First, we considered sex chromosomes in detail. It has long been known that, due to the particular evolutionary dynamics of sex chromosomes 38394041, sequences linked to the X chromosome have lower divergence rates than those linked to autosomes 314042. These results are confirmed by our analysis of human-chimpanzee pairwise divergence. Genes located in the X chromosomes presented lower synonymous substitution rates (K_S) and lower non-coding divergence (K_I) than those in autosomes, whereas non-synonymous divergence rates (K_A) did not differ (Table 2). Lineage-specific substitution rates (obtained from the second dataset; see Materials and methods) showed the same trends, although significance was lost is some comparisons (Table A1 in Additional data file 1). As usually done in previous studies 2272931, we removed genes linked to sex chromosomes from further analysis.

Next we dealt with segmental duplications (SDs), since they are known to be associated with higher rates of molecular evolution 314344. In the pairwise dataset, divergence rates in the non-coding regions of genes involved in SDs (either in the chimpanzee or in the human lineage) are not different from divergence rates of single-copy genes. This is also the case for K_A and the K_A/K_I ratio (Table 2). Surprisingly, however, K_S is significantly lower in genes within SDs. To explore this discrepancy with the previous literature referenced above, we split genes overlapping SDs in three main categories: those genes that overlap SDs shared by the human and the chimpanzee lineages; genes that overlap human SDs but not chimpanzee SDs; and genes that overlap chimpanzee SDs but not human SDs (Table 3). As expected, genes overlapping human SDs showed higher divergence than genes that do not overlap with SDs. On the other hand, genes overlapping chimpanzee SDs present the opposite pattern, that is, evolutionary rates are significantly lower for coding evolutionary rates. Finally, for those genes that overlap SDs and are shared by the human and chimpanzee lineages, only synonymous divergence is lower within shared SDs. This suggests that the lower rates of divergence for genes overlapping SDs that were detected in the overall analysis may be an artifact of the preliminary state of the annotation of chimpanzee SDs. At any rate, we excluded from further analysis any gene overlapping SDs.

The chimpanzee genome project unveiled higher human-chimpanzee divergence within 10 Mb of the telomeres 2. This effect can be detected in both the pairwise and the lineage-specific datasets (Table 2) and for both exonic and non-coding divergence. This is a particularly important factor, since nine out of the ten major rearrangements separating the two species are pericentric inversions, that is, they exclude telomeres. Thus, considering genes in telomeres might lead to under-estimation of divergence within rearrangements. To avoid such bias, genes within 10 Mb of the telomeres were removed from further analysis.

Recent evidence suggests that, just as telomeres do, centromeric and centromeric transition regions exhibit unique organizational and evolutionary characteristics 454647. In our pairwise dataset, genes located within 5 Mb of pericentromeric regions at each side of centromeres showed significantly lower divergence rates than genes elsewhere in the genome (Table 2). In contrast, there are no significant lineage-specific differences in substitution rates between genes located in centromeric regions and genes in other parts of the genome (Table A1 in Additional data file 1). Given these interesting but potentially confusing patterns, genes in centromeric regions were removed from our dataset.

Finally, human chromosome 19 (HSA19) has been reported to present peculiar divergence and nucleotide composition patterns 48. Our results also pinpoint this chromosome as an outlier. All neutral divergence measures in the pairwise dataset are markedly higher in HSA19 (Table 2). Differences in lineage-specific substitution rates are not as striking. Still, significant differences for K_S in the human and chimpanzee lineages and for K_A in the hominid lineage can be found (Table A1in Additional data file 1). Thus, genes located in this chromosome were also removed from our dataset.

The successive removal of all the genes whose divergence values could be affected by any of the aforementioned confounding factors left 5,804 genes for pairwise analysis (dataset 1) and 2,742 in the lineage-specific analysis (dataset 2). Such filtered datasets, even if dramatically reducing our sample size, allow for a detailed testing of the hypothesis of an association between chromosomal rearrangements and genic divergence rates. A graphic overview of the regions that were included in the following analysis or excluded from it is presented in Figure 1.

As a rough preliminary test, we repeated the comparison between rearranged and collinear chromosomes in this filtered dataset. Human-chimpanzee pairwise divergence rates are not different for synonymous sites (K_S) or for the K_A/K_I ratio (Table 4). In contrast to these results and to all previous literature, average rates of non-coding, K_I, and non-synonymous divergence, K_A, are significantly lower in rearranged chromosomes (Table 4). That is, the original trends detected in the unfiltered dataset remain, albeit with weaker statistical support. None of the comparisons performed upon lineage-specific rates are strikingly different. Only non-synonymous divergence for humans and neutral divergence in the hominid branches present marginal differences, being lower in rearranged chromosomes.

We then focused on rearranged chromosomes themselves and compared genes within inversions against genes outside them. In the pairwise dataset, non-coding sequences showed significantly lower divergence within rearrangements than outside them (0.0120 versus 0.0117, P value < 0.001) whereas no significant divergence differences were detected for K_A, K_S and the K_A/K_I ratio (Table 4). No general pattern was detected in the lineage-specific analysis, even if genes within rearrangements show marginally lower rates in some cases (K_A in human branch, K_S in the chimpanzee branch and both K_A and K_S in the hominid lineage; Table A2 in Additional data file 1). This suggests that the association between rearranged chromosomes and lower divergence rates reported above is mainly due to genes within the rearrangements themselves. However, when the analysis is repeated removing genes within rearrangements, divergence is still lower in genes located in rearranged chromosomes (but outside rearrangements; Table 4).

These results cannot be biased by the strict filtering applied before our main analysis. Equivalent, although stronger, trends were obtained before filtering when all genes were included in the analysis (data not shown). It is interesting, however, to consider the relative contributions of the various factors under study upon the divergence patterns between the two species. To do so, we used K_I, since it is based on much larger amounts of data and, thus, it is less noisy than the other measures (K_I is computed for a 250 kb window centered in each gene; see Materials and methods for details). A simple regression analysis allows us to see that, altogether, the location of genes in sex chromosomes, telomeres, centromeres, SDs, HSA19 or within rearrangements explains only about 37% of the variance in K_I (R² = 0.372). This shows that, as expected, other smaller-scale factors, including the individual history of each gene, have a considerable influence on nucleotide divergence patterns. All the studied factors present highly significant regression coefficients (P values < 0.001) with the exception of centromeres, whose effect is non-significant under our linear regression model. Among the remaining factors, telomeres, HSA19 and sex chromosomes show the largest standardized regression coefficients (β = 0.488, -0.274 and 0.143, respectively; with approximately 27% of the variance explained by telomeres alone), while the fact of a gene being within rearrangements or segmental duplications has much smaller power to predict divergence values (β = -0.054 and 0.036, respectively).

If rearrangements did affect divergence rates due to their recombination-reducing effect (including effects due to speciation-related processes), their effect should be maximum around the rearrangement breakpoints, where recombination between different chromosomal arrangements is most strongly reduced 49. To test for this possibility, we defined windows of 2 Mb around each rearrangement breakpoint (1 Mb at each side). Then, we compared genes within these windows against all genes in rearranged chromosomes (Table 5). In the pairwise analysis, we detected lower divergence in non-coding regions surrounding the evolutionary breakpoints. Exons also show lower K_S and K_A values near breakpoints when compared to the rest of the chromosome, although neither of these results are statistically significant (Table 5). None of these differences can be detected in lineage-specific substitution rates (Table A3 in Additional data file 1).

It would thus seem that evolutionary rates of genes close to breakpoints follow the same trend as genes within rearrangements. To check whether these two trends are independent, we removed genes surrounding breakpoints and repeated the main analysis comparing divergence within and outside rearrangements. Results did not change: in the pairwise analysis, genes within rearrangements displayed lower non-coding divergence than the rest of the rearranged chromosomes (Table 6), even if reduced sample size limits our power and some results are not significant anymore (Table A4 in Additional data file 1).

Finally, the accumulation of genes with K_A/K_S > 1 in colinear chromosomes reported by Zhang et al. 28 can also be detected in our pairwise dataset, although K_A/K_I is used instead of the 'standard' K_A/K_S ratio. When focusing on rearranged chromosomes alone, no significant accumulation of genes with K_A/K_I > 1 was found either within or outside rearrangements (Table A9 in Additional data file 1).

As explained above, genes located near the centromere had lower divergence than genes elsewhere in the genome (Table 2). This suggests that a possible explanation for our observation of lower divergence within rearrangements could be related to the fact that all the rearrangements analyzed are pericentric inversions. It is thus possible that removing genes in the centromeres and within a 5 Mb pericentromeric region on each side, as we did, is not enough to control for any potential centromere-related effects.

To test this hypothesis, we defined virtual pericentric inversions in colinear chromosomes, spanning the same average proportion of each chromosome as the real nine major inversions do in rearranged chromosomes. We compared genes within these virtual regions with genes outside them but in the same chromosomes. Table 7 shows that divergence patterns in these virtual rearrangements are similar to those in real rearranged chromosomes. In the pairwise comparison, non-coding divergence is also lower within virtual inversions (Table 7) and, again, no pattern can be detected in the lineage-specific analysis (Table A5 in Additional data file 1). This suggests that centromere-related effects extending beyond the 5 Mb windows we considered may be responsible for some, even if not all, of our observations.

All the above results refer to the ten major rearrangements separating humans and chimpanzees. More detailed information on the structural changes between the two species has recently become available by means of mapping chimpanzee fosmid paired-end sequences against the human genome 50. This analysis unveiled 37 smaller rearrangements (usually < 1 Mb) which, in contrast to the major ones, do not include centromeric regions and, thus, allow the exclusion of any potential bias caused by centromeres. We compared substitution rates of genes overlapping these rearrangements with genes in colinear regions. Pairwise non-coding substitution rates were found to be marginally higher within these rearrangements (K_I = 0.0121 versus 0.0128, P value = 0.020; Table 8) whereas other divergence measures do not present significant differences. This observation can not be retrieved in the lineage-specific analysis but, in any case, the sample size for this kind of approach is really small and should be treated with caution (Table A6 in Additional data file 1).

So far, all the tests presented here were performed by pooling all rearranged chromosomes together. It is clear, however, that no chromosomal speciation model proposes that every single rearrangement ought to have played a relevant role in the speciation processes that separated humans and chimpanzees. In fact, it is reasonable to assume that most rearrangements would have appeared and become fixed along the evolutionary history of lineages (anagenesis) and not during the relatively shorter cladogenic periods 2526. It is thus possible that a majority of speciation-unrelated rearrangements could be masking the molecular signature of chromosomal speciation in the few rearrangements involved in such processes. Provided, of course, that there are any speciation-related rearrangements at all. In fact, a recent comparative gene-expression study hints at some chromosomes (such as HSA4, HSA5, HSA9, HSA15 and HSA16) as the most different in terms of differences in expression pattern 29.

Thus, we repeated all previous analyses on a chromosome-per-chromosome basis (Table 9; Table A7 in Additional data file 1). In most cases, the small sample size caused by our extremely conservative filtering process precludes the detection of any trend or even the performance of tests (for example, no genes from chromosomes HSA15, HSA16 or HSA 18 are included in our dataset after filtering). For the rest of the chromosomes, the trends reported after filtering were similar to those obtained with the unfiltered dataset (not shown) but, of course, lower divergence in genes within pericentric rearrangements is to be expected if, for example, the highly divergent telomeres are not filtered-out.

HSA 4 clearly stands out in the pairwise comparison. It presents statistically higher K_A, K_I and K_A/K_I within the inversion (having removed the breakpoints). The centromeric region of HSA4 presents the usual lower divergence, thus confirming that the effect of HSA4 was not due to any special properties of its centromere extending beyond 5 Mb. In contrast to other chromosomes, genes outside the inversion in HSA4 also present higher divergence than genes in colinear chromosomes.

The other chromosome that stands out in the analysis is HSA12, which presents lower divergence, both for genes within its inversion relative to those outside it and for genes outside the inversion relative to genes in colinear chromosomes (data not shown). HSA15 presents the same trend, although with less statistical strength. Together, these two chromosomes are the major contributors to the observation of lower divergence for genes outside rearrangements than for genes in colinear chromosomes.

Recombination rates have been shown to correlate positively with divergence 51. We first examined the relationship between recombination and the factors we have excluded from our analysis. All figures are given in cM·Mb^-1. In our dataset, recombination rates are higher for genes located in the X chromosome than for genes elsewhere in the genome (1.43 versus 1.21, P value 0.027). This is also the case for genes in telomeric regions (1.09 versus 1.97, P value < 0.001) and in HSA19 (1.08 versus 1.57, P value < 0.001). All these results are congruent with previous observations 52. Recombination rates are also lower for genes located in SDs (1.28 versus 1.04, P value < 0.001) and centromeric regions (1.10 versus 0.82, P value = 0.002).

We then focused on chromosomal rearrangements. Recombination rates for both classes of chromosomes (colinear and rearranged) are very similar (1.06 versus 1.09, P value not significant). Within rearranged chromosomes, recombination rates are significantly higher within inversions than in regions outside the inversion, but marginally so (1.07 versus 1.24, P value = 0.07). Also, regions surrounding breakpoints show higher levels of recombination than the rest of their chromosome (1.91 versus 1.08, P value = 0.002).

To see whether rearrangements were enriched in genes with functions leading to reproductive isolation, we performed an analysis of Gene Ontology (GO) 53 terms. In our dataset, several GO categories are overrepresented in rearranged regions (Table A10 in Additional data file 1). Some of the functions, such as cytokine activity, G-protein-coupled receptor binding or immune response have been previously pinpointed as enriched in genes presenting positive selection along the human lineage 2545556. Interestingly, genes related to 'behavior' are also found more often within the inverted regions than expected by chance. Finally, in the specific inversion of HSA4, only the category of response to biotic stimulus is overrepresented.

All data analyzed were retrieved from the initial chimpanzee genome sequence 2 and the methods therein should be consulted. In summary, two databases were used. First, a set of more than 13,000 unambiguous human-chimpanzee orthologous genes filtered to avoid overrepresentation of gene families. From that initial dataset, only those genes with unequivocal coordinates in both species were kept. The chromosomal position of the sequences is a key parameter of our analysis, and, thus, genes in random chromosomes were also removed from our analysis, leaving a total of 12,135 genes.

For every coding sequence, several conventional indexes of molecular evolution, such as the number of non-synonymous substitutions per non-synonymous site (K_A), the number synonymous substitutions per silent site (K_S), and their ratio (K_A/K_S) were estimated using the maximum likelihood method implemented in the package PAML 61. Substitution rates for non-coding sequence were calculated as K_I, the number of substitutions per non-coding nucleotide. A K_I value was obtained for a window of 250 kb, centered on each gene. We used K_A/K_I instead of K_A/K_S as the measure of rates of protein evolution, because of the close proximity between human and chimpanzees, which results quite often in a K_S equal to 0. The averages for K_A, K_S, K_I, and the ratio K_A/K_I are 0.00317, 0.0142, 0.0126 and 0.2483, respectively. Because of the strict criteria defined to retrieve the set of orthologous genes, the maximum values of each index are not high enough to be suspicious of false orthology or misalignment (K_S < 0.32, K_A < 0.055 and K_I < 0.0259)

A second dataset was used to calculate lineage specific evolutionary rates. More than 7,000 unambiguously orthologous genes were recovered for 4 species (human, chimpanzee, rat and mouse). We applied the same filtering criteria as in the previous dataset and were left with a set of 4,905 orthologous genes with coordinates in both species and evolutionary rates for every branch in the non-rooted tree. Finally, the lineage specific evolutionary rates were estimated using a non-rooted tree in PAML.

Polymorphism data were gathered from the SeattleSNP webpage 57. Briefly, we downloaded nucleotide diversity measures for 256 genes. These measures have been obtained from full resequenceing of 24 African-American and 23 European (Centre d'Etude du Polymorphisme Humain (CEPH)) subjects.

Human recombination rates, measured in cM·Mb^-1, were obtained from the fine-resolution recombination map in the USCS genome browser by selecting the track SNP Recombination Rates. Estimates are based on the HapMap phase I data, release 16a, and Perlegen data 62. Fine scale recombination maps are not yet available for chimpanzees. All genes were assigned a recombination rate computed as the average of all SNPs included within them. Any genes for which recombination rates could not be determined were removed from any recombination-based analysis.

Coordinates of telomeres and centromeres of all chromosomes were obtained from Build 34 of the human genome 63 and NCBI Build 1 of the chimpanzee genome 63. We considered as rearranged chromosomes all those for which major chromosomal rearrangements in either the human or the chimpanzee lineages have been indicated by recent in silico 27 or cytological data 8910111213. This comprised human chromosomes 1, 4, 5, 9,12,15, 16, 17 and 18, which differ by a pericentric inversion, and human chromosome 2, which has been generated by an ancestral telomere-telomere fusion 6. For all chromosomes, all in silico-estimated coordinates were compared with newly available cytological data in order to confirm inversion coordinates. The most remarkable difference from both methodologies comes from chromosome 1, in which an inversion of about 30 Mb was detected in silico that has not been detected by cytological approaches (Table A9 in Additional data file 1).

Location information was derived from both humans and chimpanzees. When genes located in different genomic regions of interest (such as sex chromosomes, SDs or telomeres) were studied, being in one such region in either human or chimpanzee was enough to classify a gene as located in such a region. Location was established sequentially as shown in the Results section.

Genes in different categories were compared by means of pairwise permutation tests (based on 1,000 permutations). P values are calculated as the proportion of times that the difference of averages between two categories in a permuted dataset is equal to or larger than the observed difference.

Functional annotations of genes based on GO 53 were extracted from 66 for the three ontologies Molecular function, Biological process and Cellular component. GO terms are organized into hierarchical structures such that a specialized term can be associated with several less specialized terms. We used an inclusive analysis, in which genes annotated with terms that are descendant of a term corresponding to a given level take their annotation from their parent.

To test whether there was a significant deviation from random expectation for distribution of GO annotations for genes in colinear chromosomes compared to genes in rearranged chromosomes or genes within the inverted zone compared to genes outside of the inversion, we used the Z-score transformation:

Z x = ( X − μ x ) / σ x ¯ MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfeBSjuyZL2yd9gzLbvyNv2Caerbhv2BYDwAHbqedmvETj2BSbqee0evGueE0jxyaibaiKI8=vI8GiVeY=Pipec8Eeeu0xXdbba9frFj0xb9Lqpepeea0xd9q8qiYRWxGi6xij=hbbc9s8aq0=yqpe0xbbG8A8frFve9Fve9Fj0dmeaabaqaciaacaGaaeqabaqabeGadaaakeaacaWGAbWaaSbaaSqaaiaadIhaaeqaaOGaeyypa0JaaiikaiaadIfacqGHsisliiGacqWF8oqBdaWgaaWcbaGaamiEaaqabaGccaGGPaGaai4laiab=n8aZnaaBaaaleaaceWG4bGbaebaaeqaaaaa@3D20@

where μ_x = mean and σx¯ MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfeBSjuyZL2yd9gzLbvyNv2Caerbhv2BYDwAHbqedmvETj2BSbqee0evGueE0jxyaibaiKI8=vI8viVeY=Nipec8Eeeu0xXdbba9frFj0xb9qqpG0dXdb9aspeI8k8fiI+fsY=rqGqVepae9pg0db9vqaiVgFr0xfr=xfr=xc9adbaqaaeGacaGaaiaabeqaaeqabiWaaaGcbaacciGae83Wdm3aaSbaaSqaaiqadIhagaqeaaqabaaaaa@32AC@ = standard error). σx¯ MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfeBSjuyZL2yd9gzLbvyNv2Caerbhv2BYDwAHbqedmvETj2BSbqee0evGueE0jxyaibaiKI8=vI8viVeY=Nipec8Eeeu0xXdbba9frFj0xb9qqpG0dXdb9aspeI8k8fiI+fsY=rqGqVepae9pg0db9vqaiVgFr0xfr=xfr=xc9adbaqaaeGacaGaaiaabeqaaeqabiWaaaGcbaacciGae83Wdm3aaSbaaSqaaiqadIhagaqeaaqabaaaaa@32AC@ was calculated as:

σ x ¯ = p ( 1 − p ) N MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfeBSjuyZL2yd9gzLbvyNv2Caerbhv2BYDwAHbqedmvETj2BSbqee0evGueE0jxyaibaiKI8=vI8GiVeY=Pipec8Eeeu0xXdbba9frFj0xb9Lqpepeea0xd9q8qiYRWxGi6xij=hbbc9s8aq0=yqpe0xbbG8A8frFve9Fve9Fj0dmeaabaqaciaacaGaaeqabaqabeGadaaakeaaiiGacqWFdpWCdaWgaaWcbaGabmiEayaaraaabeaakiabg2da9maakaaabaqcfa4aaSaaaeaacaWGWbGaaiikaiaaigdacqGHsislcaWGWbGaaiykaaqaaiaad6eaaaaaleqaaaaa@3AD5@

where p = proportion of genes in the category in question and N = number of genes in the category. If several inclusive categories were found overrepresented in the regions of study, we picked up the significant GO category with higher hierarchical level. P values were estimated from Z-score using the algorithm described in 67. Only significant values after Bonferroni correction for multiple testing were considered.