Expression Analysis Uncovers Three Distinct Forms of Alzheimer’s

Alzheimer’s disease comes in distinct molecular flavors, according to a study in the January 6 Science Advances. Researchers led by Bin Zhang at Icahn School of Medicine at Mount Sinai, New York, compared bulk RNA-Seq data from AD and control brains and identified three main forms of the disease, characterized by different gene expression profiles and molecular drivers. One of these, dubbed type C, fit a classic Alzheimer’s pathological profile, with numerous plaques and tangles as well as neuroinflammation. In types A and B, however, tau pathology predominated, with fewer plaques evident. Type A was distinguished by neuronal hyperexcitability, B by loss of oligodendrocytes. At a more granular level, B and C could be broken down further into two related groups, for a total of five molecular subtypes.

  • RNA-Seq analysis defines three distinct varieties of Alzheimer’s disease.
  • One fits the classic profile, with neuronal loss, neuroinflammation, and high plaque load.
  • In the other two, tau pathology predominates over plaques.

Zhang noted that each subtype seems to be driven by distinct biological and molecular factors. Because the data come from postmortem brain, it is unclear how the subtypes might relate to clinical symptoms and progression. “These subtypes should be considered in future mechanistic studies in AD drug development, and in clinical trials,” Zhang told Alzforum.

Nicholas Seyfried at Emory University, Atlanta, said that if the expression findings replicate at the protein level, they could provide biomarkers that would allow researchers to identify these subtypes in living patients. That, in turn, could help stratify clinical trials by the patients most likely to respond to a particular therapy, he suggested.

Genetic Drivers. Network analysis identifies key genes that are down (left) or up (right) in the five molecular subtypes of AD. [Courtesy of Neff et al., Science Advances/AAAS.]

Previous studies of gene expression in the brain have compared AD with age-matched controls, but have not looked for different forms of AD (May 2019 news; Nov 2019 news). To do so, first author Ryan Neff analyzed data from 364 human brains in the Mount Sinai brain bank. RNA-Seq data was available from four different brain regions: frontal pole, superior temporal gyrus, inferior frontal gyrus, and parahippocampal gyrus. In the frontal pole and superior temporal gyrus, gene expression was almost unchanged in AD brain compared to control, while in inferior frontal gyrus, changes were modest. However, the parahippocampal gyrus (PHG), which surrounds the hippocampus and includes the entorhinal and perirhinal cortex, was strongly affected by AD, with expression changes in 3,571 genes.

The authors thus focused on PHG, for which they had samples from 151 AD and 64 control brains. AD was clinically defined as a CDR greater than 1 at the time of death. Pathologically, all AD brains contained plaques and tangles, but no Lewy bodies or vascular disease. The authors searched for patterns in the gene expression changes using weighted sample gene network analysis; this method clusters genes that change together. They corrected data for age at death, sex, race, postmortem interval, and RNA integrity. They also corrected their findings for clinical severity and neuronal loss, to ensure the expression changes did not simply reflect disease stage. Supporting this, Braak scores did not associate with any of the subtypes that came out of this analysis.

Zhang and colleagues identified three distinct patterns among the 151 AD parahippocampal gyri. Nearly an equal number of gyri fell into each group, with 47 As, 54 Bs, and 50 Cs. Intriguingly, there was no difference in cognition, as measured by CDR score, among the three types.

In type C, genes responsible for Aβ binding, clearance, and fibril formation were highly expressed in the gyrus, as were neuroinflammation and endothelial genes. Apoptosis genes were also up. Meanwhile, synaptic transmission genes were dampened. Pathological findings supported this, with these brains having about 50 percent greater plaque density than types A and B. C brains had substantial neuronal loss in the PHG, but more astrocytes, oligodendrocytes, and endothelial cells compared to controls. Network analysis pointed to the neuronal genes GABRB2, SCN2A, and amphiphysin 1, all of which were down in C brains, as potential drivers of this phenotype.

The scientists subdivided Type C into two subtypes, C1 and C2. At a genetic level, C1 was more likely than the other subtypes to be found in APOE4 carriers; this allele associates with plaque burden and AD risk. Conversely, C2 was not enriched among APOE4 carriers. Overall, this subtype had fewer expression changes than did C1, with Aβ and apoptosis gene pathways the same as in control brain.

Curiously, A brains were in some ways the opposite of C brains, despite the shared Alzheimer’s diagnosis. Neuronal and synaptic genes were up, and immune and endothelial genes down. Key regulators of this type included an increase in neuronal genes GABRB2, SCN2A, and SYT1, and a decrease in glial genes TLN1, LRP10, and NOTCH1. These brains also had revved-up expression of genes encoding parts of the protein degradation machinery in the PHG. This may explain why these samples had fewer neurofibrillary tangles in the inferior parietal lobe and medial frontal gyrus compared to B and C brains. Mirroring the expression changes, type A brains had slightly more neurons than controls, but fewer astrocytes, oligodendrocytes, and endothelial cells. This type may represent a neuronal excitability phenotype, the authors suggested.

Finally, demyelination may drive type B. Oligodendrocyte genes that regulate myelin formation and maintenance, such as PLP1, Ermin, and Quaking, were down in AD compared to controls, and these brain samples had fewer oligodendrocytes. Meanwhile, genes responsible for acid secretion were up, hinting at possible dysregulation of lysosome acidification. This type also broke into two subtypes. The B1 subtype turned down immune and oxidative stress genes, while B2 turned up immune genes. B2 was more commonly found in APOE2 carriers.

How robust are these findings? To get a sense of this, the authors evaluated RNA-Seq data from the dorsolateral prefrontal cortices (DLPFC) of 391 AD, 64 mild cognitive impairment, and 160 control brains from the Religious Orders Study–Memory and Aging Project (ROSMAP). This cohort includes participants from across the U.S. and is more ethnically diverse than the Mount Sinai brain bank samples. Even so, the samples predominantly fell into the same three types, with 46 percent of them C, 24 percent A, and 21 percent B. Nine percent of the ROSMAP brains had a distinct gene expression pattern that did not match any of these types. MCI brains exhibited the same patterns as AD brains, again showing that these subtypes do not depend on disease stage.

Finally, Zhang and colleagues investigated whether different mouse models might resemble specific subtypes of AD. They analyzed RNA-Seq data from 19 mouse models collected by the Accelerated Medicines Partnership for AD (AMP-AD). Amyloidosis models, such as 5XFAD, APP Swedish, and APP Dutch mice, had a profile resembling type C. The tauopathy mouse P301L matched type A. Other mice that feature tau-related neurodegeneration, such as those carrying mutant clusterin, BIN1, and CD2AP, were the best matches for type B. The data may help explain why therapies that work in specific mouse models do not translate well to heterogenous human populations, Zhang proposed.

Other researchers have sought to classify AD using proteins rather than expression data. Pieter Jelle Visser and Betty Tijms at Amsterdam University Medical Center recently profiled cerebrospinal fluid from AD patients and controls using 556 proteins. They identified three AD subtypes by this method (Aug 2019 conference news; Tijms et al., 2020). The data dovetail strikingly well with Zhang’s expression data, Visser told Alzforum. For example, their subtype 1 has high levels of synaptic proteins, like Zhang’s type A. Subtype 2 has high immune genes, like B2, and subtype 3 has low synaptic proteins and high blood-brain barrier proteins, like C.

“The convergence … strongly supports the presence of AD subtypes with different underlying molecular pathophysiologies. This will have major clinical implications, as different subtypes may need different treatments,” Visser wrote to Alzforum.

Researchers agreed on the need for new fluid biomarkers to distinguish subtypes. The current ATN markers of amyloid, tau, and NfL cannot capture the complexity of the disease, Seyfried said. Zhang plans to incorporate peripheral biomarkers and clinical features into the characterization of his subtypes.—Madolyn Bowman Rogers

Comments

  1. Biological heterogeneity of Alzheimer’s disease is a relatively unexplored territory, even though there is ample evidence based on clinical and imaging studies that heterogeneity does exist. Here, Neff and colleagues explored heterogeneity in gene expression profiles in brain tissue of individuals with pathologically confirmed AD.

    Clustering people based on their gene expression profiles revealed at least three molecular subgroups (A, B, C) of which B and C were subdivided (B1, B2, C1, C2). Subtypes A and B1 showed, relative to control brains, increased expression of genes involved in synaptic activity and regulation of tau, as well as lower expression of genes involved in immune function and the blood brain barrier (BBB). Subtype B2 showed increased expression of synaptic proteins and immune/BBB genes. Subtypes C1 and C2 mirrored subtype A and B1 by showing decreased expression of synaptic proteins and increased expression of immune/BBB genes.

    It is striking that some of these subtypes resemble the three subtypes we recently identified in individuals with AD using cerebrospinal fluid (CSF) proteomic analysis of 556 proteins (Tijms et al., 2020). Our subtype 1 showed increased expression of synaptic and plasticity proteins, like subtype A and B1. Our subtype 2 showed increased levels of proteins involved in synaptic activity and immune function, like subtype B2. This subtype had higher AD polygenic risk scores of immune related genes compared to the other subtypes. Our subtype 3 showed decreased concentrations of proteins involved in synaptic function and increased concentrations of proteins associated with BBB dysfunction, like subtype C1 and C2.

    We found that proteins increased in our subtype 1 were enriched for regulation by transcription factors REST and SUZ12, which are important regulators of synaptic plasticity. Using the top 100 genes with highest expression in subtype A from the Neff paper, we found that these genes were also regulated by REST/SUZ12 while the top 100 genes with lowest expression in subtype C were regulated by REST/SUZ12 as well. 

    The convergence between our CSF proteomic subtypes and Neff et al. gene expression subtypes strongly support the presence of AD subtypes with a different underlying molecular pathophysiology. This will have major clinical implications as the different subtypes may need different treatments.

    Also, the current staging system for AD based on amyloid, tau, and neuronal injury markers may need to be reconsidered as this system does not capture these subtypes. Because a wide range of biological processes seems to cluster around specific subtypes, a system may be needed that allows for subtype allocation, rather than scoring individuals on single biological processes. In such a scenario, amyloid pathology could be considered the defining feature of AD with a set of other markers to specify the subtype.

    View all comments by Pieter Jelle Visser View all comments by Betty Tijms

  2. In my opinion, this is the clearest indication that Alzheimer's disease is not a single entity, despite the ubiquity of amyloid plaques, and should not be treated as one. Therapies that seek to slow progression of dementia are unlikely to be effective in all subtypes. Biomarkers of all subtypes are less useful than biomarkers of individual subtypes; these signals can be diluted by inclusion of other subtypes.

    View all comments by Peter Lansbury

Make a Comment

To make a comment you must login or register.

References

News Citations

  1. When It Comes to Alzheimer’s Disease, Do Human Microglia Even Give a DAM?
  2. Single-Cell Expression Atlas Charts Changes in Alzheimer’s Entorhinal Cortex
  3. Proteomics Uncovers Potential Markers, Subtypes of Alzheimer’s

Paper Citations

  1. Tijms BM, Gobom J, Reus L, Jansen I, Hong S, Dobricic V, Kilpert F, Ten Kate M, Barkhof F, Tsolaki M, Verhey FR, Popp J, Martinez-Lage P, Vandenberghe R, Lleó A, Molinuevo JL, Engelborghs S, Bertram L, Lovestone S, Streffer J, Vos S, Bos I, Alzheimer’s Disease Neuroimaging Initiative (ADNI), Blennow K, Scheltens P, Teunissen CE, Zetterberg H, Visser PJ. Pathophysiological subtypes of Alzheimer's disease based on cerebrospinal fluid proteomics. Brain. 2020 Dec 1;143(12):3776-3792. PubMed.

Further Reading

Primary Papers

  1. Neff RA, Wang M, Vatansever S, Guo L, Ming C, Wang Q, Wang E, Horgusluoglu-Moloch E, Song WM, Li A, Castranio EL, Tcw J, Ho L, Goate A, Fossati V, Noggle S, Gandy S, Ehrlich ME, Katsel P, Schadt E, Cai D, Brennand KJ, Haroutunian V, Zhang B. Molecular subtyping of Alzheimer's disease using RNA sequencing data reveals novel mechanisms and targets. Sci Adv. 2021 Jan;7(2) Print 2021 Jan PubMed.

Annotate

To make an annotation you must Login or Register.