Transgenic mice fail to capture many aspects of Alzheimer’s disease, and a new transcriptional study lays bare one reason. Brain cells in the two species, it turns out, respond quite differently to amyloidosis. In the January 13 Nature Medicine, researchers led by Marco Colonna and Maxim Artyomov at Washington University School of Medicine in St. Louis report distinct transcriptional changes in several cell types in human AD brains compared with those of 5XFAD mice.

  • Microglia in AD brain turn homeostatic genes up, not down as in mice.
  • In both species, TREM2 mutation or deletion weakens microglial activation.
  • Astrocyte and oligodendrocyte responses in AD may be secondary to neuronal death.

In particular, human microglia ramped up a transcriptional program controlled by interferon regulatory factor 8 (IRF8) that had little overlap with the disease-associated microglial (DAM) signature seen in mice. However, in both species, deletion or mutation of TREM2 muted the response to amyloidosis. “This supports the idea that TREM2 mutations impair microglial responses and facilitate the progression of disease,” Colonna told Alzforum. This study is the first to report single-nucleus expression data from people with TREM2 mutations. Some of these data were previously presented at last year’s Brain Conference in Rungstedgaard, Denmark (May 2019 conference news). 

“Among the now several ‘omics’ papers on Alzheimer’s disease that have emerged in the last few years, this new report by Zhou et al. stands out because of its comprehensive analysis of mouse and human tissue at the single cell level,” Ashley Bush and Scott Ayton at the University of Melbourne, Australia, wrote to Alzforum (full comment below).

Shane Liddelow at New York University said the field needs to do more such comparative mouse-human studies to parse out what cell types in rodent models best reflect the human disease process. “This paper produces an incredibly rich dataset for the field that will be mined heavily to determine new testable hypotheses … studies like this one pave the way for expanding our combined community dataset so that we can mitigate future mistakes of studying the wrong cellular responses,” he wrote to Alzforum (full comment below).

Species Differences. Gene-expression patterns cluster into distinct cellular groups in 5xFAD mouse (top) and human AD brains (bottom). [Courtesy of Zhou et al., Nature Medicine.]

With the advent of techniques like single-nucleus RNA-Seq, researchers have started to delineate how specific cell types react to disease. Initial single-nucleus transcriptome studies from Li-Huei Tsai and Manolis Kellis at MIT and others reported little evidence for a DAM-like signature in microglia from AD brain, with increased expression of only a few DAM genes, such as ApoE (May 2019 news). 

Colonna and colleagues focused on the role of TREM2. For mice, first author Yingyue Zhou ran RNA-Seq on a total of 73,419 nuclei isolated from the cortices of four different strains: wild-type, TREM2 knockout, 5XFAD, and 5XFAD mice lacking TREM2. Zhou and colleagues analyzed nuclei from three mice from each genotype, all of them 7 months old. By similarity, the gene-expression patterns clustered into 10 groups, representing five neuronal and five non-neuronal cell types. The non-neuronal comprised microglia, astrocytes, oligodendrocytes, oligodendrocyte precursor cells (OPCs), and endothelial cells (see image above).

To compare with human brain, the authors isolated cells from dorsolateral prefrontal cortex samples taken postmortem from 11 control and 11 AD brains, as well as 10 brains of AD patients with a R62H TREM2 mutation. Samples came from the Rush Memory and Aging cohort. Altogether, the authors analyzed 66,311 nuclei. Gene-expression profiles clustered differently than in mice, falling into three neuronal and two oligodendrocyte subtypes, and one each for microglia, astrocytes, OPCs, and endothelial cells (see image above).

The greatest species differences in the amyloid response were in microglia. In mice with amyloidosis, these immune cells dialed down homeostatic genes and turned up DAM genes, in agreement with previous studies (May 2017 news; Jul 2018 conference news). Lack of TREM2 muted the DAM response, and reduced the number of microglia seen in 5XFAD cortex.

In human AD brain, by contrast, microglia turned homeostatic genes up, not down. This seemed to be controlled by increased expression of the transcription factor IRF8, which is known to activate homeostatic genes. Human microglia bumped up expression of AD risk genes that were not changed in the DAM profile: SORL1, alpha-2 macroglobulin, and chitinase-3-like protein 1, also known as YKL-40. The only overlap with the DAM transcriptome was an increase in TREM2, ApoE, CD68, and MHC-II genes. Overall, the human microglial profile resembled an IRF8-driven response seen in spinal cord microglia after peripheral nerve injury (Masuda et al., 2012). The R62H TREM2 variant, an AD risk factor, softened this response. “The effect of the mutation was not dramatic, but likely snowballs over the lifespan of an individual,” Colonna said.

Oleg Butovsky at Brigham and Women’s Hospital, Boston, called the differences between mouse and human homeostatic responses fascinating, noting that they might represent a biological dichotomy between species. “This needs to be explored and understood mechanistically. That will give us additional directions in terms of treatment,” he said.

Parsing the data by biological processes, microglia in AD brain most altered expression of genes involved in iron metabolism. They suppressed transcription of genes involved in iron transport and storage, such as SLC25A37, HAMP, and FTH1. Iron buildup has been linked to cognitive decline in AD (Mar 2019 news).

After microglia, oligodendrocytes changed most in AD and in mouse models, but again profiles were species-specific. In 5XFAD mice, these cells boosted expression of the complement factor C4b and the serine peptidase inhibitor Serpina3n, as well as the immune histocompatibility gene H2-d1. The increases were smaller in 5XFAD mice lacking TREM2, suggesting these changes partially depended on TREM2 signaling. The effects of these changes were unclear, although the authors found that C4b and Serpina3n independently promoted Aβ aggregation in vitro. In 5XFAD brain, Serpina3n+ oligodendrocytes tended to cluster near plaques. In addition, more oligodendrocytes populated 5XFAD brains than controls, suggesting enhanced proliferation of these cells.

In AD brain, on the other hand, oligodendrocytes took on a different profile, turning down genes involved in myelination, axon guidance, and differentiation, while ramping up genes for oxidative stress and lipid accumulation. The changes may be a response to axon loss and the subsequent buildup of debris in AD brain, the authors suggested. These oligodendrocyte changes were muted in R62H carriers. Curiously, C4b and Serpina3n were not expressed in human oligodendrocytes, but in astrocytes instead.

In mouse brain, astrocytes and other cell types were remarkably similar in wild-type and 5XFAD animals. In human brain, however, astrocytes turned up C4b expression and dampened Serpina3 in AD. It is unclear why Serpina3 changed in the opposite direction and in a different glial cell type than in mice. Overall, AD astrocytes dialed down expression of genes responsible for coordinating lipid metabolism with neurons, and bumped up expression of genes for extracellular matrix. Normally, astrocytes provide metabolic support to neurons, aiding in fatty-acid storage and detoxifying reactive oxygen species.

The findings dovetail with a previous study that reported astrocytes in AD brain poorly handle lipids and overexpress matrix proteins (Aug 2019 news). The human astrocyte-expression profile was distinct from a previously described inflammatory astrocyte signature, the authors noted (Jan 2017 news). The R62H variant had little effect on astrocytes.

RNA-Seq only captures a small portion of the total cells in a tissue, and requires nucleotide amplification, which can introduce errors. Therefore the authors carried out a quantitative analysis of gene expression in bulk AD brain tissue, as well. They found a dearth of transcripts for genes involved in synaptic transmission, axon guidance, and cell survival. These changes probably reflect neuron loss, Colonna said. He believes the alterations in astrocytes and oligodendrocytes may be secondary to neuronal loss as well.

“I think the major difference between human AD and mouse models is that there is much more neuronal cell death in human disease,” Colonna said. Liddelow thinks this has broad implications for understanding disease. “Changes in one cell type may not necessarily mean these cells are … driving disease pathology, but instead may simply be responding to the changing metabolic needs of the cells around them,” he wrote.

To look more specifically at the effects of TREM2 mutations, the authors analyzed bulk gene expression of parietal cortex samples from five carriers of the R47H TREM2 mutation who had AD, and compared the findings with five control AD brains. These samples came from the Alzheimer’s Disease Research Center at WashU. R47H carriers had a less-pronounced microglial response to disease, with lower levels of expression of IRF8 and downstream genes. The effect of R47H was stronger than that of R62H, the authors noted, in agreement with the greater AD risk conferred by the former.

Intriguingly, some expression differences between TREM2 genotypes did emerge, noted Zhuoran Yin in Butovsky’s group. In the Rush R62H AD samples, some homeostatic microglial genes, such as TMEM119 and TGFB1, were suppressed compared to AD brain without a TREM2 mutation, she pointed out. In the WashU R47H cohort, this did not happen. It is unclear if the difference is due to TREM2 genotype, the brain region examined, the demographics of the cohort, or some other reason.

Overall, data from this study agrees with previous single-nucleus RNA studies of human brain. There are some differences, though. For example, a recent study of expression in AD entorhinal cortex reported a downregulation of microglial homeostatic genes, and an upregulation of myelination genes in oligodendrocytes, opposite to the effects seen by Colonna and colleagues (Nov 2019 news). 

Colonna is now examining larger, more diverse human brain samples, hoping to parse out differences due to demographics and brain region. He will also try to correlate gene-expression changes with disease progression, to see if expression changes substantially from one stage to the next.

Lastly, the work might uncover new biomarkers. Several of the genes identified as altered in AD brain, such as NfL, YKL-40, SORL1, BDNF, and ADCYAP1, encode known fluid biomarkers for the disease. “By examining this transcriptome database, we may identify additional biomarkers for disease progression,” Colonna suggested.—Madolyn Bowman Rogers

Comments

  1. This study from Yingyue Zhou from the Colonna lab is exactly the sort of study that the field should be doing more of—comparing actual human transcriptomic analyses with those from animal models is the key to determining which cell types (or subtypes of cells) we are effectively able to model in a disease state in rodents. There are so many different AD models available to researchers, with more being produced each year, but researchers need to remember that not all models are equally appropriate for studying any one particular cell, pathway, or state of disease. This comparison of human and 5XFAD mouse cells is important, and something that should be done more often. We need to have a more comprehensive understanding of the subsets of disease-associated cells (e.g., microglia, astrocytes, etc.) that are properly recapitulated in current animal models. When we find a subset that is similar, we should study it, but if we do not find similarities we should not despair, but instead look for more appropriate models. Human responses must be the gold standard for studying disease-relevant functional deficits.

    Two things struck me in particular that will no doubt spark much discussion and further investigation:

    1. It was surprising, after making such a biologically important comparison, how the discussion of C4b+Serpina3n+ oligodendrocytes was conducted … "In the 5XFAD model, oligodendrocytes adopted a reactive signature including C4b and Serpina3n that was not evident in human AD samples." To me this highlights a point that the field must start to take notice of and address together. If a phenotype/subtype/transcriptomic signature is not present in human patients, is it a viable discovery for future investigation into the disease? What do we do now with the knowledge of a rodent-specific response? Is this a response that does not occur in human patients, or is this subtype functionally similar to one in humans, but with different markers and mechanisms of activation?

    2. "Neuronal loss in AD may dampen the need for astrocytic scavenger functions devoted to disposal of neuronal toxic waste. In parallel, astrocytes upregulated a signature indicative of extracellular matrix protein synthesis related to glial scarring." Again this is an interesting finding—and the biological implications are likely to open up a new avenue of investigation. It highlights that changes in one cell type may not necessarily mean these cells are reactive in a way that is driving disease pathology, but instead may simply be responding to the changing metabolic needs of the cells around them. An alternative reason for this may be that the samples analyzed are not 100 percent comparable—that is, whole cortex from mouse compared with a small region of human brain that may contain amyloid plaques that have a known glial scarring component associated with them. It will be interesting to determine if these signatures are common across the different pathological regions of the brain—or even which subsets of these astrocytes (or microglia) are present in early versus late stages of disease. This is not to be taken as a shortcoming of the paper at all, but just the nature of such studies that we all as a field must deal with.

    Overall this paper produces an incredibly rich data set for the field, one that will be mined heavily to determine new testable hypotheses, and particularly provide a comparative basis for any group that completes a similar single-cell analysis of other rodent AD models. The study also highlights some of the major problems we face in this field—that of cross-species differences, models that do not completely recapitulate all aspects of a disease, and differences in human postmortem sample pathologies. None of these difficulties are contained to a single laboratory, but studies like this one pave the way for expanding our combined community data set so that we can mitigate future mistakes of studying the wrong cellular responses.

  2. Among the now several “-omics” papers on Alzheimer’s disease that have emerged in the last few years, this new report by Zhou et al. stands out because of its comprehensive analysis of mouse and human tissue at the single -cell level. One of the more obvious and important results of this new report is the marked difference in gene-transcript changes that are observed in human and the mouse model of disease. These findings add to a growing literature that brings into question the validity of these AD mouse models, which has worrying implications for drug discovery and disease modelling in the laboratory.

    We noted that gene-ontology analysis in microglia identified metal-ion (especially iron) homeostasis as the most significant affected pathway in microglia of AD patients. This is perhaps surprising since most iron-related genes are more prominently regulated by iron-response binding proteins at the level of translation, whereas transcripts were used in the analysis of this paper. Perhaps even more striking changes would be observed if protein were measured? Downregulation of this pathway leading to “metalostasis”—impaired trafficking of metal ions—can have substantial impact on brain health, and we believe impacts on AD pathogenesis. Indeed, our recent clinical findings (Ayton et al., 2019Ayton et al., 2017; Ayton et al., 2017; Ayton et al., 2015) and prior laboratory work (Duce et al., 2010; Lei et al., 2012) support a substantial role for the damaging effects of impaired iron trafficking in AD. The fact that oxidative pathways also emerged as significant in oligodendrocytes also suggests an important role for damaging iron chemistry in AD. Ferroptosis, a recently described cell-death pathway that is dependent on iron-induced lipid peroxidation, is a potential mechanism of neurodegeneration in AD. SEPP1, a selenium transporting protein that is required for the functioning of the master regulator of ferroptosis, GPX4, also changed significantly, and further supports this pathway in the neurodegenerative mechanism of AD.

    References:

    . Brain iron is associated with accelerated cognitive decline in people with Alzheimer pathology. Mol Psychiatry. 2019 Feb 18; PubMed.

    . Association of Cerebrospinal Fluid Ferritin Level With Preclinical Cognitive Decline in APOE-ε4 Carriers. JAMA Neurol. 2017 Jan 1;74(1):122-125. PubMed.

    . Cerebral quantitative susceptibility mapping predicts amyloid-β-related cognitive decline. Brain. 2017 Aug 1;140(8):2112-2119. PubMed.

    . Ferritin levels in the cerebrospinal fluid predict Alzheimer's disease outcomes and are regulated by APOE. Nat Commun. 2015 May 19;6:6760. PubMed.

    . Iron-export ferroxidase activity of β-amyloid precursor protein is inhibited by zinc in Alzheimer's disease. Cell. 2010 Sep 17;142(6):857-67. PubMed.

    . Tau deficiency induces parkinsonism with dementia by impairing APP-mediated iron export. Nat Med. 2012 Feb;18(2):291-5. PubMed.

  3. I wish to congratulate Zhou et al. for this exceptional article. One of the genes that attracted my attention was SERPINA3 (aka alpha-1-antichymotrypsin). We published that alpha-1-antichymotrypsin (ACT) was intimately associated with amyloid plaques in AD brains (Abraham et al., 1988), and that bigenic mice overexpressing APP and ACT develop twice as many plaques as APP tg mice (Mucke et al., 2000).

    We also reported that ACT is primarily produced in reactive astrocytes (Pasternack et al., 1989; Koo et al., 1991). Finally, we showed that ACT and Aβ form a stable complex in vitro (Potter et al., 1991). We hypothesized at that time that ACT is involved in the stability of the Aβ plaques.

    References:

    . Immunochemical identification of the serine protease inhibitor alpha 1-antichymotrypsin in the brain amyloid deposits of Alzheimer's disease. Cell. 1988 Feb 26;52(4):487-501. PubMed.

    . Astroglial expression of human alpha(1)-antichymotrypsin enhances alzheimer-like pathology in amyloid protein precursor transgenic mice. Am J Pathol. 2000 Dec;157(6):2003-10. PubMed.

    . Astrocytes in Alzheimer's disease gray matter express alpha 1-antichymotrypsin mRNA. Am J Pathol. 1989 Nov;135(5):827-34. PubMed.

    . The Alzheimer amyloid components α1 antichymotrypsin and β-protein form a stable complex in vitro. Alzheimer's Disease: Basic Mechanisms, Diagnosis and Therapeutic Strategies: edited by K. Iqbal, D.R.C. McLachlan, B. Winblad and H. M. Wisniewski. 1991

    . Developmental expression of alpha 1-antichymotrypsin in brain may be related to astrogliosis. Neurobiol Aging. 1991 Sep-Oct;12(5):495-501. PubMed.

  4. This is a very informative and important paper. We should remember that there are no "mouse models of AD." Maybe we will never see any animal model that fully mimics Alzheimer’s disease (see Walker and Jucker, 2017).

    What we have are mouse models that allow us to dissect many of the molecular and cellular mechanisms that drive Aβ-amyloidosis and tauopathy. As the authors point out, the microglial transcriptome in AD and murine models of Aβ-amyloidosis only partially overlap, and this finding will hopefully help to dissect the contribution of Aβ-amyloidosis to AD.

    References:

    . The Exceptional Vulnerability of Humans to Alzheimer's Disease. Trends Mol Med. 2017 Jun;23(6):534-545. Epub 2017 May 5 PubMed.

  5. This is an excellent study providing new insights and a rich resource on TREM2-dependent glial responses in AD in mice and humans. One highlight of the study is the description of reactive oligodendrocytes in the 5xFAD mouse model of AD. Using snRNAseq, the authors discovered a Serpina3n+ population of oligodendrocytes reactive to Aβ plaques. These cells were also positive for carbonic anhydrase 2, providing evidence these cells are mature oligodendrocytes.

    While there is extensive literature on reactive microglia and astrocytes, we know very little about reactive oligodendrocytes. How are they formed? What are their functions? Zhou et al. show that Serpina3n+ oligodendrocytes are partially dependent on TREM2, suggesting that microglia activation is necessary to trigger reactive oligodendrocytes. Furthermore, they found that Serpina3n+ oligodendrocytes express C4b, a factor that promoted aggregation of Aβ, suggesting that the formation of reactive oligodendrocytes is a maladaptive response in the context of AD.

    When the authors compared glial responses in mouse and human, they observed important differences. Also, in human AD, oligodendrocytes appeared to be reactive, but these cells upregulated a distinct set of genes. These were genes involved in the control of osmotic, oxidative, and lipid metabolic pathways. It is, thus, conceivable that reactive oligodendrocytes also have beneficial functions, for example, by responding to neuronal injury by upregulating pathways that promote trophic and neuroprotective pathways.

    Are reactive oligodendroglia also present in other diseases? We know very little about reactive oligodendrocytes, but activated oligodendrocyte precursor cells have been described. For example, recently, the lab of Goncalo Castelo-Branco and the lab of Peter Calabresi defined activated oligodendrocyte precursor cells in models of multiple sclerosis. These cells express MHC-II and are able to activate memory and effector CD4-positive T cells. It will now be interesting to dig deeper into these largely unknown functions of reactive oligodendroglia and into the signaling pathways that lead to their activation.

Make a Comment

To make a comment you must login or register.

References

News Citations

  1. Brain Conference Spotlights Transcriptomics, Therapeutic Strategies
  2. When It Comes to Alzheimer’s Disease, Do Human Microglia Even Give a DAM?
  3. Paper Alert: TREM2 Crucial for Microglial Activation
  4. TREM2: Diehard Microglial Supporter, Consequences Be DAMed
  5. Does High Iron Push a Person With Pathology Into Dementia?
  6. ApoE4 Glia Bungle Lipid Processing, Mess with the Matrisome
  7. Microglia Give Astrocytes License to Kill
  8. Single-Cell Expression Atlas Charts Changes in Alzheimer’s Entorhinal Cortex

Research Models Citations

  1. Trem2 KO (Colonna)
  2. 5xFAD (C57BL6)
  3. Trem2 KO (Colonna) x 5XFAD

Mutations Citations

  1. TREM2 R62H
  2. TREM2 R47H

Paper Citations

  1. . IRF8 is a critical transcription factor for transforming microglia into a reactive phenotype. Cell Rep. 2012 Apr 19;1(4):334-340. Epub 2012 Apr 5 PubMed.

Further Reading

Primary Papers

  1. . 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.