A resolution revolution is happening to spatial transcriptomics. At this year’s Alzheimer’s Association International Conference, held July 31 to August 4 in San Diego, California, and online, Shane Liddelow of New York University presented stunning images of expression within mouse brain slices. Combining single-cell RNA sequencing and in situ hybridization to map gene expression in individual cells, his high-resolution method detected a small but specialized population of astrocytes that flock to the blood-brain barrier after inflammatory insult. He also zoomed in on expression changes around amyloid plaques, finding 400 plaque-induced genes, up from the almost 60 previously reported. The findings help researchers understand how dysfunction spreads from pathology to nearby cells to the entire brain.

  • A new algorithm mapped spatial transcriptomic gene expression in high-res.
  • It spied a subset of inflammatory astrocytes lining brain ventricles.
  • It detected expression of 403 plaque-associated genes with stunning clarity.

“Location dictates biology—how cells talk to each other, how they coordinate functions—especially in an organ as sophisticated as the brain,” Andrew Yang, University of California, San Francisco, told Alzforum. “Both single-cell and spatial transcriptomics provide the solid biological foundation that the dementia field needs by characterizing the diversity of cell types, states, and locations in different diseases.”

Bayesian Breakdown. The BayesSpace algorithm improves spatial transcriptomics resolution by dividing pixels (hexagons) into sections (bottom), then comparing the transcriptomes of each section (red) to its neighbors (blue). [Courtesy of Zhao et al., Nature Biotechnology, 2021.]

The need for high-res spatial transcriptomics was born from an experiment exploring astrocyte subtypes in mice. Philip Hasel of Liddelow’s lab had curated a massive dataset of about 80,000 astrocyte transcriptomes from wild-type mice that had been injected with lipopolysaccharide to stimulate acute inflammation (Hasel et al., 2021). The astrocytes fell into 10 subtypes. One, accounting for just 2 to 3 percent of all reactive astrocytes, included cells that were unusually interferon-responsive, suggesting provocation by peripheral inflammation. Curious about where these interferon-responsive reactive astrocytes (IRRAs) resided in the mouse brain, the scientists first used in situ hybridization to locate expression of Igtp, an interferon response gene expressed by this astrocyte subtype. They found it in outermost cortical layer next to Cldn5-positive blood vessels.

In situ hybridization only captures a small window of the brain and limits detection to a single marker. Liddelow wondered if spatial transcriptomics would offer a better brain-wide view of the IRRAs. First, the scientists spatially mapped approximately 200 genes differentially expressed in the IRRAs. Hotspots of these differentially expressed genes (DEGs) appeared around the outer surface of the brain and in what seemed to be the brain ventricles, but the resolution was too poor to precisely discern these locations.

Next, the researchers reanalyzed the image with BayesSpace, a recently created algorithm that uses Bayesian statistics to enhance image resolution (Zhao et al., 2021). Rather than comparing gene expression between the pixels of a spatial-transcriptomics picture, the algorithm divides each pixel into portions and compares expression between each portion, effectively reducing pixel size (see image above). This improved the resolution dramatically, clearly showing that IRRAs reside around ventricles and in the outer cortical layer (see image below).

High Definition. To get a higher-resolution spatial transcriptomics image of IRRAs (yellow), researchers progressed from searching for just Igtp expression (top left) to all 200 IRRA DEGs (top right). Reanalyzing the images with BayesSpace vastly improved resolution of Igtp (bottom left) and DEG (bottom right) expression. [Courtesy of Shane Liddelow, New York University.]

Liddelow thinks the location of IRRAs at the blood vessels and ventricles suggests that they are responding to peripheral immune cells that secrete interferon. IRRAs also express genes involved in antigen processing and presentation, which may beckon other immune cells to the blood-brain barrier. Upregulation of similar peripheral sirens have been reported in other types of inflammatory astrocytes (Jun 2022 news). “We think the strategic location of IRRAs also hint that they are integral to the infiltration of peripheral immune cells,” Liddelow said. In some neurodegenerative diseases, T cells can pass through blood-vessel walls and hang out in the glia limitans, the thin layer of astrocytic feet that coat the cortex, or enter the cerebrospinal fluid by infiltrating the choroid plexus (Oct 2021 news; Jun 2021 news).

Yang agreed that these astrocytes may be sentinels of the brain, and was struck by their exquisite localization. Intriguingly, Hasel and colleagues also detected IRRAs in snRNA-Seq data from mouse models of amyloidosis and multiple sclerosis, suggesting that peripheral inflammation also drives astrocyte reactivity in those diseases.

Liddelow is tracking down similar cells in people. Jessica Sadick and Michael O’Dea in his lab ran RNA-Seq of single cells from 10 postmortem AD and six control prefrontal cortex samples, isolating 41,000 astrocytes and clustering them by their transcriptomes into eight subpopulations (Sadick et al., 2022). In a control who had vascular dementia, as determined by neuropathology and clinical evaluations, one cluster had a similar gene-expression profile to the IRRAs found in mice. The scientists are currently collecting more data and working on high-res human brain tissue spatial transcriptomics.

Plaque Transcriptomics
Could high-res images clarify the location of other gene-expression changes within the brain, such as those surrounding amyloid plaques? Previously, researchers led by Bart De Strooper, UK Dementia Research Institute, London, identified 57 plaque-induced genes by spatial transcriptomics. These PIGs were up- or downregulated in microglia and astrocytes surrounding amyloid plaques in old APP knock-in mice and in people who had had AD (Jul 2020 news; Aug 2019 news).

Hasel and Emilie Castranio of Mount Sinai, New York, took a similar approach, collecting single-cell and spatial transcriptomics data on brain tissue of 6-month-old PS/APP transgenic mice (Castranio et al., 2022). But they analyzed the data differently, using the spatial prowess bestowed by BayesSpace analysis. De Strooper clustered genes based on weighted gene co-expression network analysis (WGCNA), which groups genes that are up- or downregulated in concert. Hasel and Castranio grouped gene expression based on proximity. They compared gene expression between cells from wild-type and PS/APP mice to identify DEGs, then clustered pixels of the spatial transcriptomics image based on the DEGs expressed in each pixel.

Clustering of the DEGs revealed 27 subtypes of pixels. Each occupied a unique spatial niche in the brain. One subtype seamlessly overlapped with plaques labeled by the antibody 6E10 (see image below). Cells within these pixels shared 403 DEGs, which the authors named newPIGs.

High-Res PIGs. Combining snRNA-Seq and spatial transcriptomics of brain tissue from APP mice pinpoints DEGs surrounding plaques (red/yellow pixels, left). BayesSpace (middle) improves the resolution, showing PIGs (yellow) coinciding with plaques (red, right). [Courtesy of Shane Liddelow, New York University.]

What were these newPIGs, and which cells expressed them? In addition to the 57 original PIGs, many of the remainder are expressed in IRRAs and disease-associated microglia (DAMs). Wei-Ting Chen of KU Leuven, first author of the original PIGs paper, was pleased to see that the same genes she had identified in her study turned up here, too. Liddelow noted that the strong co-localization with PIG genes hint that DAM microglia and IRRA astrocytes interact around plaques (see image below).

Do PIGs Give a DAM? High-res spatial transcriptomics show co-localized expression of PIG (left), DAM (middle), and IRRA (right) genes around plaques. [Courtesy of Castranio et al., 2022, bioRxiv.]

One newPIG stood out to the scientists because of its stark specificity for plaques. CST7, which encodes the protease inhibitor cystatin F, was expressed in every pixel surrounding plaques and none away from plaques. This gene was also the most upregulated of the previously identified PIGs, Chen noted. Liddelow thinks it may be a novel plaque marker.

All told, high spatial resolution clarifies relationships between pathology and gene expression, affording a clearer understanding of disease biology. Another beauty of BayesSpace is that any researcher can use it right now—they just need to plug existing spatial transcriptomics data into the algorithm to get high-resolution images. “The most amazing thing is that we didn’t do anything different in how we generated the data; we just changed how we analyzed it,” Liddelow said. He encourages other researchers to use his lab’s spatial transcriptomics data, including the new set from the PS/APP mice.—Chelsea Weidman Burke


No Available Comments

Make a Comment

To make a comment you must login or register.


News Citations

  1. Astrocyte Reactivity: Opposing States Emerge
  2. Intruder Alert: Inflammatory T Cells Lurk Near Lewy Bodies, Neurons
  3. COVID-19 Prompts Choroid Plexus to Ring Alarm Bell
  4. Paper Alert: Those PIGs! Spatial Transcriptomics Add Human Data
  5. Spatial Transcriptomics Uncovers Coordinated Cell Responses to Amyloid

Research Models Citations

  1. PS/APP

Paper Citations

  1. . Neuroinflammatory astrocyte subtypes in the mouse brain. Nat Neurosci. 2021 Oct;24(10):1475-1487. Epub 2021 Aug 19 PubMed.
  2. . Spatial transcriptomics at subspot resolution with BayesSpace. Nat Biotechnol. 2021 Nov;39(11):1375-1384. Epub 2021 Jun 3 PubMed.
  3. . Astrocytes and oligodendrocytes undergo subtype-specific transcriptional changes in Alzheimer's disease. Neuron. 2022 Jun 1;110(11):1788-1805.e10. Epub 2022 Apr 4 PubMed.
  4. . INPP5D limits plaque formation and glial reactivity in the APP/PS1 mouse model of Alzheimer’s disease. bioRxiv, April 30, 2022 bioRxiv

External Citations

  1. spatial transcriptomics data
  2. PS/APP mice

Further Reading

Primary Papers

  1. . Astrocytes and oligodendrocytes undergo subtype-specific transcriptional changes in Alzheimer's disease. Neuron. 2022 Jun 1;110(11):1788-1805.e10. Epub 2022 Apr 4 PubMed.