Detecting Sources of Transcriptional Heterogeneity in Large-Scale RNA-Seq Data Sets

Transcriptome diversity is a systematic source of variation in RNA-sequencing data

RNA sequencing has been widely used as an essential tool to probe gene expression. While standard practices have been established to analyze RNA-seq data, it is still challenging to interpret and remove artifactual signals. Several biological and technical factors such as sex, age, batches, and sequencing technology have been found to bias these estimates. Probabilistic estimation of expression residuals (PEER), which infers broad variance components in gene expression measurements, has been used to account for some systematic effects, but it has remained challenging to interpret these PEER factors. Here we show that transcriptome diversity-a simple metric based on Shannon entropy-explains a large portion of variability in gene expression and is the strongest known factor encoded in PEER factors. We then show that transcriptome diversity has significant associations with multiple technical and biological variables across diverse organisms and datasets. In sum, transcriptome diversity provides a simple explanation for a major source of variation in both gene expression estimates and PEER covariates.

Verifying explainability of a deep learning tissue classifier trained on RNA-seq data

For complex machine learning (ML) algorithms to gain widespread acceptance in decision making, we must be able to identify the features driving the predictions. Explainability models allow transparency of ML algorithms, however their reliability within high-dimensional data is unclear. To test the reliability of the explainability model SHapley Additive exPlanations (SHAP), we developed a convolutional neural network to predict tissue classification from Genotype-Tissue Expression (GTEx) RNA-seq data representing 16,651 samples from 47 tissues. Our classifier achieved an average F1 score of 96.1% on held-out GTEx samples. Using SHAP values, we identified the 2423 most discriminatory genes, of which 98.6% were also identified by differential expression analysis across all tissues. The SHAP genes reflected expected biological processes involved in tissue differentiation and function. Moreover, SHAP genes clustered tissue types with superior performance when compared to all genes, genes detected by differential expression analysis, or random genes. We demonstrate the utility and reliability of SHAP to explain a deep learning model and highlight the strengths of applying ML to transcriptome data.

Epigenetics of Muscle- and Brain-Specific Expression of KLHL Family Genes

KLHL and the related KBTBD genes encode components of the Cullin-E3 ubiquitin ligase complex and typically target tissue-specific proteins for degradation, thereby affecting differentiation, homeostasis, metabolism, cell signaling, and the oxidative stress response. Despite their importance in cell function and disease (especially, KLHL40, KLHL41, KBTBD13, KEAP1, and ENC1), previous studies of epigenetic factors that affect transcription were predominantly limited to promoter DNA methylation. Using diverse tissue and cell culture whole-genome profiles, we examined 17 KLHL or KBTBD genes preferentially expressed in skeletal muscle or brain to identify tissue-specific enhancer and promoter chromatin, open chromatin (DNaseI hypersensitivity), and DNA hypomethylation. Sixteen of the 17 genes displayed muscle- or brain-specific enhancer chromatin in their gene bodies, and most exhibited specific intergenic enhancer chromatin as well. Seven genes were embedded in super-enhancers (particularly strong, tissue-specific clusters of enhancers). The enhancer chromatin regions typically displayed foci of DNA hypomethylation at peaks of open chromatin. In addition, we found evidence for an intragenic enhancer in one gene upregulating expression of its neighboring gene, specifically for KLHL40/HHATL and KLHL38/FBXO32 gene pairs. Many KLHL/KBTBD genes had tissue-specific promoter chromatin at their 5' ends, but surprisingly, two (KBTBD11 and KLHL31) had constitutively unmethylated promoter chromatin in their 3' exons that overlaps a retrotransposed KLHL gene. Our findings demonstrate the importance of expanding epigenetic analyses beyond the 5' ends of genes in studies of normal and abnormal gene regulation.

TGFB1-Mediated Gliosis in Multiple Sclerosis Spinal Cords Is Favored by the Regionalized Expression of HOXA5 and the Age-Dependent Decline in Androgen Receptor Ligands

In multiple sclerosis (MS) patients with a progressive form of the disease, spinal cord (SC) functions slowly deteriorate beyond age 40. We previously showed that in the SC of these patients, large areas of incomplete demyelination extend distance away from plaque borders and are characterized by a unique progliotic TGFB1 (Transforming Growth Factor Beta 1) genomic signature. Here, we attempted to determine whether region- and age-specific physiological parameters could promote the progression of SC periplaques in MS patients beyond age 40. An analysis of transcriptomics databases showed that, under physiological conditions, a set of 10 homeobox (HOX) genes are highly significantly overexpressed in the human SC as compared to distinct brain regions. Among these HOX genes, a survey of the human proteome showed that only HOXA5 encodes a protein which interacts with a member of the TGF-beta signaling pathway, namely SMAD1 (SMAD family member 1). Moreover, HOXA5 was previously found to promote the TGF-beta pathway. Interestingly, SMAD1 is also a protein partner of the androgen receptor (AR) and an unsupervised analysis of gene ontology terms indicates that the AR pathway antagonizes the TGF-beta/SMAD pathway. Retrieval of promoter analysis data further confirmed that AR negatively regulates the transcription of several members of the TGF-beta/SMAD pathway. On this basis, we propose that in progressive MS patients, the physiological SC overexpression of HOXA5 combined with the age-dependent decline in AR ligands may favor the slow progression of TGFB1-mediated gliosis. Potential therapeutic implications are discussed.

Resolving the fibrotic niche of human liver cirrhosis at single-cell level

Liver cirrhosis is a major cause of death worldwide and is characterized by extensive fibrosis. There are currently no effective antifibrotic therapies available. To obtain a better understanding of the cellular and molecular mechanisms involved in disease pathogenesis and enable the discovery of therapeutic targets, here we profile the transcriptomes of more than 100,000 single human cells, yielding molecular definitions for non-parenchymal cell types that are found in healthy and cirrhotic human liver. We identify a scar-associated TREM2+CD9+ subpopulation of macrophages, which expands in liver fibrosis, differentiates from circulating monocytes and is pro-fibrogenic. We also define ACKR1+ and PLVAP+ endothelial cells that expand in cirrhosis, are topographically restricted to the fibrotic niche and enhance the transmigration of leucocytes. Multi-lineage modelling of ligand and receptor interactions between the scar-associated macrophages, endothelial cells and PDGFRα+ collagen-producing mesenchymal cells reveals intra-scar activity of several pro-fibrogenic pathways including TNFRSF12A, PDGFR and NOTCH signalling. Our work dissects unanticipated aspects of the cellular and molecular basis of human organ fibrosis at a single-cell level, and provides a conceptual framework for the discovery of rational therapeutic targets in liver cirrhosis.

Beyond Genomic Prediction: Combining Different Types of omics Data Can Improve Prediction of Hybrid Performance in Maize

The ability to predict the agronomic performance of single-crosses with high precision is essential for selecting superior candidates for hybrid breeding. With recent technological advances, thousands of new parent lines, and, consequently, millions of new hybrid combinations are possible in each breeding cycle, yet only a few hundred can be produced and phenotyped in multi-environment yield trials. Well established prediction approaches such as best linear unbiased prediction (BLUP) using pedigree data and whole-genome prediction using genomic data are limited in capturing epistasis and interactions occurring within and among downstream biological strata such as transcriptome and metabolome. Because mRNA and small RNA (sRNA) sequences are involved in transcriptional, translational and post-translational processes, we expect them to provide information influencing several biological strata. However, using sRNA data of parent lines to predict hybrid performance has not yet been addressed. Here, we gathered genomic, transcriptomic (mRNA and sRNA) and metabolomic data of parent lines to evaluate the ability of the data to predict the performance of untested hybrids for important agronomic traits in grain maize. We found a considerable interaction for predictive ability between predictor and trait, with mRNA data being a superior predictor for grain yield and genomic data for grain dry matter content, while sRNA performed relatively poorly for both traits. Combining mRNA and genomic data as predictors resulted in high predictive abilities across both traits and combining other predictors improved prediction over that of the individual predictors alone. We conclude that downstream "omics" can complement genomics for hybrid prediction, and, thereby, contribute to more efficient selection of hybrid candidates.

The effects of death and post-mortem cold ischemia on human tissue transcriptomes

Post-mortem tissues samples are a key resource for investigating patterns of gene expression. However, the processes triggered by death and the post-mortem interval (PMI) can significantly alter physiologically normal RNA levels. We investigate the impact of PMI on gene expression using data from multiple tissues of post-mortem donors obtained from the GTEx project. We find that many genes change expression over relatively short PMIs in a tissue-specific manner, but this potentially confounding effect in a biological analysis can be minimized by taking into account appropriate covariates. By comparing ante- and post-mortem blood samples, we identify the cascade of transcriptional events triggered by death of the organism. These events do not appear to simply reflect stochastic variation resulting from mRNA degradation, but active and ongoing regulation of transcription. Finally, we develop a model to predict the time since death from the analysis of the transcriptome of a few readily accessible tissues.

RNA levels in post-mortem tissue can differ greatly from those before death. Studying the effect of post-mortem interval on the transcriptome in 36 human tissues, Ferreira et al. find that the response to death is largely tissue-specific and develop a model to predict time since death based on RNA data.

Omics-based hybrid prediction in maize

Complementing genomic data with other "omics" predictors can increase the probability of success for predicting the best hybrid combinations using complex agronomic traits. Accurate prediction of traits with complex genetic architecture is crucial for selecting superior candidates in animal and plant breeding and for guiding decisions in personalized medicine. Whole-genome prediction has revolutionized these areas but has inherent limitations in incorporating intricate epistatic interactions. Downstream "omics" data are expected to integrate interactions within and between different biological strata and provide the opportunity to improve trait prediction. Yet, predicting traits from parents to progeny has not been addressed by a combination of "omics" data. Here, we evaluate several "omics" predictors-genomic, transcriptomic and metabolic data-measured on parent lines at early developmental stages and demonstrate that the integration of transcriptomic with genomic data leads to higher success rates in the correct prediction of untested hybrid combinations in maize. Despite the high predictive ability of genomic data, transcriptomic data alone outperformed them and other predictors for the most complex heterotic trait, dry matter yield. An eQTL analysis revealed that transcriptomic data integrate genomic information from both, adjacent and distant sites relative to the expressed genes. Together, these findings suggest that downstream predictors capture physiological epistasis that is transmitted from parents to their hybrid offspring. We conclude that the use of downstream "omics" data in prediction can exploit important information beyond structural genomics for leveraging the efficiency of hybrid breeding.