Amyotrophic lateral sclerosis (ALS) brutally targets nerve cells. The disease first causes progressive muscle weakness and, typically within three years, results in paralysis and death. Baseball player Lou Gehrig and scientist Stephen Hawking both contracted ALS, and today over 200,000 people worldwide live with the disease. While scientists have linked several genes to ALS, core questions remain about what sets off the cascade of neuron damage and how the disease progresses.
The answers to such questions may be within reach, thanks to a pioneering approach in which researchers are able to examine whole slices of spinal cord tissue — throughout the course of the disease — and to study how the various cell types present there interact and contribute to the disease’s progression. And for the first time, technology exists that allows researchers to see gene expression patterns at a high resolution throughout the spinal cord.
“We use sophisticated computational methods to study human disease,” says Tarmo Äijö, a data scientist at the Flatiron Institute’s Center for Computational Biology (CCB) and co-lead author on a study published in the April 5, 2019, issue of Science. Silas Maniatis of the New York Genome Center and Sanja Vickovic of the Broad Institute of the Massachusetts Institute of Technology and Harvard University and the New York Genome Center co-led the study with Äijö.
In this work, Vickovic and colleagues from the KTH Royal Institute of Technology in Stockholm together with Joakim Lundeberg’s group at Science for Life Laboratory in Solna, Sweden, applied novel technology they developed for studying spatial gene expression. This technology, along with new computational modeling techniques, enabled the researchers to study the co-expression of different groups of genes throughout the diseased spinal cord and, for the first time, see how the location and extent of gene expression changed as the disease progressed.
Spinal cord tissue from control mice and mouse models of ALS, taken over time as the symptoms progressed, were mounted onto glass slides, each covered with 1,007 tiny spots. The spots contained molecules that captured the tissue’s mRNA; mRNA is used as a measure of gene expression levels. The captured mRNA was then copied and embedded with unique identifiers that recorded its spatial location in the tissue. The researchers then analyzed all the spots together, with gene expression results tied back to the mRNA’s original location within the tissue.
The researchers repeated the process with post-mortem human spinal cord tissue samples from seven ALS patients.
With over 130,000 spatial gene expression measurements from about 1,300 slices of spinal cord tissue, the study far surpassed the depth and scale of the next largest comparable study, which included only a dozen tissue sections from a single time point. “We had multiple time points, genotypes, patients and animals: This was the first use of spatial gene expression analysis at scale,” says Maniatis.
“We know now ‘neighborhood matters’ in ALS,” says Hemali Phatnani, director of the Center for Genomics of Neurodegenerative Disease at the New York Genome Center, referring to how non-neuronal cells can affect neurons’ vulnerability.
Incorporating this spatial component into the data analysis presented significant difficulty. Leveraging his prior work on a time-series analysis of the microbiome, Äijö developed a model for the data in collaboration with Maniatis and Vickovic, which went through multiple iterations and was guided by biology-driven questions: What are the dynamics of ALS pathology across time and space? How does the disease start and spread? And why are only motor neurons impacted? Aiming for a balance between resolution and statistical power, the group settled on defining 11 regions of the spinal cord to guide the spatial interpretation of the data.
The results suggested that microglia — specialized immune cells of the central nervous system that remove damaged or dead cells — show dysfunction near motor neurons even before ALS symptoms begin. When the researchers looked at an array of genes known to be affected in neurodegenerative disease, they saw a sequence of gene expression changes unfold and, for the first time, could see the order in which gene changes occurred. For example, in the mice, an increase in expression of one of the genes, Tyrobp, occurred before symptom onset in particular regions of the spinal cord. It was followed by an increase in expression of the gene Trem2 in the same regions. The expression of both genes increased further as the symptoms progressed. These temporal snapshots are the closest researchers have to real-time video of ALS progression in the spinal cord.
The researchers also identified 31 ‘co-expression modules,’ groups of genes in the spinal cord with similar expression profiles in space and time. “Within the modules, we see the genes in glial cell subpopulations behaving differently at different places in the spinal column, and they seem to be acting in concert,” says Phatnani. The researchers found that in some gene groups, the natural cleaning up of damaged cells — known as autophagy — increased, diminishing the impact of ALS. In other modules, though, autophagy worsened ALS’s impact. “We can now generate testable hypotheses about how autophagy impacts the progression of the disease,” Phatnani says. While the researchers could not study multiple disease time points in the human samples, a spatial relationship between the site of symptom onset and the locations of gene-expression changes in the spine post-mortem was still evident.
Regarding the sheer scale of the data, Maniatis says, “We are continuing to find things in this vast dataset.” He’s most excited, though, by the contribution that this data resource makes to the broader scientific community, via an interactive data browser developed by Äijö that specialists and nonspecialists alike can quickly and easily interrogate.
“In high-throughput biology, genomics and imaging haven’t had a way of connecting until now,” says Richard Bonneau, group leader for systems biology at the CCB. In complicated tissue containing multiple cell types, images are a powerful addition to genomic data, but manual image analysis can be tedious and rate limiting. Aidan Daly, a research fellow in systems biology at the CCB, is developing a two-stage machine learning tool to identify geographic regions in the most complicated tissue, like that found in the spinal cord and brain and in tumors lacking a stereotyped anatomy. “One of the reasons we’re targeting diseases of the central nervous system is because of the inherently challenging tissue. If this can work in these tissues, it can work in any tissues,” says Bonneau.
Patterns in the data consistent with earlier ALS research provide proof of concept for mapping a disease in space and time. However, the identification of potential new ALS mechanisms, through the identification of co-expression modules over the entire spinal cord, adds a dimension of discovery to the work. The next step will be a larger-scale human study, in which the researchers aim to study both the brain and the spinal column, with the hope of uncovering spatially anchored regulatory networks involved in the disease. The ultimate goal is single-cell resolution, says Maniatis. “You would know without ambiguity what cell type is responsible for the signals in the data.”
The scientists agree that without computational and biological scientists interfacing early in the research, this result almost surely could not have occurred. Äijö recalls that hearing Maniatis give a talk at the Flatiron Institute first sparked his interest in solving the computational piece of the puzzle. “This work shows the value of enterprises like the Flatiron Institute and the New York Genome Center,” says Phatnani. “You have the technology, the biology, the computation and the people all together — the research would not have been possible in any other place.”
An interactive data exploration portal is publicly available at http://als-st.nygenome.org/.