Identification of brain cell types underlying genetic association with word reading and correlated traits

Reading disability (RD) is a common, polygenic neurocognitive disorder characterized by difficulties in word reading, with clinical and genetic overlap with other neurodevelopmental disorders. While GWAS have begun to identify risk loci for reading and related skills, most associated variants are non-coding and likely act through tissue- and cell-type-specific regulatory mechanisms, complicating interpretation. Neuroimaging implicates multiple cortical regions in reading, and hypotheses such as disrupted neuronal migration and excitatory-inhibitory (E/I) imbalance suggest neuronal cell types may be central to pathology. The research question is which specific brain cell types underlie the genetic associations with word reading and with genetically correlated traits (ADHD, educational attainment, cognitive ability). The purpose is to leverage LDSC with human adult and fetal brain gene expression to identify cell types enriched for GWAS heritability, thereby informing mechanisms and prioritizing targets for functional follow-up. The study is important because identifying relevant cell types bridges GWAS signals to neurobiology and may clarify the cellular basis of reading and correlated cognitive traits.

Prior GWAS and candidate studies have identified suggestive and significant associations for reading and component skills (e.g., RPL7P34, MIR924HG, DOCK7) and numerous loci for self-reported dyslexia. Yet functional mechanisms remain unclear due to non-coding localization of variants and complex enhancer-promoter relationships. The disrupted neuronal migration hypothesis is supported by post-mortem findings (polymicrogyria, heterotopias, dyslamination) and some genetic data. Magnetic resonance spectroscopy studies implicate E/I imbalance (elevated cortical glutamate) in RD and in ASD and ADHD, underpinning the neural noise hypothesis. Polygenic overlap and genetic correlations link reading-related traits with ADHD, educational attainment, and cognitive ability. Previous LDSC/MAGMA studies using adult and limited fetal RNA-seq implicated adult cortical excitatory and inhibitory neurons, and for education/intelligence also astrocytes and oligodendrocytes; fetal midbrain/PFC GABAergic, fetal PFC glutamatergic neurons, and fetal quiescent neural stem cells have been reported for education/intelligence. However, cell types specifically underlying word reading and broader coverage of fetal cortical cell classes remained underexplored.

Study design: Partitioned heritability analysis using LDSC to test enrichment of GWAS SNP-heritability within sets of genes highly expressed in specific brain cell types from adult and fetal datasets. GWAS datasets: a meta-analysis GWAS of word reading (n=5054; Toronto family-based sample with reading difficulties and siblings; PNC population-based cohort), and large external GWAS for ADHD, educational attainment, and cognitive ability (European ancestry). Genotyping used Michigan Imputation Server with HRC r1.1; QC included removing SNPs with imputation r2 <0.30–0.70 (trait-specific thresholds), MAF <1–5%, Hardy–Weinberg disequilibrium, and MHC region (chr6:26–33Mb). Heritability estimation: LDSC ‘ldsc.py’ to compute SNP-heritability (h2) and z-scores for each GWAS. Gene expression datasets: Adult snRNA-seq from Allen Brain Bank (ABB; ~49,000 NeuN-sorted nuclei from multiple cortical areas, ages 16–68), fetal scRNA-seq from Kriegstein lab (~4,000 cells from V1, PFC, MGE, 5–37 post-conception weeks), and fetal bulk RNA-seq from Shen lab (FAC-sorted excitatory neurons, interneurons, radial glia, intermediate progenitors from 15–22 gestational weeks). Data processing and QC: For ABB, used scrattch.io for processing, pseudo-bulking by cluster, removing donors/outliers; excluded cell types with <100 nuclei; retained non-neural endothelial/pericytes; Seurat used for QC, excluding cells with <1000 genes detected or >10% mitochondrial/ribosomal gene expression. Kriegstein scRNA-seq included 12 classes (≥100 cells), excluding low-count classes; similar QC criteria applied. Shen bulk RNA-seq: removed lowly expressed genes (<10 counts in <3 samples), checked for mitochondrial/ribosomal overrepresentation. Building LDSC annotations: For each dataset, computed mean expression (primary analysis) and specificity (secondary) per gene per cell type using EWCE for single-cell/nucleus data (drop uninformative genes via ANOVA; generate cell-type mean/specific expression). For bulk data (Shen), computed in R without EWCE. Selected the top decile (top 10%) of genes by mean expression (or by specificity in secondary analyses) to define cell-type gene sets. Obtained gene coordinates (hg19) ±100 kb via biomaRt; created LDSC annotation files with make_annot.py, ensuring consistency with 1000 Genomes Phase 3 reference used for LD scores. LDSC partitioned heritability: Computed LD scores (1 cM window) using HapMap3 SNPs and baseline annotations as controls; used weights to address heteroscedasticity; accounted for annotation overlap. For computational efficiency, prioritized regression coefficients. Multiple testing correction: Applied FDR (q<0.05) per RNA-seq dataset and per analysis type (mean vs specificity). Outcome measures: Significant enrichment indicated cell types whose top-expressed genes contribute disproportionately to GWAS heritability.

- Heritability: Estimated SNP-heritability (approximate) and z-scores: word reading h2 ~0.25 (z ≈ 3), ADHD h2 ~0.24 (z ≈ 12), educational attainment h2 ~0.11 (z ≈ 41), cognitive ability h2 ~0.18 (z ≈ 24). - Adult major classes (ABB, mean expression): Word reading showed significant enrichment for excitatory and inhibitory neurons (FDR<0.05). Educational attainment and cognitive ability were enriched for excitatory and inhibitory neurons, astrocytes, and oligodendrocytes. ADHD showed no significant enrichment at the major class level. - Adult subclasses (ABB, mean expression): Word reading was enriched in excitatory subclasses expressing FEZF2 (L6b FEZF2; L5/6 NP FEZF2; L6 CT FEZF2) and IT neurons marked by LINC00507/THEMIS/RORB (including L5/6 IT Car3 THEMIS), and in inhibitory subclasses VIP, SST, and PVALB (all FDR<0.05). ADHD showed enrichment in L4 IT RORB excitatory neurons (FDR<0.05). Educational attainment and cognitive ability showed broad enrichment across multiple excitatory subclasses (L6b FEZF2, L5/6 NP FEZF2, IT LINC00507/THEMIS/RORB, L6 CT FEZF2, L5 ET FEZF2, L5/6 IT Car3 THEMIS, L4 IT RORB) and inhibitory subclasses (PAX6, LAMP5, VIP, SST, PVALB), as well as astrocytes and oligodendrocytes (FDR<0.05). - Adult specificity analyses: Enrichment was more limited; excitatory neurons enriched for word reading, educational attainment, and cognitive ability; inhibitory neurons met FDR for cognitive ability. - Fetal cell types (Kriegstein, mean expression): Educational attainment showed significant enrichment across all neural subclasses, with the strongest effect for MGE-derived cortical plate inhibitory neurons; cognitive ability showed enrichment for excitatory and inhibitory neurons and intermediate progenitor cells (IPCs), but not for radial glia or MGE progenitors. Word reading and ADHD showed no fetal enrichments in this dataset. - Fetal cell types (Shen, mean expression): Educational attainment and cognitive ability showed significant enrichment in excitatory and inhibitory neurons, IPCs, and radial glia; word reading and ADHD did not reach significance. Specificity analyses in fetal datasets yielded fewer significant categories (notably excitatory neurons for education and cognitive ability). - Overall interpretation: Word reading heritability is enriched in adult cortical excitatory and inhibitory neuronal subclasses, aligning with hypotheses of E/I imbalance in RD; ADHD shows enrichment in a specific adult L4 IT excitatory subclass; education and cognitive ability display broad enrichment across adult neuronal and glial types and multiple fetal cortical neuronal and progenitor populations.

The study addresses which brain cell types mediate genetic risk for word reading and genetically correlated traits by integrating GWAS with adult and fetal brain expression profiles via LDSC. The enrichment of word reading heritability in adult cortical excitatory and inhibitory neuronal subclasses (including FEZF2-expressing deep-layer projection neurons and IT neuron populations, and inhibitory VIP, SST, PVALB interneurons) links genetic risk to specific neuronal phenotypes within cortical reading networks. This aligns with prior neurometabolite evidence of E/I imbalance in RD and supports neuronal circuitry-focused hypotheses such as disrupted neuronal migration and neural noise. ADHD’s enrichment in L4 IT RORB excitatory neurons suggests a layer- and subtype-specific excitatory contribution. For educational attainment and cognitive ability, broad enrichment across adult excitatory and inhibitory subclasses and glial cells replicates prior findings and extends them by identifying multiple fetal cortical populations (excitatory and inhibitory neurons, IPCs, radial glia), highlighting developmental windows and progenitor populations relevant to higher-order cognition. Together, these results help bridge GWAS signals to cellular substrates, prioritize cell types for functional studies (e.g., iPSC-derived neuronal models), and suggest that both mature neuronal circuits and developmental progenitor processes contribute to cognitive phenotypes.

This work identifies specific adult cortical excitatory and inhibitory neuronal subclasses underlying genetic associations with word reading and indicates a specific excitatory subclass (L4 IT RORB) for ADHD. It confirms and extends prior findings for educational attainment and cognitive ability, revealing broad adult neuronal and glial contributions and novel fetal cortical enrichments including excitatory/inhibitory neurons, intermediate progenitors, and radial glia. These insights refine the neurobiological framework for reading and related traits, supporting roles for E/I balance and cortical circuit development. Future research should leverage larger GWAS for reading/ADHD, incorporate additional brain regions and developmental stages, and perform functional assays in prioritized cell types to elucidate mechanisms linking regulatory variants to gene expression and cellular phenotypes.

Key limitations include limited GWAS power for word reading (n≈5,054) and ADHD (≈50,000) compared with very large education and cognitive GWAS, likely reducing sensitivity to detect enriched cell types. Some cell classes/subclasses were underrepresented or excluded due to low cell/nucleus counts in expression datasets, reducing coverage and power. Analyses focused on cortical tissues; potential contributions from other brain regions (e.g., subcortical, cerebellar) were not assessed. Fetal datasets pooled samples across developmental stages, potentially obscuring stage-specific effects. Differences between snRNA-seq, scRNA-seq, and bulk RNA-seq may influence gene detection and cell-type representation. Results for word reading should be interpreted cautiously given lower h2 z-score and sample size.