An integrated single-nucleus and spatial transcriptomics atlas reveals the molecular landscape of the human hippocampus.

The hippocampus contains many unique cell types, which serve the structure's specialized functions, including learning, memory and cognition. These cells have distinct spatial topography, morphology, physiology, and connectivity, highlighting the need for transcriptome-wide profiling strategies that retain cytoarchitectural organization. Here, we generated spatially-resolved transcriptomics (SRT) and single-nucleus RNA-sequencing (snRNA-seq) data from adjacent tissue sections of the anterior human hippocampus across ten adult neurotypical donors. We defined molecular profiles for hippocampal cell types and spatial domains. Using non-negative matrix factorization and transfer learning, we integrated these data to define gene expression patterns within the snRNA-seq data and infer the expression of these patterns in the SRT data. With this approach, we leveraged existing rodent datasets that feature information on circuit connectivity and neural activity induction to make predictions about axonal projection targets and likelihood of ensemble recruitment in spatially-defined cellular populations of the human hippocampus. Finally, we integrated genome-wide association studies with transcriptomic data to identify enrichment of genetic components for neurodevelopmental, neuropsychiatric, and neurodegenerative disorders across cell types, spatial domains, and gene expression patterns of the human hippocampus. To make this comprehensive molecular atlas accessible to the scientific community, both raw and processed data are freely available, including through interactive web applications.

Analysis of gene expression in the postmortem brain of neurotypical Black Americans reveals contributions of genetic ancestry.

Ancestral differences in genomic variation affect the regulation of gene expression; however, most gene expression studies have been limited to European ancestry samples or adjusted to identify ancestry-independent associations. Here, we instead examined the impact of genetic ancestry on gene expression and DNA methylation in the postmortem brain tissue of admixed Black American neurotypical individuals to identify ancestry-dependent and ancestry-independent contributions. Ancestry-associated differentially expressed genes (DEGs), transcripts and gene networks, while notably not implicating neurons, are enriched for genes related to the immune response and vascular tissue and explain up to 26% of heritability for ischemic stroke, 27% of heritability for Parkinson disease and 30% of heritability for Alzheimer's disease. Ancestry-associated DEGs also show general enrichment for the heritability of diverse immune-related traits but depletion for psychiatric-related traits. We also compared Black and non-Hispanic white Americans, confirming most ancestry-associated DEGs. Our results delineate the extent to which genetic ancestry affects differences in gene expression in the human brain and the implications for brain illness risk.

Classification of Schizophrenia, Bipolar Disorder and Major Depressive Disorder with Comorbid Traits and Deep Learning Algorithms.

Recent GWASs have demonstrated that comorbid disorders share genetic liabilities. But whether and how these shared liabilities can be used for the classification and differentiation of comorbid disorders remains unclear. In this study, we use polygenic risk scores (PRSs) estimated from 42 comorbid traits and the deep neural networks (DNN) architecture to classify and differentiate schizophrenia (SCZ), bipolar disorder (BIP) and major depressive disorder (MDD). Multiple PRSs were obtained for individuals from the schizophrenia (SCZ) (cases = 6,317, controls = 7,240), bipolar disorder (BIP) (cases = 2,634, controls 4,425) and major depressive disorder (MDD) (cases = 1,704, controls = 3,357) datasets, and classification models were constructed with and without the inclusion of PRSs of the target (SCZ, BIP or MDD). Models with the inclusion of target PRSs performed well as expected. Surprisingly, we found that SCZ could be classified with only the PRSs from 35 comorbid traits (not including the target SCZ and directly related traits) (accuracy 0.760 ± 0.007, AUC 0.843 ± 0.005). Similar results were obtained for BIP (33 traits, accuracy 0.768 ± 0.007, AUC 0.848 ± 0.009), and MDD (36 traits, accuracy 0.794 ± 0.010, AUC 0.869 ± 0.004). Furthermore, these PRSs from comorbid traits alone could effectively differentiate unaffected controls, SCZ, BIP, and MDD patients (average categorical accuracy 0.861 ± 0.003, average AUC 0.961 ± 0.041). These results suggest that the shared liabilities from comorbid traits alone may be sufficient to classify SCZ, BIP and MDD. More importantly, these results imply that a data-driven and objective diagnosis and differentiation of SCZ, BIP and MDD may be feasible.

scMeFormer: a transformer-based deep learning model for imputing DNA methylation states in single cells enhances the detection of epigenetic alterations in schizophrenia.

DNA methylation (DNAm), a crucial epigenetic mark, plays a key role in gene regulation, mammalian development, and various human diseases. Single-cell technologies enable the profiling of DNAm states at cytosines within the DNA sequence of individual cells, but they often suffer from limited coverage of CpG sites. In this study, we introduce scMeFormer, a transformer-based deep learning model designed to impute DNAm states for each CpG site in single cells. Through comprehensive evaluations, we demonstrate the superior performance of scMeFormer compared to alternative models across four single-nucleus DNAm datasets generated by distinct technologies. Remarkably, scMeFormer exhibits high-fidelity imputation, even when dealing with significantly reduced coverage, as low as 10% of the original CpG sites. Furthermore, we applied scMeFormer to a single-nucleus DNAm dataset generated from the prefrontal cortex of four schizophrenia patients and four neurotypical controls. This enabled the identification of thousands of differentially methylated regions associated with schizophrenia that would have remained undetectable without imputation and added granularity to our understanding of epigenetic alterations in schizophrenia within specific cell types. Our study highlights the power of deep learning in imputing DNAm states in single cells, and we expect scMeFormer to be a valuable tool for single-cell DNAm studies.

Alcohol Use Disorder-Associated DNA Methylation in the Nucleus Accumbens and Dorsolateral Prefrontal Cortex.

Background: Alcohol use disorder (AUD) has a profound public health impact. However, understanding of the molecular mechanisms underlying the development and progression of AUD remain limited. Here, we interrogate AUD-associated DNA methylation (DNAm) changes within and across addiction-relevant brain regions: the nucleus accumbens (NAc) and dorsolateral prefrontal cortex (DLPFC). Methods: Illumina HumanMethylation EPIC array data from 119 decedents of European ancestry (61 cases, 58 controls) were analyzed using robust linear regression, with adjustment for technical and biological variables. Associations were characterized using integrative analyses of public gene regulatory data and published genetic and epigenetic studies. We additionally tested for brain region-shared and -specific associations using mixed effects modeling and assessed implications of these results using public gene expression data. Results: At a false discovery rate ≤ 0.05, we identified 53 CpGs significantly associated with AUD status for NAc and 31 CpGs for DLPFC. In a meta-analysis across the regions, we identified an additional 21 CpGs associated with AUD, for a total of 105 unique AUD-associated CpGs (120 genes). AUD-associated CpGs were enriched in histone marks that tag active promoters and our strongest signals were specific to a single brain region. Of the 120 genes, 23 overlapped with previous genetic associations for substance use behaviors; all others represent novel associations. Conclusions: Our findings identify AUD-associated methylation signals, the majority of which are specific within NAc or DLPFC. Some signals may constitute predisposing genetic and epigenetic variation, though more work is needed to further disentangle the neurobiological gene regulatory differences associated with AUD.

Deep learning predicts DNA methylation regulatory variants in specific brain cell types and enhances fine mapping for brain disorders.

DNA methylation (DNAm) is essential for brain development and function and potentially mediates the effects of genetic risk variants underlying brain disorders. We present INTERACT, a transformer-based deep learning model to predict regulatory variants impacting DNAm levels in specific brain cell types, leveraging existing single-nucleus DNAm data from the human brain. We show that INTERACT accurately predicts cell type-specific DNAm profiles, achieving an average area under the Receiver Operating Characteristic curve of 0.98 across cell types. Furthermore, INTERACT predicts cell type-specific DNAm regulatory variants, which reflect cellular context and enrich the heritability of brain-related traits in relevant cell types. Importantly, we demonstrate that incorporating predicted variant effects and DNAm levels of CpG sites enhances the fine mapping for three brain disorders-schizophrenia, depression, and Alzheimer's disease-and facilitates mapping causal genes to particular cell types. Our study highlights the power of deep learning in identifying cell type-specific regulatory variants, which will enhance our understanding of the genetics of complex traits.

Smoking-informed methylation and expression QTLs in human brain and colocalization with smoking-associated genetic loci.

Smoking is a leading cause of preventable morbidity and mortality. Smoking is heritable, and genome-wide association studies (GWAS) of smoking behaviors have identified hundreds of significant loci. Most GWAS-identified variants are noncoding with unknown neurobiological effects. We used genome-wide genotype, DNA methylation, and RNA sequencing data in postmortem human nucleus accumbens (NAc) to identify cis-methylation/expression quantitative trait loci (meQTLs/eQTLs), investigate variant-by-cigarette smoking interactions across the genome, and overlay QTL evidence at smoking GWAS-identified loci to evaluate their regulatory potential. Active smokers (N=52) and nonsmokers (N=171) were defined based on cotinine biomarker levels and next-of-kin reporting. We simultaneously tested variant and variant-by-smoking interaction effects on methylation and expression, separately, adjusting for biological and technical covariates and using a two-stage multiple testing approach with eigenMT and Bonferroni corrections. We found >2 million significant meQTL variants (padj<0.05) corresponding to 41,695 unique CpGs. Results were largely driven by main effects; five meQTLs, mapping to NUDT12, FAM53B, RNF39, and ADRA1B, showed a significant interaction with smoking. We found 57,683 significant eQTLs for 958 unique eGenes (padj<0.05) and no smoking interactions. Colocalization analyses identified loci with smoking-associated GWAS variants that overlapped meQTLs/eQTLs, suggesting that these heritable factors may influence smoking behaviors through functional effects on methylation/expression. One locus containing MUSTIN1 and ITIH4 colocalized across all data types (GWAS + meQTL + eQTL). In this first genome-wide meQTL map in the human NAc, the enriched overlap with smoking GWAS-identified genetic loci provides evidence that gene regulation in the brain helps explain the neurobiology of smoking behaviors.

Investigating trait variability of gene co-expression network architecture in brain by controlling for genomic risk of schizophrenia.

The effect of schizophrenia (SCZ) genetic risk on gene expression in brain remains elusive. A popular approach to this problem has been the application of gene co-expression network algorithms (e.g., WGCNA). To improve reliability with this method it is critical to remove unwanted sources of variance while also preserving biological signals of interest. In this WCGNA study of RNA-Seq data from postmortem prefrontal cortex (78 neurotypical donors, EUR ancestry), we tested the effects of SCZ genetic risk on co-expression networks. Specifically, we implemented a novel design in which gene expression was adjusted by linear regression models to preserve or remove variance explained by biological signal of interest (GWAS genomic scores for SCZ risk-(GS-SCZ), and genomic scores- GS of height (GS-Ht) as a negative control), while removing variance explained by covariates of non-interest. We calculated co-expression networks from adjusted expression (GS-SCZ and GS-Ht preserved or removed), and consensus between them (representative of a "background" network free of genomic scores effects). We then tested the overlap between GS-SCZ preserved modules and background networks reasoning that modules with reduced overlap would be most affected by GS-SCZ biology. Additionally, we tested these modules for convergence of SCZ risk (i.e., enrichment in PGC3 SCZ GWAS priority genes, enrichment in SCZ risk heritability and relevant biological ontologies. Our results highlight key aspects of GS-SCZ effects on brain co-expression networks, specifically: 1) preserving/removing SCZ genetic risk alters the co-expression modules; 2) biological pathways enriched in modules affected by GS-SCZ implicate processes of transcription, translation and metabolism that converge to influence synaptic transmission; 3) priority PGC3 SCZ GWAS genes and SCZ risk heritability are enriched in modules associated with GS-SCZ effects. Overall, our results indicate that gene co-expression networks that selectively integrate information about genetic risk can reveal novel combinations of biological pathways involved in schizophrenia.

Prioritization of potential causative genes for schizophrenia in placenta.

Our earlier work has shown that genomic risk for schizophrenia converges with early life complications in affecting risk for the disorder and sex-biased neurodevelopmental trajectories. Here, we identify specific genes and potential mechanisms that, in placenta, may mediate such outcomes. We performed TWAS in healthy term placentae (N = 147) to derive candidate placental causal genes that we confirmed with SMR; to search for placenta and schizophrenia-specific associations, we performed an analogous analysis in fetal brain (N = 166) and additional placenta TWAS for other disorders/traits. The analyses in the whole sample and stratifying by sex ultimately highlight 139 placenta and schizophrenia-specific risk genes, many being sex-biased; the candidate molecular mechanisms converge on the nutrient-sensing capabilities of placenta and trophoblast invasiveness. These genes also implicate the Coronavirus-pathogenesis pathway and showed increased expression in placentae from a small sample of SARS-CoV-2-positive pregnancies. Investigating placental risk genes for schizophrenia and candidate mechanisms may lead to opportunities for prevention that would not be suggested by study of the brain alone.

Genetic and environmental contributions to ancestry differences in gene expression in the human brain.

Ancestral differences in genomic variation are determining factors in gene regulation; however, most gene expression studies have been limited to European ancestry samples or adjusted for ancestry to identify ancestry-independent associations. We instead examined the impact of genetic ancestry on gene expression and DNA methylation (DNAm) in admixed African/Black American neurotypical individuals to untangle effects of genetic and environmental factors. Ancestry-associated differentially expressed genes (DEGs), transcripts, and gene networks, while notably not implicating neurons, are enriched for genes related to immune response and vascular tissue and explain up to 26% of heritability for ischemic stroke, 27% of heritability for Parkinson's disease, and 30% of heritability for Alzhemier's disease. Ancestry-associated DEGs also show general enrichment for heritability of diverse immune-related traits but depletion for psychiatric-related traits. The cell-type enrichments and direction of effects vary by brain region. These DEGs are less evolutionarily constrained and are largely explained by genetic variations; roughly 15% are predicted by DNAm variation implicating environmental exposures. We also compared Black and White Americans, confirming most of these ancestry-associated DEGs. Our results highlight how environment and genetic background affect genetic ancestry differences in gene expression in the human brain and affect risk for brain illness.

A self-inspected adaptive SMOTE algorithm (SASMOTE) for highly imbalanced data classification in healthcare.

In many healthcare applications, datasets for classification may be highly imbalanced due to the rare occurrence of target events such as disease onset. The SMOTE (Synthetic Minority Over-sampling Technique) algorithm has been developed as an effective resampling method for imbalanced data classification by oversampling samples from the minority class. However, samples generated by SMOTE may be ambiguous, low-quality and non-separable with the majority class. To enhance the quality of generated samples, we proposed a novel self-inspected adaptive SMOTE (SASMOTE) model that leverages an adaptive nearest neighborhood selection algorithm to identify the "visible" nearest neighbors, which are used to generate samples likely to fall into the minority class. To further enhance the quality of the generated samples, an uncertainty elimination via self-inspection approach is introduced in the proposed SASMOTE model. Its objective is to filter out the generated samples that are highly uncertain and inseparable with the majority class. The effectiveness of the proposed algorithm is compared with existing SMOTE-based algorithms and demonstrated through two real-world case studies in healthcare, including risk gene discovery and fatal congenital heart disease prediction. By generating the higher quality synthetic samples, the proposed algorithm is able to help achieve better prediction performance (in terms of F1 score) on average compared to the other methods, which is promising to enhance the usability of machine learning models on highly imbalanced healthcare data.

A multimodal deep learning model to infer cell-type-specific functional gene networks.

BACKGROUND: Functional gene networks (FGNs) capture functional relationships among genes that vary across tissues and cell types. Construction of cell-type-specific FGNs enables the understanding of cell-type-specific functional gene relationships and insights into genetic mechanisms of human diseases in disease-relevant cell types. However, most existing FGNs were developed without consideration of specific cell types within tissues. RESULTS: In this study, we created a multimodal deep learning model (MDLCN) to predict cell-type-specific FGNs in the human brain by integrating single-nuclei gene expression data with global protein interaction networks. We systematically evaluated the prediction performance of the MDLCN and showed its superior performance compared to two baseline models (boosting tree and convolutional neural network). Based on the predicted cell-type-specific FGNs, we observed that cell-type marker genes had a higher level of hubness than non-marker genes in their corresponding cell type. Furthermore, we showed that risk genes underlying autism and Alzheimer's disease were more strongly connected in disease-relevant cell types, supporting the cellular context of predicted cell-type-specific FGNs. CONCLUSIONS: Our study proposes a powerful deep learning approach (MDLCN) to predict FGNs underlying a diverse set of cell types in human brain. The MDLCN model enhances prediction accuracy of cell-type-specific FGNs compared to single modality convolutional neural network (CNN) and boosting tree models, as shown by higher areas under both receiver operating characteristic (ROC) and precision-recall curves for different levels of independent test datasets. The predicted FGNs also show evidence for the cellular context and distinct topological features (i.e. higher hubness and topological score) of cell-type marker genes. Moreover, we observed stronger modularity among disease-associated risk genes in FGNs of disease-relevant cell types. For example, the strength of connectivity among autism risk genes was stronger in neurons, but risk genes underlying Alzheimer's disease were more connected in microglia.

Deep learning predicts DNA methylation regulatory variants in the human brain and elucidates the genetics of psychiatric disorders.

There is growing evidence for the role of DNA methylation (DNAm) quantitative trait loci (mQTLs) in the genetics of complex traits, including psychiatric disorders. However, due to extensive linkage disequilibrium (LD) of the genome, it is challenging to identify causal genetic variations that drive DNAm levels by population-based genetic association studies. This limits the utility of mQTLs for fine-mapping risk loci underlying psychiatric disorders identified by genome-wide association studies (GWAS). Here we present INTERACT, a deep learning model that integrates convolutional neural networks with transformer, to predict effects of genetic variations on DNAm levels at CpG sites in the human brain. We show that INTERACT-derived DNAm regulatory variants are not confounded by LD, are concentrated in regulatory genomic regions in the human brain, and are convergent with mQTL evidence from genetic association analysis. We further demonstrate that predicted DNAm regulatory variants are enriched for heritability of brain-related traits and improve polygenic risk prediction for schizophrenia across diverse ancestry samples. Finally, we applied predicted DNAm regulatory variants for fine-mapping schizophrenia GWAS risk loci to identify potential novel risk genes. Our study shows the power of a deep learning approach to identify functional regulatory variants that may elucidate the genetic basis of complex traits.

Estimation of Causal Effect of Age at Menarche on Pubertal Height Growth Using Mendelian Randomization.

We use Mendelian randomization to estimate the causal effect of age at menarche on late pubertal height growth and total pubertal height growth. The instrument SNPs selected from the exposure genome-wide association study (GWAS) are validated in additional population-matched exposure GWASs. Based on the inverse variance weighting method, there is a positive causal relationship of age at menarche on late pubertal growth (ß^=0.56, 95% CI: (0.34, 0.78), p=3.16×10-7) and on total pubertal growth (ß^=0.36, 95% CI: (0.14, 0.58), p=1.30×10-3). If the instrument SNPs are not validated in additional exposure GWASs, the estimated effect on late pubertal height growth increases by 3.6% to ß^=0.58 (95% CI: (0.42, 0.73), p=4.38×10-13) while the estimates on total pubertal height growth increases by 41.7% to ß^=0.51 (95% CI: (0.35, 0.67), p=2.96×10-11).

Molecular phenotypes associated with antipsychotic drugs in the human caudate nucleus.

Antipsychotic drugs are the current first-line of treatment for schizophrenia and other psychotic conditions. However, their molecular effects on the human brain are poorly studied, due to difficulty of tissue access and confounders associated with disease status. Here we examine differences in gene expression and DNA methylation associated with positive antipsychotic drug toxicology status in the human caudate nucleus. We find no genome-wide significant differences in DNA methylation, but abundant differences in gene expression. These gene expression differences are overall quite similar to gene expression differences between schizophrenia cases and controls. Interestingly, gene expression differences based on antipsychotic toxicology are different between brain regions, potentially due to affected cell type differences. We finally assess similarities with effects in a mouse model, which finds some overlapping effects but many differences as well. As a first look at the molecular effects of antipsychotics in the human brain, the lack of epigenetic effects is unexpected, possibly because long term treatment effects may be relatively stable for extended periods.

Air pollution interacts with genetic risk to influence cortical networks implicated in depression.

Air pollution is a reversible cause of significant global mortality and morbidity. Epidemiological evidence suggests associations between air pollution exposure and impaired cognition and increased risk for major depressive disorders. However, the neural bases of these associations have been unclear. Here, in healthy human subjects exposed to relatively high air pollution and controlling for socioeconomic, genomic, and other confounders, we examine across multiple levels of brain network function the extent to which particulate matter (PM2.5) exposure influences putative genetic risk mechanisms associated with depression. Increased ambient PM2.5 exposure was associated with poorer reasoning and problem solving and higher-trait anxiety/depression. Working memory and stress-related information transfer (effective connectivity) across cortical and subcortical brain networks were influenced by PM2.5 exposure to differing extents depending on the polygenic risk for depression in gene-by-environment interactions. Effective connectivity patterns from individuals with higher polygenic risk for depression and higher exposures with PM2.5, but not from those with lower genetic risk or lower exposures, correlated spatially with the coexpression of depression-associated genes across corresponding brain regions in the Allen Brain Atlas. These converging data suggest that PM2.5 exposure affects brain network functions implicated in the genetic mechanisms of depression.

Genome-wide sequencing-based identification of methylation quantitative trait loci and their role in schizophrenia risk.

DNA methylation (DNAm) is an epigenetic regulator of gene expression and a hallmark of gene-environment interaction. Using whole-genome bisulfite sequencing, we have surveyed DNAm in 344 samples of human postmortem brain tissue from neurotypical subjects and individuals with schizophrenia. We identify genetic influence on local methylation levels throughout the genome, both at CpG sites and CpH sites, with 86% of SNPs and 55% of CpGs being part of methylation quantitative trait loci (meQTLs). These associations can further be clustered into regions that are differentially methylated by a given SNP, highlighting the genes and regions with which these loci are epigenetically associated. These findings can be used to better characterize schizophrenia GWAS-identified variants as epigenetic risk variants. Regions differentially methylated by schizophrenia risk-SNPs explain much of the heritability associated with risk loci, despite covering only a fraction of the genomic space. We provide a comprehensive, single base resolution view of association between genetic variation and genomic methylation, and implicate schizophrenia GWAS-associated variants as influencing the epigenetic plasticity of the brain.

Effect of selection bias on two sample summary data based Mendelian randomization.

Mendelian randomization (MR) is becoming more and more popular for inferring causal relationship between an exposure and a trait. Typically, instrument SNPs are selected from an exposure GWAS based on their summary statistics and the same summary statistics on the selected SNPs are used for subsequent analyses. However, this practice suffers from selection bias and can invalidate MR methods, as showcased via two popular methods: the summary data-based MR (SMR) method and the two-sample MR Steiger method. The SMR method is conservative while the MR Steiger method can be either conservative or liberal. A simple and yet more powerful alternative to SMR is proposed.

Body Part Regression With Self-Supervision.

Body part regression is a promising new technique that enables content navigation through self-supervised learning. Using this technique, the global quantitative spatial location for each axial view slice is obtained from computed tomography (CT). However, it is challenging to define a unified global coordinate system for body CT scans due to the large variabilities in image resolution, contrasts, sequences, and patient anatomy. Therefore, the widely used supervised learning approach cannot be easily deployed. To address these concerns, we propose an annotation-free method named blind-unsupervised-supervision network (BUSN). The contributions of the work are in four folds: (1) 1030 multi-center CT scans are used in developing BUSN without any manual annotation. (2) the proposed BUSN corrects the predictions from unsupervised learning and uses the corrected results as the new supervision; (3) to improve the consistency of predictions, we propose a novel neighbor message passing (NMP) scheme that is integrated with BUSN as a statistical learning based correction; and (4) we introduce a new pre-processing pipeline with inclusion of the BUSN, which is validated on 3D multi-organ segmentation. The proposed method is trained on 1,030 whole body CT scans (230,650 slices) from five datasets, as well as an independent external validation cohort with 100 scans. From the body part regression results, the proposed BUSN achieved significantly higher median R-squared score (=0.9089) than the state-of-the-art unsupervised method (=0.7153). When introducing BUSN as a preprocessing stage in volumetric segmentation, the proposed pre-processing pipeline using BUSN approach increases the total mean Dice score of the 3D abdominal multi-organ segmentation from 0.7991 to 0.8145.