Research Roundup: April 15, 2022

Welcome to the April 15, 2022 installment of Research Roundup, a recurring snapshot of recent studies published by scientists at the Broad Institute and their collaborators.

A watershed moment for schizophrenia research

In a landmark 121,000-subject study published in Nature, TJ Singh, institute member Ben Neale, institute member Mark Daly, and colleagues in the international SCHEMA Consortium have identified ultra-rare protein-disrupting mutations in 10 genes that strongly increase an individual's risk of developing schizophrenia. The findings complement those of a 320,000-person genome-wide association study from the Psychiatric Genomics Consortium, also published in Nature, and underscore an emerging view of schizophrenia as a breakdown in communication at the synapse. As Morgan Sheng, co-director of the Stanley Center for Psychiatric Research explains in a Q&A, the results also provide actionable targets that can inform new therapeutics and diagnostic biomarkers. Learn more in a Broad press release and video, Singh's tweetorial, and coverage in the Boston Globe and Washington Post.

The first strong genetic risk factor for bipolar disorder

An international team led by Duncan Palmer, institute member Ben Neale, and others in the Stanley Center for Psychiatric Research and the Program in Medical and Population Genetics studied DNA from thousands of people with bipolar disorder to uncover a gene called AKAP11 as a robust risk factor for both bipolar disorder and schizophrenia. The findings may provide clues to how lithium improves symptoms for some. The results have already sparked new research focused on molecular mechanisms, which could one day reveal biomarkers to match patients with treatments and open doors to novel therapies. Read more in Nature Genetics and a Broad news story.

To infinity and beyond

In the first scientific paper to come from the Eric and Wendy Schmidt Center, core institute member Caroline Uhler, Adit Radhakrishnan, and others built a machine learning framework for addressing complex matrix completion tasks. Based on infinitely large neural networks, their method is simple, fast, and flexible, and can be easily deployed on a standard laptop for a range of tasks. They demonstrated the method by performing virtual drug screening using Connectivity Map data and filling in missing areas of digital images, with proficiency rivaling more computationally costly methods. Read more in PNAS and a Broad news story.

Taking advantage of antibiotic resistance enzymes

Broad-spectrum antibiotics can disrupt a patient's gut microbiome, fueling inflammatory problems, secondary infections, and antibiotic resistance. To protect the microbiome, Andrés Cubillos-Ruiz, research scientist Julian Avila-Pacheco, institute member Jim Collins, and colleagues turned a strain of Lactococcus lactis (widely used in cheese, yogurt, and sauerkraut production) into a beta-lactamase-producing engineered live biotherapeutic product (eLBP). When fed to ampicillin-treated mice, the eLBP helped maintain the gut microbiome without affecting serum antibiotic levels and guarded against colonization by opportunistic bugs. The team's use of a two-gene strategy to encode beta-lactamase kept the engineered bacteria from sharing the enzyme with other species. Learn more in Nature Biomedical Engineering (paywall), a tweetorial by Cubillos-Ruiz, and an MIT News story.

A look at lactation

Accessing human mammary gland cells during lactation is a challenge, but recent work has shown that breast milk may provide a noninvasive way to access these cells. In light of this, Sarah Nyquist, institute member Alex Shalek, associate member Bonnie Berger, Brittany Goods (Dartmouth), and their team collected breast milk from 15 nursing mothers and analyzed the samples using single-cell RNA-sequencing. They uncovered various cell types, including fibroblast cells, epithelial cells, and immune cells. Most abundant were lactocytes, a type of epithelial cell, which expressed genes for proteins found in breast milk and transporters needed to secrete breast milk components. This study offers cellular and molecular insights into the components and production of breast milk. Read more in PNAS and an MIT News story.

Synthetic switches sway cellular signaling

Molecular switches that turn particular proteins or pathways on or off can help researchers trace and understand cellular signaling networks, and can also have applications in synthetic biology. Mikołaj Słabicki teamed up with Lena Nitsch and Patrizia Jensen of DFKZ and colleagues to engineer BTBBCL6, a domain of the BCL6 transcription factor, into a reversible, compound-triggered molecular switch. Fusion proteins made of BTBBCL6 and the EGFR membrane receptor induce strong downstream activity and cell growth when turned on or off with a small molecule-controlled polymerization switch. Learn more in Cell Reports Methods and a tweetorial by Słabicki.

The key to CAR T-cell therapy in solid tumors

Chimeric antigen receptor (CAR) T-cell therapy has transformed the treatment of some blood cancers but hasn’t worked as well for solid tumors. To look for possible resistance pathways, Rebecca Larson, associate member Marcela Maus (Massachusetts General Hospital), and colleagues conducted a genome-wide CRISPR knockout screen in glioblastoma. They found that the loss of genes in the interferon-γ receptor (IFNγR) signaling pathway made glioblastoma and other solid tumors more resistant to CAR T-cell killing in vitro and in vivo, by reducing binding of CAR T cells. Moreover, IFNγR signaling was required for sufficient CAR T-cell binding and killing in glioblastoma. The findings suggest that boosting CAR T-cell binding could make solid tumors more responsive to this therapy. Read more in Nature. 

Droids and JEDI: A new hope for addressing bias in EHRs

Electronic health record (EHR) databases are large-scale tools that aggregate data including laboratory images, risk factors, and notes. EHRs can provide valuable statistically-powered clinical insight but may be subject to bias based on patient selection, misclassification, and data acquisition. In Nature Digital Medicine, Shaan Khurshid, Chris Reeder, Puneet Batra of the Data Sciences Platform, associate member Steven Lubitz of the Program in Medical and Population Genetics, and colleagues report the Community Care Cohort Project (C3PO), an EHR-based study structured to reduce bias and missing data in cardiovascular disease research. The team used natural language processing, a machine learning model that processes language, to recover missing data. JEDI, an open-source pipeline for processing EHR data, could help researchers develop accurate predictive and classification models using EHR data.

Incorporating human genetics data in animal model studies

Peter Dornbos, Preeti Singh, Dong-Keun Jang, associate member Jason Flannick, and their team recently published a forum article in Cell Metabolism discussing how much human genetics are used to assess translation potential from animal models to humans. After reviewing three years’ worth of publications from top-tier journals, they found that very few experimental studies incorporate human genetic data. They propose publicly available genetic associations to determine what they call a Human Genetics Evidence (HuGE) score to determine the probability of true association. The team hopes that their HuGE calculator tool will help researchers incorporate human genetic data into their studies on model organisms.

Closer look at genome loops

Animal genomes are folded into loops and domains by the proteins CTCF and cohesin. These loops and domains play critical roles in multiple nuclear processes, including regulation of gene expression and DNA repair. However, whether these loops are stable or dynamic was unknown. Michele Gabriele, Hugo Brandão, Simon Grosse-Holz (MIT), Leonid Mirny (MIT), Christoph Zechner (Max Planck Institute), associate member Anders Hansen of the Epigenomics Program, and colleagues used super resolution live-cell imaging to study these loops in mice stem cells. Their findings suggest that these loops are very dynamic and shorter-lived than previously thought. Read more in Science and an MIT news story.