The million-molecule challenge: a moonshot project to rapidly advance longevity intervention discovery

Abstract

Targeting aging is the future of twenty-first century preventative medicine. Small molecule interventions that promote healthy longevity are known, but few are well-developed and discovery of novel, robust interventions has stagnated. To accelerate longevity intervention discovery and development, high-throughput systems are needed that can perform unbiased drug screening and directly measure lifespan and healthspan metrics in whole animals. C. elegans is a powerful model system for this type of drug discovery. Combined with automated data capture and analysis technologies, truly high-throughput longevity drug discovery is possible. In this perspective, we propose the “million-molecule challenge”, an effort to quantitatively assess 1,000,000 interventions for longevity within five years. The WormBot-AI, our best-in-class robotics and AI data analysis platform, provides a tool to achieve the million-molecule challenge for pennies per animal tested.

Introduction

Targeting aging is the future of twenty-first century medicine [52]. Geroscience connects mechanisms of biological aging with age-related functional declines and disease, thereby providing a unifying approach to increase healthspan, the period of life free from chronic disease and disability [61]. Age-associated chronic illnesses commonly manifest by an individual’s sixth decade of life, the halfway point of the maximal human lifespan as we understand it today. Proactive lifestyle approaches to maintain health as we age, like improving diet, sleep, and exercise, can be effective at delaying age-related dysfunction. Lifestyle changes like these, however, are difficult for most people to maintain across a lifespan and their optimization is still poorly defined, particularly at the individual level. For example, effective personalized nutrition is still in its infancy and, despite decades of study, the most effective diets for maximizing longevity remain unclear even in laboratory animals [71]. Further, it seems likely that even an optimal lifestyle may be insufficient to allow most people to achieve centenarian healthspan, free from chronic disease and disability at age 100 and beyond. Additional approaches and interventions that target biological aging are necessary for the field to fulfill its promise of achieving the fullest healthspan potential for humans and other animals who we wish to live the longest, healthiest lives possible.

The first step in the pursuit of effective longevity medicine is the discovery and validation of interventions that increase healthspan of large mammals. While the few promising longevity interventions currently known suggest a large potential impact, there is a dearth of good candidates. Inhibiting mTOR signaling via small molecules like rapamycin is currently the gold standard for pharmacological longevity interventions [50, 62]. mTOR inhibition improves healthy survival across vast evolutionary distances and in every model system in which it has been tested [14, 44, 55, 120]. In rodents, mTOR inhibitors delay or attenuate age-associated diseases like cancer [15], neurodegeneration [54, 117], obesity [21], kidney disease [57], cardiac dysfunction [34, 116], and periodontal disease [8], as well as delay or reverse molecular and functional declines in nearly every tissue and organ examined. In addition to mTOR inhibitors, small molecules that target and clear senescent cells (senolytics) [20], increase NAD availability (NAD boosters) [93], or improve glucose homeostasis (e.g. metformin) [94] also promote healthspan in multiple studies. However, the magnitude of effect for these interventions is smaller than for rapamycin and their ability to increase lifespan is less consistent.

A small number of large-scale efforts seek to identify additional longevity interventions with translational potential. The National Institute on Aging Interventions Testing Program (ITP) is the flagship example of this [92, 97, 98]. Anyone can nominate an intervention to be tested by the ITP for effects on mouse lifespan. After a peer-review process, selected interventions are tested in parallel at three different sites in large cohorts of male and female UMHET3 mice. While of excellent scientific quality, the ITP is slowed by the realities of using mice to assess longevity promoting compounds and can only test between three and six interventions each year. In operation since 2004 the ITP has tested 65 compounds in various treatments and identified only three able to significantly increase lifespan in both male and female mice [72, 91]. Rapamycin was published by the ITP to extend lifespan in 2009 [44], and since then none of the ITP-tested interventions have rivaled the effect size or consistency of rapamycin.

A related effort, also funded by the National Institute on Aging, is the Caenorhabditis Interventions Testing Program (CITP). The CITP tests putative longevity interventions in three closely related species of the nematode Caenorhabditis. While throughput is theoretically much greater with this system, the CITP has only reported on 55 compounds and extracts since beginning in 2013 and none consistently extend lifespan across Caenorhabditis strains [101, 134]. This is disappointingly low for an animal system where a lifespan experiment can be conducted in just one month.

Efforts to catalog interventions tested for longevity highlight the slow pace of intervention discovery. The largest publicly available dataset, DrugAge, currently lists only 1,097 unique interventions tested across 37 model systems [11]. In addition to the small size of the dataset, the experiments captured are difficult to compare given technical variation and variation in quality. These confounding problems make it difficult to know what interventions to pursue for follow up studies and limit the utility of AI-based predictive approaches for novel intervention discovery, which rely on quality input data to make quality predictive outputs [37, 133].

There are numerous reasons why so few longevity interventions are at an advanced stage of development. Most notably, funding for the biology of aging has lagged far behind disease-centric areas of research and drug discovery such as cancer, heart disease, diabetes, and others for many decades [53]. Even within the National Institute on Aging, more than half of the budget is directed toward Alzheimer’s disease research rather than biological aging, despite the fact that age is by far the greatest risk factor for Alzheimer’s disease as well as most other major causes of morbidity and mortality [53]. Additionally, the reward system for academic research is generally not aligned with translational drug discovery and development. Academic researchers are often required to focus on programs that are fundable under the current National Institutes of Health framework, which tends to favor low-risk, mechanism-focused research and “grantsmanship” rather than innovative discovery science. Until recently, a lack of enthusiasm among the biotechnology community to develop longevity therapeutics has stalled private drug development. While that is starting to change, a pipeline of novel interventions does not exist, and most companies focus on a small number of targets based on preclinical work generally centered around nutrient sensing and developmental growth pathways.

Two powerful forces are converging that create a significant longevity intervention “supply and demand” problem, where the number of entities seeking to advance healthspan therapeutics will greatly outstrip the availability of novel preclinical and early clinical stage candidates. The first force is the stagnation of intervention discovery across the field of geroscience. As evidence for this stagnation, we simply note that the most effective non-genetic longevity intervention to date is caloric restriction [141], which was discovered more than 80 years ago [83], and the most effective pharmacological longevity intervention is rapamycin, which, as discussed above, was first shown to increase lifespan in yeast in 2006 [115] and mice in 2009 [44]. In the past 14 years of aging research, nothing as effective as rapamycin at increasing lifespan and delaying or reversing decline across numerous healthspan metrics has been identified.

The second force is the rapid growth of longevity biotechnology over the last five years [28]. Several companies are beginning to prioritize development of healthspan interventions. This trend will likely increase as regulatory strategies for geroscience clinical trials become better established, both in humans and companion animals. There is a similar growing appetite for natural product and GRAS compound formulations that can bypass the protracted FDA approval process and go straight to market. This is amplified by the current regulatory landscape where claims regarding slowed aging or improved healthspan are not under the purview of the FDA. As longevity interventions, particularly rapamycin, continue to show efficacy in large mammals over the next five years, direct-to-consumer, biotech, and pharmaceutical companies will all be seeking novel longevity interventions (both single compound and drug-drug combinations) and targets, preferably with larger effect sizes than rapamycin.

Toward massively high-throughput longevity drug discovery

To overcome intervention stagnation within the field and meet growing demand, a return to unbiased discovery science is needed. Other than genetic loss of function studies, which have been performed in a genome-wide manner in both yeast [16, 85] and C. elegans [30, 42, 43, 74], the longevity intervention space remains largely unexplored. Much of what we know about biological aging today came from these unbiased genetic studies, and we have no idea how much remains to be discovered. For example, genetic overexpression, essential genes, and natural genetic variation remain largely unexplored in this context, with several studies suggesting these represent rich, untapped sources of new insight into mechanisms of biological aging [49, 58, 59, 102, 121, 142].

Small molecule, natural product, and drug-drug combination screens in live animals are an even greater underexplored area of longevity intervention discovery. Several studies have tested small molecules and natural products across different model systems for longevity [18, 73, 130, 147]. The largest of these efforts, to our knowledge, was published 16 years ago [109]. In this study, 88,000 compounds were screened in liquid culture for longevity in C. elegans. A different study, again in C. elegans, screened 30,000 longevity interventions [78]. Both studies reported novel lifespan extending interventions, but the actual number of compounds tested, their source and purity, and overall quality of the data generated, however, are impossible to evaluate because these datasets and compound information have not been released. Considering both studies together, 172 compounds are reported as extending lifespan (0.14% hit rate) but only three have been successfully independently reproduced, and these show inconsistent results [4, 22, 79, 90, 151].

This highlights a major issue with prior longevity intervention screening efforts—the variable quality and difficulty identifying and organizing information from published screens. As a research pursuit today, screening (both genetic and chemical) is typically underpowered and often only mentioned in publications as the starting point for mechanistic studies. The goal of screening, as it currently stands, is not to produce high-quality datasets but instead to identify a “hit” that can be replicated and pursued mechanistically. Lack of quality control across screens limit interpretability of the results and leads to low utility when compiled into databases like DrugAge. This also limits AI approaches that use existing datasets for prediction of novel compounds or drug targets. The disorganization of screening data is another major limiting factor. While the above publications suggest upwards of 100,000 interventions have been screened, curated longevity intervention databases like DrugAge have captured fewer than 1,100 [11]. Beyond single intervention screens, few studies have focused on identifying intervention combinations that extend lifespan [1, 19, 33, 127, 131]. No large-scale efforts to identify drug-drug combinations that extend lifespan have been pursued.

Chemical space is vast and largely unexplored for longevity interventions. Typical FDA approved drug libraries contain 2000–3000 compounds. Millions of small molecules are directly available from commercial suppliers in screening libraries that focus on bioactivity, molecular and therapeutic targets, and chemical structure. Natural product libraries capturing bacterial, plant, fungal, marine, and other extracts are available from multiple sources. Custom and medicinal chemistry services are also widely available to perform custom synthesis de novo or around small molecule scaffolds. To seriously engage and accelerate longevity interventions discovery, massively high-throughput, high-quality unbiased drug screening strategies are necessary. We propose the “million-molecule challenge”, a large-scale drug discovery project that quantitatively measures lifespan and healthspan in a statistically robust manner with a goal of testing 1,000,000 drug treatments (including single compound and drug-drug combinations across different doses) within five years.

How do we meet the million-molecule challenge?

Rapidly screening through one million interventions to identify longevity therapeutics for further development is no small task. The first step is to identify a model system amenable to this level of high-throughput screening. While cell culture models are appealing with regard to throughput, whole animal systems that allow lifespan and healthspan phenotypes to be measured are critical. Cell culture models are well-suited to follow up mechanistic studies and secondary target-based screening approaches once lifespan and healthspan enhancing treatments are identified. Without substantial investment, this level of screening is clearly not feasible in rodent systems like mice. Even testing 100,000 treatments in 40 middle-aged mice per treatment would cost over $1.2 billion just in animal costs alone. Including costs and logistics associated with personnel, drug treating animals, as well as building and maintaining facilities to perform such a massive operation, an estimated cost of over $2 billion is likely a substantial underestimate. Invertebrate model systems, therefore, are best suited to this kind of large-scale screening.

Of the three major non-mammalian systems (yeast, worms, and flies) used in longevity research, C. elegans is the best option for massively large-scale drug discovery to identify lifespan and healthspan extending interventions. C. elegans is short-lived, can be grown in large populations cost effectively, is an excellent system to identify and track age-related changes in movement and behavior, and boasts a wealth of genetic and molecular resources for mechanistic studies [87]. Animals do not require regular manual maintenance and drug studies can be performed by adding compounds to media only once, making large-scale automation possible [132]. C. elegans is an excellent model system to study evolutionarily conserved aspects of muscular (Gieseler K 2018), digestive [31], nervous [100, 108, 126], and reproductive systems [10], as well as metabolism and mitochondrial biology [138, 140]. Well-developed disease models are also readily available for several age-associated, rare, and other disease states including: Alzheimer’s, amyotrophic lateral sclerosis, frontotemporal dementia, Huntington’s, Parkinson’s, cancer, cardiovascular, mitochondrial, metabolic, obesity, diabetes, kidney, sarcopenia, frailty, and many more [17, 23, 32, 36, 60, 67, 68, 80, 82, 95, 96, 114, 125].

As a translational model system for biology of aging, the worm is impressive. The first age-regulating genes were identified in C. elegans [35, 42, 64]. Through the lens of the Hallmarks of Aging [77], genome instability [47, 99], epigenetic changes [39, 75, 88], disrupted proteostasis [137, 153], disabled macroautophagy [7, 86], deregulated nutrient-sensing [48, 56, 139], mitochondrial dysfunction [13, 27], stem cell exhaustion [70, 135], altered intercellular communication [45, 89], and dysbiosis [41, 124] all contribute to establishing lifespan. Numerous evolutionarily conserved cellular and molecular mechanisms drive aging from worms to mammals. Insulin signaling [51, 63], mTOR signaling [104], AMPK signaling [123], HIF signaling [3, 76], G protein-coupled receptor signaling [69], calcium signaling [6, 118], endocytosis [5, 106, 128], autophagy [7], the unfolded protein response [143], DNA repair [99, 148], transcription [29], protein translation [144], nuclear translocation [81], mitochondrial biogenesis and metabolism [26, 40, 103], and DNA methylation and acetylation [107, 122, 150] are just a subset of highly conserved mechanisms associated with aging and lifespan. Indeed, it is easier to ask what is not shared between worms and mammals than what is shared with regards to longevity.

In terms of small molecule longevity interventions, It is well-established that the most robust and reproducible intervention in this category, rapamycin, works in C. elegans and large mammals [44, 120, 136]. Several other small molecule interventions of interest in the field similarly impact lifespan and healthspan metrics in C. elegans and mice. These include metformin [111], alpha-ketoglutarate [9, 145], spermidine [46], urolithin A [25], fisetin [105, 149], and several others. A recent study attempted to quantify the degree of overlap among small-molecule longevity interventions from worms to mice, but failed to arrive at a clear conclusion due to the very small number of interventions tested in both systems [12]. This is not surprising, as a statistically valid comparison would require much larger datasets of compounds tested under controlled conditions in both species, such as prior work demonstrating quantitative overlap between budding yeast replicative lifespan and C. elegans lifespan from genome-wide deletion and RNAi datasets [129]. Nonetheless, as alluded to above, the numerous cases where interventions that increase lifespan in C. elegans positively modulate the biology of aging to increase lifespan and/or healthspan metrics in mammals provides high confidence that hits from worms will often translate to mice, and ultimately humans. Thus, there is strong justification to move forward with large-scale screening efforts to identify other highly conserved longevity interventions.

To facilitate high-throughput longevity drug discovery, we built the WormBot-AI platform. The WormBot-AI is a high-precision, high-throughput robotic system (WormBot) coupled with AI-powered data collection and analytics (WormBot-YOLO). The WormBot is a robotic image capture platform that performs automated data capture of up to 144 C. elegans populations under standard growth conditions using timelapse imaging and short videos [112]. Images captured by the WormBot (roughly one image every 10 min) are compiled into a single timelapse video that encompasses the entire duration of population survival. Several optional features expand the WormBot data capture capabilities, like fluorescence data capture which allows high-resolution physiological changes to be measured along with health and survival in animals expressing transgenic fluorescent proteins or age-associated autofluorescence [66, 110]. Originally built using commercial robotics hardware, the current version of the WormBot is made from 100% in-house custom designed robotics to maximize scalability and versatility. WormBot-YOLO is a neural network powered platform for analyzing image data captured by the WormBot using “you only look once” (YOLO)-based machine learning object detection [119]. WormBot-YOLO identifies, detects, and tracks individual animals through the compiled timelapse image series to record death events and other disease end-states. This tracking is also used to assess features of individual animal health, like movement, velocity, and how these features change with age and drug treatment. Health and survival events are compiled and statistically analyzed to identify differences between drug-treated populations and control (untreated) animals. As a robust, versatile high-throughput screening platform that measures animals in a non-invasive manner and matches standard growth conditions without cumbersome overengineering, the WormBot-AI is, in our opinion, best-in-class for automated lifespan technology [24, 65, 113, 132, 146].

As a proof-of-principle screen, we used the WormBot-AI to screen small molecules for delayed paralysis in GMC101, a C. elegans strain that expresses Aβ in body wall muscles [84]. We screened compounds from an FDA-approved compound library in single and in combination with 25 mM metformin, which delays paralysis in a similar Aβ body wall-expressing background [2] and which we confirmed delays paralysis by about 20% in GMC101 at 25 mM. Using the equivalent of 1.5 of our full-size WormBot platforms, we screened a total of 1266 interventions (633 individual compounds, with and without metformin) over the period of one month (Fig. 1). This is 169 more interventions than the total number of unique compounds within the DrugAge database. Excitingly, we identified several novel interventions that delay paralysis alone and/or in combination with metformin. As we move forward with screening, we are also developing these novel interventions by testing in other disease models and normative aging backgrounds.

Fig. 1.

Summary plots of proof-of-principle FDA-approved drug screen for single and metformin drug combinations that delay Aβ-mediated paralysis. Percent change in paralysis relative to untreated animals for 633 drug treatments, alone (A) and combined with 25 mM metformin (B). Drug treatments ranked by percent change in paralysis for single and drug combination separately. Black circle = no change in paralysis. Blue circle = average paralysis delay produced by metformin alone (20% delay) averaged across all experiments. Plots generated using RCircos [152]

As we have continued to innovate around the WormBot-AI and develop high-throughput protocols, we estimate with 50 robots we can screen 1,000,000 interventions for extended C. elegans lifespan within three years. This will produce the world’s largest longevity therapeutics database, being more than 300-fold larger than the number of assays listed in DrugAge and more than tenfold larger than any dataset described in the literature. Importantly, unlike prior attempts, this breakthrough in scale will not come at the cost of sacrificing quality; interventions will be tested in duplicate populations of 25–30 animals, providing statistical power to resolve as low as a 15% change in lifespan. Further, because data collection occurs via automated image capture and analysis, human bias is removed from the system and long-term archiving of raw data is straightforward. This large-scale, high-quality dataset opens the door for next generation predictive AI to reveal new interventions, drug targets, and combination therapeutics.

Conclusion

Longevity medicine has the potential to revolutionize healthcare for humans and provide benefits in the spaces of veterinary care, agriculture, and other yet to be identified areas. To enable this revolution, the field needs to quickly identify novel longevity interventions, new aging targets, and develop combination therapies to maximize healthspan in large mammals. To accelerate longevity intervention discovery and development, high-throughput, unbiased screening is needed. Given the complex nature of aging, whole animal studies that simultaneously capture health and age-associated behavior changes must be engaged. Mammalian systems, like cell culture or mice, cannot provide the type of high-throughput, whole animal phenotypic analyses that are necessary for fundamental, unbiased longevity drug discovery. For these reasons, C. elegans is optimal as a foundational longevity drug discovery system. Combined with automated analysis technologies, like WormBot-AI, chemical space can be combed through much more quickly to find next generation preventative medicines. Identifying these interventions, however, is only the beginning. Validation across disease models and mammalian systems is needed to show the versatility and translatability of longevity interventions discovered in worms. Precision medicine approaches that identify the interventions most well-suited to extend healthy lifespan will be needed to maximize individual healthspan depending on genetic and environmental context. Finally, mechanisms to make these interventions widely available and understood must be created. This includes growing geroscience-tailored clinical pathways, developing and committing to high-quality, science-backed natural products, and broadly educating the public on the value, importance, and attainability of healthy aging. A million small intervention screening steps can lead to one giant leap in longevity medicine.

Declarations

Ethical statement/Conflict of interest

M.L., B.B., and M.K. are co-founders of Ora Biomedical, Inc., a for-profit company that specializes in longevity drug discovery and development. M.L., B.B., and M.M. are employed by Ora Biomedical, Inc. All authors hold equity stake in Ora Biomedical, Inc.