Third session of the Longevity Therapeutics summit. For coverage of previous sessions:
Hacking Aging: A Data-Driven Approach to Healthy Life Extension (Peter Fedichev, Founder & CSO, Gero LLC)
Gero employs dynamic systems and physicial kinetics theory, in conjunction with advanced machine learning, to understand human healthspan. The company’s perspective is based on the idea that aging emerges from molecular-level processes, and that the state of an organisms depends on a few “master variables” that could conceivably be detected and measured as biomarkers.
The company has defined a dynamic frailty index (DFI) defined in terms of the principal components of a multidimensional analysis of functional metrics; the distance along the DFI trajectory over the lifespan is a biomarker of age, and hence a predictor of the risk of age-related disease. Indeed, the biomarker developed by Gero is even more useful because it is effectively a metric of the amount of life remaining, rather than the amount of time that has passed since birth. At least one small molecule trug against selected targets decreased the DFI.
This analysis led to discovery of genomic regions that influence the rate of aging, described in the company’s recent paper, “Identification of 12 genetic loci associated with human healthspan.”
A key issue that came up during the question period is that mice and humans age very differently. As Fedichev put it, “Mice are born frail and die of frailty. Humans are born and remain non-frail for a long time.” The upshot of this is that a critical aspect of the aging process in humans simply cannot be studied in rodents—emphasizing the importance of human biomarker studies.
AI-Powered End-To-End Pipelines for Drug Discovery (Alex Zhavoronkov, Founder & CEO, Insilico Medicine)
Insilico Medicine is seeking to increase the success rate of the drug development pipeline by taking advantage of big data and artificial intelligence. They collaborate with a wide range of companies around the world, including large multinationals such as GSK, Novartis, and Johnson & Johnson.
The company has designed several ‘age clocks’ that uses various types of data to predict a person’s age, including gene expression data, microbiome sequence information, and even facial images. Their results indicate that transcriptomic data is the most valuable, but it is even more useful in combination with phenotype, longitudinal, and chemistry data. Earlier this year, the company published a paper showing that a very simple hematological clock could reveal the acceleration of aging due to cigarette smoking.
Currently, the company is seeking to identify natural compounds that target senescence pathways, with the ultimate goal of releasing them as nutraceuticals.
An excellent summary of the company’s work, as well as the efforts of others in the field, appears in a recent review by Zhavoronkov and colleages: “Artificial intelligence for aging and longevity research: Recent advances and perspectives.”
Data-driven Approaches to Identifying Circulating Factors That Drive Aging (Kristen Fortney, CEO, BioAge Labs)
BioAge is a company devoted to discovering and developing proprietary drugs for aging approach. They take a data-driven approach to target discovery, but because one can only go so far in silico, they also make considerable investment in in vivo validation.
We are in a unique time right now, Fortney argues, in which the rate of progress in aging research is increasing exponentialy: just 30 years ago we became able to genetically manipulate the lifespans of simple model animals, but now we are entering the era of human aging research.
Essential to BioAge’s perspective is the idea that animals don’t age the same way as humans, making it critical to study aging in the most important model system of all. The company starts with human aging data to find targets that are translationally relevant. Specifically, they use human aging cohorts for which both biological samples and clinical outcomes are available.
They use these data to characterize the metabolome and looked for compounds that correlated with how long the subject lived after the sample was taken. The first panel of biomarkers, discovered using a healthy aging cohort from Estonia, outperformed single markers such as telomere length, or even complex physiological outcomes like frailty. Work with a second cohort from Hawaii is currently underway. From the large, high-dimensional data sets, they use machine learning approaches to identify individual molecules that are likely to play causative roles in determining the aging process. After these targets are selected, the next step is validation in mice, leading ultimately to the identification of drug candidates.
As many as 20% of the lipids identified in metabolomics experiments were associated with all-causes mortality, and Bioage also discovered several intriguing patterns connecting the chemical features of lipids, their levels in the blood, and the risk of age-related disease. The lipid risk score built using the Estonian data replicated in a completely different population from the US, arguing strongly that BioAge has identified molecules with important involvement in aging.
Another direction for the company was the development of a new method for measuring compound functional similarity based on gene expression data, described in their recent paper, “Drug Repurposing Using Deep Embeddings of Gene Expression Profiles”.
Fortney also surveyed the company’s public projects, such as MortalityPredictors.org, a manually curated database of published biomarkers of all-cause mortality.
Prioritizing Drugs & Targets in Aging Research (João Pedro de Magalhães, Reader, Integrative Genomics of Ageing Group, University of Liverpool)
The aging process is plastic, and can be manipulated by altering genes and other types of interventions such as drug treatments. We know of more than 2000 genesthat, when manipulated in model organisms (knocked out or overexpressed), can dramatically affect lifespan—but of course, genes do not work by themselves, but in interactions with each other and their environment. Therefore, de Magalhães seeks to understand how genes work together to determine given traits such as lifespan.
Those 2000 genes (compiled at João’s website GenAge) can be grouped into pathways with directional effects on aging: the 53 pathway and cell cycle genes and pro-longevity, whereas the mTOR signaling pathway is anti-longevity. Based on this idea, the group uses network analyses to make “guilt-by-association” connections between genes that are known to affect aging and those that function in similar pathways, and might therefore be affected to also influence longevity. João has also assembled a similar set of information for senesence, available at CellAge.
Genetic and gene expression data can be repurposed to identify life-extension compounds. Testing of five compounds revealed four that extended lifespan in worms, and the group is currently working to characterize their modes of action.
Yet another database is DrugAge, a database of aging-relatd drugs comprising 1316 entries from 29 model organisms. Using these data, João used machine learning to identify key features of anti-aging drugs, with the hope that these could be combined in novel ways to create new longevity-extending compounds.
João closed with a few comments regarding the burgeoning anti-aging biotech sector. Only a few percent of pharma companies eventually become profitable, and only 1 in 5000 drug candidates obtain approval. In addition, aging-related endpoints require long validation times, increasing the expense of studies if not outright decreasing the likelihood of success. Worse, the magnitude of drug effects on longevity seem to decrease as we get evolutionarily closer to humans—a species in which we still have a poor biological understanding of aging. Collectively, this paints a somewhat pessimistic picture about the prospects for the explosively growing field of anti-aging medicine.