Structural connectivity associations with dyslexia-linked genes identifies the left insula as the key phonological hub

Structural connectivity associations with dyslexia-linked genes identifies the left insula as the key phonological hub | Research Square Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Help Center Sign In Submit a Preprint Cite Share Download PDF Article Structural connectivity associations with dyslexia-linked genes identifies the left insula as the key phonological hub Jingjing Zhao, Yueye Zhao, Colin Buchanan, Hayley Mountford, Joanna Moodie, and 7 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8187892/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Dyslexia is characterized by difficulties acquiring fluent reading skills, it is heritable, possibly involving neural disconnection. Results from well-powered genome-wide association studies can be harnessed to uncover associations between dyslexia genetic liability with the macroscale white matter connectome to pinpoint neurobiology. Using multimodal magnetic resonance imaging data from 40,372 adults (44-85 years) in the UK Biobank, we examined associations between dyslexia polygenic indices (PGI) and regional grey matter structure and hub-based white matter connectivity across 85 regions (68 cortical; 17 subcortical). The left insula showed negative PGI associations across grey and white matter measures, with replication in 1,068 adults (26-84 years) from the Generation Scotland: Scottish Family Health Study. Connectivity of the left insula mediated the association between dyslexia PGI and phonological short-term memory/reading skill in both cohorts. Polygenic dyslexia predisposition is thus reflected in both structural and connectomic properties of the left insula and consistent with phonological deficit theory. Biological sciences/Genetics Biological sciences/Neuroscience dyslexia polygenic indices connectivity hub phonology reading Figures Figure 1 Figure 2 Figure 3 Figure 4 Introduction Developmental dyslexia is a specific learning difficulty characterized by significant difficulties in reading and writing development despite access to conventional education 1 , and affects 1.3-17.2% of school-aged children depending on definition 2,3 . It is strongly heritable with a neurobiological basis that affects cognitive processes underlying reading acquisition. Twin and family studies estimate its heritability at 40-60%, and recent genome-wide association studies (GWAS) have identified multiple loci linked to dyslexia and dyslexia-related traits although SNP-based heritability is ~.20 4,5,6 . In terms of the neuro-cognitive basis of developmental dyslexia, cognitive studies have long observed phonological deficit as a core cognitive feature in both dyslexic adults 7 and children 8 from alphabetic languages, and more recently from non-alphabetic languages such as Chinese 9 . Yet the neural underpinnings of developmental dyslexia as a neural disconnection syndrome have only been considered in the last decade 10,11,12,13 . Neuroimaging evidence suggests that dyslexia may largely stem from difficulties in accessing phonological representations due to disrupted structural and functional connectivity among regions in inferior frontal and temporal lobes involved in phonological processing 10,11,12,13 . The disrupted structural connectivity may be further linked to the severity of phonological deficits in developmental dyslexia 14 . Despite advances in our genetic and neuro-cognitive understanding of developmental dyslexia, the neurobiological mechanisms by which genetic liability translates into clinical and behavioral phenotype remain poorly defined 15 . A recent imaging genetic study using data-driven voxel-wise and fixel-wise analyses revealed widespread associations between a polygenic index (PGI) for dyslexia and both grey matter volume and white matter microstructure in UK Biobank (UKB; N up to ~35,000 adults) 16 . But the focus on localized microstructural properties which separated grey and white matter modules limited investigation of integrative large-scale structural connectivity and network level organizational patterns. Further, there was no attempt to link gene-brain associations with behavior to understand cognitive mechanisms. Here, we combine polygenic indices constructed from weights estimated using separate GWAS studies of dyslexia and dyslexia-related traits 4,5 with multimodal neuroimaging datasets across two independent adult cohorts with cognitive and reading phenotypes to investigate the key neural hubs associated with dyslexia-related polygenic liability. We first parcellated the cortex into 85 macro-regions and tested associations between dyslexia-related PGI and grey matter structure of these regions in a large subsample from the UKB (i.e., our discovery sample). We then examined the associations between dyslexia-related PGI and nodal degree white matter metrics (representing hub-based properties of the structural connectome) of any significantly implicated macro-regions of interest (ROIs) in grey matter structure. This macro-level, hub-based alignment of grey and white matter 16 is preferred to the data-driven voxel-wise and fixel-wise separated regional metrics approach 15 because it reduces measurement noise and multiple testing corrections at the voxel and fixel levels. Moreover, the structural connectome integrates grey matter nodes with white matter connectivity in the subject’s native space, providing a holistic, anatomically-grounded framework that may more optimally represent brain-behavior relationships compared to voxel-, fixel-, or tract-based approaches which focus on more isolated features. Significant PGI-imaging findings (at a false discovery rate corrected threshold) were then validated in an independent cohort of adults from Generation Scotland: Scottish Family Health Study (GS:SFHS). Finally, we examined whether brain structure and connectivity serve as intermediate endophenotypes mediating the effect of dyslexia-related PGI on phonological and reading-related skills, thereby providing a more direct test of the extent to which the identified PGI-brain associations serve as mechanisms of cognitive disfunctions underlying dyslexia. The workflow of our approach is depicted in Figure 1. Results UKB Discovery sample We conducted regression analyses testing the association between all 6 PGIs including both dyslexia and dyslexia-related traits (dyslexia, phonological awareness, word reading, nonword reading, nonword repetition, and spelling) and brain structure with the covariates of age, sex, handedness, head position, scanning site, whole brain volume, and the first 40 genetic principal components (PC1-PC40). After false discovery rate (FDR) correction 17 , only the PGI for dyslexia was found to significantly predict all three grey matter structure measures (volume, surface area, and cortical thickness) within the same brain region. All significant associations were negative. Effect sizes of volume range from -0.0218 to -0.01. Effect sizes of surface area range from -0.0151 to -0.0135. Effect sizes of thickness range from -0.0249 to -0.0129. Full statistical results for all PGI are provided in Supplementary Table 1. Figure 2 presents significance levels for associations between dyslexia PGI and grey matter volume for the 85 brain regions (68 cortical and 17 subcortical), and surface area and cortical thickness for the 68 cortical regions across different PGI thresholds after multiple testing corrections. Results showed that the PGI of dyslexia was negatively significantly associated with grey matter volume for 5 temporal, 3 parietal, 4 occipital, and 3 frontal regions, as well as 4 subcortical areas, and brainstem. They are the left insula, bilateral superior temporal gyrus, left fusiform gyrus, bilateral lateral occipital gyrus, bilateral precentral gyrus, right transverse temporal gyrus, left rostral anterior cingulate gyrus, right postcentral gyrus, right inferior parietal lobule, left caudal middle frontal gyrus, right medial orbitofrontal cortex, right pericalcarine cortex, left thalamus, bilateral ventral diencephalon, and brainstem (Supplementary Figure 1, top panel). Surface area was negatively associated with the dyslexia-related PGI in 1 temporal, 4 occipital, and 2 frontal regions. Specifically, they are the left insula, left superior frontal gyrus, left lingual gyrus, bilateral pericalcarine cortex, left rostral anterior cingulate cortex, and right cuneus (Supplementary Figure 1, middle panel). Cortical thickness was significantly negatively associated with dyslexia-related PGI in 9 temporal, 4 parietal, 4 frontal, and 3 occipital regions. They are bilateral insula, fusiform gyrus, superior temporal gyrus, right inferior and middle temporal gyrus, left transverse temporal gyrus, right lateral orbitofrontal cortex, left inferior parietal lobule and pars opercularis, bilateral lateral occipital, postcentral gyrus, precentral, and supramarginal gyri, right cuneus, and left paracentral lobule (Supplementary Figure 1, bottom panel). Notably, the left insula was consistently associated with PGI of dyslexia across all three grey matter measures (volume, surface area, and cortical thickness) after correction. Specifically, PGI at a threshold of P£1x10 -5 predicted both its volume ( Standardized β = -0.016, p_fdr = 0.003) and surface area ( Standardized β = -0.014, p_fdr = 0.014), while cortical thickness was associated with PGI of dyslexia at a threshold of P£1x10 -2 ( Standardized β = -0.017, p_fdr = 0.023), as shown in Figure 3 (left upper panel). We next examined whether PGI of dyslexia was associated with nodal degree (a weighted network metric) of white matter connectivity of the implicated brain regions showing significant effects in grey matter measures in UKB dataset. Using nodal degree derived from structure-weighted connectomes as the dependent variable, we assessed associations across both conventional diffusion tensor imaging (DTI) metrics including fractional anisotropy (FA), streamline counts (SC), and mean diffusivity (MD) alongside neurite orientation dispersion and density imaging (NODDI)-derived measures including intra-cellular volume fraction (ICVF), isotropic volume fraction (ISOVF), and orientation dispersion (OD). PGI of dyslexia significantly predicted nodal connectivity across widespread brain regions in various white matter connectivity measures. Full results are reported in Supplementary Table 2. However, only the left insula which was consistently associated with PGI of dyslexia across all three grey matter measures also had significant associations between PGI of dyslexia and white matter connectivity measures of FA ( Standardized β = -0.0173, p = 0.002), SC ( Standardized β = -0.01, p = 0.04), MD ( Standardized β = -0.0174, p = 0.001), OD ( Standardized β = 0.02, p < 0.001), and ISOVF ( Standardized β = 0.011, p = 0.04) (Figure 3). After controlling for multiple comparisons using FDR correction, only the associations with MD ( Standardized β = -0.0181, p_fdr = 0.029) and OD ( Standardized β = 0.0199, p_fdr = 0.003) remained significant. Although ICVF was not associated with PGI of dyslexia, it was shown to be associated with PGI of phonological awareness ( Standardized β = -0.02, p < 0.001), an important cognitive phenotype of dyslexia (Figure 3, lower right panel). Supplementary Figure 2 illustrates the white matter network connected to the left insula, based on the inter-regional pathways identified by probabilistic tractography and consistency-thresholding. This network encompasses regions associated with phonological skills and reading abilities 18 including the left transverse temporal gyrus, left superior temporal sulcus, middle temporal gyrus, inferior temporal gyrus, fusiform gyrus, supramarginal gyrus, inferior parietal lobule, temporal pole, pars opercularis, pars orbitalis, pars triangularis, lateral and medial orbitofrontal cortices, rostral middle frontal gyrus, superior frontal gyrus, lateral occipital cortex, postcentral gyrus, precentral gyrus, entorhinal cortex , as well as the bilateral superior parietal lobule and precuneus. As a previous study observed that the association between the dyslexia PGI and the left anterior insula was influenced by education and intelligence 15 , we further conducted regression analyses for the measures of the left insula by regressing out socioeconomic status including both educational attainment and income, and a general measure of intelligence (i.e., verbal numerical reasoning). We found that the associations between the dyslexia PGI and the left insula measures remained significant after FDR correction (Supplementary Table 3). To further localize the effect of PGI of dyslexia on the left insula at the vertex level and compare our grey matter results in the left insula with a previous voxel-wise analysis of volumetric measures in UKB samples which only observed the effect of the left insula at the anterior portion 15 , we conducted vertex-wise surface-based regression analyses within the insula. Controlling for all covariates, we mapped the spatial distribution of associations between the PGI of dyslexia and morphometric measures (volume and surface area) of the bilateral insula. For the left insula, the mean beta values were -0.0084 for volume and -0.0070 for surface area, but no associations survived FDR correction. The associations between the PGI of dyslexia and the volume and surface area of the left insula were slightly stronger in the anterior portion (mean beta: -0.0093 for volume, -0.0078 for area) than in the posterior portion (mean beta: -0.0074 for volume, -0.0063 for area), but Williams tests 19 , which account for the dependency between overlapping correlations, indicated that these differences were not significant (see Supplementary Figure 3). Cross-Dataset validation from a subsample of GS:SFHS In the validation analysis of grey matter measures in the GS:SFHS dataset, the association between the left insula and PGI of dyslexia was consistently replicated across two structural metrics: volume ( Standardized β = -0.064, p = 0.01) and area ( Standardized β = -0.061, p = 0.016), as shown in Figure 3 (left lower panel). For all other regions associated with the PGI of dyslexia, significance was observed in only one structural metric (Supplementary Table 4). In the white matter validation analysis of the GS:SFHS dataset, the left insula was also replicated in the association between the PGI of dyslexia and SC ( Standardized β = -0.084, p = 0.012) (Supplementary Table 5). While some other regions also showed significant effects, as shown in Supplementary Tables 3 and 4, the directions of these effects were inconsistent with those observed in the UKB dataset and were therefore not considered successful replications. Additionally, we conducted regression analyses including educational attainment as an additional covariate. The results showed that the association between the PGI of dyslexia and volume and surface area of the left insula remained significant, while SC became non-significant and MD emerged as significant (Supplementary Table 6). In sum, these results indicate that the PGI of dyslexia consistently predicts both grey matter structure and white matter connectivity of the left insula in both the exploratory UKB dataset and the validatory GS:SFHS dataset. Mediation analysis of the left insula Given the insula’s role in verbal processing including phonological discrimination 20 we further examined the mediating role of grey matter structure and white matter connectivity of the left insula between dyslexia PGI and available phonological and reading behavior phenotypes in both cohorts, we conducted two mediation analyses. In the UKB dataset, we tested the mediating role of the left insula in the association between dyslexia PGI and numeric short-term memory measured by a visual digit span task, with the same covariates as in regression analyses. All three grey matter measures significantly mediated the association between dyslexia PGI and digit span (volume: ACME = -0.00359, 95% CI = [-0.00522, -0.00179], p < 0.001; area: ACME = -0.00271, 95% CI = [-0.00406, -0.00141], p < 0.001; thickness: ACME = -0.00116, 95% CI = [-0.00198, -0.00037], p = 0.004) (Figure 4A, top panel). For white matter measures, nodal degree values of FA and SC also significantly mediated the association between dyslexia PGI and digit span (FA: ACME = -0.00086, 95% CI = [-0.0015, -0.00031], p < 0.001; SC: ACME = -0.00124, 95% CI = [-0.00228, -0.00029], p = 0.014) (Figure 4A, bottom panel). In the GS:SFHS dataset, we tested the mediating role of the left insula in the association between dyslexia PGI and tests of reading ability measured by regular word reading, irregular reading, and nonword reading, as well as digit span, nonword repetition and spelling. The mediation effect of the left insula was not replicated in any test after covariates (i.e., age, sex, scanning site, intracranial volume, and the first ten genetic principal components) were controlled. However, without including covariates, for regular word reading, the SC of the left insula showed a significant mediating effect (ACME = -0.022952, 95% CI = [-0.05294, -0.00134], p = 0.032) (figure 4B, top panel). The surface area of the left insula significantly mediated the association between dyslexia PGI and irregular word reading (ACME = -0.03448, 95% CI = [-0.08299, -0.00264], p = 0.024) (figure 4B, top panel). Both the surface area and volume of grey matter showed significant mediating effects for nonword reading (volume: ACME = -0.039196, 95% CI = [-0.0804, -0.0075], p = 0.014; area: ACME = -0.033106, 95% CI = [-0.07294, -0.03667], p = 0.018) (figure 4B, lower panel). No significant mediation effects were found for digit span, spelling, or nonword repetition, even without covariate adjustment. Full results of mediation analysis are shown in Supplementary Tables 7 and 8. Discussion The current study examined the relationship between dyslexia and dyslexia-related PGI and ROI-based grey matter structure and white matter connectivity across two independent datasets (UKB and GS:SFHS). Several phonological and reading-related brain regions were identified; however, only the left insula consistently showed a significant association with dyslexia PGI across multiple structural measures in both samples with the same direction of effect: the higher dyslexia PGI, the lower grey matter volume and surface area of the left insula and the lower white matter connectivity of the subnetwork centered around the left insula. More importantly, when the structure and connectivity of the left insula were considered as neurogenetic endophenotypes of dyslexia in mediation analyses, results showed that they significantly mediated the effects of dyslexia-related genetic liability on phonological memory and reading ability. Our results indicated that the grey matter structure and white matter connectivity of the left insula were significantly associated with genetic liability for dyslexia. This suggests that the left insula may act as a hub within a broader network, where dyslexia-related common genetic variants contribute to disrupted structural connectivity. More interestingly, the grey matter structure and white matter connectivity of the left insula played a mediating role in linking the PGI of dyslexia with numeric short- term memory and reading ability (but not spelling and nonword repetition). The latter findings were in a relatively small sample (~ 200 participants) so the inconsistency of mediation across dyslexia-related phenotypes might be due to increased sampling error, especially for spelling and nonword repetition tests, where items were administered verbally over the phone. These findings further elucidate the pathways from gene to brain structural connectivity and then to cognitive or behavioral phenotypes, providing more comprehensive evidence for the phonological deficit theory of dyslexia 6 , 7 , 8 , 9 , 21 . Indeed, our data showed that the white matter network with the hub at the left insula included important phonological processing regions key for dyslexia such as the left superior temporal gyrus 11 . The mediation role of structural connectivity of the left insula between dyslexia genetic predisposition and short-term memory/reading abilities therefore might be consistent with a notable finding which suggested that dyslexia stems from difficulties in accessing phonological representations due to disrupted connectivity among regions involved in phonological processing 13 . Our findings support the left insula as the key phonological hub influenced by genes associated with dyslexia. Previous research has observed that learning to read improves the structural connectivity 22 . Therefore, there could be an alternative pathway that phonological and reading abilities mediate the link between dyslexia genetic predisposition and structural connectivity of the left insula. The left insula has been found to be involved in awareness of body movement 23 . Children with dyslexia may struggle to perceive the position and movement of their articulatory organs during speech, making it more difficult to associate these articulatory gestures with the corresponding speech sounds. This, in turn, may weaken grapheme-phoneme mapping 24 . Therefore, our findings suggest a potential pathway through which dyslexia-related genetic predisposition may influence the left insula’s role in the internal monitoring of speech production, potentially disrupting grapheme-phoneme mapping involving more distributed phonological networks and contributing to the phonological deficit characteristic of dyslexia 25 . Beyond its role in phonological processing, the left insula is also a core node within the salience network, which is responsible for detecting and evaluating the most behaviorally relevant stimuli from internal and external environments 26 . Specifically, the anterior part of the insula can identify salient stimuli from a large volume of incoming information and recruit attention, working memory, and higher-order cognitive regions for further processing 27 . It has also been implicated in the initiation and articulation of language via organization and planning processes 23 . Structural and connectivity alterations of the left insula associated with dyslexia-related genes may therefore impair individual’s ability to sustain attention and regulate cognitive status during reading. Moreover, such disruptions could hinder adaptive strategy adjustment in response to reading difficulties or failures, resulting in poor reading ability/language articulation. Our results of the left insula contrasted with the findings from a prior study 15 that showed a non-significant association between the dyslexia PGI and the left anterior insula after controlling education and intelligence. We found that the dyslexia PGI associated with the entire left insula as a macro-region and no difference for the association strength was observed between its anterior and posterior portions whereas the previous study only observed significant association between the dyslexia PGI and the left anterior insula. These differences likely stem from the increased power of our approach. Rather than focusing on vertex- and fixel-wise structure, our study tested macro-regional white matter connectivity of reading-related region associations with dyslexia PGI. This approach reduced the measurement and registration noise, as well as the multiple corrections for vertex- and fixel-wise comparisons. By treating the left insula as an entire macro-region, we observed more powerful results in the left insula which were less influenced by education and intelligence, and our results suggested little difference between the anterior portion and the posterior portion. Importantly, using DTI and NODDI, we captured inter-regional connection strength and identified network-level disruptions in white matter microstructural connectivity centered on the left insula as a hub, which aligned with previous research that suggested white matter connectivity was more sensitive to dyslexia predisposition, phonological processing, and reading ability 12 , 13 , 14 . Another possible source of differences between studies is in covariates (e.g., number of ancestry PCs) included in the regression analysis. Given our robust results in the left insula, which had replication level support, we believe that our findings may offer unique implications for future research. The left insula, identified as a central hub in the reading network and consistently associated with dyslexia-related genetic variation and reading behaviors, emerges as a promising target for complementing behavioral, cognitive and genetic markers to support more comprehensive and thus reliable profiling of children with suspected dyslexia. The left insula might also be a promising target for neural intervention where behavioral interventions have failed. While non-invasive transcranial current stimulation has previously been applied to reading-related regions such as the left temporal cortex with measurable behavioral improvements 28 , 29 , the insula’s deep anatomical location poses a challenge for direct stimulation 30 . However, recent studies employing repetitive transcranial magnetic stimulation have demonstrated the feasibility of modulating insular activity 31 , suggesting that targeted stimulation of the left insula holds potential for enhancing reading-related outcomes. The effect sizes observed in the UK Biobank were generally small, which is expected given that our previous dyslexia GWAS study showed that dyslexia PGIs do not explain more than 6% of variance in quantitative measures of dyslexia itself 4 . Our replication cohort (GS:SFHS) was much smaller than the UK Biobank and was thus underpowered to detect the very small effects. Nonetheless, many non-significant findings showed consistent direction of effects and similar effect sizes (e.g., the negative associations between brain structure/connectivity and dyslexia PGI), supporting the robustness of the associations despite low statistical power (supplementary Tables 4 and 5; the correlation between effect sizes for ROIs between UKB and GS:SFHS was significant (p < .001; r = .23). Interestingly, some associations exhibited larger effect sizes in the replication sample, which may reflect sampling variability, small sample size making it susceptible to effect size inflation, or potentially stronger PGI-phenotype relationships within GS:SFHS. Despite valuable findings, the current study includes several limitations. Firstly, the datasets we used were from general population rather than dyslexic participants. As a result, we were unable to examine whether the associations observed in the general population also apply specifically to individuals with dyslexia. Additionally, due to the limited set of cognitive measures available, we only tested one phonological dimension: verbal/numeric short-term memory. Other core components of phonological processing, such as phonological awareness and rapid automatized naming, both of which have been identified as key dimensions of phonological deficits in dyslexia 7 , were not available. Compared to the dyslexia PGI, the phoneme awareness PGI showed weaker associations with grey matter measures, with no significant results surviving correction. However, since the PGI of phonological awareness was derived from a low powered GWAS 6 , this is not strong evidence against our interpretation. Thirdly, our study was conducted on adult participants rather than children, which limits the ability to fully capture the early neurodevelopmental mechanisms underlying dyslexia. Future research could utilize longitudinal data from children to examine the dynamic developmental trajectory of the genetic and neural mechanisms underlying dyslexia. Additionally, network properties are known to depend on MRI acquisition, tractography algorithm, cortical parcellation, and thresholding strategy 32 , therefore results may vary with different methodological choices. It might also be valuable to examine functional connectivity to validate the current structural connectivity findings. Conclusion The study provides the first empirical evidence for associations between dyslexia-related genetic liability and brain structure and white matter connectivity of the left insula, which is in turn related to phonological memory and reading ability in adults. Our findings support the phonological deficit hypothesis of dyslexia identifying a specific brain mechanism that might mediate genetic liability for the phonological deficit. Materials and Methods Study population The UK Biobank is a large-scale epidemiological cohort study that recruited approximately 500,000 adults, aged 40–69 years across Great Britain between 2006 and 2010 33 . Data collection has included extensive demographic, medical, and cognitive data. At the time of image processing, a subset of ~ 40,000 participants had undergone MRI brain scanning at four UK imaging centers 34 . Scans were acquired for the first time when participants were aged 44–85 years, with scans taking place 9.60 ± 1.10 years after baseline assessment. The current study includes participants who had at least one brain MRI (n = 40,380). Among the participants, 40,372 have structural MRI (sMRI) data, and 37,284 have diffusion MRI (dMRI) data. Ethical approval was granted by the North West Multi-centre Research Ethics Committee (11/NW/0382). The validation dataset was derived from a subset of the GS:SFHS, including participants who underwent dMRI and sMRI collection. The GS:SFHS is a large-scale cohort study involving more than 24,000 individuals who underwent comprehensive phenotyping between 2006 and 2011. A subset of these participants was recontacted between 2015 and 2019 for additional data collection including neuroimaging, cognitive assessment, and health measures. This specific subset forms the basis of our current study and has been extensively described elsewhere 35 , 36 . The validation dataset included 1,068 participants with complete sMRI FreeSurfer data and 938 participants with complete connectomes after dMRI quality control. Ethical approval was obtained from the NHS Tayside Research Ethics Committee (14/SS/0039). In both cohorts, although participants were generally healthy, no exclusions were applied for neurological conditions. Polygenic Indices (PGI) In the current study, genome-wide PGI were constructed using summary statistics for dyslexia 4 and reading-related cognitive traits (phonological awareness, word reading, nonword reading, nonword repetition, and spelling) 5 from GWAS summary statistics based on samples of European ancestry that did not include UKB or GS:SFHS. Specifically, from a single sample GWAS for self-reported dyslexia diagnosis in 23andMe participants and from a meta-analysis GWAS of European, North American and Australian samples for quantitative traits. Summary statistics without genomic control were used and underwent quality control to exclude rare SNPs (minor allele frequency (MAF) < 0.01) and SNPs with low quality imputation scores (INFO < 0.8). Indels, multiallelic sites, duplicate and ambiguous SNPs were also excluded. Following QC, summary statistics contained > 5 million autosomal SNPs (dyslexia = 5,829,028, WR = 7,532,716, NREAD = 7,480,144, SP = 7,315,150, PA = 7,497,763, and NREP = 7,133,624). UK Biobank target genotypes consisted of directly typed SNPs in unrelated British participants only (> 3rd degree based on kinship file) (N = 408,087), as previously described 37 . GS:SFHS target data was imputed to HRC.r1-1 (N = 20,032) as previously described 38 . Both target populations underwent standard QC using PRSice2 best practice 39 (genotyping rate > 0.99, sample missingness < 0.01, Hardy-Weinberg Equilibrium P > 1x10 − 6 , heterozygosity < 3 standard deviations of the mean, MAF > 0.01 and imputation INFO score > 0.8). PGIs were calculated using autosomal SNPs with PRSice 2.3.5 with default clumping settings applied (clumping window of 250kb, SNPs with r2 > 0.1 with the index SNP excluded, and all SNPs used in the analysis) and using the sum method which directly sums the beta weights. Following clumping, PGIs calculated in UK Biobank contained > 121,000 autosomal SNPs (dyslexia = 121,903, WR = 217,877, NREAD = 217,706, SP = 217,177, PA = 217,701, NREP = 216,360) and GS:SFHS contained > 180,000 autosomal SNPs (dyslexia = 181,946, WR = 320,991, NREAD = 320,666, SP = 6261,807, PA = 320,466, and NREP = 320,126). Informed by the p-value for SNP–trait association supplied by GWAS summary statistics, we calculated several PGI for different significant thresholds (P ≤5x10 − 08 , P ≤1x10 − 05 , P ≤0.001, P ≤0.01, P ≤0.05, P ≤0.1, P ≤0.5, P ≤1), with less stringent thresholds incorporating a greater number of SNPs. After z-standardization (M = 0, SD = 1), a PGI above zero for the reading related cognitive measures reflects a higher genetic predisposition for increased reading-related skill relative to the population mean; whereas a score below zero indicates a lower predisposition. A higher PGI for dyslexia is related to increased likelihood of self-reported dyslexia diagnosis. To control for population stratification, we regressed the PGI on the first 40 genetic principal components and used the residuals in all subsequent analyses. Cognitive and behavioral phenotypes A total of 26,412 participants with genetic and imaging data from the UKB dataset completed a numeric memory test (field 4282) at baseline similar to the digit span task but with a visual rather than auditory presentation of stimuli. The maximum length of a number sequence accurately recalled served as an index of phonological memory. It was used as a cognitive test closely related to reading ability for subsequent mediation analysis and has been shown to significantly correlate ( r ~ .45) with WAIS digit span tasks 40 . A total of 203 participants completed the phonological and reading tasks in the GS:SFHS dataset. The tasks included WAIS-III digit span test, nonword repetition (CNRep), regular word reading, irregular word reading, nonword reading Castles and Coltheart test 2 adults (CC2A) and spelling (Components of Reading Examination spelling test). These were participants who had participated in an earlier brain imaging study 35 in the years 2015 to 2017 and in a more recent reading and language study (Mountford H. S., Campbell A., Marioni R. E., Bates T. C., Whalley H. C., Luciano M., manuscript in preparation) from 2020 to 2023. MRI acquisition The UKB employed identical Siemens Skyra 3T scanners and a standard 32-channel head coil at all four imaging sites (Cheadle, Newcastle upon Tyne, Reading, and Bristol). For sMRI, 3D T1-weighted MPRAGE sequences with a 1×1×1 mm³ resolution and a 208×256×256 matrix field-of-view (FOV) were acquired. The acquisition is performed in the sagittal plane with an in-plane acceleration (iPAT = 2) and a prescan normalization to ensure consistent image quality across participants. This high-contrast acquisition enables precise volumetric analysis and derivation of brain structure and pathology 34 . More information about MRI acquisitions in the UKB are available here https://www.fmrib.ox.ac.uk/ukbiobank/protocol/V4_23092014.pdf . The dMRI in the UK Biobank was performed using a spin-echo echo-planar imaging (SE-EPI) sequence. The imaging parameters include a resolution of 2×2×2 mm³ with a FOV of 104×104×72 matrix, and the scan duration is around 7 minutes. The sequence involves 5 non-diffusion-weighted images volumes (b = 0 s/mm²) and 2 diffusion shells (each comprising 50 diffusion directions at b = 1000 and 2000 s/mm²), resulting in 100 distinct diffusion-encoding directions. The gradient timings are δ = 21.4 ms and Δ = 45.5 ms, with a spoiler b-value of 3.3 s/mm². The sequence employs SE-EPI with ×3 multislice acceleration, fat saturation, in-plane acceleration R = 1, and TE/TR = 92/3600 ms. Diffusion weighting followed a standard Stejskal-Tanner pulse sequence with a shorter echo time (TE = 92 ms), enhancing signal-to-noise ratio. The increased susceptibility to eddy current distortions was later corrected in the image processing pipeline. Image acquisition for the GS:SFHS sample was conducted at two sites: Aberdeen (N = 541) and Dundee (N = 523). In Aberdeen, imaging was performed using a Philips 3T Achieva TX-series MRI system (Philips Healthcare, Best, Netherlands) equipped with a 32-channel phased-array head coil. In Dundee, a Siemens 3T PrismaFIT scanner (Siemens, Erlangen, Germany) with a 20-channel head and neck phased-array coil was used. At both sites, high-resolution 3D T1-weighted structural images were acquired using magnetization-prepared gradient echo sequences with isotropic voxel dimensions of 1×1×1 mm³ and acquisition parameters differing slightly between sites (e.g., TR/TE/TI = 8.2/3.8/1031 ms in Aberdeen vs. 1740/2.62/900 ms in Dundee). The dMRI protocols were harmonized across sites, comprising 60 axial slices with a voxel size of 2.3 × 2.3 × 2.3 mm³, 64 non-collinear gradient directions (b = 1200 s/mm²), and eight non-diffusion-weighted (b = 0) images. Total acquisition time for the dMRI sequence was 9 minutes and 28 seconds in Aberdeen and 8 minutes and 54 seconds in Dundee 33 . Image processing T1-weighted sMRI processing steps were similar for both cohorts, with UKB data processed by the UKB imaging team and GS:SFHS data processed locally 41 ; cohort-specific differences are detailed below. Raw DICOM images were converted to NIFTI format (using dcm2niix). Subsequent processing relied on the FMRIB Software Library (FSL), including brain extraction with BET 42 , while detailed cortical surface modeling and subcortical structure measurements were conducted using FreeSurfer 43 . FreeSurfer versions differed by cohort: GS:SFHS data were processed locally using v5.3.0, whereas UKB data were processed by the UKB imaging team using v6.0. Manual quality checking (QC) was conducted for GS:SFHS, whereas UKB segmentations underwent automated QC. Imaging-derived phenotypes including cortical surface area, volume, and average thickness of ROIs were extracted based on the Desikan-Killiany atlas 44 . Subcortical volumes were segmented using FreeSurfer's automated segmentation (aseg) tool 45 . For the dMRI data, cohort-specific pipelines were applied due to differences in acquisition and time of processing. GS:SFHS data were processed locally using FSL v5.0.9, with brain extraction, eddy current, and head motion correction using affine registration to the first T2-weighted (b = 0) volume with ‘eddy_correct’. UKB data were processed by the UKB imaging team using FSL v6.0, exploiting multi-shell acquisition and reverse phase encoding for distortion correction. Eddy current and motion artifacts were corrected using FSL's eddy tool 46 , and fieldmap estimation and advanced distortion correction were managed by FSL's topup 47 . For both cohorts, diffusion tensor measures such as FA and MD were estimated with DTIFIT 48 . In UKB, microstructural modeling was enhanced through the NODDI and AMICO tool box 49 , providing a more sophisticated characterization of white matter microstructure. Tractography was performed using an established probabilistic algorithm (BEDPOSTX/ProbtrackX), with probability density functions estimated using a multi-fibre model (two-fibre for GS; three-fibre for UKB). In the GS:SFHS cohort, advanced distortion correction could not be performed, and NODDI was not applied, as the acquisition lacked the multi-shell data required for these steps. Network construction Whole-brain structural networks were constructed using an automated pipeline harmonized across cohorts. Tractography was performed using FSL ProbtrackX, with streamlines initiated from all white matter voxels and generated using 100 Markov Chain Monte Carlo iterations with a 0.5 mm step size, applying previously described stopping criteria 50 . Structural connectivity matrices (85×85) were then constructed by combining T1-weighted and dMRI data, with grey matter nodes defined by 85 FreeSurfer-derived regions of interest (68 cortical, 17 subcortical). Custom Matlab code was applied to identify the interconnecting streamlines between grey matter regions. Alternative network types were computed to capture different aspects of connectivity. Streamline count (SC) represents the total uncorrected number of streamlines connecting each pair of grey matter nodes. FA and MD networks were computed as the mean values of these water diffusion measures across all voxels along the streamlines connecting each node pair. In UKB only, additional edge weights were derived from NODDI-based microstructural parameters, including intra-cellular volume fraction (ICVF), isotropic volume fraction (ISOVF), and orientation dispersion (OD). For example, higher ICVF indicates greater neurite density and stronger white matter connectivity; elevated ISOVF reflects increased free water content and reduced connectivity; while higher ODI corresponds to greater dispersion of neurite orientations 51 . To minimize the impact of false positive connections and ensure a robust network representation, a consistency-thresholding strategy was implemented, retaining only the top 30% most consistently observed edges across subjects within each cohort, based on findings from a prior large-scale single-scanner study 50 . Nodal degree (a weighted network metric) was calculated for each node by summing the weights of all connected edges to that node for multiple network weightings (including DTI and NODDI models), thereby providing an index of each node’s connection strength. Nodal metrics were computed to characterize structural connectivity, and all streamline counts were normalized for ROI size 15 , thereby providing a standardized and reliable basis for the comparative analysis of the white matter network's structural connectivity. Statistical analysis A linear regression model was used to examine the association between brain structure and all PGI measures. PGI served as the independent variable along with brain structure measures as dependent variables. Covariates included age, sex, handedness, brain scan positioning, scanning site, whole brain volume, and the first 40 genetic principal components (PC1-PC40) to account for population stratification. For the discovery sample, we firstly tested the association between PGI and grey matter volume (17 subcortical and 68 cortical regions), cortical thickness (68 cortical regions), and surface area (68 cortical regions). Multiple comparison correction was performed within each PGI across the different thresholds tested, using the FDR method 17 as the primary approach and Bonferroni correction as a supplementary check. For the brain regions identified through the grey matter analysis, we further analyzed the corresponding nodal degrees within the white matter structural network, which was applied in our previous studies 15 , to explore how reading-related genetic factors influence white matter connectivity in the UKB dataset. Additionally, localized vertex-wise surface-based regression analyses of insula (volume and surface area) were conducted in the FreeSurfer average template space using the SurfStat MATLAB toolbox. To assess the replication of the significant associations in the discovery sample, we then conducted a cross-cohort replication using the GS:SFHS dataset. A mixed linear regression model was employed, with kinship (based on pedigree) included as a random effect and age, sex, scanning site, intracranial volume, and the first ten genetic principal components (PC1–PC10) included as covariates. Associations that survived multiple comparison correction in the discovery sample (UKB) were re-evaluated in the GS:SFHS dataset. Finally, for brain regions showing consistent associations, regression analyses were conducted to test whether these regions could predict phonological memory and reading-related skills (results shown in Supplementary Fig. 4, Supplementary Table 8). Mediation analyses were conducted using the mediation package in R to examine the pathways linking dyslexia-related PGI, brain structure or connectivity, and phonological memory in the UKB cohort. These findings were then validated in the GS:SFHS cohort, where additional mediation analyses were also performed using reading-related outcome variables. Mediation models included the same covariates as in the previous analyses. Indirect effects were assessed using non-parametric bootstrapping with 1,000 iterations to estimate confidence intervals and determine statistical significance. Declarations Acknowledgements The Generation Scotland: Scottish Family Health Study (GS:SFHS) was supported by grants from the Biotechnology and Biological Sciences Research Council, UK (BB/T000813/1). GS:SFHS also received core support from the Chief Scientist Office of the Scottish Government Health Directorates [CZD/16/6] and the Scottish Funding Council [HR03006] and is currently supported by the Wellcome Trust [216767/Z/19/Z]. Genotyping of the GS:SFHS samples was carried out by the Genetics Core Laboratory at the Edinburgh Clinical Research Facility, University of Edinburgh, Scotland and was funded by the Medical Research Council UK and the Wellcome Trust (Wellcome Trust Strategic Award “STratifying Resilience and Depression Longitudinally” (STRADL) Reference 104036/Z/14/Z). This project falls under UK Biobank (UKB) approved project 10279. We thank the GS:SFHS, UKB, and 23andMe research-consented participants for their involvement in the study. We would like to thank the employees of 23andMe Research Institute for making this work possible. Author contribution J.Z. conceived the study, conducted data analysis, and wrote the manuscript. H.S.M. computed the polygenic indices. C.B., X.S., and J.M. processed the neuroimaging data, and C. B. assisted the statistical analyses. Y. L. and Y.Z. assisted the statistical analyses and wrote the manuscript. The dyslexia PGI were based on GWAS summary statistics through a collaboration with 23andMe Research Institute. H.W. secured funding and provided access to the Generation Scotland imaging dataset. S.R.C. supervised processing of the UK Biobank imaging dataset and neuroimaging data analysis. SRC and EMT secured funding to support UKB neuroimaging connectome analysis. M.L. secured funding for the GS:SFHS reading and language study, coordinated the project, and supervised genetic data analysis. All authors reviewed and/or commented on the manuscript. COMPETING INTERESTS The authors declare no competing interests. References Lyon, G. R., Shaywitz, S. E. & Shaywitz, B. A. A definition of dyslexia. Ann. Dyslexia 53 , 1–14 (2003). Di Folco, C., Guez, A., Peyre, H. & Ramus, F. Epidemiology of reading disability: A comparison of DSM-5 and ICD-11 criteria. Sci. Stud. Read. 26 , 337–355 (2022). Shaywitz, S. E. & Shaywitz, B. A. Dyslexia (specific reading disability). Biol. Psychiatry 57 , 1301–1309 (2005). Doust, C. et al. Discovery of 42 genome-wide significant loci associated with dyslexia. Nat. Genet. 54 , 1621–1629 (2022). Eising, E. et al. Genome-wide analyses of individual differences in quantitatively assessed reading- and language-related skills in up to 34,000 people. Proc. Natl Acad. Sci. USA 119 , e2202764119 (2022). Mountford H. S. et al. Multivariate genome-wide association analysis of dyslexia and quantitative reading skill improves gene discovery. Translational Psychiatry 15 , 289 (2025). Ramus, F. et al. Theories of developmental dyslexia: insights from a multiple case study of dyslexic adults. Brain 126 , 841–865 (2003). Landerl, K. et al. Predictors of developmental dyslexia in European orthographies with varying complexity. J. Child Psychol. Psychiatry 54 , 686–694 (2013). Ding, N. et al. Profiles of phonological deficits and comorbidity in Chinese developmental dyslexia. Read Writ. (2024). https://doi.org/10.1007/s11145-024-10615-7 Boets, B. et al. Intact but less accessible phonetic representations in adults with dyslexia. Science 342 , 1251–1254 (2013). Lou, C. et al . White matter network connectivity deficits in developmental dyslexia. Hum. Brain Mapp. 40 , 505–516 (2019). Vandermosten, M. et al. A tractography study in dyslexia: neuroanatomic correlates of orthographic, phonological and speech processing. Brain 135 , 935–948 (2012). Zhao, J., Thiebaut de Schotten, M., Altarelli, I., Dubois, J. & Ramus, F. Altered hemispheric lateralization of white matter pathways in developmental dyslexia: evidence from spherical deconvolution tractography. Cortex 76 , 51–62 (2016). Liu, T., Thiebaut de Schotten, M., Altarelli, I., Ramus, F. & Zhao, J. Neural dissociation of visual attention span and phonological deficits in developmental dyslexia: a hub-based white matter network analysis. Hum. Brain Mapp. 43 , 5210–5219 (2022). Soheili-Nezhad, S., Schijven, D., Mars, R. B., Fisher, S. E. & Francks, C. Distinct impact modes of polygenic disposition to dyslexia in the adult brain. Sci. Adv. 10 , eadq2754 (2024). Sydnor, V. J. et al. Human thalamocortical structural connectivity develops in line with a hierarchical axis of cortical plasticity. Nat Neurosci 28 , 1772–1786 (2025). Benjamini, Y. & Hochberg, Y. Controlling the false discovery rate: a practical and powerful approach to multiple testing. J. R. Stat. Soc. B 57 , 289–300 (1995). Turker, S., Fumagalli, B., Kuhnke, P. & Hartwigsen, G. The ‘reading’ brain: Meta-analytic insight into functional activation during reading in adults. Neurosci. Biobehav. Rev. 162 , 106166 (2025). https://doi.org/10.1016/j.neubiorev.2025.106166 Williams, E. J. The comparison of regression variables. J. R. Stat. Soc. Series B (Methodol.) 21 , 396–399 (1959). Ardila, A., Bernal, B. & Rosselli, M. Participation of the insula in language revisited: a meta-analytic connectivity study. J. Neurolinguistics 29 , 31–41 (2014). Ding, N., Peng, P., Tang, J., Ding, Y. & Zhao, J. An investigation of phonological predictors in Chinese developmental dyslexia using a machine learning approach. Read. Writ. (2025). https://doi.org/10.1007/s11145-024-10618-4 Thiebaut de Schotten, M., Cohen, L., Amemiya, E., Braga, L. W. & Dehaene, S. Learning to read improves the structure of the arcuate fasciculus. Cereb. Cortex 24 , 989–995 (2014). Uddin, L. Q., Nomi, J. S., Hébert-Seropian, B., Ghaziri, J. & Boucher, O. Structure and function of the human insula. J. Clin. Neurophysiol. 34 , 300–306 (2017). Heilman, K. M., Voeller, K. & Alexander, A. W. Developmental dyslexia: a motor-articulatory feedback hypothesis. Ann. Neurol. 39 , 407–412 (1996). Cabbage, K. L., Farquharson, K., Iuzzini-Seigel, J., Zuk, J. & Hogan, T. P. Exploring the overlap between dyslexia and speech sound production deficits. Lang. Speech Hear. Serv. Sch. 49 , 774–786 (2018). Uddin, L. Q. Salience processing and insular cortical function and dysfunction. Nat. Rev. Neurosci. 16 , 55–61 (2015). Menon, V. & Uddin, L. Q. Saliency, switching, attention and control: a network model of insula function. Brain Struct. Funct. 214 , 655–667 (2010). Marchesotti, S. et al. Selective enhancement of low-gamma activity by tACS improves phonemic processing and reading accuracy in dyslexia. PLoS Biol. 18 , e3000833 (2020). Rufener, K. S., Krauel, K., Meyer, M., Heinze, H. J. & Zaehle, T. Transcranial electrical stimulation improves phoneme processing in developmental dyslexia. Brain Stimul. 12 , 930–937 (2019). Ibrahim, C. et al . The insula: a brain stimulation target for the treatment of addiction. Front. Pharmacol. 10 , 720 (2019). Citherlet, D. et al. Insular cortex modulation by repetitive transcranial magnetic stimulation with concurrent functional magnetic resonance imaging: preliminary findings. Brain Sci. 15 , 680 (2025). Qi, S., Meesters, S., Nicolay, K., ter Haar Romeny, B. M. & Ossenblok, P. The influence of construction methodology on structural brain network measures: A review. J. Neurosci. Methods. 253 , 170–182 (2015). Sudlow, C. et al. UK Biobank: an open access resource for identifying the causes of a wide range of complex diseases of middle and old age. PLoS Med. 12 , e1001779 (2015). Miller, K. L. et al. Multimodal population brain imaging in the UK Biobank prospective epidemiological study. Nat. Neurosci. 19 , 1523–1536 (2016). Habota, T. et al. Cohort profile for the STratifying Resilience and Depression Longitudinally (STRADL) study: a depression-focused investigation of Generation Scotland, using detailed clinical, cognitive, and neuroimaging assessments. Wellcome Open Res. 4 , 185 (2021). Navrady, L. B. et al. Cohort profile: stratifying resilience and depression longitudinally (STRADL): a questionnaire follow-up of Generation Scotland: Scottish Family Health Study (GS:SFHS). Int. J. Epidemiol. 47 , 13–14g (2018). Davies, G. et al. Genome-wide association study of cognitive functions and educational attainment in UK Biobank (N=112,151). Mol. Psychiatry 21 , 758–767 (2016). Smith, B. H. et al. Cohort profile: Generation Scotland: Scottish Family Health Study (GS:SFHS). The study, its participants and their potential for genetic research on health and illness. Int. J. Epidemiol. 42 , 689–700 (2013). Choi, S. W., Mak, T. S. H. & O’Reilly, P. F. Tutorial: a guide to performing polygenic risk score analyses. Nat. Protoc. 15 , 2759–2772 (2020). Fawns-Ritchie, C. & Deary, I. J. Reliability and validity of the UK Biobank cognitive tests. PLoS One 15 , e0231627 (2020). Stolicyn, A. et al. Automated classification of depression from structural brain measures across two independent community‐based cohorts. Hum. Brain Mapp. 41 , 3922–3937 (2020). Smith, S. M. Fast robust automated brain extraction. Hum. Brain Mapp. 17 , 143–155 (2002). Fischl, B., Sereno, M. I., Tootell, R. B. & Dale, A. M. High-resolution intersubject averaging and a coordinate system for the cortical surface. Hum. Brain Mapp. 8 , 272–284 (1999). Fischl, B. et al. Automatically parcellating the human cerebral cortex. Cereb. Cortex 14 , 11–22 (2004). Desikan, R. S. et al. An automated labeling system for subdividing the human cerebral cortex on MRI scans into gyral based regions of interest. Neuroimage 31 , 968–980 (2006). Fischl, B. et al. Whole brain segmentation: automated labeling of neuroanatomical structures in the human brain. Neuron 33 , 341–355 (2002). Andersson, J. L. & Sotiropoulos, S. N. Non-parametric representation and prediction of single- and multi-shell diffusion-weighted MRI data using Gaussian processes. Neuroimage 122 , 166–176 (2015). Jenkinson, M. Fast, automated, N-dimensional phase-unwrapping algorithm. Magn. Reson. Med. 49 , 193–197 (2003). Behrens, T. E. et al. Characterization and propagation of uncertainty in diffusion-weighted MR imaging. Magn. Reson. Med. 50 , 1077–1088 (2003). Buchanan, C. R. et al. The effect of network thresholding and weighting on structural brain networks in the UK Biobank. NeuroImage 211 , 116443 (2020). Behrens, T. E., Berg, H. J., Jbabdi, S., Rushworth, M. F. & Woolrich, M. W. Probabilistic diffusion tractography with multiple fibre orientations: what can we gain? Neuroimage 34 , 144–155 (2007). Additional Declarations The authors have declared there is NO conflict of interest to disclose Supplementary Files SupplementaryFigure.docx supplementary figures supplementarytable.xlsx supplementary tables Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information

Related documents

• fMRI Evidence for a Cortical Hierarchy of Pitch Pattern Processing#642897
• Statistical learning generalizes across phonological context and transfers to vowel production#434961
• BRD4 represses developmental and neuronal genes through interactions with polycomb complexes#283932
• Introduction to Evolving (Proto)Language/s#713385
• The Shape and Function of Phonology in Evolutionary Linguistics: Why We can Explore Language Origins from Extant Languages, and How#69706