.
Abstract
We aim to improve anti-ageing drug discovery, currently achieved through laborious and lengthy longevity analysis. Recent studies demonstrated that the most accurate molecular method to measure human age is based on CpG methylation profiles, as exemplified by several epigenetics clocks that can accurately predict an individual's age. Here, we developed CellAgeClock, a new epigenetic clock that measures subtle ageing changes in primary human cells in vitro. As such, it provides a unique tool to measure effects of relatively short pharmacological treatments on ageing.
We validated the CellAgeClock against known longevity drugs such as rapamycin and trametinib. Moreover, we uncovered novel anti-ageing drugs, torin2 and Dactolisib (BEZ-235), demonstrating the value of our approach as a screening and discovery platform for anti-ageing strategies. The CellAgeClock outperforms other epigenetic clocks in measuring subtle ageing changes in primary human cells in culture.
The tested drug treatments reduced senescence and other ageing markers, further consolidating our approach as a screening platform. Finally, we show that the novel anti-ageing drugs we uncovered in vitro, indeed increased longevity in vivo. Our method expands the scope of CpG methylation profiling from measuring human chronological and biological age from human samples in years, to accurately and rapidly detecting anti-ageing potential of drugs using human cells in vitro, providing a novel accelerated discovery platform to test sought after geroprotectors.
----------------------------------------------------------------------------------------------------------------------------------------------
One of the remarkable achievements of developed countries is a continuous increase in life expectancy at birth, leading to greater longevity. However, a higher proportion of elderly in modern societies is accompanied by a steep increase in people suffering from age-related diseases. For example, cancer incidence rates, currently at 17 million worldwide, are expected to increase to 6 million in 2040 (Wilson et al. 2019), and a similar rise is expected for Alzheimer’s and Parkinson’s disease (Reeve et al. 2014). Compression of late-life morbidity is, therefore, a priority to alleviate suffering in the elderly (Partridge et al. 2018) and to reduce a growing economic burden to society (Rae et al. 2010).
Critically, seminal discoveries in the biology of ageing showed that ageing is a malleable process and that down-regulation of major cellular nutrient signalling pathways, either glucose-sensing insulin signalling or amino acid-sensing target-of-rapamycin signalling, results in longevity and health improvement in all model organisms tested from yeast to mammals (Lopez-Otin et al. 2013). For instance, the long-lived mutants in C. elegans are protected from tumorous cell proliferation (Pinkston et al. 2006) and have reduced toxic protein aggregation (Cohen et al. 2006), while Drosophila show less deterioration in their hearts (Wessells et al. 2004). Long-lived mouse mutants are protected from osteoporosis, cataracts and skin pathology, as well as decline in glucose homeostasis, immune and motor function (Selman et al. 2008). The effect of these mutations is conserved from yeast to mammals, and it is, therefore, expected that if drugs replicate the biological impact of these changes, this could improve health in the elderly and prevent age-related disease. It is increasingly recognised that directly targeting ageing through pharmacological interventions, as opposed to specific age-related diseases, is a highly promising strategy for broad-spectrum disease protection (Niccoli and Partridge 2012). However, at present, there are only a handful of reliable antiageing drugs whose effects have been confirmed in mammals, such as rapamycin (Harrison et al. 2009) and metformin (Martin-Montalvo et al. 2013). Crucially, there are currently no sufficiently reliable ageing biomarkers for testing drugs on human cells in vitro, and the development of a specialised epigenetic clock is a promising approach (Castillo-Quan et al. 2015; Field et al. 2018; Horvath et al. 2018; Bell et al. 2019; Horvath et al. 2019).
To accelerate the discovery workflow for anti-ageing drugs, we took advantage of the breakthrough in the ageing field which showed that epigenetic clocks provide the most accurate asurements of human age. For instance, the approximate error rate for the Skin and Blood clock is ±2.5 years (maximal correlation coefficient 0,98) (Horvath et al. 2018). Epigenetic clocks surpass the accuracy of other ageing biomarkers such as telomere length and those based on transcriptomic, metabolomic or proteomic approaches, potentially because the latter approaches detect more transient and less stable cellular changes (Horvath 2013). Ageing is accompanied by overall CpG hypomethylation, whilst some CpG islands and gene regions become hypermethylated (Booth and Brunet 2016). Remarkably, only a small selection of the 56 million CpG sites in the diploid human genome, coupled with computational algorithms, is sufficient to provide an accurate readout of human age. One of the first epigenetic clocks was developed by Hannum using just 71 CpG sites to estimate age from blood samples (Hannum et al. 2013), while Horvath’s multi-tissue age estimator (Horvath 2013) and Skin and Blood clock (Horvath et al. 2018) use 353 and 391 CpG sites, respectively (Field et al. 2018; Horvath and Raj 2018). Even a single CpG site in the ELOVL2 gene is sufficient to determine age (Garagnani et al. 2012), albeit clocks using only a few CpG sites are less accurate and less applicable to different tissues (Horvath and Raj 2018). The epigenetic clocks measure the ageing process inherent to all our cells and tissues, irrespective of their proliferation rate (Horvath et al. 2019). As the human epigenome reflects physiological changes, epigenetic clocks cannot only predict chronological age from a human sample but also give an estimate of biological age as has widely been demonstrated by the associations of epigenetic age with morbidity and mortality (Marioni et al. 2015; Horvath and Raj 2018). Recently, valuable predictors focussing on this aspect have been developed: PhenoAge (Levine et al. 2018) and GrimAge (Lu et al. 2019), which form the best epigenetic morbidity and mortality predictors available to date.
DNA methylation also captures information on the approximate number of cell divisions a cell has been through, as has been shown by epiTOC (Yang et al. 2016), a mitotic-like clock that approximates stem cell divisions and correlates with cancer risk (Tomasetti et al. 2017), and MiAge, which also measures mitotic age (Youn and Wang 2018). The biology underlying CpG methylation alterations at the sites linked to ageing clocks is not well understood. The exception is ribosomal clock based on CpG methylation in ribosomal RNA (rRNA), which is highly conserved throughout evolution and which forms nucleolus that has itself been implicated in ageing (Tiku and Antebi 2018; Wang and Lemos 2019). Horvath suggests an interesting hypothesis that epigenetic maintenance programmes are being reflected in DNA methylation alterations (Horvath 2013; Horvath and Raj 2018; Raj and Horvath 2020). Recent findings implicate loss of H3K36 histone methyltransferase NSD1 in epigenetic ageing clock acceleration (Martin-Herranz et al. 2019). Despite the enigma regarding the mechanism of epigenetic clocks, they are reliable predictors of age and extremely useful biomarkers (Field et al. 2018; Horvath and Raj 2018; Bell et al. 2019). However, little is known so far about the performance of these clocks in in vitro ageing experiments. It has recently been shown that the rate of epigenetic ageing in cultured cells is significantly faster than in the human body (Horvath et al. 2019; Sturm et al. 2019) and that epigenetic age is retarded by rapamycin in vitro (Horvath et al. 2019), but neither of the clocks specialised for in vitro drug discovery nor were they tested on multiple anti-ageing drugs.
Therefore, we aimed to exploit the exceptional accuracy of CpG methylation clocks to uncover new anti-ageing pharmacological treatments. The current gold standard for discovering novel anti-ageing drugs are longevity experiments, which are laborious, lengthy and expensive. For instance, in mice, they take three years, thereby precluding any large scale drug screens. Existing screens in C. elegans commonly use live E. coli as food (Lucanic et al. 2013; Ye et al. 2014), which is a disadvantage as drugs are metabolised first by the bacteria making their effect on worms secondary, which may lead to confounded results (Cabreiro et al. 2013; Pryor et al. 2019). Yeast drug screens lack the crucial aspect of tissue toxicity (Zimmermann et al. 2018). In addition, all longevity assays require constant supply of the drug, making them highly expensive. Other attempts to uncover anti124 ageing effects of drugs are based on computational analysis using existing transcriptomic information on the ageing process combined with drug characteristics (Donertas et al. 2018). However, transcriptomic changes are more transient and noisy when compared to DNA methylation and are, therefore, a less consistent ageing marker (Horvath and Raj 2018).
We tested if existing epigenetic clocks could be used to measure anti-ageing drug potential in human primary cells in vitro and if we could build a new clock specialised for this purpose. Senescence is tightly associated with ageing of the organism, and because of the pronounced resemblance of ageing in primary cells in vitro to ageing in vivo, together with the evidence that human DNA methylation signatures are conserved and accelerated in cultured fibroblasts (Sturm et al. 2019), we used cultured human cells as a proxy for human ageing (Lowe et al. 2015; Horvath et al. 2019). The ability to test anti-ageing drug properties directly on human cells in vitro could considerably accelerate the discovery of new compounds promoting healthy ageing. To this end, we used normal human mammary fibroblasts (HMFs) from a healthy 16-year old donor that we cultured from passage 10 to passage 20, which is before these cells reach senescence at passage 29 (Supplemental Fig. S1A-D). To measure CpG methylation, we used EPIC Arrays (Illumina) that measure methylation at 850,000 sites.
Figure 1.
First, we tested the three most suitable existing epigenetic clocks, to determine if they could detect weekly and monthly ageing differences occurring during serial passaging of HMFs (Fig. 1A). The Multi-tissue clock (Horvath 2013) consistently predicted a higher epigenetic age, and at passage ten this was 43.6±1.0 years (Fig. 1A), consistent with what was recently reported (Sturm et al. 2019). This increased age estimate, compared to the age of the donor who was 16 years old, is in accordance with published data demonstrating that this epigenetic clock overestimates the age of mammary tissue samples (Horvath 2013). The PhenoAge clock (Levine et al. 2018), developed to predict mortality and morbidity risks, reported the epigenetic age of the donor to be 3.5±1.1 years (Fig. 1A). The most accurate age estimate, predicting the age of the donor at 23.2±0.87 years, was obtained using the Skin and Blood clock, which is specialised for determining donor age of easily accessible human tissues and cells in culture (Fig. 1A). The Multi-tissue clock and Skin and Blood clock showed a small increase in age with progressive passaging (from passage 10 to 20, age estimate increased from 43.6±1.0 to 53.9±1.7 and from 23.2±0.87 to 31.6±1.2 years, respectively), while this increase was greater for the PhenoAge clock (from 3.5±1.1 to 26.6±9.7 years). This suggested that, of the tested clocks, the PhenoAge clock captures ageing in vitro best (Fig. 1A). However, the PhenoAge clock showed substantial variability in predictions for higher passages, which would obstruct the detection of subtle ageing differences upon anti-ageing drug treatments. In conclusion, while the Skin and Blood clock (Horvath et al. 2018) measures fibroblast ageing in culture, none of the existing clocks was ideally suited to accurately measure subtle anti-ageing drug potential in human primary cells in vitro, and similar comparisons have recently been reported by others (Horvath et al. 2019; Sturm et al. 2019).
This prompted us to develop a new clock that, rather than predicting donor age in years, specialises in measuring methylation changes occurring during ageing of primary cells in culture and could differentiate DNA methylation state between each passage. To this end, we developed a clock using two different cell types, the above-mentioned HMFs and human dermal fibroblasts (HDFs), which were obtained from a different donor, have a different proliferative lifespan in vitro, and a different rate of DNA methylation change. Like the HMFs, the HDFs were serially passaged and sampled every other passage for DNA methylation analysis.
We used a total of 39 HMF and HDF samples to build the clock (see Materials and methods). To preselect informative probes, we performed a statistical test to identify CpGs undergoing DNA methylation changes with increasing cell passage using linear regression (Supplemental Fig. S2A). The resulting 2,543 CpGs were used to build the clock model by elastic net regression, similar to the method used by Horvath (Horvath 2013). The model selected 42 predictor CpGs (“clock CpGs”), shown in Fig. 1B and Supplemental Fig. S2B. Of these CpGs, 23 undergo hypomethylation and 19 become hypermethylated with increasing cell passage (Fig. 1B and Supplemental Fig. S2B). Sixteen of the CpGs are located in intergenic regions (IGRs), whereas 14 of them are located in gene bodies and 12 in promoters, respectively (Supplemental Fig. S2C). Interestingly, two of the clock CpGs map to gene GRID1, one is located in its 3’UTR and one in the gene’s body. GRID1 encodes a subunit of glutamate receptor channels. Several other clock CpGs are also located in genes implicated in cell receptor activity and metabolic processes, such as LDLRAD4 and NPSR1. Furthermore, multiple clock CpGs map to genes that play roles development as well as in the regulation of transcription and protein binding. Examples include GGN, MEIS2, NF1, PROP1, RFX4, RUNX3 and SMARCA2. The 42 CpGs together with their detailed genomic and functional annotation are available in Table 1.
After building our novel epigenetic clock to measure cell ageing in vitro, named CellAgeClock, we tested its performance using an entirely different set of samples (n=26), consisting of 22 HMF and four HDF samples. We observed accurate prediction of passage number for both HMFs and HDFs, with a Root Mean Square Error (RMSE) of 0.37 (Fig. 1C). To compare the performance of the CellAgeClock with other epigenetic age predictors, we calculated Spearman’s rank correlation coefficients between the clocks’ output and actual cell passage (see Table 2). The CellAgeClock showed the best correlation among the tested predictors, with Spearman’s Rho = 0.98 and p <2.2e190 16. We also tested the mitotic-like clocks EpiTOC and MiAge, for comparison. However, their correlation coefficients were negative, small and non-significant (Rho >-0.3, p >0.05).
Having built a precise epigenetic clock that measures methylation changes during replicative ageing of human primary cells in vitro, we tested if anti-ageing drug treatment of HMFs and HDFs decelerated the CellAgeClock. We chose an mTOR inhibitor, rapamycin, which is one of the most robust and evolutionarily conserved anti-ageing drug targets (Saxton and Sabatini 2017), and which mediates its effect through down-regulation of S6K and Pol III, and up-regulation of autophagy (Bjedov et al. 2010; Filer et al. 2017). We chose relatively low rapamycin concentration of 5nM that did not inhibit cell growth (Supplemental Fig. S1A) but moderately downregulated mTOR signalling, as evidenced by decreased pS6K and p4E-BP phosphorylation (Supplemental Fig. S3). This setup mimics the pro-longevity effects of rapamycin in vivo where it is well accepted that only mild nutrient sensing pathway inhibition increases life- and healthspan (Bjedov and Partridge 2011; Lopez-Otin et al. 2013).
DNA methylation profiles from HDF and HMFs collected following four, six and eight weeks of rapamycin treatment (passage 16, 18 and 20; Fig. 2) were analysed using the CellAgeClock and clearly demonstrated that rapamycin slows down methylation changes associated with replicative ageing. Interestingly, this clock deceleration was more pronounced upon longer treatment as shown by the gradual decrease of predicted-actual passage from 16 to 20 weeks. The low dose rapamycin treatment did not affect population doublings, confirming that the methylation changes were not a reflection of proliferation inhibition or slowing of the cell cycle (Supplemental Fig. S1). This is further evidenced by comparing the predicted passage from the CellAgeClock against cumulative population doubling, showing that rapamycin samples lie on a separate line to that of the control samples (Supplemental Fig. S4A-B). Contrarily, rapamycin samples and controls differed to considerably lesser extent when actual passage and cumulative population doublings are compared (Supplemental Fig. S4A-B). We observed a similar pattern for HMFs and HDFs (Fig. 2), suggesting that the CellAgeClock is applicable to different cells, albeit calibration is required for cells that reach senescence at different rates.
.../...
.
Edited by Engadin, 24 May 2020 - 09:46 PM.
