Relating enhancer genetic variation across mammals to complex phenotypes using machine learning

Abstract

Protein-coding differences between species often fail to explain phenotypic diversity, suggesting the involvement of genomic elements that regulate gene expression such as enhancers. Identifying associations between enhancers and phenotypes is challenging because enhancer activity can be tissue-dependent and functionally conserved despite low sequence conservation. We developed the Tissue-Aware Conservation Inference Toolkit (TACIT) to associate candidate enhancers with species’ phenotypes using predictions from machine learning models trained on specific tissues. Applying TACIT to associate motor cortex and parvalbumin-positive interneuron enhancers with neurological phenotypes revealed dozens of enhancer–phenotype associations, including brain size–associated enhancers that interact with genes implicated in microcephaly or macrocephaly. TACIT provides a foundation for identifying enhancers associated with the evolution of any convergently evolved phenotype in any large group of species with aligned genomes.

Evolutionary constraint and innovation across hundreds of placental mammals

“…Zoonomia realizes that vision. In combination with existing resources such as ENCODE 38 and TOPMed 39 and new technologies including regulatory genomic assays 94, 97 , chromatin conformation capture 27, 51 , and massively parallel reporter assays 36 , Zoonomia’s alignments of hundreds of placental species address fundamental questions in mammalian evolution.…”

“…We developed a forward genomics toolkit for associating cis -regulatory elements with specific phenotypes (detailed in 94 ). This Tissue-Aware Conservation Inference Toolkit (TACIT) uses open chromatin sequence data available for a few species to predict open chromatin (a proxy for cis -regulatory element activity) in many species at open chromatin region orthologs found by the Cactus alignment( Fig.…”

Vocal learning–associated convergent evolution in mammalian proteins and regulatory elements

“…We used the predictions from convolutional neural networks (CNNs) trained by using DNA sequence–based M1 open chromatin data obtained in this study for R. aegyptiacus with ATAC-seq and collected earlier for the mouse ( 21 ), rat, and macaque ( 20 ) to predict motor cortex open chromatin across 222 mammalian genomes (materials and methods) ( 23 ). Given that parvalbumin has been shown to be a shared marker of brain areas critical for vocal learning in songbirds and humans ( 4 ), we also used CNNs trained to predict cell type–specific open chromatin by using ATAC-seq data from mouse and human M1 parvalbumin-positive neurons (M1-PV+) ( 22 , 23 , 71 ). We identified regions whose predicted open chromatin was consistently lower or higher in vocal learners relative to vocal nonlearners by using phylogenetic logistic regression ( 72 , 73 ) with phylogenetic permutations ( 24 ) [permulations adj.…”

“…We therefore sought to extend our search for cis-regulatory elements whose evolution is associated with vocal learning behavior using a recently developed machine learning approach, TACIT (23). Given that it is infeasible to map the brains and collect motor cortex tissue from each vocal learning and closely related nonlearning species, the TACIT approach uses machine learning models (23) to predict motor cortex open chromatin across orthologous regions of the genome (66)(67)(68). TACIT then associates predictions with vocal learning in a way that corrects for phylogenetic relationships (Fig.…”

“…We accomplished this by combining anatomical tracing and electrophysiological recordings in vocalizing bats to identify a region of the motor cortex associated with vocal production. The vocalization-associated epigenomic data collected from this bat species, combined with hundreds of mammalian genomes ( 17 , 18 ), their associated reference-free whole-genome alignments ( 19 ), and high-quality epigenomic data from the motor cortex of multiple additional mammalian species ( 20 – 22 ), provided the foundation to apply a machine learning approach, the Tissue-Aware Conservation Inference Toolkit (TACIT) ( 23 ). This approach allowed us to identify putative enhancers, distal regulatory elements that tend to be highly tissue specific, associated with the convergent evolution of vocal learning.…”