Network analysis of gene expression in human and chimpanzee brains identifies distinct modules of coexpressed genes in human (A) and chimpanzee (B). (A) Dendrograms produced by average linkage hierarchical clustering of 2,241 genes based on TO (see Supporting Text). The red line in the human dendrogram indicates the height at which the tree was cut (0.95) to define modules. Modules were assigned colors as indicated in the horizontal bar beneath the human dendrogram. Genes in the chimpanzee network are depicted by using human module colors to represent the extent of module conservation. (B) Classical multidimensional scaling plots in three dimensions (color-coded as in A) depict the relative size and cohesion of modules in humans and chimpanzees.

Module visualization identifies hub genes and human-specific connections. (A) Three hundred pairs of genes with the greatest TO in humans are depicted for cortex (brown module). Genes with expression levels that are negatively correlated are connected by black lines. Where gene symbols are unknown, Affymetrix probe set IDs are shown (e.g., 37158_at). (B) Connections from A that are present in humans but absent in chimpanzees (see Materials and Methods).

Modules correspond to functional subdivisions of the brain. (A–G) (Upper) Heat maps depicting expression levels for all genes (rows) in all human and chimpanzee brain regions (columns; black labels are human samples and red are chimpanzee) for each module: turquoise (A), blue (B), brown (C), yellow (D), green (E), red (F), and black (G). Red, increased expression; black, neutral expression; green, decreased expression. (Lower) Barplots of the values of the module eigengene (i.e., the first principal component) derived from singular value decomposition are displayed for each module. Black horizontal lines beneath the barplots denote indicator variables (line = 1, no line = 0). Modules were characterized as follows (Kruskal–Wallis test): cerebellum (1,001 genes, P = 0.00013; A), cortex (360 genes, P = 0.00089; B), cortex (343 genes, P = 0.0000014; C), caudate nucleus (200 genes, P = 0.00013; D), cortex and cerebellum (126 genes, P = 0.003; E), and anterior cingulate cortex and caudate nucleus (122 genes, P = 0.008; F). The black module (G), consisting of 50 genes, is a white matter module as characterized by manual inspection of its constituent genes (see text and Table 1). To assess module conservation between humans and chimpanzees, the Spearman correlations in intramodular connectivity (k_in) were calculated for each module between the species: r = 0.55 (A), r = 0.30 (B), r = 0.39 (C), r = 0.51 (D), NS (E), r = 0.42 (F), and r = 0.62 (G). All correlations were highly significant (P < 10E-6), with the exception of the green module (P = 0.32). H1, human 1; C1, chimp 1, etc.; Broca, Broca's area; acc, anterior cingulate cortex; prv, primary visual cortex; prf, prefrontal cortex; cn, caudate nucleus; vc, cerebellum; NS, not significant. (H) Upon removal of the cerebellar samples from the dataset, an additional module specific to primary visual cortex was identified (P = 0.0011, Kruskal–Wallis test). The Spearman correlation in k_in between humans and chimpanzees was 0.54 (P = 1.36E-6).

DC between humans and chimpanzees reflects differences in gene expression and protein structure. (A) DE vs. DC for 2,152 genes expressed in brain. DE is defined as log₁₀(mean gene expression [human]/mean gene expression [chimp]) in the brain region(s) corresponding to each gene's module (as defined in Fig. 2). DC is defined as log₁₀(k_in [human]/k_in [chimp]). Colors denote modules. The Spearman correlation between DE and DC is 0.32 (P < 2.20E-16; linear least-squares regression line in red). The pairs of vertical and horizontal lines have been arbitrarily drawn to illustrate the utility of DC as a means of stratifying differentially expressed genes. (B) Genes exhibiting DC show evidence of genomic rearrangements between humans and chimpanzees. For each module, 10 genes with k_in human ∼ = k_in chimp, and 10 genes with k_in human ≫ k_in chimp were selected; genes with at least one gap in their aligned exonic sequence were compared (n = 46 [k_in human ≫ k_in chimp] and n = 44 [k_in human ∼ = k_in chimp]). Data were highly skewed and log-transformed. The mean gap percentage in aligned exonic sequence was ≈3-fold higher in DC genes (P = 0.009, Wilcoxon test), suggesting that genomic rearrangements contribute to DC. (C) Genes exhibiting DC show accelerated protein sequence divergence between humans and chimpanzees. Rates of protein sequence divergence (K_a/K_i) were obtained for 1,168 genes from ref. 8. These genes were ranked according to the absolute value of DC between humans and chimpanzees as defined in A. Mean K_a/K_i was significantly higher for the most differentially connected genes (top quintile, n = 234; μ = 0.157) compared with the least differentially connected genes (bottom quintile, n = 233; μ = 0.100; P = 0.003, Wilcoxon test). (Scale bars indicate SE.)