The Alzheimer’s disease proteome has remained mostly a mystery. At the EMBO/EMBL Symposium on Mechanisms of Neurodegeneration, held June 14-17 in Heidelberg, Germany, Nick Seyfried described one of the first major attempts to chart this unexplored territory. Working with Allan Levey at Emory University in Atlanta, Seyfried has analyzed levels of more than 10,000 proteins from about 40 control, Alzheimer's disease (AD) and Parkinson's disease (PD) patients and controls. He used bioinformatic network analysis to uncover suites of proteins that correlated with Aβ and tau pathology. This work, he hopes, will help identify drivers of disease pathology or new biomarkers. “The Emory team is the only one doing large-scale AD proteomics; they are pushing the limits of what you can do with mass spectrometry,” said meeting co-organizer Todd Golde from the University of Florida, Gainesville. “To have a centralized repository of this size and scope will be incredibly valuable to the community,” Golde said.
The project is funded by the Alzheimer’s disease initiative of the Accelerating Medicines Partnership, a joint venture between the National Institutes of Health, the Food and Drug Administration, and 10 biopharmaceutical companies and multiple nonprofits. The AMP has tasked a consortium of academic teams to build and openly share large, complex AD data sets (see Feb 2014 news).
Seyfried expects proteomics will complement existing transcriptomics data. Because translation is a regulated process, RNA transcript levels don’t always mirror the levels of the proteins they encode. Indeed, in isolated mouse brain cells, that correlation reaches 0.47 at most (Sharma et al., 2015). In addition, proteomics analyses can identify post-translational modifications, protein fragments such as soluble TREM2 or soluble APP, and proteins that find their way into the brain from the periphery. For example, Golde noted that several complement proteins are found at high levels in the cerebrospinal fluid in AD patients and probably accumulate in their brains. “Because AD is a proteinopathy, proteomics is as important, if not more so, than transcriptomics,” Golde said.
The Emory researchers have set up two analytical pipelines. One is quick, relatively inexpensive, and creates estimates of the most abundant proteins in high-throughput fashion. The second is slower and pricier, but dives more deeply into the proteome pool.
Seyfried showed how he used the quicker approach, known as label-free, single-shot proteomics, to quantify roughly 3,000 proteins in each of about 800 tissue samples. The hope is that these robust signatures could identify potential biomarkers. With its ability to process many samples at once, single-shot proteomics also facilitates protein network analyses, which rely on many replications for accuracy.
The second approach labels peptides with multiple chemical tags allowing the least abundant proteins to be retrieved and quantified. Seyfried said his team can now track about 10,000 proteins, mapping to more than 10,200 expressed genes covered in recent transcriptome efforts. Researchers called this a significant feat, considering that proteins cannot be amplified the way mRNA can.
In his talk, Seyfried covered data from the high-throughput technique (Seyfried et al., 2017). He has studied postmortem samples from the dorsolateral prefrontal cortex and the precuneus of 15 controls, 15 asymptomatic people who had AD as judged by postmortem CERAD and Braak staging scores, and 20 people who had had an AD diagnosis before death. All came from the Baltimore Longitudinal Study of Aging. The researchers identified 5,130 different proteins, but focused on 2,735 that were found in at least 90 percent of the samples from each brain region. The abundance of two peptides corresponding to residues 6-16 and 17-28 of Aβ strongly correlated with CERAD scores of amyloid pathology, lending confidence to the approach.
To better assess differences between AD and control samples, the researchers turned to weighted co-expression network analysis, following the lead of previous transcriptomics studies (Zhang and Hovarth, 2005; Miller et al., 2008; Miller et al., 2013). “This analysis correlates expression of proteins in an agnostic manner, without using information about the neuropathology of the samples,” said Seyfried. First it compares each protein, pairwise with others in each sample, and then clusters them into highly correlated groups, or modules. Seyfried found 16 such modules in the BLSA samples. Because the results from the prefrontal and precuneus regions were highly similar, he combined the data to gain power.
The largest module comprised mostly neuron-specific proteins, such as those involved in synaptic transmission and neurite biology. Other modules reflected basic cellular functions, such as protein folding, or proteins found in particular organelles, such as mitochondria. Modules enriched in proteins specific to astrocytes, oligodendrocytes, and microglia emerged, judging by similarities to proteomes from isolated mouse brain cells. “We could infer from the modules what might be happening in specific cells without having to isolate them,” said Seyfried.
As expected from transcriptome studies, the researchers saw the largest AD-associated changes in modules enriched for cell type-specific proteins. However, the correlation between the calculated “eigenprotein,” aka the first principal component of a module, and CERAD and Braak staging differed for glia and neurons. For astrocyte- and microglia-enriched modules, the eigenprotein correlation strengthened as pathology worsened, whereas for neurons, the correlation weakened (see image above). This is in line with transcriptomic data, which found increases in modules associated with immune responses, in particular microglia, in early AD.
Changes in functional modules correlated with diagnosis. Levels of proteins involved in inflammation and apoptosis were higher in AD than controls. Curiously, apoptosis hub protein levels were slightly higher in controls than in asymptomatic AD patients, even though the latter have accumulated amyloid plaques. Seyfried wondered if some protective mechanism may be at play in individuals who have amyloid in the brain but are still cognitively normal.
While cell-specific proteomic modules overlapped with previously identified RNA modules, the overall overlap of protein and RNA, including non-cell-specific modules such as apoptotic or mitochondrial, was only 39 percent. Seyfried thinks this may reflect, in part, the distance between soma and axons. Protein changes in the latter would be picked up by proteomics, but not by transcriptomics if the transcripts stay near the nucleus. Also, protein from the periphery may build up to a greater extent in AD and aging brains as a result of disruption of the blood-brain barrier, he suggested
Are these changes cause or consequence of AD? Eric Reiman of the Banner Alzheimer’s Institute in Phoenix suspects it’s a bit of both. Working with Dietrich Stephan, now of the University of Pittsburgh but previously at the Translational Genomics Research Institute in Phoenix, Reiman had found that transcripts encoding some of the neuronal and mitochondrial proteins Seyfried identified were also downregulated (Liang et al., 2008). Seyfried believes proteomic network analysis will help reveal upstream triggers of disease, but Golde cautioned that bioinformatics will go only so far. “Deconvoluting causality is asking a lot from these data. This type of data set gives us a large number of samples as reference points. We’re not there yet with respect to causality,” he said.
Is network analysis enough to tease apart the roles of different cells in AD? Bart De Strooper of the U.K. Dementia Research Center at University College London worried that analyzing tissue lysates loses valuable information about the contributions of specific cells. A recent study, for example, revealed a specific subset of microglia that surround and engulf plaques (Jun 2017 news).
Seyfried and others agreed that single-cell analysis is the way of the future. The challenge lies in the teeny amount of material in each cell. “Single-cell proteomics is advancing fast, but it’s not ready for high-throughput [analysis],” Seyfried said. Approaches to labeling proteins for capture and detection are improving, as are the mathematical tools for analysis, though Golde cautioned that every additional manipulation adds cost and contributes noise to the data.
Konrad Beyreuther from Heidelberg University was concerned that the analysis misses insoluble proteins, and Seyfried agreed this should be addressed. “The insoluble proteome likely holds a lot of clues,” said Golde, noting that changes in protein aggregation would be of interest.
Seyfried and Levey are open to sharing their data. “We want to get it out to the public as soon as possible,” said Seyfried. He plans to submit to Scientific Data Nature, an open collection of peer-reviewed data sets. Seyfried said he is also hoping Sage Bionetworks in Seattle will help link his proteomics data sets with transcriptomics data sets. Reiman and Golde were enthusiastic about such resources. “I think these rich repositories will be mined for years to come,” said Golde. “It will allow us to work faster and prevent duplicative efforts.”—Marina Chicurel
- Sharma K, Schmitt S, Bergner CG, Tyanova S, Kannaiyan N, Manrique-Hoyos N, Kongi K, Cantuti L, Hanisch UK, Philips MA, Rossner MJ, Mann M, Simons M. Cell type- and brain region-resolved mouse brain proteome. Nat Neurosci. 2015 Dec;18(12):1819-31. Epub 2015 Nov 2 PubMed.
- Seyfried NT, Dammer EB, Swarup V, Nandakumar D, Duong DM, Yin L, Deng Q, Nguyen T, Hales CM, Wingo T, Glass J, Gearing M, Thambisetty M, Troncoso JC, Geschwind DH, Lah JJ, Levey AI. A Multi-network Approach Identifies Protein-Specific Co-expression in Asymptomatic and Symptomatic Alzheimer's Disease. Cell Syst. 2017 Jan 25;4(1):60-72.e4. Epub 2016 Dec 15 PubMed.
- Zhang B, Horvath S. A general framework for weighted gene co-expression network analysis. Stat Appl Genet Mol Biol. 2005;4:Article17. PubMed.
- Miller JA, Oldham MC, Geschwind DH. A systems level analysis of transcriptional changes in Alzheimer's disease and normal aging. J Neurosci. 2008 Feb 6;28(6):1410-20. PubMed.
- Miller JA, Woltjer RL, Goodenbour JM, Horvath S, Geschwind DH. Genes and pathways underlying regional and cell type changes in Alzheimer's disease. Genome Med. 2013 May 25;5(5):48. PubMed.
- Liang WS, Dunckley T, Beach TG, Grover A, Mastroeni D, Ramsey K, Caselli RJ, Kukull WA, McKeel D, Morris JC, Hulette CM, Schmechel D, Reiman EM, Rogers J, Stephan DA. Altered neuronal gene expression in brain regions differentially affected by Alzheimer's disease: a reference data set. Physiol Genomics. 2008 Apr 22;33(2):240-56. PubMed.
- Savas JN, Wang YZ, DeNardo LA, Martinez-Bartolome S, McClatchy DB, Hark TJ, Shanks NF, Cozzolino KA, Lavallée-Adam M, Smukowski SN, Park SK, Kelly JW, Koo EH, Nakagawa T, Masliah E, Ghosh A, Yates JR 3rd. Amyloid Accumulation Drives Proteome-wide Alterations in Mouse Models of Alzheimer's Disease-like Pathology. Cell Rep. 2017 Nov 28;21(9):2614-2627. PubMed.