Research Roundup: March 11, 2022

Welcome to the March 11, 2022 installment of Research Roundup, a recurring snapshot of recent studies published by scientists at the Broad Institute and their collaborators.

ADAPTing to viral variation

Omicron’s surge has shown the importance of diagnostic tests that detect viruses even when they mutate. Hayden Metsky, institute member Pardis Sabeti in the Infectious Disease and Microbiome Program (IDMP), and colleagues have developed ADAPT, the first fully automated system that uses machine learning to design viral diagnostics. ADAPT incorporates viral variation as well as new viral genomes from public databases to help researchers quickly create tests that are sensitive, specific, and can be modified as viruses evolve. The team used ADAPT to create CRISPR-based diagnostics for all 1,933 known vertebrate-infecting viruses, and their method can also be applied to other sequence-based techniques. Read more in Nature Biotechnology, a tweetorial by Metsky, and a Broad news story.

A window into how gene regulation evolves

Predicting naturally-arising and artificially-engineered noncoding variants' effects on gene expression and cell fitness is one of the holy grails of biology. Writing in Nature, Eeshit Dhaval Vaishnav, Carl de Boer (UBC), core institute member (on leave) Aviv Regev, and colleagues including Xian Adiconis and Joshua Levin present a neural network model that allows researchers to explore the "fitness landscape" of gene promoter variants, forecast promoters' future evolution, and design promoter sequences that change gene expression in desired ways. They trained the model based on the expression effects of hundreds of millions of randomly-generated promoter sequences in yeast. Learn more in a Nature News & Views review, an MIT/Broad story, a tweetorial by Vaishnav, and coverage in Genetic Engineering & Biotechnology News.

Genetic insights into heart attack risk 

Spontaneous coronary artery dissection (SCAD) arises when a tear forms in a blood vessel in the heart and is a common cause of heart attack, especially for women under 50. To understand which genes contribute to SCAD risk, Seyedeh Maryam Zekavat, Kaavya Paruchuri, Satoshi Koyama, Md Mesbah Uddin, James Pirruccello, associate member Pradeep Natarajan of the Program in Medical and Population Genetics, associate member Mark Lindsay of the Cardiovascular Disease Initiative, and colleagues compared the exomes of 130 individuals with SCAD with those of 46,468 people without. They identified rare genetic variants in fibrillar collagen genes associated with SCAD risk. The findings could help lead to therapeutic and prevention strategies for at-risk groups. Read more in JAMA Cardiology and The Harvard Gazette.

Synthetic success

Over the course of their eons-long arms race with other microbes, bacteria have evolved numerous antibiotic compounds, many of which form the basis of antibiotics we use today. In Science, Kyan D'Angelo (MIT), Carly Schissel (MIT), Chemical Biology and Therapeutics Science Program associate member Bradley Pentelute, and Mohammad Movassaghi (MIT) report the successful chemical synthesis of himastatin, an antibiotic peptide with an unusual structure first isolated from Streptomyces bacteria. Their mastery of a key step in himastatin synthesis also allowed them to generate several derivatives of the compound, shedding light on its mechanism of action and opening the door to developing more potent versions. Learn more in an MIT News story.

A new take on teaching neural networks

Neural networks can extract massive amounts of information from electrocardiograms (ECGs), but generally require large numbers of annotated ECGs to be trained properly. Patient Contrastive Learning of Representations (PCLR, aka "Pickler") — an approach developed by Nathaniel Diamant, Erik Reinertsen (MGH), Steven Song (MGH), Aaron Aguirre (MGH), Collin Stultz (MGH/MIT), and Machine Learning for Health (ML4H) director Puneet Batra — generates biologically-informed representations of ECGs which researchers can use to train models when little annotated data is available or they lack the resources to train a network from scratch. In PLOS Computational Biology, the team showed that PCLR-trained networks make better predictions than those trained from scratch or using other pre-training approaches. Learn more in a tweetorial by Batra.

Revealing EBV’s evasive maneuvers

It’s unclear how the Epstein-Barr virus subverts host immune barriers to cause widespread, life-long infections. Using temporal proteomic maps, Stephanie Yiu (Harvard), associate member Benjamin Gewurz of Harvard Medical School and the IDMP, and others found that B cells lose the SMC5/6 cohesin complex upon viral infection and lytic reactivation. That loss depends on the major tegument protein BNRF1 — in its absence, SMC5/6 can sense viral DNA and block steps in viral reactivation. The results highlight BNRF1 and SMC5/6 as key to EBV host/pathogen interactions, indicate a novel mechanism by which EBV contributes to cancers including Hodgkin’s lymphoma, and suggest that all herpesviruses may need to counteract SMC5/6 to replicate. Read more in Cell Reports and a tweetorial by Gewurz.

Bacterial punch

Some Enterococcus species are leading causes of multi-drug resistant infections. Francois Lebreton, associate members Michael Gilmore and Jonathan Abraham, both in the IDMP and at Harvard Medical School, and colleagues have found that E. faecalis, E. faecium, and E. hirae produce pore-forming toxins, which help bacteria kill cells by forming transmembrane pores on cell surfaces. The team determined the toxins' structures and used CRISPR-Cas9 genome-wide screens to identify a receptor for two of them: MHC/HLA-I. They also showed in co-culture experiments that a toxin-producing E. faecium strain kills blood immune cells and damages intestinal organoids in a toxin-dependent manner. Read more in Cell.

Single-cell sequencing studies of schizophrenia

Mutations in the SETD1A gene are associated with schizophrenia. Studies suggest that loss-of-function mutations in just one copy of the gene are enough to cause pathology, but the molecular and cellular mechanisms underlying this relationship are unknown. In Science Advances, chief scientist emeritus Ed Scolnick of the Stanley Center for Psychiatric Research  partnered with Boston Children's Hospital's Renchao Chen, Yiqiong Liu, Mohamed Djekidel, and Yi Zhang and colleagues to study cell type-specific transcriptional changes in a Setd1a heterozygous mouse model using single-cell RNA sequencing. They observed variable transcriptional changes across different cells in the prefrontal cortex and striatum, and dysregulation of many genes that help regulate neuron morphogenesis and synaptic function. 

Exploring how bacteria get along

The human microbiome is home to bacteria from a variety of species as well as mixtures of strains within species, which are particularly difficult to study. To better understand bacterial strain mixtures, Lucas van Dijk, Bruce Walker, institute scientist Ashlee Earl of the IDMP, and their team developed the Strain Genome Explorer (StrainGE) toolkit. StrainGE allows users to detect and characterize low abundance strains using whole metagenomic sequencing data, at sequencing coverages as low 0.1x. The team demonstrate StrainGE’s efficacy on strains of E. coli and Enterococcus, and encourage users to apply it to all bacterial samples where same species dynamics are of interest. Read more about the tool in Genome Biology.

To learn more about research conducted at the Broad, visit broadinstitute.org/publications, and keep an eye on broadinstitute.org/news.