Molecular adaptations in response to exercise training are associated with tissue-specific transcriptomic and epigenomic signatures.

Regular exercise has many physical and brain health benefits, yet the molecular mechanisms mediating exercise effects across tissues remain poorly understood. Here we analyzed 400 high-quality DNA methylation, ATAC-seq, and RNA-seq datasets from eight tissues from control and endurance exercise-trained (EET) rats. Integration of baseline datasets mapped the gene location dependence of epigenetic control features and identified differing regulatory landscapes in each tissue. The transcriptional responses to 8 weeks of EET showed little overlap across tissues and predominantly comprised tissue-type enriched genes. We identified sex differences in the transcriptomic and epigenomic changes induced by EET. However, the sex-biased gene responses were linked to shared signaling pathways. We found that many G protein-coupled receptor-encoding genes are regulated by EET, suggesting a role for these receptors in mediating the molecular adaptations to training across tissues. Our findings provide new insights into the mechanisms underlying EET-induced health benefits across organs.

MousiPLIER: A Mouse Pathway-Level Information Extractor Model.

High throughput gene expression profiling measures individual gene expression across conditions. However, genes are regulated in complex networks, not as individual entities, limiting the interpretability of gene expression data. Machine learning models that incorporate prior biological knowledge are a powerful tool to extract meaningful biology from gene expression data. Pathway-level information extractor (PLIER) is an unsupervised machine learning method that defines biological pathways by leveraging the vast amount of published transcriptomic data. PLIER converts gene expression data into known pathway gene sets, termed latent variables (LVs), to substantially reduce data dimensionality and improve interpretability. In the current study, we trained the first mouse PLIER model on 190,111 mouse brain RNA-sequencing samples, the greatest amount of training data ever used by PLIER. We then validated the mousiPLIER approach in a study of microglia and astrocyte gene expression across mouse brain aging. mousiPLIER identified biological pathways that are significantly associated with aging, including one latent variable (LV41) corresponding to striatal signal. To gain further insight into the genes contained in LV41, we performed k-means clustering on the training data to identify studies that respond strongly to LV41. We found that the variable was relevant to striatum and aging across the scientific literature. Finally, we built a web server (http://mousiplier.greenelab.com/) for users to easily explore the learned latent variables. Taken together this study defines mousiPLIER as a method to uncover meaningful biological processes in mouse brain transcriptomic studies.Significance statement RNA-sequencing studies define differential expression of individual genes across conditions. However, genes are regulated in complex networks, not as individual entities. Machine learning models that incorporate biological pathway information are a powerful tool to analyze human gene expression. However, such models are lacking for mouse, despite the vast number of mouse RNA-seq datasets. We trained a mouse pathway-level information extractor model (mousiPLIER) to reduce data dimensionality from over 10,000 genes to 196 'latent variables' that map to known biological pathways. To validate this approach, we applied mousiPLIER to differential expression across mouse brain aging. We identified 26 functional pathways (latent variables) that varied across aging. Finally, we developed a web server to facilitate use of mousiPLIER by the scientific community.

Evolutionary rate covariation is a reliable predictor of co-functional interactions but not necessarily physical interactions.

Co-functional proteins tend to have rates of evolution that covary over time. This correlation between evolutionary rates can be measured over the branches of a phylogenetic tree through methods such as evolutionary rate covariation (ERC), and then used to construct gene networks by the identification of proteins with functional interactions. The cause of this correlation has been hypothesized to result from both compensatory coevolution at physical interfaces and nonphysical forces such as shared changes in selective pressure. This study explores whether coevolution due to compensatory mutations has a measurable effect on the ERC signal. We examined the difference in ERC signal between physically interacting protein domains within complexes compared to domains of the same proteins that do not physically interact. We found no generalizable relationship between physical interaction and high ERC, although a few complexes ranked physical interactions higher than nonphysical interactions. Therefore, we conclude that coevolution due to physical interaction is weak, but present in the signal captured by ERC, and we hypothesize that the stronger signal instead comes from selective pressures on the protein as a whole and maintenance of the general function.

Vocal learning-associated convergent evolution in mammalian proteins and regulatory elements.

Vocal production learning ("vocal learning") is a convergently evolved trait in vertebrates. To identify brain genomic elements associated with mammalian vocal learning, we integrated genomic, anatomical, and neurophysiological data from the Egyptian fruit bat (Rousettus aegyptiacus) with analyses of the genomes of 215 placental mammals. First, we identified a set of proteins evolving more slowly in vocal learners. Then, we discovered a vocal motor cortical region in the Egyptian fruit bat, an emergent vocal learner, and leveraged that knowledge to identify active cis-regulatory elements in the motor cortex of vocal learners. Machine learning methods applied to motor cortex open chromatin revealed 50 enhancers robustly associated with vocal learning whose activity tended to be lower in vocal learners. Our research implicates convergent losses of motor cortex regulatory elements in mammalian vocal learning evolution.

Optimizer's dilemma: optimization strongly influences model selection in transcriptomic prediction.

Motivation: Most models can be fit to data using various optimization approaches. While model choice is frequently reported in machine-learning-based research, optimizers are not often noted. We applied two different implementations of LASSO logistic regression implemented in Python's scikit-learn package, using two different optimization approaches (coordinate descent, implemented in the liblinear library, and stochastic gradient descent, or SGD), to predict mutation status and gene essentiality from gene expression across a variety of pan-cancer driver genes. For varying levels of regularization, we compared performance and model sparsity between optimizers. Results: After model selection and tuning, we found that liblinear and SGD tended to perform comparably. liblinear models required more extensive tuning of regularization strength, performing best for high model sparsities (more nonzero coefficients), but did not require selection of a learning rate parameter. SGD models required tuning of the learning rate to perform well, but generally performed more robustly across different model sparsities as regularization strength decreased. Given these tradeoffs, we believe that the choice of optimizers should be clearly reported as a part of the model selection and validation process, to allow readers and reviewers to better understand the context in which results have been generated. Availability and implementation: The code used to carry out the analyses in this study is available at https://github.com/greenelab/pancancer-evaluation/tree/master/01_stratified_classification. Performance/regularization strength curves for all genes in the Vogelstein et al. (2013) dataset are available at https://doi.org/10.6084/m9.figshare.22728644.

Memory B cell subsets have divergent developmental origins that are coupled to distinct imprinted epigenetic states.

Memory B cells (MBCs) are phenotypically and functionally diverse, but their developmental origins remain undefined. Murine MBCs can be divided into subsets by expression of CD80 and PD-L2. Upon re-immunization, CD80/PD-L2 double-negative (DN) MBCs spawn germinal center B cells (GCBCs), whereas CD80/PD-L2 double-positive (DP) MBCs generate plasmablasts but not GCBCs. Using multiple approaches, including generation of an inducible GCBC-lineage reporter mouse, we demonstrate in a T cell-dependent response that DN cells formed independently of the germinal center (GC), whereas DP cells exhibited either extrafollicular (DPEX) or GCBC (DPGC) origins. Chromatin and transcriptional profiling revealed similarity of DN cells with an early memory precursor. Reciprocally, GCBC-derived DP cells shared distinct genomic features with GCBCs, while DPEX cells had hybrid features. Upon restimulation, DPEX cells were more prone to divide, while DPGC cells differentiated toward IgG1+ plasmablasts. Thus, MBC functional diversity is generated through distinct developmental histories, which imprint characteristic epigenetic patterns onto their progeny, thereby programming them for divergent functional responses.

RERconverge Expansion: Using Relative Evolutionary Rates to Study Complex Categorical Trait Evolution.

Comparative genomics approaches seek to associate evolutionary genetic changes with the evolution of phenotypes across a phylogeny. Many of these methods, including our evolutionary rates based method, RERconverge, lack the capability of analyzing non-ordinal, multicategorical traits. To address this limitation, we introduce an expansion to RERconverge that associates shifts in evolutionary rates with the convergent evolution of multi-categorical traits. The categorical RERconverge expansion includes methods for performing categorical ancestral state reconstruction, statistical tests for associating relative evolutionary rates with categorical variables, and a new method for performing phylogenetic permulations on multi-categorical traits. In addition to demonstrating our new method on a three-category diet phenotype, we compare its performance to naive pairwise binary RERconverge analyses and two existing methods for comparative genomic analyses of categorical traits: phylogenetic simulations and a phylogenetic signal based method. We also present a diagnostic analysis of the new permulations approach demonstrating how the method scales with the number of species and the number of categories included in the analysis. Our results show that our new categorical method outperforms phylogenetic simulations at identifying genes and enriched pathways significantly associated with the diet phenotype and that the new ancestral reconstruction drives an improvement in our ability to capture diet-related enriched pathways. Our categorical permulations were able to account for non-uniform null distributions and correct for non-independence in gene rank during pathway enrichment analysis. The categorical expansion to RERconverge will provide a strong foundation for applying the comparative method to categorical traits on larger data sets with more species and more complex trait evolution.

Benchmarking of deep neural networks for predicting personal gene expression from DNA sequence highlights shortcomings.

Deep learning methods have recently become the state of the art in a variety of regulatory genomic tasks1-6, including the prediction of gene expression from genomic DNA. As such, these methods promise to serve as important tools in interpreting the full spectrum of genetic variation observed in personal genomes. Previous evaluation strategies have assessed their predictions of gene expression across genomic regions; however, systematic benchmarking is lacking to assess their predictions across individuals, which would directly evaluate their utility as personal DNA interpreters. We used paired whole genome sequencing and gene expression from 839 individuals in the ROSMAP study7 to evaluate the ability of current methods to predict gene expression variation across individuals at varied loci. Our approach identifies a limitation of current methods to correctly predict the direction of variant effects. We show that this limitation stems from insufficiently learned sequence motif grammar and suggest new model training strategies to improve performance.

MousiPLIER: A Mouse Pathway-Level Information Extractor Model.

High throughput gene expression profiling is a powerful approach to generate hypotheses on the underlying causes of biological function and disease. Yet this approach is limited by its ability to infer underlying biological pathways and burden of testing tens of thousands of individual genes. Machine learning models that incorporate prior biological knowledge are necessary to extract meaningful pathways and generate rational hypothesis from the vast amount of gene expression data generated to date. We adopted an unsupervised machine learning method, Pathway-level information extractor (PLIER), to train the first mouse PLIER model on 190,111 mouse brain RNA-sequencing samples, the greatest amount of training data ever used by PLIER. mousiPLER converted gene expression data into a latent variables that align to known pathway or cell maker gene sets, substantially reducing data dimensionality and improving interpretability. To determine the utility of mousiPLIER, we applied it to a mouse brain aging study of microglia and astrocyte transcriptomic profiling. We found a specific set of latent variables that are significantly associated with aging, including one latent variable (LV41) corresponding to striatal signal. We next performed k-means clustering on the training data to identify studies that respond strongly to LV41, finding that the variable is relevant to striatum and aging across the scientific literature. Finally, we built a web server (http://mousiplier.greenelab.com/) for users to easily explore the learned latent variables. Taken together this study provides proof of concept that mousiPLIER can uncover meaningful biological processes in mouse transcriptomic studies.

An intrinsically interpretable neural network architecture for sequence-to-function learning.

MOTIVATION: Sequence-based deep learning approaches have been shown to predict a multitude of functional genomic readouts, including regions of open chromatin and RNA expression of genes. However, a major limitation of current methods is that model interpretation relies on computationally demanding post hoc analyses, and even then, one can often not explain the internal mechanics of highly parameterized models. Here, we introduce a deep learning architecture called totally interpretable sequence-to-function model (tiSFM). tiSFM improves upon the performance of standard multilayer convolutional models while using fewer parameters. Additionally, while tiSFM is itself technically a multilayer neural network, internal model parameters are intrinsically interpretable in terms of relevant sequence motifs. RESULTS: We analyze published open chromatin measurements across hematopoietic lineage cell-types and demonstrate that tiSFM outperforms a state-of-the-art convolutional neural network model custom-tailored to this dataset. We also show that it correctly identifies context-specific activities of transcription factors with known roles in hematopoietic differentiation, including Pax5 and Ebf1 for B-cells, and Rorc for innate lymphoid cells. tiSFM's model parameters have biologically meaningful interpretations, and we show the utility of our approach on a complex task of predicting the change in epigenetic state as a function of developmental transition. AVAILABILITY AND IMPLEMENTATION: The source code, including scripts for the analysis of key findings, can be found at https://github.com/boooooogey/ATAConv, implemented in Python.

Benchmarking of deep neural networks for predicting personal gene expression from DNA sequence highlights shortcomings.

Deep learning methods have recently become the state-of-the-art in a variety of regulatory genomic tasks1-6 including the prediction of gene expression from genomic DNA. As such, these methods promise to serve as important tools in interpreting the full spectrum of genetic variation observed in personal genomes. Previous evaluation strategies have assessed their predictions of gene expression across genomic regions, however, systematic benchmarking is lacking to assess their predictions across individuals, which would directly evaluates their utility as personal DNA interpreters. We used paired Whole Genome Sequencing and gene expression from 839 individuals in the ROSMAP study7 to evaluate the ability of current methods to predict gene expression variation across individuals at varied loci. Our approach identifies a limitation of current methods to correctly predict the direction of variant effects. We show that this limitation stems from insufficiently learnt sequence motif grammar, and suggest new model training strategies to improve performance.

A methylation clock model of mild SARS-CoV-2 infection provides insight into immune dysregulation.

DNA methylation comprises a cumulative record of lifetime exposures superimposed on genetically determined markers. Little is known about methylation dynamics in humans following an acute perturbation, such as infection. We characterized the temporal trajectory of blood epigenetic remodeling in 133 participants in a prospective study of young adults before, during, and after asymptomatic and mildly symptomatic SARS-CoV-2 infection. The differential methylation caused by asymptomatic or mildly symptomatic infections was indistinguishable. While differential gene expression largely returned to baseline levels after the virus became undetectable, some differentially methylated sites persisted for months of follow-up, with a pattern resembling autoimmune or inflammatory disease. We leveraged these responses to construct methylation-based machine learning models that distinguished samples from pre-, during-, and postinfection time periods, and quantitatively predicted the time since infection. The clinical trajectory in the young adults and in a diverse cohort with more severe outcomes was predicted by the similarity of methylation before or early after SARS-CoV-2 infection to the model-defined postinfection state. Unlike the phenomenon of trained immunity, the postacute SARS-CoV-2 epigenetic landscape we identify is antiprotective.

IL-21R signal reprogramming cooperates with CD40 and BCR signals to select and differentiate germinal center B cells.

Both B cell receptor (BCR) and CD40 signaling are rewired in germinal center (GC) B cells (GCBCs) to synergistically induce c-MYC and phosphorylated S6 ribosomal protein (p-S6), markers of positive selection. How interleukin-21 (IL-21), a key T follicular helper (TFH)-derived cytokine, affects GCBCs is unclear. Like BCR and CD40 signals, IL-21 receptor (IL-21R) plus CD40 signals also synergize to induce c-MYC and p-S6 in GCBCs. However, IL-21R plus CD40 stimulation differentially affects GCBC fate compared with BCR plus CD40 ligation-engaging unique molecular mechanisms-as revealed by bulk RNA sequencing (RNA-seq), single-cell RNA-seq, and flow cytometry of GCBCs in vitro and in vivo. Whereas both signal pairs induced BLIMP1 in some GCBCs, only the IL-21R/CD40 combination induced IRF4hi/CD138+ cells, indicative of plasma cell differentiation, along with CCR6+/CD38+ memory B cell precursors. These findings reveal a second positive selection pathway in GCBCs, document rewired IL-21R signaling in GCBCs, and link specific TFH- and Ag-derived signals to GCBC differentiation.

An intrinsically interpretable neural network architecture for sequence to function learning.

MOTIVATION: Sequence-based deep learning approaches have been shown to predict a multitude of functional genomic readouts, including regions of open chromatin and RNA expression of genes. However, a major limitation of current methods is that model interpretation relies on computationally demanding post hoc analyses, and even then, one can often not explain the internal mechanics of highly parameterized models. Here, we introduce a deep learning architecture called tiSFM (totally interpretable sequence to function model). tiSFM improves upon the performance of standard multi-layer convolutional models while using fewer parameters. Additionally, while tiSFM is itself technically a multi-layer neural network, internal model parameters are intrinsically interpretable in terms of relevant sequence motifs. RESULTS: We analyze published open chromatin measurements across hematopoietic lineage cell-types and demonstrate that tiSFM outperforms a state-of-the-art convolutional neural network model custom-tailored to this dataset. We also show that it correctly identifies context specific activities of transcription factors with known roles in hematopoietic differentiation, including Pax5 and Ebf1 for B-cells, and Rorc for innate lymphoid cells. tiSFM's model parameters have biologically meaningful interpretations, and we show the utility of our approach on a complex task of predicting the change in epigenetic state as a function of developmental transition. AVAILABILITY AND IMPLEMENTATION: The source code, including scripts for the analysis of key findings, can be found at https://github.com/boooooogey/ATAConv, implemented in Python.

Multi-objective optimization identifies a specific and interpretable COVID-19 host response signature.

The identification of a COVID-19 host response signature in blood can increase the understanding of SARS-CoV-2 pathogenesis and improve diagnostic tools. Applying a multi-objective optimization framework to both massive public and new multi-omics data, we identified a COVID-19 signature regulated at both transcriptional and epigenetic levels. We validated the signature's robustness in multiple independent COVID-19 cohorts. Using public data from 8,630 subjects and 53 conditions, we demonstrated no cross-reactivity with other viral and bacterial infections, COVID-19 comorbidities, or confounders. In contrast, previously reported COVID-19 signatures were associated with significant cross-reactivity. The signature's interpretation, based on cell-type deconvolution and single-cell data analysis, revealed prominent yet complementary roles for plasmablasts and memory T cells. Although the signal from plasmablasts mediated COVID-19 detection, the signal from memory T cells controlled against cross-reactivity with other viral infections. This framework identified a robust, interpretable COVID-19 signature and is broadly applicable in other disease contexts. A record of this paper's transparent peer review process is included in the supplemental information.

Complementary evolution of coding and noncoding sequence underlies mammalian hairlessness.

Body hair is a defining mammalian characteristic, but several mammals, such as whales, naked mole-rats, and humans, have notably less hair. To find the genetic basis of reduced hair quantity, we used our evolutionary-rates-based method, RERconverge, to identify coding and noncoding sequences that evolve at significantly different rates in so-called hairless mammals compared to hairy mammals. Using RERconverge, we performed a genome-wide scan over 62 mammal species using 19,149 genes and 343,598 conserved noncoding regions. In addition to detecting known and potential novel hair-related genes, we also discovered hundreds of putative hair-related regulatory elements. Computational investigation revealed that genes and their associated noncoding regions show different evolutionary patterns and influence different aspects of hair growth and development. Many genes under accelerated evolution are associated with the structure of the hair shaft itself, while evolutionary rate shifts in noncoding regions also included the dermal papilla and matrix regions of the hair follicle that contribute to hair growth and cycling. Genes that were top ranked for coding sequence acceleration included known hair and skin genes KRT2, KRT35, PKP1, and PTPRM that surprisingly showed no signals of evolutionary rate shifts in nearby noncoding regions. Conversely, accelerated noncoding regions are most strongly enriched near regulatory hair-related genes and microRNAs, such as mir205, ELF3, and FOXC1, that themselves do not show rate shifts in their protein-coding sequences. Such dichotomy highlights the interplay between the evolution of protein sequence and regulatory sequence to contribute to the emergence of a convergent phenotype.

Whales, elephants, humans, and naked mole-rats all share a somewhat rare trait for mammals: their bodies are covered with little to no hair. The common ancestors of each of these species are considerably hairier which must mean that hairlessness evolved multiple times independently. When distantly related species evolve similar traits, it can be interpreted as a certain aspect of their evolution repeating itself. This process is called 'convergent evolution' and may provide insights about how different species were able to arrive at the same outcome. One possibility is that they have undergone similar genetic changes such as turning on or off key genes that play a role in the trait's development. Kowalczyk et al. set out to identify what genetic changes may have contributed to the convergent evolution of hairlessness in unrelated species of mammals. By looking at the genomes of 62 mammalian species, they hoped to link specific genomic elements to the origins of the hairless trait. The genetic sequences under investigation included nearly 20,000 genes that encode information about how to make proteins, as well as 350,000 regulatory sequences composed of non-coding DNA, which specify when and how genes are activated. This marks the first time genetic mechanisms behind various hair traits have been studied in such a diverse group of mammals. Using a computational approach, Kowalczyk et al. identified parts of the genome that have evolved similarly in mammalian species that have lost their hair. They found that genes and regulatory sequences, that had been previously associated with hair growth, accumulated mutations at significantly different rates in hairless versus hairy mammals. This indicates that these regions associated hair growth are also related to evolution of hairlessness. This includes several genes that encode keratin proteins, the main material that makes up hair. The team also reported an increased rate of evolution in genes and regulatory sequences that were not previously known to be involved in hair growth or hairlessness in mammals. Together these results suggest that a specific set of genetic changes have occurred several times in different mammalian lineages to drive the evolution of hairlessness in unrelated species. Kowalczyk et al. describe the parts of the genome that may be involved in controlling hair growth. Once their findings are validated, they could be used to develop treatments for hair loss in humans. Additionally, their computational approach could be applied to other examples of convergent evolution where genomic data is available, allowing scientists to better understand how the same traits evolve in different species.

Pre-infection antiviral innate immunity contributes to sex differences in SARS-CoV-2 infection.

Male sex is a major risk factor for SARS-CoV-2 infection severity. To understand the basis for this sex difference, we studied SARS-CoV-2 infection in a young adult cohort of United States Marine recruits. Among 2,641 male and 244 female unvaccinated and seronegative recruits studied longitudinally, SARS-CoV-2 infections occurred in 1,033 males and 137 females. We identified sex differences in symptoms, viral load, blood transcriptome, RNA splicing, and proteomic signatures. Females had higher pre-infection expression of antiviral interferon-stimulated gene (ISG) programs. Causal mediation analysis implicated ISG differences in number of symptoms, levels of ISGs, and differential splicing of CD45 lymphocyte phosphatase during infection. Our results indicate that the antiviral innate immunity set point causally contributes to sex differences in response to SARS-CoV-2 infection. A record of this paper's transparent peer review process is included in the supplemental information.

Re-analysis on the statistical sampling biases of a mask promotion trial in Bangladesh: a statistical replication.

A recent randomized trial evaluated the impact of mask promotion on COVID-19-related outcomes. We find that staff behavior in both unblinded and supposedly blinded steps caused large and statistically significant imbalances in population sizes. These denominator differences constitute the rate differences observed in the trial, complicating inferences of causality.

A Quantitative Systems Pharmacology Platform Reveals NAFLD Pathophysiological States and Targeting Strategies.

Non-alcoholic fatty liver disease (NAFLD) has a high global prevalence with a heterogeneous and complex pathophysiology that presents barriers to traditional targeted therapeutic approaches. We describe an integrated quantitative systems pharmacology (QSP) platform that comprehensively and unbiasedly defines disease states, in contrast to just individual genes or pathways, that promote NAFLD progression. The QSP platform can be used to predict drugs that normalize these disease states and experimentally test predictions in a human liver acinus microphysiology system (LAMPS) that recapitulates key aspects of NAFLD. Analysis of a 182 patient-derived hepatic RNA-sequencing dataset generated 12 gene signatures mirroring these states. Screening against the LINCS L1000 database led to the identification of drugs predicted to revert these signatures and corresponding disease states. A proof-of-concept study in LAMPS demonstrated mitigation of steatosis, inflammation, and fibrosis, especially with drug combinations. Mechanistically, several structurally diverse drugs were predicted to interact with a subnetwork of nuclear receptors, including pregnane X receptor (PXR; NR1I2), that has evolved to respond to both xenobiotic and endogenous ligands and is intrinsic to NAFLD-associated transcription dysregulation. In conjunction with iPSC-derived cells, this platform has the potential for developing personalized NAFLD therapeutic strategies, informing disease mechanisms, and defining optimal cohorts of patients for clinical trials.