As ever-improving analytical techniques enable scientists to better understand gene-expression changes in Alzheimer’s disease, Vivek Swarup, University of California, Irvine, and colleagues realized they could double up on analytical power by combining multiple approaches. They paired single-nucleus Assay for Transposase-Accessible Chromatin (ATAC) sequencing with RNA sequencing. This allowed them to assess both how accessible genes are for transcription and which ones are actually being transcribed in the same brain samples from people who had died with advanced AD.

  • Together, single-nucleus ATAC and RNA-Seq map gene expression in AD brains.
  • Oligodendrocyte, astrocyte, microglia signals stood out.
  • AD GWAS loci linked to glial genes based on DNA accessibility and transcription.
  • Transcription factor SREBF1 activates target genes in oligodendrocytes.

In the July 8 Nature Genetics, they described subsets of glial cells that are unique to healthy or AD tissue. How? They matched sections of DNA that were open for transcription-factor binding to upregulated genes. By finding active genes near AD risk loci in cells from AD tissue, the scientists linked these loci to the genes they modify. “This provides a finer map of molecular changes that occur in this complex condition,” Nilüfer Ertekin-Taner, Mayo Clinic, Jacksonville, Florida, wrote (full comment below). The authors also modeled how microglia, astrocytes, and oligodendrocytes might have become diseased.

This is the first time single-nucleus RNA and ATAC-Seq have been used together in AD tissue. “This was not possible three years ago,” Carlos Cruchaga, Washington University, St. Louis, told Alzforum. Marcos Costa, Institut Pasteur de Lille in France, confirmed the study’s novelty. “This work provides a valuable tool to the scientific community, as there are only three to four single-nucleus RNA-Seq datasets on AD, to my knowledge, none of which combine chromatin accessibility,” he wrote (full comment below). Ryan Corces, University of California, San Francisco, agreed. “This will undoubtably be a resource for my lab and for other researchers,” he told Alzforum.

One recent study combined bulk RNA-Seq and ATAC-Seq to analyze hippocampal tissue from APP/PS1 mice (Wang et al., 2020). Another had done the same in primary astrocytes and three types of induced-pluripotent-stem-cell-derived neurons from a healthy person (Song et al., 2019). Two more studies tested human iPSC-derived neurons created from familial AD donors and human-embryonic-stem-cell-derived microglia with AD variants (Caldwell et al., 2020; Liu et al., 2020). However, none of those were done in tissue from a person with AD.

Co-first authors Samuel Morabito, Emily Miyoshi, Neethu Michael, and colleagues harnessed the precision of single-nucleus RNA and ATAC sequencing. In the latter technique, an enzyme binds to open chromatin and inserts a tag sequence (Oct 2020 news). Sequencing the DNA to reveal these tags shows researchers which bits are accessible to transcription factors and are likely being expressed. RNA-Seq confirms gene expression.

From Brain to Information. After being isolated from AD brain tissue, nuclei were sequenced and sorted by cell type. Gene expression and chromatin openness were assessed. Transcription factor networks connected open chromatin to regulated genes, cell changes were modeled from healthy to diseased, and GWAS loci were paired with the genes they influenced. [Courtesy of Morabito et al., Nature Genetics, 2021.]

The researchers obtained postmortem samples from the UC Irvine Alzheimer's Disease Research Center. They used prefrontal cortices from 12 people who had died with late-stage AD, and from eight older controls. They isolated, and analyzed chromatin accessibility in, a total of 130,418 nuclei and sequenced the RNA of 61,472 nuclei. This collection contained excitatory and inhibitory neurons, astrocytes, microglia, oligodendrocytes, and oligodendrocyte precursor cells. The scientists compiled their sequencing data into a database for analysis and turned it into a searchable web app.

What did they find? In AD tissue, each cell type comprised distinct subpopulations with differentially expressed genes (DEGs) and differentially accessible chromatin regions. For example, there were more astrocytes expressing lots of GFAP and CHI3L, the latter encoding the inflammation marker and AD biomarker YKL-40, while there were fewer astrocytes expressing little GFAP, the Wnt inhibitor WIF1, and the metallopeptidase ADAMTS17. Two related microglial subsets were more numerous in AD, both of which expressed the perivascular macrophage marker CD163 and high levels of the transcriptional activation marker SPP1.

To learn what was happening to gene transcription in advanced AD, the scientists scoured healthy and diseased genomes for cis-regulatory elements. CREs are bits of noncoding DNA that regulate transcription of nearby genes. ATAC-Seq and RNA-Seq offer complementary intel on CREs: If a region of DNA turns up in both, then it can be transcribed and actually is. If that region contained a promoter, then the researchers tagged this as a CRE-gene pair. They found 56,552 CREs near 11,440 genes, averaging four CREs per gene.

More than half the CREs sat in introns. Some genes were expressed in multiple cell types but were controlled by a different CRE in each cell type; many were DEGs or genes upregulated in AD (see image below).

Gene Overlap. In each brain-cell type, some genes controlled by a cis-regulatory element (large circles) were unique to the cell type (medium circles) and/or altered in AD (small circles). The overlap between large and small circles indicates genes that are differently expressed and regulated in AD. [Courtesy of Morabito et al., Nature Genetics, 2021.]

To estimate how well transcription factors were able to access DNA at their target genes, the authors searched open regions for binding sites. Several were enriched in astrocytes, microglia, and excitatory neurons from AD tissue. The scientists focused on two genes: SPI1, which encodes the master transcriptional regulator PU.1 in microglia, and Nuclear Respiratory Factor (NRF1), which regulates mitochondrial function in oligodendrocytes. The two microglial subpopulations that were more numerous in advanced AD had more open SPI1 binding sites, yet fewer of its target genes expressed. This suggests that, in late-stage AD, SPI1 acts as a transcriptional repressor, the authors wrote. As for NRF1, it was dysregulated in certain oligodendrocyte subgroups, hinting at a possible role in mitochondrial dysfunction in AD, the authors suggest.

The researchers created cell-specific networks of transcription factors and genes they target, focusing on those that were open for expression in AD cells. SPI1 and NRF1 each regulate multiple AD DEGs and genes near AD risk loci in microglia and oligodendrocytes, respectively (see image below).

Transcription Networks. Late-stage AD microglia (left) and oligodendrocytes (right) have different transcription factors (big blue circles) and predicted target genes. [Courtesy of Morabito et al., Nature Genetics, 2021.]

How might the cells have morphed from healthy to diseased? The scientists strung together transcriptional and epigenetic information from microglia, astrocytes, and oligodendrocytes from cognitively normal people and from people with late-stage AD. Some of the former had mild plaque and tangle pathology, giving the scientists an intermediate point in a hypothetical cellular trajectory. Then, they used mathematical modeling to identify DEGs in each cell type at each stage. In this way, they constructed a disease scale for each cell type. “Every sample is a snapshot, so combining them into a continuum from healthy to AD allows us to figure out where cells are on their trajectory,” Swarup told Alzforum.

First, disease-associated oligodendrocytes. Such a subpopulation was recently identified in cortex tissue of 5xFAD mice and people with AD (Jan 2020 news). Morabito and colleagues split healthy and AD oligodendrocytes into three subpopulations based on unique gene expression: newly formed, myelin-forming, and mature cells. As cells progressed from healthy to diseased in the model, there were fewer newly formed and myelin-forming oligodendrocytes and more mature cells.

The researchers then scrutinized NRF1 and Sterol Regulatory Element Binding Factor 1 (SREBF1) in oligodendrocytes. They compared expression of both these transcription factors to expression of their respective target genes across the oligodendrocyte healthy-to-disease continuum. Oligodendrocytes from AD brain tissue upregulated NRF1 but expressed fewer of its target genes. This indicates that NRF1 acts as a repressor, the authors wrote.

As for SREBF1, it and its target genes were both downregulated in AD brain oligodendrocytes; expression going in the same direction indicates the factor normally acts as an activator. This was corroborated by two more analyses. A protein-protein interaction network in AD oligodendrocytes revealed fewer accessible SREBF1 binding sites and fewer copies of the protein itself. Likewise, single-nucleus, weighted gene co-expression analysis, aka WGCNA, also identified three subpopulations of oligodendrocytes from AD tissue with reduced expression of SREBF1 target genes and proteins. The authors noted that a physiological function of SREBF1 is to regulate cholesterol homeostasis, and Aβ has been previously proposed to inhibit its activation (Mohamed et al., 2018). 

Likewise, the scientists spotted disease-associated astrocytes in these AD brains. Identified in 5xFAD amyloidosis mice, DAAs are a GFAP-rich subset of astrocytes that express a unique set of genes (Habib et al., 2020). To study what went on in the DAAs, the researchers correlated DEGs to two transcription factors: the master chromatin regulator CCCTC-binding factor (CTCF), and FOSL2, which encodes a subunit of the cell proliferation and differentiation regulator AP.1. CTCF was associated with DEGs in GFAP-low, aka healthy, astrocytes. It was downregulated in DAAs, while FOSL2 was up. Taken together, the authors propose that CTCF may promote homeostatic astrocytes while FOSL2 may stimulate DAA production.

Regarding disease-associated microglia, aka DAMs, in AD tissue they congregate around plaques and can be activated with or without TREM2 (June 2017 newsSep 2017 news). In the late-stage AD brains analyzed here, microglia downregulated homeostatic genes; the number of TREM2-independent DAMs grew while TREM2-dependent cells became fewer.

What about AD risk loci? Compared to microglia from healthy controls, microglia from AD tissue, including DAMs, held more GWAS SNPs, including variants in APOE, BIN1, ADAM10, and SLC24A4. Lo and behold, gene regulation was different at these loci of microglia from AD brains.

These techniques could advance other questions in AD biology by using them on different samples. For example, Costa is interested in gene-expression changes in other brain regions, such as the entorhinal cortex and hippocampus. Those areas undergo more significant gene-expression alterations than the prefrontal cortex, Costa wrote, hence the authors may have missed important changes that could characterize advanced stages of AD pathology (full comment below). Moreover, that this study found its strongest signals in glia may reflect the extensive neuronal loss and gliosis in advanced AD. To this, Swarup replied that a follow-up paper on the neuronal results of this study is in the works.

Ertekin-Taner and Cruchaga also consider changes in preclinical or mild AD a priority. “We know there may be other processes ongoing at early disease stages, so it would be nice to see similar single-nucleus multi-omics studies on preclinical AD,” Cruchaga said.

Corces further mentioned a new commercial kit enabling researchers to do RNA and ATAC-Seq on the same nuclei, rather than splitting nuclei from a tissue sample between the two assays then manually integrating the data, as the authors had to do in this paper. “While this data was probably generated over a few years, this type of data can now be generated within a few months using the combination kit,” he told Alzforum. Ever-improving analytics, indeed.—Chelsea Weidman Burke

Comments

  1. This work is an important step toward filling three important gaps in the study of gene expression alterations in Alzheimer’s disease:

    • Understanding whether gene expression alterations observed in previous studies based on tissue analysis (bulk RNA-seq) are a consequence of changes in cellular composition (neurodegeneration plus gliosis, for example).
    • Identifying cell-type specific changes in gene expression.
    • Characterizing possible epigenetic regulators of gene expression, which is a prerequisite for linking GWAS hits to gene-expression alterations observed in the diseased brain.

    In principle, the methodology used is appropriate to address these points, though two aspects of the work temper my enthusiasm. First, the brain region used to perform the snATAC-Seq is perhaps not ideal. We and other researchers have shown that gene-expression alterations in the prefrontal cortex are less significant than those observed in brain regions affected at earlier stages of AD, such as the entorhinal cortex or hippocampus. Thus, the authors are likely missing important gene-expression alterations that could characterize AD pathology at advanced stages. Second, there is a strong bias toward glial cells, mainly because these cells (especially oligodendrocytes) are overrepresented in the samples. Consequently, there is little discussion about gene-expression alterations in neurons, which are the functional units of our brains. Glial cells are important, but I wonder whether a “transcriptomic characterization of the AD brain” can neglect neuronal cells.

    On the other hand, the data generated and now publicly available, is an interesting resource to evaluate gene expression alterations in specific cell types of the AD brain. To my knowledge, there are only three to four snRNA-Seq datasets on AD, and none of them combine chromatin accessibility. Thus, the work provides a valuable tool to the scientific community.  

    The work also has merit in raising discussion about the identification of disease-associated microglia (DAM) in the human brain using snRNA-Seq, which has been recently challenged by other researchers.

    As for the link between GWAS hits and gene-expression alterations observed in the AD brain, I think we need to await new studies on other brain regions and with larger number of brain samples obtained from patients at different stages of the disease. With the data presented in this work, possible causal links between the genetic variants identified in GWAS and the gene-expression alterations in AD remain elusive.

  2. In this paper, the authors provide a comprehensive characterization of human prefrontal cortex tissue from 12 Alzheimer’s disease and eight control brains by both single-nucleus assay for transposase-accessible chromatin sequencing (snATAC-Seq) and RNA sequencing (RNA-Seq, subset of 11 AD and seven control brains). Through a series of bioinformatics analyses, they integrate these data modalities from 191,890 nuclei (130,418 for snATAC-Seq and 61,472 for snRNA-Seq):

    1. to identify cell cluster-specific, differentially expressed genes (DEGs) and differentially accessible chromatin regions in AD;
    2. to build cis co-accessibility networks (CCANs) of regulatory elements and their target genes;
    3. to construct cell-type specific transcription factor (TF) regulatory networks with a particular focus on TFs SPI1 in microglial and NRF1 in oligodendrocyte networks;
    4. to perform pseudotime trajectory analysis and investigate correlations of select TFs with genes at different pseudotime stages of cell-specific trajectories;
    5. to integrate their single-nucleus data with disease GWAS variants; and
    6. to generate co-expression networks using aggregate expression levels from their single nucleus data.

    Their findings build upon prior work on the transcriptional and genomic landscape of AD brains by adding an important layer of chromatin-accessibility data, performing both snATAC-Seq and snRNA-Seq in the same samples and focusing on both cis and trans regulatory elements.

    Prior brain gene-expression quantitative trait loci (eQTL) studies integrated with AD risk GWAS have discovered many significant eQTL that also influence AD risk, highlighting the potential role of regulatory disease risk variants in AD (Zou et al., 2010; Zou et al., 2012; Allen et al., 2015; Carrasquillo et al., 2017). Brain transcriptome studies utilizing bulk RNA-Seq data (Allen et al., 2016; Wang et al., 2018; De Jager et al., 2018) revealed myriad perturbations in transcripts and transcriptional networks implicating pathways and cell types including myelination/oligodendrocytes (McKenzie et al., 2017; Allen et al., 2018) and innate immunity/microglia (Zhang et al., 2013). 

    Both deconvolution analyses of bulk transcriptome data (McKenzie et al., 2018; Wang et al., 2020) and recent single-cell/single-nucleus transcriptome studies (Mathys et al., 2019; Zhou et al., 2020; Lau et al., 2020; Olah et al., 2020) are beginning to uncover the complexity of the cell-specific expression changes in AD. These studies underscore the variety of cellular subtypes annotated as clusters in sn/scRNA-Seq data and their distinct transcriptional changes in AD, thereby providing a finer map of the molecular changes that occur in this complex condition.   

    This study confirms prior observations pertinent to cell proportion and cell-specific transcriptional changes in AD, but extends beyond these observations to bioinformatic “matching” of transcripts to both cis- and trans-regulatory elements. This provides potential, testable regulatory mechanisms that underlie transcriptional dysregulation in AD. The regulatory elements, their target genes, and biological pathways identified in this study could serve as novel therapeutic or biomarker targets and also be leveraged to model various cell-specific dysfunction in AD.

    Despite the wealth of data and analytic paradigms afforded by this study, its main shortcoming is the small sample size. This limits generalization with respect to the full spectrum of AD. The extent to which the transcriptome and epigenetic changes in end-stage autopsy brain samples reflect disease mechanisms that initiate and propagate AD is an ongoing question we must collectively try to answer in this field.

    Future studies on larger autopsy cohorts, racially and ethnically diverse patient populations, as well as longitudinal peripheral tissue samples are necessary to address these concerns. In the meantime, integrative multi-omics studies at the single-cell level can be expected to continue to expand our understanding of the molecular complexities of AD.

    References:

    . Gene expression levels as endophenotypes in genome-wide association studies of Alzheimer disease. Neurology. 2010 Feb 9;74(6):480-6. PubMed.

    . Brain expression genome-wide association study (eGWAS) identifies human disease-associated variants. PLoS Genet. 2012;8(6):e1002707. PubMed.

    . Late-onset Alzheimer disease risk variants mark brain regulatory loci. Neurol Genet. 2015 Aug;1(2):e15. Epub 2015 Jul 23 PubMed.

    . A candidate regulatory variant at the TREM gene cluster associates with decreased Alzheimer's disease risk and increased TREML1 and TREM2 brain gene expression. Alzheimers Dement. 2017 Jun;13(6):663-673. Epub 2016 Dec 8 PubMed.

    . Human whole genome genotype and transcriptome data for Alzheimer's and other neurodegenerative diseases. Sci Data. 2016 Oct 11;3:160089. PubMed.

    . The Mount Sinai cohort of large-scale genomic, transcriptomic and proteomic data in Alzheimer's disease. Sci Data. 2018 Sep 11;5:180185. PubMed.

    . A multi-omic atlas of the human frontal cortex for aging and Alzheimer's disease research. Sci Data. 2018 Aug 7;5:180142. PubMed.

    . Multiscale network modeling of oligodendrocytes reveals molecular components of myelin dysregulation in Alzheimer's disease. Mol Neurodegener. 2017 Nov 6;12(1):82. PubMed.

    . Conserved brain myelination networks are altered in Alzheimer's and other neurodegenerative diseases. Alzheimers Dement. 2018 Mar;14(3):352-366. Epub 2017 Oct 31 PubMed.

    . Integrated systems approach identifies genetic nodes and networks in late-onset Alzheimer's disease. Cell. 2013 Apr 25;153(3):707-20. PubMed.

    . Brain Cell Type Specific Gene Expression and Co-expression Network Architectures. Sci Rep. 2018 Jun 11;8(1):8868. PubMed.

    . Deciphering cellular transcriptional alterations in Alzheimer's disease brains. Mol Neurodegener. 2020 Jul 13;15(1):38. PubMed.

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

    . 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.

    . 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.

    . Single cell RNA sequencing of human microglia uncovers a subset associated with Alzheimer's disease. Nat Commun. 2020 Nov 30;11(1):6129. PubMed.

Make a Comment

To make a comment you must login or register.

References

Research Models Citations

  1. APPswe/PSEN1dE9 (C57BL6)
  2. 5xFAD (B6SJL)

News Citations

  1. Epigenomic Roadmap Points to Causal Genes
  2. Human and Mouse Microglia React Differently to Amyloid
  3. Hot DAM: Specific Microglia Engulf Plaques
  4. ApoE and Trem2 Flip a Microglial Switch in Neurodegenerative Disease

Paper Citations

  1. . Characterization of the chromatin accessibility in an Alzheimer's disease (AD) mouse model. Alzheimers Res Ther. 2020 Mar 23;12(1):29. PubMed.
  2. . Mapping cis-regulatory chromatin contacts in neural cells links neuropsychiatric disorder risk variants to target genes. Nat Genet. 2019 Aug;51(8):1252-1262. Epub 2019 Jul 31 PubMed.
  3. . Dedifferentiation and neuronal repression define familial Alzheimer's disease. Sci Adv. 2020 Nov;6(46) Print 2020 Nov PubMed.
  4. . Multi-omic comparison of Alzheimer's variants in human ESC-derived microglia reveals convergence at APOE. J Exp Med. 2020 Dec 7;217(12) PubMed.
  5. . Aβ inhibits SREBP-2 activation through Akt inhibition. J Lipid Res. 2018 Jan;59(1):1-13. Epub 2017 Nov 9 PubMed.
  6. . Disease-associated astrocytes in Alzheimer's disease and aging. Nat Neurosci. 2020 Jun;23(6):701-706. Epub 2020 Apr 27 PubMed.

External Citations

  1. searchable web app

Further Reading

Papers

  1. . Conserved brain myelination networks are altered in Alzheimer's and other neurodegenerative diseases. Alzheimers Dement. 2018 Mar;14(3):352-366. Epub 2017 Oct 31 PubMed.

Primary Papers

  1. . Single-nucleus chromatin accessibility and transcriptomic characterization of Alzheimer's disease. Nat Genet. 2021 Aug;53(8):1143-1155. Epub 2021 Jul 8 PubMed.