Most studies of gene-expression changes in Alzheimer’s brain analyze tissue samples containing a mix of different cell types. This makes it impossible to definitively figure out the contribution of specific cells. In the November 25 Nature Neuroscience, researchers led by Jose Polo at Monash University in Clayton, Australia, and Enrico Petretto and Owen Rackham at Duke-National University of Singapore Medical School now present cell-type-specific gene-expression data from the entorhinal cortices of late-stage AD and control brains. The researchers sequenced RNA from 13,214 cells representing six types: neurons, astrocytes, oligodendrocytes, oligodendrocyte progenitor cells (OPCs), endothelial cells, and microglia. They found distinct gene-expression differences between AD and controls, not only in each cell type, but also in subtypes. In addition, they identified transcription factors that may control suites of differentially expressed genes. “Taken together, these observations will allow us to better understand how AD progresses and, as a result, find new ways to tackle this debilitating disease,” the authors wrote to Alzforum. The data are available to researchers on a searchable web interface.

  • Single-cell RNA-Seq offers a glimpse of how different cell types change in AD brain.
  • Data largely agree with prior studies, add new details about transcriptional control.
  • In AD astrocytes, lysosomal TFEB regulates 10 AD-associated genes.

“This study provides direct evidence of significant gene-expression changes in all major cell types in AD,” noted Jeremy Miller at the Allen Institute for Brain Science in Seattle (full comment below).

Previously, researchers led by Bin Zhang at Mount Sinai Medical School, New York, and Valur Emilsson at the University of Iceland, Kopavogur, used bulk transcriptomic analysis of cortical samples from 376 AD patients and 173 controls to pinpoint immune genes as the network most altered in AD. At the time, these researchers could not parse out cell-specific changes (Zhang et al., 2013). Earlier this year, Li-Huei Tsai and Manolis Kellis and colleagues at the Massachusetts Institute of Technology reported single-cell RNA-Seq data from the prefrontal cortices of 24 AD patients and 24 controls, opening a more granular view. They found that microglia, astrocytes, and oligodendrocytes harbored the majority of the differentially expressed genes (May 2019 news). 

Gene-Expression Fingerprints. Neuronal gene expression patterns (top) cluster into six distinct subtypes, while astrocytes (bottom) cluster into eight. Some subtypes, like a1, a2, and n1, are found only in AD brain. [Courtesy of Grubman et al., Nature Neuroscience.]

Polo and colleagues focused on a different brain region, the entorhinal cortex, one of the first areas to lose neurons in Alzheimer’s disease. First authors Alexandra Grubman, Gabriel Chew, and John Ouyang isolated nuclei from the postmortem brains of six AD patients at Braak stage VI and six controls, who had died at an average age of 78. In total, they obtained 7,432 oligodendrocytes, 2,171 astrocytes, 1,078 OPCs, 656 neurons, 449 microglia, and 98 endothelial cells, as well as 1,330 cells that could not be clearly sorted into one of these categories. Analysis of expression profiles further subdivided each cell type. Altogether, the authors delineated six oligodendrocyte subtypes, eight astrocyte, four OPC, six neuronal, five microglial, and two endothelial subtypes. For all cells except neurons, these subtypes segregated cleanly into either AD or control brain. For example, two astrocyte subtypes were found only in AD brain, the remaining six only in controls.

The changes in AD varied by subtype. Excitatory neurons turned down synaptic transmission genes, while inhibitory neurons dialed back genes involved in ion transport and memory. Oligodendrocytes boosted genes responsible for myelination, perhaps as a compensatory response to myelin loss in AD, the authors speculated. As expected, astrocytes, microglia, and endothelial cells turned up inflammatory genes. At the same time, microglia dampened genes involved in homeostasis, cell adhesion, and lipid metabolism, in agreement with other studies (Sep 2017 news; Aug 2019 news). Some expression changes were common to multiple cell types. Glial cell types in general turned down cell-death pathways, perhaps to protect damaged cells, the authors noted. And most of the cells that remained in these late-stage AD brain samples had ramped up pathways for dealing with misfolded proteins and cellular stress.

The authors paid particular attention to the expression of about 1,000 genes that have been implicated as AD risk or protective factors by GWAS with a p value of 9 x 10-6 or better. In some cases, they found the gene was expressed in only a single cell type in both AD and control brain. For example, the endocytosis gene RIN3 and the vasoconstrictor TBXAS1 were expressed only in microglia, a new finding. In other cases, the authors found AD differential expression in only one cell type; for example, MS4A6A expression in AD increased only in microglia.

Expression of some AD risk genes varied by cell type. ApoE went up in microglia and down in astrocytes, oligodendrocytes, and OPCs. BIN1 went up in one astrocyte subtype and down in one neuronal one. Curiously, the authors found no change in microglial BIN1 expression, even though recent studies have placed the AD risk variant for this gene in a microglia-specific enhancer (Nov 2019 news). 

The authors also identified coordinated changes in multiple genes controlled by single transcription factors. This analysis suggested key genes that may drive a cell’s transition to an AD state. For example, the transcription factor AEBP1, which goes up with amyloid plaque burden, likely directs many of the expression changes in astrocytes (Hokama et al., 2013; Shijo et al., 2016). HIF3A, which inhibits hypoxia-induced genes, seemed to be responsible for some neuronal transitions. The lysosomal master transcription factor TFEB rises in AD astrocytes, where it controls expression of 10 loci that associated with AD in GWAS—BIN1, CLDN11, POLN, STK32B, EDIL3, AKAP12, HECW1, WDR5, LEMD2, and DLC1.

“The link to TFEB is really interesting, and speaks to the role of lysosome dysfunction in AD,” Fenghua Hu at Cornell University in Ithaca, New York, wrote to Alzforum.

Despite the wealth of data, the study’s findings are limited by the small sample size, the authors acknowledge. Many more samples will be needed to parse out how age, disease stage, and individual genetic variation affect gene expression. Curiously, the specific genes identified as changing in AD in this study showed little overlap with those found by Tsai and Kellis. For example, among 182 differentially expressed microglial genes in this dataset, only 11 also turned up in the earlier study. Numbers were similar for the other cell types. Nonetheless, the two datasets do broadly agree on which processes are up- or downregulated in specific cell types.—Madolyn Bowman Rogers


  1. In this paper Grubman and colleagues present the second published study of human AD using snRNA-Seq, focusing on one of the earliest brain regions affected by the disease. Despite the relatively small sample size (six control and six AD), this study highlights some of the advantages of single-cell vs. bulk-transcriptomic data. In particular, bulk-transcriptomics data cannot separate gene-expression changes within specific cell types from changes in proportions of cells in a tissue due to neurodegeneration or gliosis.

    This study identified many examples of genes differentially expressed between control and AD in one or more cell types, and even identified examples of genes which show different changes with disease in different cell types. For example, APOE showed increased expression in microglia with AD, but decreased expression in other glial types, consistent with published results in prefrontal cortex (Mathys et al., 2019).

    More generally, this study provides direct evidence of significant gene-expression changes in all major cell types in AD. To quote the authors: "[T]ogether, these data highlight the advantage of studying single-cell data to understand the effect of disease gene variants on cell subtype-specific genetic susceptibility, and may explain why conventional (whole-body) rather than conditional or cell-type-specific gene knockouts in Alzheimer’s disease models have often yielded discrepant results."

    Grubman and colleagues also perform computational analyses that present some interesting hypotheses that connect GWAS genes, transcription factors, and cell types for future validation. In particular, they find that the master lysosomal regulator TFEB acts upstream of 10 AD GWAS loci (BIN1, CLDN11, POLN, STK32B, EDIL3, AKAP12, HECW1, WDR5, LEMD2, and DLC1) to control the transition from control to AD in astrocytes. Finally, this study presents a user-friendly web interface to allow the community to browse their data.


    . Single-cell transcriptomic analysis of Alzheimer's disease. Nature. 2019 Jun;570(7761):332-337. Epub 2019 May 1 PubMed.

  2. Once again, I feel compelled to point out that TFEB has been connected to APOE genotype in AD. Essentially, compelling evidence indicates that ApoE4 binds (evidently, with higher affinity than ApoE3) to the "CLEAR" enhancers that comprise the cis elements to which TFEB binds (Parcon et al., 2018). In doing so, it competes with TFEB and reduces expression of at least some of its target genes.

    At first blush, this may seem contrary to the data presented by Grubman et al. However, it should be noted that Parcon et al. found activation of TFEB and elevation of three of its target genes in Alzheimer brains that were APOE4-negative. TFEB also seemed to be activated in APOE4 homozygotes, though in a futile sense: Its target genes were squelched.

    One interpretation is that the activation of TFEB is a compensatory response to be expected in a condition where proteins accumulate in a manner that would alleviated by autophagy. This pathway seems to work reasonably well in persons devoid of an APOE4 allele, who succumb for other reasons, presumbably. But a compromise of TFEB's induction of autophagy in APOE4 carriers seems almost certain to contribute to their more common and aggressive disease course. It may also be the case that subtle distinctions in the sequence of individual CLEAR sites create a differential inhibition by ApoE4 at different genes; subsets of TFEB-regulated genes may be impacted by ApoE4 less than are key autophagy genes such as SQSTM1, MAP1LC3B, and LAMP2.


    . Apolipoprotein E4 inhibits autophagy gene products through direct, specific binding to CLEAR motifs. Alzheimers Dement. 2018 Feb;14(2):230-242. Epub 2017 Sep 22 PubMed.

  3. The researchers carried out single-nucleus RNA sequencing in entorhinal cortex samples of Alzheimer’s disease patients and were able to identify cell-type-specific expression patterns corresponding to six cell types, microglia, neurons, astrocytes, oligodendrocytes, oligodendrocyte progenitor cells, and endothelial cells. In further analysis of these clusters, between four and eight molecular subclusters were identified for each cell type. Interestingly, except for neurons, all Alzheimer’s disease and control cells segregated into different subclusters matching with specific disease-associated transcription patterns.

    An absolute strength of the study is the correlation of genetic data obtained by GWAS with the expression results. So far, the function of the majority of disease-associated genes identified by GWAS remains unknown, but the present study revealed the exact cell subpopulations where these GWAS genes operate in the Alzheimer’s disease brain.

    And, the researchers went beyond the reporting their results and mapping GWAS genetic data onto expression results, by developing a web-based tool, the Single-cell atlas of the Entorhinal Cortex in Human Alzheimer’s Disease that, once finished, will be extremely useful for putting future expression and GWAS data into an appropriate context.

    The study additionally revealed how healthy cell populations transform into Alzheimer’s disease cell populations. The researchers were able to build gene regulatory networks and identify several transcription factors responsible for this transition. Regulatory networks driven by specific transcription factors were integrated with gene targets identified in Alzheimer’s disease GWAS so that subcluster-specific regulation of Alzheimer’s disease genetic susceptibility could be identified. One example is the transcription factor HIF3A, which regulates multiple GWAS genes driving the transition from healthy neurons to Alzheimer disease neurons. Another is the transcription factor EB (TFEB). TFEB is the master regulator of lysosome biogenesis and autophagy, and as revealed by this study, it is upregulated in AD astrocytes and acts upstream of 10 deregulated GWAS genes. The upregulation of the regulatory module controlled by TFEB affects specific astrocyte subpopulations driving their transition from healthy to Alzheimer disease states.

    The authors concluded that there is a functional link between specific astrocyte subpopulations and Alzheimer’s disease, suggesting that susceptibility genes identified by GWAS play a determined role in Alzheimer’s disease development, which, at the same time, is mediated by TFEB.

    Over the past 10 years we have learned that brain-area specific drug delivery is mandatory to develop effective treatments for neurological disorders. But this study tells us that we have to go even further to make possible treatments successful, and shows that cell-type directed therapies may represent the future of Alzheimer disease prevention.

    The study shows that TFEB drives conversion of healthy into diseased astrocytes, suggesting that it may represent a new target for effective therapeutic interventions to prevent Alzheimer’s disease. But precise fine-tuning of TFEB inhibition will be necessary to avoid blocking lysosomal and autophagy systems in healthy cells. Of note, there is also the need to target specific astrocyte populations to prevent the TFEB-driven machinery from transforming healthy cells toward diseased ones.

    Although further studies need to confirm the results in larger cohorts, and probably not all Alzheimer disease patients develop the disease by the mechanisms identified, the results of this study by Grubman et al. represent a step forward in our understanding of Alzheimer disease development and line out new directions to pursue preventive therapies.

  4. Grubman et al. present a data set of single nuclei sequencing of entorhinal cortex tissue derived from the brains of six Alzheimer’s disease (AD) patients and six age-matched control patients. This report, combined with the data set recently published by Mathys et al. using single-nuclei sequencing to analyze the prefrontal cortices from AD patients and controls, provides a remarkable resource for understanding the cell-specific changes that occur in the brains of AD patients. These data sets provide insight into how each of the different major cell types in the brain changes during AD, how different GWAS hits may act in specific cells, and the transcriptional networks that may control cellular transitions during AD.

    What immediately stood out was the differential response between non-neuronal cells (astrocytes, oligodendrocyte lineage cells, endothelial cells, and microglia) and neuronal cells. The most coordinated changes were observed in non-neuronal cells, and each non-neuronal population largely subclustered into AD and non-AD clusters, whereas neuronal subclusters contained cells from both AD and non-AD tissue. This could be due to the larger baseline heterogeneity of neurons or could indicate that glial cells specifically take on a unique phenotype during AD. Perhaps most surprising was the finding that one astrocyte subcluster had more similarities to oligodendrocytes than to other astrocyte subclusters. Could this indicate de-differentiation of astrocytes, or even oligodendrocyte lineage cells, into a common glial progenitor in AD? If so, do these cells contribute to the pathophysiology of AD or are they part of the repair process?

    This data set was also able to pinpoint the cell-type-specific expression of different GWAS hits, and how their expression changes during AD. This confirmed the cell-specific expression of many GWAS hits within microglia, further highlighting the importance of these innate immune cells in AD pathophysiology, and was also able to identify new microglial-specific GWAS hits. Of particular interest was the identification that the GWAS hit TBXAS1 was specifically expressed in microglia. TBXAS1 is a cytochrome P450 enzyme that generates the potent vasoconstrictor thromboxane A2, and this linkage provides evidence that aberrant microglial activity may be involved the hypoperfusion observed in early AD patients. Interestingly, many GWAS hits were expressed by multiple cell populations, and this data set gave insight into how the expression of these genes may change within each cell type during AD. In some cases, such as BIN1, the GWAS hit changed similarly in multiple cell populations, whereas in other cases, such as APOE, the GWAS hit changed differentially in different cell types. While this adds to the mystery of how APOE regulates AD risk, it also highlights the importance of understanding the role of APOE in the context of the whole brain.

    Perhaps most clinically relevant is the authors’ use of CellRouter to predict transcription factors that drive cell transitions during AD, as these pathways may be targeted to inhibit widespread changes to cellular phenotypes. In particular, the finding that AEABP—a gene associated with amyloid deposition—drives astrocyte subclusters toward the astrocyte/olidodenrocyte hybrid cluster may provide crucial information into the role of these hybrid cells. Moreover, this analysis predicted transcription factors that drive cell transitions involving the expression of multiple GWAS hits, such as TFEB driving astrocyte transitions and HIF3 driving a neuronal transition. Because HIF3 is a hypoxia-inducible factor, this finding may provide insight into how cardiovascular risk factors, including hypoperfusion, may be causative factors in AD.

    From their data sets, the authors have created a remarkable public resource—a searchable online atlas, allowing users to analyze their cell or gene of interest in the context of AD. While these data sets provide a fascinating glimpse into the pathophysiology of AD, there is still much work to be done. Going forward, it will be essential to determine which of these changes are causal, which are part of the pathophysiology, and which are part of a response to restrict the pathophysiology. Once the effects of these changes are fully understood, the field will be able to identify which have the greatest potential as therapeutic targets.


    . Single-cell transcriptomic analysis of Alzheimer's disease. Nature. 2019 Jun;570(7761):332-337. Epub 2019 May 1 PubMed.

Make a Comment

To make a comment you must login or register.


News Citations

  1. When It Comes to Alzheimer’s Disease, Do Human Microglia Even Give a DAM?
  2. ApoE and Trem2 Flip a Microglial Switch in Neurodegenerative Disease
  3. ApoE4 Glia Bungle Lipid Processing, Mess with the Matrisome
  4. Cell-Specific Enhancer Atlas Centers AD Risk in Microglia. Again.

Paper Citations

  1. . Integrated systems approach identifies genetic nodes and networks in late-onset Alzheimer's disease. Cell. 2013 Apr 25;153(3):707-20. PubMed.
  2. . Altered Expression of Diabetes-Related Genes in Alzheimer's Disease Brains: The Hisayama Study. Cereb Cortex. 2013 Apr 17; PubMed.
  3. . Association of adipocyte enhancer-binding protein 1 with Alzheimer's disease pathology in human hippocampi. Brain Pathol. 2016 Dec 20; PubMed.

External Citations

  1. searchable web interface

Further Reading

Primary Papers

  1. . A single-cell atlas of entorhinal cortex from individuals with Alzheimer's disease reveals cell-type-specific gene expression regulation. Nat Neurosci. 2019 Dec;22(12):2087-2097. PubMed.