrated radiocarbon date) Neandertal from Croatia (Vi33.15, Vindija Cave). The DNA libraries from these two Neandertals contain 0.2% and 0.5% endogenous DNA, respectively, and the capture approach enriched the endogenous DNA 325-fold and 153-fold, resulting in an average coverage of 12.5-fold and 42.0-fold for the El Sidrón and Vindija exomes, respectively (Table 1). Present-day human mitochondrial contamination in the enriched libraries is estimated to be between 0.24% [confidence interval (CI): 0.23–0.24%] and 0.40% (CI: 0.38–0.42%) for El Sidrón and between 0.28% (CI: 0.27–0.28%) and 1.08% (CI: 1.05–1.08%) for Vindija. Before genotypes were called, we lowered the base quality scores toward the ends of the Neandertal DNA sequences to compensate for substitutions potentially due to cytosine deaminations (Methods and Fig. S2).

We also retrieved the exomes from the genome sequences of a Neandertal (10) (Table 1) and a Denisovan (13) individual, both from Denisova Cave in the Altai Mountains, Siberia, as well as from nine present-day humans (Yoruba, Mandenka, and Dinka from Africa; French, Sardinian, and Italian American from Europe; Han and Dai from Asia; and Papuan from Oceania), which have been sequenced to a quality similar to that of the two archaic genomes (13, 14) (SI Appendix and Table S7). A neighbor-joining tree based on pairwise differences (Fig. 1A) shows that the three Neandertals form a clade, whereas the Denisovan individual is a sister group to the Neandertals, in agreement with previous results (8, 13, 15).

• Download figure
• Open in new tab
• Download powerpoint

Fig. 1.

(A) Neighbor-joining tree based on the number of pairwise differences between individuals. Phenotypic categories enriched in amino acid-changing substitutions on the lines to Neandertals and to modern humans, respectively, are indicated. The bar indicates 50 inferred substitutions per megabase. (B) Neighbor-joining tree of pairwise F_ST values for the same individuals. The bar indicates an F_ST of 0.05.

As expected for closely related groups, a substantial amount of the variation is shared between Neandertals and present-day humans. Thus, about 33% of the derived alleles seen in the Neandertals are shared with the present-day humans and about 19%, 24%, and 24% of the derived alleles seen in the Africans, Europeans, and Asians, respectively, are shared with the Neandertals (Table S14). The number of nucleotide differences within an individual (heterozygosity) is 0.143 per thousand coding bases in the El Sidrón Neandertal, 0.127 in the Vindija Neandertal, and 0.113 in the Altai Neandertal. The average heterozygosity among the three Neandertals (0.128) is 25%, 33%, and 36% of the average heterozygosity in the three African (0.507), European (0.387), and Asian (0.358) individuals, respectively (Table S12), and thus significantly smaller than that in humans today (P = 4.5 × 10⁻³; Mann–Whitney U test). The three Neandertals also have longer runs of homozygosity (Fig. S9), suggesting that mating among related individuals may have been more common in Neandertals than in present-day humans, compatible with the observations in the Altai Neandertal (10). Also compatible with these observations is that the genetic differentiation among individuals, as measured by the pairwise fixation index (F_ST) (Fig. 1B), is greater among Neandertals than among the Africans, Europeans, and Asians (P = 1.3 × 10⁻²; Mann–Whitney U test), suggesting that mating in small and isolated populations may have caused Neandertal populations to be more differentiated from one another than what is typical for present-day humans.

To investigate the extent to which the population history reflected in the low genetic diversity affected the evolution of coding regions, we investigated derived alleles present in Neandertals and compared them with those of the present-day Africans, Europeans, and Asians (Table 2). We found that the fraction of all derived SNPs inferred to change amino acids in Neandertals (51.1%) is larger than that in Africans (45.7%; P = 3.1 × 10⁻¹¹; G-test), Europeans (46.4%; P = 6.9 × 10⁻¹⁰), and Asians (46.7%; P = 4.5 × 10⁻⁷) (Table 2). We used PolyPhen-2 (16) to assess whether the amino acid-changing SNPs in Neandertals are enriched in alleles inferred to affect protein function or structure and thus likely to be slightly deleterious. The proportion of such SNPs in the Neandertals (45.4%) is larger than that in Africans (36.4%; G-test; P = 1.6 × 10⁻¹⁴), Europeans (37.9%; P = 4.5 × 10⁻⁹), and Asians (38.1%; P = 1.3 × 10⁻⁹) (Table 2). Similarly, the proportion of such SNPs in conserved protein sites [PhastCons (17)] is larger in the Neandertals (50.1%) than in the Africans (38.1%; G-test; P < 10⁻²⁰), Europeans (40.2%; P = 1.1 × 10⁻¹⁵), and Asians (39.7%; P < 10⁻²⁰) (Table 2). Furthermore, a physicochemical classification of amino acid substitutions, the Grantham score (GS) (18), shows that the proportion of derived SNPs in the Neandertals that cause radical (GS >100) amino acid changes is larger than that in present-day humans (Table 2). Derived alleles in conserved protein sites that are homozygous in all individuals analyzed also are more frequent in Neandertals (39.6%) than in Africans (G-test; 29.4%; P = 2.1 × 10⁻³), Europeans (27.8%; P = 1.6 × 10⁻⁵), and Asians (26.9%; P = 5.9 × 10⁻⁶) (Table 2).

Although only six Neandertal chromosomes are available, we note that among derived amino acid-changing alleles likely to have been at low frequency in Neandertals because they are seen once among the six chromosomes (Fig. 2), a higher proportion is inferred to alter protein structure or function in Neandertals (49.1% by PolyPhen-2; similar results with PhastCons) than in Africans (G-test; 38.3%; P = 2.2 × 10⁻¹²), Europeans (40.6%; P = 8.8 × 10⁻⁷), and Asians (41.8%; P = 5.1 × 10⁻⁶). Furthermore, we note that these low-frequency, putatively deleterious alleles have a stronger inferred impact on proteins in Neandertals than in Africans (P < 2.2 × 10⁻¹⁶), Europeans (P = 6.7 × 10⁻⁹), or Asians (P = 2.0 × 10⁻⁷) (Fig. 2 and SI Appendix).

• Download figure
• Open in new tab
• Download powerpoint

Fig. 2.

Frequency spectra of nonsynonymous derived alleles classified by PolyPhen-2 as either benign or deleterious in Neandertals and present-day humans. The ratio of deleterious to benign derived alleles is higher in Neandertals than in the present-day humans. See Table S19 for the actual allele counts and Fig. S10 for the frequency spectra of nonsynonymous derived alleles assessed by PhastCons and GS.

When analyzed individually, each of the three Neandertal individuals carries a larger fraction of derived amino acid-changing alleles (P = 7.6 × 10⁻³ for SNPs; P = 7.0 × 10⁻³ for homozygous; Mann–Whitney U test) as well as alleles inferred to change protein function (PhastCons; P = 8.0 × 10⁻³ for heterozygous, P = 8.0 × 10⁻³ for homozygous) than any of the present-day individuals analyzed (Tables S15 and S16). Thus, not only Neandertals as a group, but also each Neandertal individual analyzed, carried a larger fraction of putatively deleterious alleles than present-day humans. This also is the case for the Denisovan individual, who represents a little-known Asian population related to Neandertals.

We conclude that the patterns of coding variation in Neandertals differ strongly from those in present-day humans, a fact likely to be the result of differences in their demographic histories. Inferences from the complete Neandertal genome suggest that sometime after 0.5–1.0 Mya, the ancestral Neandertal population decreased in size, whereas the ancestors of present-day humans increased in number (10). A low population size over a long time would reduce the efficacy of purifying selection and contribute to a larger fraction of slightly deleterious alleles, particularly at low frequency, but not necessarily to the absolute number of slightly deleterious alleles per individual (19) (SI Appendix). The fact that the three Neandertals carry longer tracts of homozygosity and differ more from one another than present-day humans within continents further suggests that Neandertals may have been subdivided in small local populations. However, we note that the three Neandertal individuals analyzed are not contemporaneous but differ in time by perhaps as much as 20,000 y. It therefore is crucial for DNA sequences from more Neandertals to be generated when suitable specimens become available. It is also crucial for more complex demographic scenarios, for example involving recurrent population reductions and expansions, to be explored to fully understand the differences in population history between Neandertals and modern humans.

We previously used the Neandertal and Denisova genomes (8, 10, 13) to identify amino acid changes that are shared by all humans today but do not occur in the archaic genomes and therefore are inferred to have risen to high frequency on the modern human lineage. The exomes of the three Neandertals and the Denisovan individual allow us, for the first time, to identify derived amino acid changes shared by the three Neandertals as well as the Denisovan individual that are not seen, or occur at low frequency, in present-day humans. Such changes are of interest because they may underlie phenotypes specific to the archaic populations. We calculated the fraction of all amino acid changes specific to either the archaic or present-day human lineages for each phenotype category of genes in the Human Phenotype Ontology (20), and compared it with the fraction of all amino acid changes in the same phenotype categories that occurred on the human lineage after its divergence from the chimpanzee lineage but before the divergence of modern and archaic humans (SI Appendix). Thus, unlike in previous approaches (8, 10, 13), we assessed the enrichment of amino acid changes in phenotypes across multiple archaic individuals while correcting for the size and base composition of genes across the genome (Methods and SI Appendix).

For the archaic lineage before the divergence of Neandertals and Denisovans, we find an overrepresentation of changes in genes related to metabolism, the cardiovascular system, hair distribution, and morphology (genitals, palate, face, extremities, joints, digits, thorax, orbital, and occipital skull regions and general mobility) [Table S28; binomial test; P values and false discovery rates (FDRs) in the table]. Thus, it may be that future functional investigations of the changes in the 22 genes identified in enriched categories (Table S29) will reveal the genetic basis for some features shared by Neandertals and Denisovans, although almost no morphologic information currently is available for the latter (15).

In the Neandertal lineage after the split from Denisovans, only the phenotype category “hyperlordosis” is overrepresented (Fig. 1A; binomial test; P = 4.5 × 10⁻⁵; FDR = 2.1 × 10⁻²). This is interesting because Neandertals had a reduced lordotic curvature relative to other archaic hominins as well as to modern humans (21). Changes in the six genes associated with this phenotype (CUL7, GLB1, NEB, COL2A1, HSPG2, and VPS13B) thus may have contributed to this feature of the Neandertal morphology.

Finally, the modern human lineage has an overrepresentation of amino acid-changing substitutions in genes related to pigmentation (Fig. 1A; P = 3.2 × 10⁻⁵; FDR = 5.9 × 10⁻²) and behavioral traits (Fig. 1A; P < 5 × 10⁻⁴; FDR < 7.8 × 10⁻²). The overrepresentation in the pigmentation category is the result of amino acid substitutions in two genes involved in melanosome function (GPR143 and LYST), for which the derived alleles are at high frequency in Africa and fixed outside Africa. These mutations may contribute to differences in pigmentation among present-day humans (22, 23). In contrast, the behavioral category is the result of substitutions that seem to be fixed in all present-day humans and occur in the three genes adenylosuccinate lyase (ADSL), glycine dehydrogenase (GLDC), and transmembrane protein SLITRK1. These genes are associated with de novo purine biosynthesis, mitochondrial glycine degradation, and neurite outgrowth, respectively. In the case of ADSL, mutations known to impair the gene result in varying degrees of psychomotor retardation, autism, and muscle wasting, whereas in the case of GLDC, mutations result in glycine encephalopathy and in the case of SLITRK1, in Tourette syndrome, characterized by motoric and vocal ticks. In the Human Phenotype Ontology (20), all these genes are annotated as being involved in “hyperactivity” and “aggressive behavior.” Thus, it is possible that the mutations in these genes may have affected some component of behavior unique to modern humans. However, we note that even if they did, the way in which they occurred is unknown. For example, if they affected activity or aggression levels, it is unclear whether they increased or decreased such traits. In any event, these changes clearly warrant further investigation.

In conclusion, we show that the genetic diversity of late Neandertals across their entire geographical range was lower than that of present-day humans and that there is evidence suggesting that Neandertal populations may have been small and isolated from one another. We also show that genes involved in morphology may have changed more on the Neandertal and Denisova lineages than on the preceding lineage from the common ancestor shared with chimpanzees, whereas genes involved in behavior may have changed more on the modern human lineage.