The primary aim of this proposal is a proof-of-concept experiment to mimic one of the most impressive genetic lifespan synergies in a model organism using human prescription drugs for which transcriptional signatures are available. The ultimate aim is to develop a more general technique for in silico design of pharmacological interventions that mimic benefits of ageing mutations.
The question we address in our research is what drives ageing? Our approach is to focus on a limited number of evolutionarily conserved and interconnected processes linked to ageing. This approach defines a new way of studying and preventing age-dependent diseases. Instead of viewing each age-dependent disease as the result of a separate and independent chain of events, we view age-dependent diseases in general as driven by a limited set of common processes which, in turn, are governed by a highly-conserved set of longevity assurance and stress response pathways. Mutations affecting these pathways typically extend lifespan of model organisms by between 30% and 100%. These same pathways also contribute to disease resistance and extreme longevity in humans, with some variants being associated with increased old age survival and better functional status as well as resistance to diabetes and certain cancers. A major challenge for the field has been to translate these genetic and mechanistic insights into pharmacological interventions suitable for use in humans. Human ageing interventions will likely be, at least initially, preventative, requiring the long-term or repeat treatment of large numbers of healthy adults. Such interventions require a high safety margin and would ideally involve GRAS (generally recognized as safe) compounds or repurposing of well-understood drugs, rather than novel drugs that may have unexpected effects when taken over long periods of time. These safety considerations mean that novel inhibitors that directly target newly identified ageing genes face a long and expensive development process. Moreover, even when targeting the same genetic pathways, lifespan effects of pharmacological interventions are typically much smaller than those of ageing mutations, even in model organisms. A major recent discovery shows how insights from model organisms could be translated to humans. Simultaneous genetic perturbation of distinct ageing pathways can result in synergistic lifespan extension effects, where the benefits of the combined perturbations are larger than the sum of individual lifespan benefits. Pathway synergies suggest a potential strategy for the design of novel pharmacological interventions to extend healthy lifespan in humans by repurposing existing drugs or supplements that have been developed (or can be shown) to target pathways involved in such synergistic benefits. We and others previously explored this approach with some success, but these projects were based on detailed prior knowledge of specific pathways and/or screening, making them difficult to generalize.
Approach to drug discovery In pharmacology a “pharmacophore” is an abstraction of the specific molecular features of a drug molecule or class of drugs. The pharmacophore comprises those molecular interactions with the target molecule (e.g. enzyme or receptor) that are required for the drug effect on its biological target. Only these specific interactions between the pharmacophore and the target molecule are required for the biological effects of a drug. Medicinal chemists routinely create structurally diverse drug molecules aimed at the same target. By ensuring that such molecules have similar pharmacophores, diverse chemical structures can be designed to have similar biological effects. The pharmacophore concept therefore generalizes the molecular structure of drugs by concerning itself exclusively with those molecular interactions required for binding to a given target. On an even more general level, drug effects can be described as targeted perturbations of the physiological “state” of a biological system (organism, tissue, pathway) with the aim of moving it from a less desirable state (e.g. disease, disease-promoting or symptomatic state) to a more desirable (healthy, disease-resistant, more functional) state. Given an appropriate definition of “state”, we would expect that two pharmacological interventions (drugs or drug combinations) that cause comparable changes in systems state would cause similar biological effects (have comparable benefits), even if they achieve these changes in state by targeting different primary nodes in the underlying gene regulatory network. We know of many examples where this is indeed true. For example, as mentioned, in the above example of the IGF/mTOR synergy, the benefits of rsks-1 knock-down can be partially reproduced by chemical inhibition of mTOR using Rapamycin. While the primary molecular target for each intervention (drug, mutation) may be different (different gene products are initially inhibited or blocked), the global change in systems state is mediated through the conserved IGF and mTOR networks and the resulting changes (e.g., from normal-ageing to slow-ageing) are consistent. Based on this idea, we can define a “network pharmacophore” (NP) as a specific set of changes in state variables that are necessary and sufficient to cause a specific phenotype. By this definition, two interventions have the same NP if they impact the same critical set of state variables in a similar way. State variables In principle, any number of variables could be used to define the “state” of a system. The goal of this approach is not to “solve” the biological system and explain all interactions (e.g., generate a full model of the system) but to extract sufficient information to identify state changes that are required for the phenotype of interest. For this purpose, transcriptional changes are the most obvious choice for several reasons. First, transcriptional changes provide a high-resolution snapshot of changes to a system’s state. Any change in physiological state involves changes of the level of transcription of some key genes. While such transcriptional changes alone do not fully determine the physiological state of an organism (because physiological responses also involve changes in translation, protein degradation and post-translational modifications etc.), transcriptional changes are relatively easily detectable indicators for responses mediated by the gene regulatory network controlling the global state of any biological system or organism. Second, transcriptomics is rapid and now comparatively cheap. With the development of Next Generation Sequencing (NGSs), genome-wide transcriptional changes can now be rapidly determined using RNAseq. Due to the dramatic decrease in the cost of NGS, it is now possible to determine the transcriptional state (transcription levels of most genes in the genome) for as little as $250 per sample, making studies involving hundreds of samples feasible. Network pharmacophore via transcriptomics The idea of a transcriptomics-based NP is closely related to gene sets and pathway analysis in that it includes enumerating sets of genes that are involved in a given perturbation (differentially expressed genes analysis) as well as analysis of correlation structure (partial correlation analysis, eigengenes, gene set/pathway enrichment) and statistical identification of genes correlated with the desired phenotype (regression models, transcriptional clocks). In this sense, identification and refinement of the NP involves the same tools and approaches as mechanistic studies aimed at explaining (creating a model) of the gene regulatory network mediating the effect. However, the purpose of this analysis is entirely different. The NP is not intended to explain the effect but to act as scoring function to compare the desired state (lifespan synergy) to any observed states (e.g., transcriptional profiles of drugs of unknown efficacy vs ageing). This then allows drug repurposing based on evaluating transcriptional changes instead of lifespan studies. Data collection For this study, the nodes of the NP are defined by genes that experience significant changes in transcriptional levels in daf-2 rsks-1 or daf-2 Rap only but are not similarly impacted in WT. A simple analysis of differentially expressed genes can be used initially to identify candidate genes for inclusion. We are engaged in a mechanistic study of the daf-2 rsks-1 interaction and have collected data comparing transcriptional levels and lifespan effects in wild-type or daf-2 mutants with and without Rap treatment. To further refine the nodes to include in the NP, we have used the additional pre-existing knowledge regarding the dependence of the effect on additional pathways, in particular on aak-2 and daf-16. We have collected data on transcriptional changes daf-2 with Rap with and without RNAi against daf-16 or aak-2. Transcriptional clocks can be extracted from longitudinal data by following a cohort of ageing animals throughout life. Transcriptional levels are then determined at several timepoints throughout the lifespan of the cohort to quantify change in system state. Processes with characteristic times on the order of lifespan include: growth and development, disease processes and ageing itself. Such genes identify subspaces of total transcriptional state space that show high variance across age groups. Genes involved in long-term/slow dynamics can be extracted using statistical or linear algebra techniques, including Singular Value Decomposition (SVD) and various regression techniques. We have constructed such clocks based on different biomarkers / state parameters, including transcriptional changes and we will use them to refine our candidate NP. This work is ongoing in the lab with data collected and analysis in progress with a view to generating a better mechanistic understanding of the pathways involved. However, the same data can also be used to define, refine and test the NP and this is the starting point of this project. The techniques needed for the proposal are in routine use in my laboratory for mechanistic studies. We have two “wormbot” robotic lifespan systems developed by the Kaeberlein group at the University of Washington Nathan Shock Center of Excellence in the Basic Biology of Aging. These systems are suitable for high-throughput (robotic) lifespan studies, and we routinely carry out ageing intervention testing and associated biochemical and RNAseq analysis (e.g., to explore drug mode of action in nematodes and fruit flies). But here, the goal is not to explain mode of action but to identify nodes and edges of the NP.
We aim to select 50 compounds (40 candidate drugs, 10 negative control) for testing and analysis. It is difficult to exactly estimate the cost per compound as pharmaceutical grade drugs can range widely in price. However, from experience we estimate that a relatively small sample for initial testing will on average cost no more than 500 USD. We therefore estimate a total budget to obtain compounds for testing to be 25,000 USD.
We will initially test 50 (40 candidate drugs, 10 negative control) compounds at 3 different concentrations (orders of magnitude, centered on published dose used in tissue culture / clinical practice for each drug). For the 10 positive control compounds we will use a single “optimal” concentration as established in our lab. Following these initial 160 first-pass trials (3 wells each, 480 wells in total), we will carry out a more high-power (8 repeat wells) confirmation at a single concentration compounds that showed promising / no toxicity. All in all, we expect to carry out approximately 180 total robotic lifespan studies comprising a total of 640. Estimated cost for this number of samples using our robotic system (for plastic ware, nematode growth medium, chemicals and reagents, storage medium for datafiles) is approximately 3500 USD (excluding systems cost and/or repairs. We assume here that nothing will break).
Ultimately, we expect to carry out transcriptomics analysis for between 40 and 60 drugs (depending on screening results). In addition, we expect to carry out a small number of controls (to control for batch effects, untreated controls) and repeats (e.g., when quality control fails, validation of surprising results). We expect to run about 70 RNAseq conditions, with each sample being run in quadruplicate – for a total of 280 RNAseq samples. Current price for RNAseq in Singapore is 220 USD per sample. In addition to sequencing, we require budget to grow worm cohorts for RNA extraction, extract RNA to generate samples. Overall, we estimate that cost per sample will be 280 USD per sample for a total sequencing budget of 78,400 USD.
The data analysis, bioinformatics and experimental design will be carried out by the team members and existing staff with the possible involvement of current and future PhD and Hons students. However, a significant fraction of the work is routine and requires a high degree of standardization (lifespan studies, RNA sample generation). We would therefore like to hire a full-time research assistant (RA) for approximately one year to ensure quality and reproducibility of samples. At current rates, a fresh graduate RA position under standard NUS contract requires approximately 45,000 USD pa to fund.