Comparative study maps fast-evolving brain genes in humans

A comparative analysis of human and macaque brain genes has identified 984 brain-expressed genes with unusually high evolutionary rates, according to a study published in Frontiers in Neuroscience. The researchers found 450 genes with elevated interspecies divergence, measured by the ratio of nonsynonymous to synonymous substitutions, and 534 genes with elevated variation within human populations. Protein-interaction network analysis further identified 97 and 137 linker genes, respectively, that connect the fast-evolving genes into functional subnetworks associated with immune signaling, metabolism, nervous-system development and neurological disease pathways.

The work is an in silico exploration of coding-sequence evolution rather than an experiment demonstrating that any of the identified genes causes a distinct human trait. It combines published human and macaque brain transcriptomes, coding sequences from Ensembl, population variation data from the 1000 Genomes Project and a human protein-protein interaction network. The researchers then used network topology and single-cell brain data to examine where fast-evolving genes sit, which functions accompany them and whether the broader species-level expression signature appears in particular cell populations.

For the interspecies comparison, the team began with 23,467 genes expressed in adult human brains and 17,102 expressed in macaque brains, retaining genes with positive cell-type specificity scores. Homologous coding sequences with valid evolutionary-rate estimates produced a shared set of 13,522 genes expressed in both species. The resulting analysis combined dN/dS, which compares nonsynonymous and synonymous substitutions between species, with pN/pS, which was estimated from synonymous and nonsynonymous variant sites within human populations. The population calculation used 84.4 million variants from 2,504 people representing 26 populations, retaining both common and rare variants.

To isolate unusually fast-evolving genes, the researchers applied the conventional 1.5-times-interquartile-range outlier rule. Because the lower boundaries calculated from the interquartile ranges were negative while both evolutionary metrics were positive, only upper outliers were retained. Genes above the dN/dS threshold were classified as DEinter, representing differential evolutionary constraints between species, while those above the pN/pS threshold were classified as DEintra, representing constraints within the human lineage. This statistical definition is important: the sets identify values that depart markedly from the distributions of other brain-expressed genes, but the analysis alone does not establish the precise selective mechanism responsible.

Protein-interaction topology revealed a contrast between the fast-evolving genes and the broader brain-expression background. Of the 450 DEinter genes, 323 could be mapped to the network, where their median degree was 5. Of the 534 DEintra genes, 415 mapped to the network, with a median degree of 6. By comparison, 11,712 of the 13,522 background genes mapped to the network and had a median degree of 10. Most nodes in all three groups had only a few connections, although a small number had more than 500. The authors interpret the low-degree outliers as peripheral nodes that may experience weaker purifying selection than highly connected, more conserved hubs.

The relationships between sequence evolution and cellular expression were mixed rather than uniformly strong. Both dN/dS and pN/pS were negatively correlated with maximum cell-type specificity scores, indicating that more cell-specific genes in this dataset tended to have lower measured evolutionary-rate ratios. The two fast-evolving groups also differed significantly from the background distribution of specificity scores, but not significantly from one another. Coding-sequence length and GC content showed weak negative correlations with both metrics, while genes affected by synonymous-substitution saturation generally had lower rates. The researchers reported that no additional outliers remained after accounting for these covariates, underscoring how easily such factors can affect conclusions derived from coding sequences.

Functional enrichment concentrated the interspecies outliers in 13 KEGG pathways retained at a false-discovery rate below 0.2. These included cytokine-cytokine receptor interaction, cell-adhesion molecules, oxidative phosphorylation and pathways related to Alzheimer’s, Parkinson’s and Huntington’s diseases. The set also included hematopoietic-cell lineage, NOD-like receptor signaling and natural-killer-cell-mediated cytotoxicity, alongside pathways linked to measles, malaria and influenza. GO biological-process terms added associations with taste and smell stimuli. No KEGG pathway was detected for the DEintra set, although its GO terms included immune response, energy metabolism and processing of genetic information, with “response to auditory stimulus” among the reported enrichments.

To examine how the outlier genes might operate together, the team used a Steiner minimal-tree algorithm to construct compact subnetworks from the human protein-interaction network. The DEinter network contained 397 nodes and 883 edges, while the DEintra network contained 534 nodes and 1,136 edges. Their clustering coefficients were higher and their average shortest-path lengths shorter than those of 1,000 Erdős-Rényi-style simulated networks, a result the authors describe as consistent with scale-free architecture. Twenty-three high-degree genes were shared by the two sets of added linker genes. Candidates with prior evidence of brain-expression or evolutionary divergence included SRC, EP300, CCR5, EEF1D, CFTR, RAD51, GFAP, MMP9, IL4 and COPS5, but their presence in a network does not by itself establish a brain-specific adaptive function.

Exploratory single-cell analysis brought the broad evolutionary pattern into a cellular context. The researchers combined two human brain-organoid datasets with two macaque brain datasets, producing an integrated collection of 131,759 cells after processing and batch correction. Single-sample gene-set enrichment scores for the DEinter signature were higher in human cells than in macaque cells, with the difference assessed at P less than 2.22 × 10^-16. Much of the signal was associated with excitatory neurons and non-neuronal cells. Eight of the 13 enriched pathways were recovered in the single-cell data, including cell-adhesion molecules and Alzheimer’s disease pathways, which the analysis indicated were expressed at lower levels in the human organoids.

Several limitations constrain how the findings can be interpreted. The human-population analysis pooled 26 populations rather than calculating population-specific evolutionary rates, so it cannot identify which groups carry particular variants. Genes with very few synonymous variants may produce unstable pN/pS estimates. The specificity filter could have excluded meaningful information from genes expressed broadly across the brain. In addition, matched adult human and macaque multi-tissue samples were unavailable, preventing the researchers from excluding selection acting in peripheral organs, a particular concern because many proteins have functions beyond the nervous system. The single-cell component was also exploratory: organoid transcriptomes differ from those of adult brains, so the findings should not be treated as definitive cross-species validation.

Despite these restrictions, the study offers a gene-level complement to earlier pathway-wide analyses. It shows that evolutionary signals associated with neurological disease, immunity and metabolism can be concentrated in small sets of unusually fast-evolving genes, many of which occupy peripheral rather than hub positions in protein-interaction networks. The most defensible conclusion is that these genes and their network neighbors mark routes of brain-related genetic variation between humans and macaques and among human populations. Establishing whether individual genes are products of positive selection, relaxed purifying selection or another evolutionary process will require population-stratified analyses, matched adult primate tissues and direct functional experiments.

Subject of Research: Evolutionary analysis of human and macaque brain-expressed genes using protein-interaction networks and single-cell transcriptomics

Article Title: Network-based analysis reveals genetic diversity of human-brain expressed genes with high evolutionary rates

Article References: Ma, Y., Yuan, J., & Wang, J. (2026). Network-based analysis reveals genetic diversity of human-brain expressed genes with high evolutionary rates. Frontiers in Neuroscience, 20, Article 1866277. https://doi.org/10.3389/fnins.2026.1866277

Image Credits: AI Generated

DOI: 10.3389/fnins.2026.1866277

Keywords: human brain evolution, macaque comparison, fast-evolving genes, protein interaction networks, single-cell transcriptomics, dN/dS, pN/pS, cell-type specificity, genetic variation, neurological disease pathways, Network-based, analysis