A comparative-genomic analysis of ten genes implicated in language impairment (LI) and developmental dyslexia (DD) — Mozzi and colleagues, 2016 — identified positively selected amino-acid substitutions in coding sequences, including human-lineage substitutions in KIAA0319, and high-frequency modern-human non-coding variants that rose after the split from archaic hominins at CNTNAP2, FOXP2, ROBO1 and ROBO2. Whether these selected changes alter protein function, gene regulation, neural development, or behavior, and which phenotypes (if any) natural selection actually acted on at these loci, remains experimentally unresolved. The problem asks for functional dissection of the identified selected variants and identification of the causal phenotypes of the detected selective sweeps, with pleiotropy (selection on traits unrelated to LI or DD) and demographic artefact as the competing explanations against which any proposed functional effect or phenotypic target must be tested.

Background

Human speech, language and reading are heritable capacities whose disruption gives rise to common neurodevelopmental disorders: language impairment (LI, also called specific language impairment or developmental language disorder) and developmental dyslexia (DD), an unexpected difficulty in learning to read despite normal instruction and opportunity. Family and association studies have implicated a set of candidate genes, including FOXP2 (a transcription factor — a protein that regulates the expression of other genes — whose disruption causes a severe speech and language disorder), CNTNAP2, ROBO1 and ROBO2 (components of axon-guidance and cell-adhesion pathways involved in wiring the developing brain), KIAA0319, DYX1C1, DCDC2, ATP2C2, CMIP and NFXL1. The FOXP2 story set the template for the field: two amino-acid substitutions accumulated along the human lineage since the split from chimpanzees, functional work showed that humanized mice carrying them display altered cortico-basal ganglia circuitry (connections between the cerebral cortex and the basal ganglia, deep brain structures involved in motor-skill and habit learning) and learning, and a regulatory substitution was reported to be nearly fixed in modern humans but absent in Neanderthals and Denisovans — suggesting both coding and regulatory changes contributed to modern human capacities. Mozzi and colleagues (2016) extended evolutionary analysis from FOXP2 to ten LI/DD genes. Their method combined codon-based phylogenetic tests — comparing the rate of non-synonymous to synonymous substitutions (the dN/dS ratio), with branch-site models that allow a proportion of sites to evolve with dN/dS above 1 (indicating diversifying positive selection) on specified lineages — across mammals and birds including vocal-learning species (species that learn their vocalizations by imitation, such as songbirds, parrots, hummingbirds and some bats and cetaceans), plus population-genetic scans for recent selective sweeps (the rapid rise of a beneficial variant to high frequency, which leaves reduced nearby genetic variation) in human populations, and comparison with archaic hominin genomes. They found: diversifying selection affecting a small minority of sites in about half the genes; positively selected sites in the human lineage concentrated in KIAA0319, two of which (N364 and R865) are human-specific — meaning that all other mammals sequenced to date carry the same ancestral residue — and were already present in Neanderthals and Denisovans, suggesting that any phenotypic change they entailed was shared between modern humans and archaic hominins; episodic selection at CNTNAP2 in the bat lineage and at FOXP2 in three non-vocal-learner bird species (most of the selected sites lying within or in the vicinity of the leucine-zipper dimerization motif, a protein segment whose helices zipper together in pairs so that the protein binds DNA as a two-part complex); and, within modern humans, non-coding variants at high frequency at CNTNAP2, FOXP2, ROBO1 (a cluster surrounding the transcription start site of the ROBO1b isoform, which is differentially regulated in fetal brain areas related to hearing and speech) and ROBO2, all post-dating the split from archaic hominins. The central unresolved issue is that a statistical selection signature does not say what the selected change does or what selection was acting on: most of these genes are pleiotropic (they act in many tissues and processes), so the detected sweeps may have targeted traits unrelated to language or reading — for instance, one CNTNAP2 signal is in linkage disequilibrium (the statistical co-inheritance of nearby DNA variants on the same chromosome) with a protective allele for schizophrenia and bipolar disorder. In the authors' closing formulation, experimental analyses will be necessary to address the functional role of the selected changes they report and the phenotype they modulate, and the lack of suitable experimental models for the study of human-specific traits will make this task difficult to accomplish. The same study also noted that most of the selected alleles are rare or absent in populations of non-African ancestry, the ones most often analyzed in genetic studies, so that association analysis of these variants would require very large subject samples and/or cohorts of African or mixed African ancestry. Reviews of language genetics through 2025 continue to treat the mapping from evolutionary selection signals to molecular function and phenotype as an open problem.

Problem Statement

Determine the functional consequences of the positively selected variants identified in the language-impairment and dyslexia gene set, and identify the phenotypes that natural selection acted on at these loci. Concretely, for (i) the amino-acid substitutions flagged by codon-based selection tests — including the human-lineage KIAA0319 substitutions N364 and R865, the selected sites within or in the vicinity of the leucine-zipper motif of avian FOXP2, and the bat-lineage CNTNAP2 sites — establish whether they alter protein function, gene regulation, neural development or behavior; and (ii) for the high-frequency modern-human non-coding variants and haplotypes at CNTNAP2, ROBO1 (including the cluster surrounding the ROBO1b transcription start site), ROBO2 and FOXP2, establish whether they change gene expression or regulation in relevant neural contexts, and which phenotypes they modulate. Overarching both parts, decide whether LI and DD themselves, other cognitive traits, or unrelated pleiotropic traits were the direct targets of the detected selective sweeps; the minor allele frequency of most of the selected alleles is very low or zero in several human populations, so association evidence bearing on this decision is difficult to obtain. A finding that particular selected variants are functionally neutral, or that a reported sweep signal reflects demographic history or sample composition rather than selection on a trait, resolves the question for those loci in the negative when it rests on locus-specific experimental or population-genetic evidence, and is part of the same objective: converting statistical selection signatures at these genes into demonstrated functional effects and identified phenotypic targets. The overarching target decision is to be reached through these per-locus characterizations: resolving individual loci, including in the negative, is a valid partial resolution, and the overarching answer is not required to outrun the per-locus evidence.

References

  1. Mozzi A, Forni D, Clerici M, Pozzoli U, Mascheretti S, Guerini FR, Riva S, Bresolin N, Cagliani R, Sironi M. The evolutionary history of genes involved in spoken and written language: beyond FOXP2. Scientific Reports 6:22157 (2016). DOI: 10.1038/srep22157. https://doi.org/10.1038/srep22157
  2. Atkinson EG, Audesse AJ, Palacios JA, Bobo DM, Webb AE, Ramachandran S, Henn BM. No Evidence for Recent Selection at FOXP2 among Diverse Human Populations. Cell 174:1424-1435 (2018). DOI: 10.1016/j.cell.2018.06.048. https://doi.org/10.1016/j.cell.2018.06.048
  3. Bornschein U, Zeberg H, Enard W, Hevers W, Pääbo S. Functional dissection of two amino acid substitutions unique to the human FOXP2 protein. Scientific Reports 13:3747 (2023). DOI: 10.1038/s41598-023-30663-3. https://doi.org/10.1038/s41598-023-30663-3
  4. Doust C, Fontanillas P, Eising E, Gordon SD, Wang Z, Alagöz G, Molz B, Pourcain BS, Francks C, Marioni RE, Zhao J, Paracchini S, Talcott JB, Monaco AP, Stein JF, Gruen JR, Olson RK, Willcutt EG, DeFries JC, Pennington BF, Smith SD, Wright MJ, Martin NG, Auton A, Bates TC, Fisher SE, Luciano M. Discovery of 42 genome-wide significant loci associated with dyslexia. Nature Genetics 54:1621-1629 (2022). DOI: 10.1038/s41588-022-01192-y. https://doi.org/10.1038/s41588-022-01192-y
  5. DeSalle R, Lepski G, Arévalo A, Bobo D, LaVergne N, Miyagawa S, Tattersall I. Natural selection and language genes in humans. Scientific Reports 16:9382 (2026). DOI: 10.1038/s41598-026-39032-2. https://doi.org/10.1038/s41598-026-39032-2
  6. Kuhlwilm M. The evolution of FOXP2 in the light of admixture. Current Opinion in Behavioral Sciences 21:120-126 (2018). DOI: 10.1016/j.cobeha.2018.04.006. https://doi.org/10.1016/j.cobeha.2018.04.006
  7. St George-Hyslop F, Kivisild T, Livesey FJ. The role of contactin-associated protein-like 2 in neurodevelopmental disease and human cerebral cortex evolution. Frontiers in Molecular Neuroscience 15:1017144 (2022). DOI: 10.3389/fnmol.2022.1017144. https://doi.org/10.3389/fnmol.2022.1017144
  8. Lei H, Yan Z, Sun X, Zhang Y, Wang J, Ma C, Xu Q, Wang R, Jarvis ED, Sun Z. Axon guidance pathways served as common targets for human speech/language evolution and related disorders. Brain and Language 174:1-8 (2017). DOI: 10.1016/j.bandl.2017.06.007. https://doi.org/10.1016/j.bandl.2017.06.007
  9. Nishiyama KV, Satta Y, Gojobori J. Do Genes Associated with Dyslexia of Chinese Characters Evolve Neutrally? Genes 11:658 (2020). DOI: 10.3390/genes11060658. https://doi.org/10.3390/genes11060658
  10. Li X, Zhu K, Zhen Y. A versatile pipeline to identify convergently lost ancestral conserved fragments associated with convergent evolution of vocal learning. Briefings in Bioinformatics 26:bbae614 (2024). DOI: 10.1093/bib/bbae614. https://doi.org/10.1093/bib/bbae614
  11. Deriziotis P, Fisher SE. Speech and Language: Translating the Genome. Trends in Genetics 33:642-656 (2017). DOI: 10.1016/j.tig.2017.07.002. https://doi.org/10.1016/j.tig.2017.07.002
  12. Dobrynin P, Zeng Y, Norkina M, Fedorova A, Zhuk A, Grigorenko EL. Four Decades of Inquiry Into the Genetic Bases of Specific Reading Disability. Journal of Speech, Language, and Hearing Research 68:5158-5173 (2025). DOI: 10.1044/2025_JSLHR-25-00050. https://doi.org/10.1044/2025_jslhr-25-00050