Multi-omics Analysis Sheds Light on Drivers of Alzheimer’s Disease, Other Neurodegeneration

By Kyle Proffitt 

May 24, 2022 | “I want to disabuse you of any illusion there exist black-box technologies where you can feed in data from multiple omics inputs and out comes interesting biology,” Shankar Subramaniam told the crowd in Boston earlier this month at the Bio-IT World Conference and Expo. “That’s a myth.”  

Subramaniam, Distinguished Professor of Bioengineering, Professor of Chemistry and Biochemistry and Nanotechnology, at the University of California, San Diego, knows biological context is important, and the devil is in the details. 

The challenge in biology, he warned, is not so much the scale of data but its heterogeneity—imaging, handwritten physicians’ notes, the variety of sequencing technologies currently available—and the lack of well-defined structures and standards. While he ascribes essential value to these data streams, he is wary of hype suggesting that simply collecting the data and giving it an AI fix will solve our problems.  

With this as framework, he discussed the audacious efforts and challenges of the UK Biobank and All of Us projects, intending to sequence up to 1 million genomes while including associated environmental and lifestyle data, but he countered these with a more granular study of a different sort—a detailed tracing of a family affected by early-onset Alzheimer’s Disease (AD). In this example, local psychiatric hospital records from Girifalco, Italy, dating to the 16th century, combined with a genealogy of a 331-member family, contributed to the discovery of Presenilin 1 (PSEN1) as the primary causal gene for familial Alzheimer’s Disease (FAD). Specific and distinct mutations in PSEN1 are essentially guaranteed to cause FAD. Subramaniam showed a world map with dots representing the linkages of different families with four distinct PSEN1 mutations, made possible through these microcosm-level genomics analyses. 

Despite the strong penetrance of PSEN1 mutations, just how they cause FAD is less clear. PSEN1 is a member of the 4-protein gamma secretase complex, perhaps best known for its role in processing amyloid precursor protein (APP). Gamma secretase dysfunction results in the accumulation of different versions of amyloid-beta (β) peptide, a hallmark of AD brains that was long believed to drive disease pathology.  

However, Subramaniam reminded the audience that effective gamma-secretase inhibitors have met with consistent failure in clinical tests, indicating that the buildup of β plaques is an inadequate explanation for disease etiology. He cited a 2012 Forbes article written on the heels of multiple canceled trials that stated boldly: “the amyloid hypothesis is dead”.  

Rather than a cause, “it’s a consequence of AD,” Subramaniam said, and he indicated the same opinion with regard to tau tangles. Beyond APP processing, PSEN1 activity includes Notch intracellular domain processing and β-catenin interactions, and these activities have important roles in disease, as Subramaniam went on to discuss. 

De-Differentiation Hypothesis 

Subramaniam’s lab became involved in AD research five to six years ago, but they needed a new angle to identify underlying mechanisms. They applied a focused multi-omic analysis, starting with FAD because of the direct line between PSEN1 mutations and disease and the availability of patient samples with these mutations. With sporadic AD, Subramaniam says “there’s no question there is a genetic background, but if you don’t sequence a billion people, we’re not going to know what are these mutations.” This provides support for All of Us-level and much larger population genomics efforts. 

For their studies, Subramaniam’s lab used fibroblasts from non-demented controls and FAD patients with different disease-causing PSEN1 mutations to create induced pluripotent stem cells, iPSCs, which were then differentiated into neurons in culture. At this point, the differentiated neurons were subjected to a battery of RNA-seq, ATAC-seq, and ChIP-seq analyses to assess changes to transcription, the chromatin landscape (open and closed regions), and epigenetic modification (histone methylation). Collecting and overlaying these datasets revealed common and significant changes; PSEN1-mutant neurons consistently displayed a more de-differentiated state, repressed neuronal identity, re-entry of the cell cycle, and a more inflammatory profile. “Your neurons cease to behave like neurons, they try to reverse back into some plastic state,” Subramaniam described.  

The transcriptional changes lined up in many cases with the chromatin landscape and histone methylation markers. Subramaniam believes that the “chromatin really defines the transcription, the state of the cell”. His group went further to identify the basis of chromatin alteration, and he showed a slide based on transcription factor target enrichment in which a group of just 10 miRNAs and transcription factors explain the networks of altered chromatin and transcription yielding distinct AD endotypes—the mechanistic, pathobiological bases for disease pathology, such as downregulated neuronal lineage. These nodes driving the endotypes represent potential targets.  

Importantly, comparison of these results to prior transcriptional analyses of PSEN1-mutant human brain samples revealed strong overlap. Subramaniam also applied these methods to sporadic AD brain samples. “The brains from human postmortem samples showed exactly the same endotypes,” he found. Based on these correlations, although much of his work has focused on early-onset FAD, which accounts for only 5% of AD, Subramaniam believes his results are informative for all AD cases.  

Subramaniam also tested differentiated neurons with known FAD-causative mutations in APP and PSEN2, revealing the same patterns. Including these datasets, he identified correlation between the time of FAD onset associated with a particular mutation (more “severe” mutations show earlier onset) and the magnitude of neuronal repression. This suggests that fibroblasts could be isolated from AD patients by skin biopsy and used to establish iPSCs and then neurons, and the chromatin landscape and transcriptional profile could be assessed to provide a differentiation “score”. Less neuronal dedication would be indicative of a worse prognosis, and this method could provide a metric for disease staging. 

With an eye toward improved treatment, Subramaniam returned to the failed gamma secretase inhibitors to understand their effects on chromatin and transcription in PSEN1-mutant neurons. The inhibitors only partially restored neuronal differentiation, and Subramaniam believes that this incomplete restoration explains the drugs’ ultimate lack of success. However, by focusing on the AD endotypes and the ability of treatments to reverse them, he believes better drugs or combinations can be more easily identified. 

Perhaps even more exciting, Subramaniam is finding similar phenomenology in Parkinson’s disease, Huntington’s disease, traumatic brain injury, even glioma—an “extreme case of de-differentiation”. Subramaniam sees a common underlying theme to neurodegenerative disorders in which cells trace their way back from a defined lineage-specified state to a de-differentiated, intermediate, plastic state.  

He shared his pet theory that de-differentiation is fundamentally a protective mechanism of human physiology—something that occurs in wound healing, for instance. “When there is stress in cells and tissues, there’s a programming message to go back into a protective state.” However, in the case of neurons, there is an inability at some point to recover the differentiated state; the loss of synapses is irrecoverable. But Subramaniam remains hopeful: “If we can figure out how to address this path, prevent it from going from the neuronal state, with early detection and intervention at this level, we should be able to reverse the path of neurodegeneration.”