The gene variant APOE4 is the most common genetic risk factor associated with late-onset Alzheimer's disease. A comprehensive, multilayer study reveals the molecular and cellular signatures of APOE4 in humans. See Article p.45
Alzheimer's disease is a genetically complex neurodegenerative condition typically associated with ageing. Reporting in this issue, Rhinn et al.1 (page 45) use an integrative genomic approach based on analysis of transcriptional networks in human brain to uncover new relationships between this disorder, ageing, and the gene apolipoprotein E, or APOE, shedding light on molecular mechanisms that connect this gene to the processing and transport of amyloid precursor protein — considered to be a primary effector of the disease. To validate their hypotheses, the authors follow bioinformatic analyses with multiple levels of experimentation, impressively demonstrating the power of network biologyFootnote 1.
Recent genome-wide association studies (GWAS) and transcriptomic studies have implicated2,3 a wide set of pathophysiological mechanisms in Alzheimer's disease, including innate immunity, cellular uptake of molecules by the process of endocytosis and their intracellular trafficking. However, a variant of APOE called APOE4 remains the most common genetic risk factor, with the largest effect in late-onset Alzheimer's disease (LOAD)4. Therefore, understanding how APOE4 conveys disease risk and its relationship with ageing remains a research imperative.
Rhinn et al. start with a simple, but previously unexplored, set of comparisons using published data on gene expression in the cerebral cortex region of the human brain to address a question widely debated in the field: do the molecular mechanisms underlying LOAD differ between those who carry APOE4 and those who do not?
To discern the independent effects of disease and APOE status, the authors compare APOE4-negative healthy individuals with both APOE4-negative LOAD patients and APOE4-positive healthy individuals. Remarkably, they find that the total cellular transcript (transcriptome) profile of healthy APOE4 carriers overlaps to a large extent with that of LOAD patients, and that the overlapping set of transcripts (designated A/E) most closely resembles the changes in transcript profile observed during the transition from normal to incipient LOAD — an early phase of Alzheimer's disease. These data indicate that the APOE4 status of an individual reflects a prodromal, or 'primed', state for developing LOAD.
Notably, transcriptional changes observed during normal ageing (through comparison of those under 75 years of age with those over 85) overlapped significantly with LOAD, but not with changes specific to APOE4 carriers. That certain transcriptional alterations in LOAD are seen either with ageing or with APOE4 status suggests that the contributions of APOE4 to LOAD are mechanistically distinct from those of ageing. A potential caveat of this analysis, however, is the use of a hard categorical cutoff of age 75, rather than, for example, 65 — the age below which Alzheimer's disease is considered early onset. In this regard, a non-categorical quantitative analysis of transcripts associated with ageing should provide maximum analytical power.
Network-level co-expression analyses have been used successfully before to investigate disease-related molecular pathways in various neuropsychiatric5,6 and neurodegenerative disorders, including Alzheimer's disease3,7; such analyses permit an unbiased, systems-level understanding of disease mechanism. Rhinn and colleagues' work differs from other network-based co-expression approaches in focusing on differential co-expression — that is, on changes in co-expression patterns observed between control and disease samples. The authors, who used a transcriptome-wide network-analysis tool called differential co-expression correlation analysis (DCA), provide extensive experimental support for the targets identified, which adds an unprecedented level of validation.
Of the 215 differentially expressed genes Rhinn et al. found by comparing APOE4 carriers and non-carriers, 20 are top candidate 'network node' genes on the basis of their DCA ranking. Many of the node genes had been previously implicated in regulating amyloid precursor protein (APP) processing and intracellular sorting8, supporting their relevance.
The authors performed various in vitro experiments to validate several of the node genes, in particular SV2A and RNF219. In cell lines and in human induced neurons, obtained by the process of cellular reprogramming, they show that SV2A affects APP processing through the regulation of endocytosis and intracellular trafficking. They also find that RNF219, which had not been previously associated with APP processing or Alzheimer's disease, acts as a nuclear regulatory mediator of APOE4-induced APP endocytosis and its processing by the β-secretase enzyme.
The researchers muster and analyse multiple independent data sets, including published GWAS, which provide further support for the key role of the DCA hits in APOE4-mediated susceptibility to Alzheimer's disease, and make a case for further testing and replication studies using this gene set. For example, they show that, together with the APOE4 status, common genetic variants of RNF219 and FYN — another node gene with a role in endocytosis — are correlated with age of LOAD onset. Indeed, using publicly available human positron-emission tomography imaging data from the Alzheimer's Disease Neuroimaging Initiative, Rhinn et al. find that variations in RNF219 significantly affect APP processing in APOE4-negative healthy individuals — consistent with a prodromal role of this, and possibly other, A/E genes.
Overall, the present paper significantly advances our knowledge of molecular and genetic mechanisms that modify LOAD risk, further implicating defects in endocytosis and intracellular trafficking in Alzheimer's disease development9,10. The identification of key regulatory molecules and pathways involved in APOE-based risk for LOAD sets the stage for investigating their link to other deficits in this disorder, including immune function and metabolism of hyperphosphorylated tau protein, which forms the disease's characteristic neurofibrillary tangles. Ultimately, such investigations should lead to molecularly targeted therapeutics.
By demonstrating distinct mechanisms that depend on APOE status in LOAD, another implication of this work is that different genetic forms or subtypes of Alzheimer's disease may have specific underlying mechanisms and respond differently to treatment. If so, as with cancer, therapies based on genetic and genomic diagnostics may be most effective.
This work is also exemplary in demonstrating the extraordinary value of publicly available data resources. Published data on human gene expression, Alzheimer's disease GWAS and neuroimaging provide the pillars of Rhinn and collaborators' paper. Integrative analyses of these data by the authors, and previously by others3,7, weaken the view that substantive biological experimentation only takes place at the wet bench, and highlight the value of innovative re-analyses of existing data6.
*This article and the paper under discussion1 were published online on 24 July 2013.
Rhinn, H. et al. Nature 500, 45–50 (2013).
Bettens, K., Sleegers, K. & Van Broeckhoven, C. Lancet Neurol. 12, 92–104 (2013).
Zhang, B. et al. Cell 153, 707–720 (2013).
Wijsman, E. M. et al. PLoS Genet. 7, e1001308 (2011).
Voineagu, I. et al. Nature 474, 380–384 (2011).
Geschwind, D. H. & Konopka, G. Nature 461, 908–915 (2009).
Miller, J. A., Oldham, M. C. & Geschwind, D. H. J. Neurosci. 28, 1410–1420 (2008).
Minami, S. S., Clifford, T. G., Hoe, H. S., Matsuoka, Y. & Rebeck, G. W. Neurobiol. Aging 33, 825.e15–825.e24 (2012).
Ihara, Y., Morishima-Kawashima, M. & Nixon, R. Cold Spring Harb. Persp. Med. 2, a006361 (2012).
Small, S. A. & Gandy, S. Neuron 52, 15–31 (2006).
About this article
Advancing computer-aided drug discovery (CADD) by big data and data-driven machine learning modeling
Drug Discovery Today (2020)
Annual Review of Pharmacology and Toxicology (2020)
Machine learning and big data: Implications for disease modeling and therapeutic discovery in psychiatry
Artificial Intelligence in Medicine (2019)
Frontiers in Medicine (2018)