Research Roundup: April 2, 2021

Welcome to the April 2, 2021 installment of Research Roundup, a recurring snapshot of recent studies published by scientists at the Broad Institute and their collaborators.

Microbes' gene swaps speed up with industrialization

Bacterial species share genes for a variety of traits (antibiotic resistance, etc.) through horizontal gene transfer (HGT). The rate at which HGTs occur in the gut microbiome, and industrialization's impact on that rate, has been unclear. In Cell, Matthieu Groussin, Mathilde Poyet, and institute member Eric Alm of the Infectious Disease and Microbiome Program and collaborators in the international Global Microbiome Conservancy report that HGTs occur frequently within individuals — more so in industrialized settings — and that many have accumulated within populations over the last two to three human generations. They came to these conclusions after sequencing thousands of bacterial genomes sampled from people representing 15 populations spanning a range of industrialization. Learn more in an MIT news story.

Targeting cellular stress response in AML

Though many cancer drugs target proteins that directly drive tumor progression, these drugs don’t work for everyone, and tumors frequently develop resistance. Other crucial survival pathways, such as the cellular stress response pathway, have become attractive drug targets as a result. In Science Translational Medicine, Blandine Roux (Université de Paris), Camille Vaganay (Paris), Alexandre Puissant (Paris), institute member Kimberly Stegmaier, Lina Benajiba (Paris), and colleagues performed a pooled screen in a mouse model of acute myeloid leukemia (AML) to identify stress response genes necessary for AML survival. They identified the protein VCP, involved in DNA repair, as a potential target, for which they developed a drug that could decrease AML growth in mice.

A hotspot mutation in the driver’s seat

An IKZF3 hotspot mutation may drive chronic lymphocytic leukemia (CLL), but its function is unknown. A team led by Gregory Lazarian, Shanye Yin, Elisa ten Hacken, and institute member Catherine Wu of the Cancer Program, Dana-Farber Cancer Institute, and Harvard Medical School showed in a mouse model that the mutant gene disrupts DNA binding specificity and target selection, leading to CLL-like disease in elderly mice. Human tumor cells carrying the mutation have altered B cell receptor and NF-kappaB signaling, and reduced drug sensitivity. The work highlights IKZF3’s role as an oncogene via transcriptional dysregulation, and suggests combination therapy could help overcome drug resistance. Read more in Cancer Cell.

Hunting for biases in machine learning models

Training machine learning models using data with biases can lead to inaccurate performance and data interpretation. Fatma-Elzahraa Eid, Haitham Elmarakeby, Yujia Alina Chan, Nadine Fornelos, associate member Eliezer Van Allen of the Cancer Program, associate member Kasper Lage of the Stanley Center for Psychiatric Research, and colleagues have developed an approach to audit machine learning models in biology. They used this method to examine three models and identified unrecognized biases that reduced model performance on new datasets. The team concluded that models learn biases from data when signals from data the models are learning from are weak. The authors provide tools to tailor their auditing framework to other biomedical applications. Learn more from Communications Biology and Psychology Today.

Therapeutic target for breast cancer brain metastases

HER2+ breast cancer cells that metastasize to the brain are resistant to therapies that otherwise control disease at extracranial sites. Associate member Rakesh Jain, institute member Matthew Vander Heiden of the Cancer Program, Gino Ferraro (MGH), Ahmed Ali, Alba Luengo (MIT), and colleagues in the Metabolomics Platform and elsewhere studied how metabolism differs between breast tumors within and outside the brain and found that fatty acid synthesis is elevated in brain metastases. Their findings, reported in Nature Cancer, suggest that genetic and chemical inhibition of fatty acid synthase suppresses breast cancer growth in the brain, thus highlighting a potential therapeutic approach. 

Detecting genetic differences between psychiatric disorders

Psychiatric disorders are highly correlated, and as a result many studies have focused on their genetic similarities. Comparatively few studies have examined their differences, because such analyses would require collecting and matching individual-level case-case data. Postdoctoral scholar Wouter Peyrot and associate member Alkes Price of the Medical and Population Genetics Program have developed a new method called a case-case GWAS that looks for differences in allele frequencies between cases of two different disorders by analyzing their case-control GWAS summary statistics. In Nature Genetics, they apply case-case GWAS to data for eight psychiatric disorders and identify 196 loci with different allele frequency among cases, offering new biological insight into the differences between these disorders.

Setting benchmarks for structural variant detection in sequencing technologies 

Structural variants (genomic alterations greater than 50 base pairs in length) are more difficult to detect with short-read than long-read whole-genome sequencing (srWGS vs. lrWGS), but lrWGS is slower and more expensive. In The American Journal of Human Genetics, Xuefang Zhao, Harrison Brand, institute member Michael Talkowski of the Medical and Population Genetics Program, and colleagues compare structural variant detection from srWGS and lrWGS to establish expectations for and quantify unique advantages of each technology. They estimate that srWGS can detect over 10,000 structural variants in a human genome, but show that lrWGS offers superior detection of insertions and repeat-mediated variation.

Predicting cell health from image profiling

A team led by Gregory Way, Center for the Development of Therapeutics senior research scientist Maria Kost-Alimova, Imaging Platform senior director and institute scientist Anne Carpenter, Cancer Dependency Map Project associate director Francisca Vazquez, and Shantanu Singh, senior group leader also in the Imaging Platform, has developed two new microscopy assays to collectively measure 70 different indicators of cell health — including proliferation, apoptosis, DNA damage, and cell cycle stage. The researchers further paired the Cell Painting technique with these assays and determined that simple machine learning algorithms can predict many of the cell health readouts directly from Cell Painting images. They successfully validated predictions of cell health outcomes from a set of 1,500+ compound perturbations imaged using Cell Painting, and developed a web app to browse predictions. Learn more in Molecular Biology of the Cell.

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