09 Apr 2022

Hundreds, if not thousands, of transcripts and proteins get up- or downregulated in Alzheimer’s disease, but how this broadly affects the brain remains a mystery. Scientists are trying to solve it by simultaneously isolating millions of single cells or single nuclei from brain samples and scrutinizing patterns that correlate with disease traits. Such studies illuminate the cellular state of the brain, but not how it got there.

At this year’s AD/PD meeting, held March 15-20 in Barcelona, Spain, Philip de Jager, Columbia University, New York, described an unbiased look at how communities of cells change during age and disease. It turns out some cell types wax and wane in unison, and some of their shifts correlate with amyloid plaques, neurofibrillary tangles, and cognitive decline.

Also working to figure out what governs en masse changes, Ricardo D’Oliveira Albanus, a postdoctoral fellow in the lab of Oscar Harari at Washington University, St. Louis, analyzed ligand-receptor pairs to infer thousands of cell-cell interactions. Albanus predicted that the number of liaisons between cell types in the brain changes as AD worsens, with most of those changes involving microglia. Many of the proteins involved have cropped up in genome-wide association studies. He linked some of these proteins to signaling networks that correlated with pathology, but more correlated with resilience. “This could help us find new drug targets that slow or prevent AD,” said Albanus.

Together, the two approaches exemplify the power of merging single-cell and interactome analyses in scientists' quest to understand the cellular phases of AD and related diseases.

Cells of a Feather
Unlike birds, it turns out that cells in the AD brain not only flock with their own kind, but with other cells as well. That’s what de Jager, collaborating with Vilas Menon at Columbia and Naomi Habib, Hebrew University of Jerusalem, discovered when they looked at how the amounts of different cells change over time. The scientists counted cells in prefrontal cortex tissue taken postmortem from volunteers in the Religious Orders Study and the Memory and Aging Project at Rush University, Chicago (see image below). Some of the data was recently posted on bioRxiv (Cain et al., 2020). 

Cell Tracking. Single-nuclei RNA-Seq of 24 ROSMAP samples (top) yielded high-resolution transcriptome data that could be used to identify proportions of brain cells in a larger sample of 640 individuals (bottom). [Courtesy of Cain et al., bioRxiv 2022.]

First, the scientists ran a pilot study of 24 ROSMAP participants, from whom they obtained more than 170,000 single-nuclei transcriptomes total. They reflect all the eight major cell subtypes of the brain, though the proportions were not the same in each person. Indeed, the relative frequency of certain cell types seemed highly coordinated. For example, when the number of microglia was high in a given person, so was that of oligodendrocytes, and vice versa. “We could see that there was structure in the data,” said de Jager (see image below).

Two main groups of cells seemed to fluctuate in unison. One comprised microglia, oligodendrocytes, pericytes, and endothelial cells; the other included astrocytes, oligodendrocyte precursor cells, and inhibitory neurons. Curiously, excitatory neuron counts did not track with any other cell type.

Because two dozen samples harbor little statistical power, the scientists next turned to bulk RNA-Seq data from 640 people. Menon devised a method called CelMod, aka Cellular Landscapes Modeling by Deconvolution. It infers cell type and subtype from snRNA-Seq data. CelMod first matches snRNA-Seq and bulk RNA-Seq from the same person, and then uses that benchmark to infer cell types and subtypes for other samples where only bulk data is available. The scientists found cells behaved similarly in this larger data set, with the two main groups still changing in unison, although they were now able to identify 37 different cell subtypes (see image above).

Do these cellular undulations correlate with AD pathology? On this question, two things stood out. First, cell type compositions that correlated with cognitive decline and neurofibrillary tangle pathology were very similar to each other, but distinct from cells that correlated with amyloid plaques. Second, while some cell numbers bloomed with these pathology traits, others, often of subsets of the same cell type, shrank. For example, one subset of astrocytes expanded in the presence of tau pathology or cognitive impairment, whereas another subset of astrocytes waned. Neither subset correlated with plaques. When plaques were present, a subset of inhibitory neurons swelled in number, whereas subsets of endothelial cells and excitatory neurons dwindled. In this analysis, none of the cells that correlated with plaques seemed to be affected by tangles or cognition.

The scientists took their work a step further, obtaining snRNA-Seq data from 424 ROSMAP participants. This yielded 1.7 million transcriptomes, identifying 92 subsets of cells. With this higher resolution, the scientists have begun to home in on cells that change most as pathology advances. Again, some changed more with tangles than with plaques. One subset of excitatory neurons grew when tangles did, whereas another shrank as plaque burden increased. The most dramatic change was in a subcluster of microglia, dubbed M13. Numbers of these cells grew dramatically as both tangle and plaque burden rose, hinting at a connection. A subset of astrocytes, A10, also swelled with both pathologies.

What does this mean for AD progression? De Jager said it’s too early to tell. Meanwhile the scientists are beginning to add detail to the picture. Using mediation analysis, they predict that M13 may be bad news because as these microglia respond to Aβ they exacerbate tau pathology. Likewise, A10 astrocytes seem to exacerbate the effect of tau on cognitive decline. The researchers are now looking at how an individual’s cell profile might predict whether they had been on a path to disease before they died, or whether they might have been more resilient.

Cell-Cell Cross Talk Resilience emerged as an important cellular state in Albanus’s analysis. He also used snRNA-Seq data, this time from parietal cortex tissue donated by 67 people who had either been in the DIAN study of autosomal-dominant AD, or had been treated at the Knight Alzheimer’s Disease Research Center at WashU. Logan Brase in Hariri’s lab had previously used this data to uncover transcriptome differences between ADAD and late-onset AD (see Aug 2021 conference news).

Albanus wanted to know how the different subtypes of cells communicate with each other, and how that changes with disease. For this, he turned on CellPhoneDB. This is not the latest smart phone, but an algorithm that predicts cellular cross talk based on expression of ligands and receptors in transcriptome data (Efremova et al., 2020). It enabled Albanus to evaluate more than a million potential interactions between thousands of ligand pairs expressed by hundreds of cell subtypes. From control transcriptomes, he found 31,000l ligand/receptor pairs that drive interactions between different cells. From this he could predict patterns of cellular cross talk. For example, oligodendrocyte precursor cells shared more ligand/receptor pairs with endothelial cells than with any other cell type, and astrocytes had more lines of communication with microglia than with neurons or oligodendrocytes.  

These patterns were different in Alzheimer's samples, i.e., both the types and number of interactions between cells had changed with disease. Interestingly, endothelial cells doubled their repertoire, going from fewer than 100 ligand/pair interactions with oligodendrocytes to almost 200. Their interactions between other cells increased similarly. Researchers hearing the talk were intrigued by this response and asked if it might be due to the number of cell adhesion molecules endothelial cells produce. Albanus said this is possible, but hard to study because the dataset contains so few of these cells. He agreed that it would be interesting to run a similar analysis of other tissues to see if this endothelial response is specific to the brain.

More statistically powerful data came from analysis of other cells. Interactions between microglia and excitatory neurons changed the most of all cell types, with the former bumping up interactions, and the latter tamping them down, by about 30 percent. Curiously, people with AD who carried a TREM2 mutation were an exception. Their microglial interactions were unchanged from controls. This may not be surprising, Albanus agreed, since TREM2 is a cell-surface signaling receptor on microglia, and pathogenic AD variants cause a loss of function, i.e., dampen microglial responses.

Alzheimer's Chatter. Ligand/receptor cross talk (left) involving AD-related genes buzzes more often between microglia and other cells, including neurons. In AD cases, those interactions occur even more often (right). [Courtesy of Ricardo Albanus, Washington University.]

Which of the thousands of ligand/pair interactions might be important in AD? To address this, Albanus looked to 800 genes that have popped up in AD genome-wide association and functional studies, and found that much of the overall cross talk involved AD-linked genes, for example TREM2/semaphorin and APP/CD74. These interactions were enriched in microglia, were mostly between microglia and either excitatory or inhibitory neurons, and occurred more often in tissue samples from AD brain.

Given that such interactions are sprouting in AD, Albanus next wanted to know what effect they might be having. CellPhoneDB offers little help in this regard, so Albanus instead used the CytoTalk bioinformatics algorithm, which reconstructs signaling networks downstream of ligand-receptor interactions (Hu et al., 2021). Albanus has only just begun to figure out what these networks do. For example, CytoTalk linked the TREM2/semaphorin cross-talk pair to a massive spiderweb of signaling networks inside both microglia and neurons. Intriguingly, known AD genes cropped up mostly in the microglial networks (see image below). Albanus found very similar patterns when he applied the same analysis to datasets from Marco Colonna’s lab at WashU (Zhou et al., 2020).

To better understand how these networks are linked to disease, Albanus has focused in on small subnetworks, many of which are enriched in AD-related genes, but not all of which are directly connected to the ligand/pair interactions identified by CellPhoneDB. One such subnetwork ties TREM2/semaphorin directly to ApoE and HLA immune pathways that regulate microglial responses to amyloid and other pathological changes in the brain. He found similar subnetworks in prefrontal cortex microglia when he analyzes snRNA-Seq data from Nancy Ip’s lab at Hong Kong University of Science and Technology (Lau et al., 2020). Albanus has identified 360 potential subnetworks with CytoTalk.

Cross Talk Networks. CytoTalk links ligand/receptor pairs identified by CellPhoneDB (center) to large protein-protein interaction networks in cells, including microglia (left) and excitatory neurons (right). AD-linked genes (circled red) crop up mostly in microglia, and cluster in sub-networks. [Image courtesy Ricardo Albanus, WashU.]

To get a handle on what these subnetworks might be doing in AD, he correlated their expression with Braak stage. Some, for example those involving APP, and presenilins, positively correlated and so likely increase risk for AD, Albanus said. But many others, including the TREM2 subnetwork, negatively correlated, suggesting they reflect cellular resilience. Intriguingly, only some of these subnetworks mediate ligand/pair crosstalk among cells, and most of them associated with resilience.

“This is why we are so excited about this cross talk,” Albanus told Alzforum. “Proteins that are tuned to those network signals are mostly intracellular and not easy targets to regulate therapeutically. But if we can find a shortcut where we can directly modulate the network by changing the cross-talk interaction on the cell surface, that could help us find new drug targets for AD,” he said.—Tom Fagan

Comments

No Available Comments

Make a Comment

To make a comment you must login or register.

References

News Citations

1. ADAD and LOAD: At Cellular Level, They Are Not the Same 20 Aug 2021

Paper Citations

1. Cain A, Taga M, McCabe C, Hekselman I, White CC, Green G, Rozenblatt-Rosen O, Zhang F, Yeger-Lotem E, Bennett DA, Yang HS, Regev A, Menon V, Habib N, De Jager PL. Multi-cellular communities are perturbed in the aging human brain and with Alzheimer’s disease. medRxiv. December 23, 2020 medRxiv
2. Efremova M, Vento-Tormo M, Teichmann SA, Vento-Tormo R. CellPhoneDB: inferring cell-cell communication from combined expression of multi-subunit ligand-receptor complexes. Nat Protoc. 2020 Apr;15(4):1484-1506. Epub 2020 Feb 26 PubMed.
3. Hu Y, Peng T, Gao L, Tan K. CytoTalk: De novo construction of signal transduction networks using single-cell transcriptomic data. Sci Adv. 2021 Apr;7(16) Print 2021 Apr PubMed.
4. Zhou Y, Song WM, Andhey PS, Swain A, Levy T, Miller KR, Poliani PL, Cominelli M, Grover S, Gilfillan S, Cella M, Ulland TK, Zaitsev K, Miyashita A, Ikeuchi T, Sainouchi M, Kakita A, Bennett DA, Schneider JA, Nichols MR, Beausoleil SA, Ulrich JD, Holtzman DM, Artyomov MN, Colonna M. Human and mouse single-nucleus transcriptomics reveal TREM2-dependent and TREM2-independent cellular responses in Alzheimer's disease. Nat Med. 2020 Jan;26(1):131-142. Epub 2020 Jan 13 PubMed. Correction.
5. Lau SF, Cao H, Fu AK, Ip NY. Single-nucleus transcriptome analysis reveals dysregulation of angiogenic endothelial cells and neuroprotective glia in Alzheimer's disease. Proc Natl Acad Sci U S A. 2020 Oct 13;117(41):25800-25809. Epub 2020 Sep 28 PubMed.

Further Reading

Papers

1. Mostafavi S, Gaiteri C, Sullivan SE, White CC, Tasaki S, Xu J, Taga M, Klein HU, Patrick E, Komashko V, McCabe C, Smith R, Bradshaw EM, Root DE, Regev A, Yu L, Chibnik LB, Schneider JA, Young-Pearse TL, Bennett DA, De Jager PL. A molecular network of the aging human brain provides insights into the pathology and cognitive decline of Alzheimer's disease. Nat Neurosci. 2018 Jun;21(6):811-819. Epub 2018 May 25 PubMed.

News

1. Most Detailed Look Yet at Activation States of Human Microglia 11 Dec 2020
2. Single-nucleus RNA Sequencing Misses Activation of Human Microglia 9 Oct 2020
3. Human and Mouse Microglia React Differently to Amyloid 17 Jan 2020