Gene expression‐based drug repurposing to target aging

Abstract Aging is the largest risk factor for a variety of noncommunicable diseases. Model organism studies have shown that genetic and chemical perturbations can extend both lifespan and healthspan. Aging is a complex process, with parallel and interacting mechanisms contributing to its aetiology, posing a challenge for the discovery of new pharmacological candidates to ameliorate its effects. In this study, instead of a target‐centric approach, we adopt a systems level drug repurposing methodology to discover drugs that could combat aging in human brain. Using multiple gene expression data sets from brain tissue, taken from patients of different ages, we first identified the expression changes that characterize aging. Then, we compared these changes in gene expression with drug‐perturbed expression profiles in the Connectivity Map. We thus identified 24 drugs with significantly associated changes. Some of these drugs may function as antiaging drugs by reversing the detrimental changes that occur during aging, others by mimicking the cellular defence mechanisms. The drugs that we identified included significant number of already identified prolongevity drugs, indicating that the method can discover de novo drugs that meliorate aging. The approach has the advantages that using data from human brain aging data, it focuses on processes relevant in human aging and that it is unbiased, making it possible to discover new targets for aging studies.

: a) Age distribution of the sub-datasets. b) PCA plots for the sub-datasets (after outlier removal).       Characterising the potential effect of each drug on ageing Literature search:

Eleven of the drugs have targets previously associated with ageing.
Vorinostat is a histone deacetylase (HDAC) inhibitor used for the treatment of cutaneous T-cell lymphoma ("Vorinostat," n.d.). Although not reported in the DrugAge database, vorinostat has already been tested on Drosophila for lifespan extension and shown to increase both mortality rate and survival when the drug is given during 'midto late-life' (McDonald, Maizi, & Arking, 2013). This drug had the most pro-longevity drug-like profile ( Figure S9, cluster 6) based on our analysis, suggesting that the methodology, as well as the interpretation, yields biologically relevant results. Quinostatin, targeting the catalytic subunit of PI3K, had the highest CMap score with percent similarity above 50% to the pro-longevity drug profile for all four categories ( Figure S9). Considering that drugs targeting PI3K, such as LY-294002 and wortmannin, extend lifespan in worms and flies , quinostatin is a strong anti-ageing drug candidate. Alvespimycin and tanespimycin inhibit the heat shock protein HSP90, which is also inhibited by geldanamycin. Heat shock proteins are implicated in ageing based on both experiments on worms and flies (Tacutu et al., 2017) and human expression studies. Protein aggregation and disrupted proteostasis are a hallmark of ageing (López-Otín, Blasco, Partridge, Serrano, & Kroemer, 2013). It is thus plausible that increased activity of HSP90 would reverse the effects of ageing by restoring proteostasis, although its downstream effects might result in reverse (Fuhrmann-Stroissnigg et al., 2017;McClellan et al., 2007). Tretinoin is a retinoic acid receptor (RAR) agonist widely studied for skin (Mukherjee et al., 2006) and brain ageing (Enderlin et al., 1997). RAR genes in are implicated in synaptic plasticity, learning, memory, and pathological conditions such as Alzheimer's disease (Lane & Bailey, 2005). GW-8510 is a cyclin-dependent kinase 2 / 5 inhibitor and was suggested to be neuroprotective (Johnson et al., 2005).15-d prostaglandin J2 activates PPARG, which shows decreased expression with age, which can be restored by DR (Tacutu et al., 2017). Camptothecin and irinotecan both target TOP1, which alters the topological state of DNA during transcription and can inhibit Warner syndrome protein (WRN), which functions in DNA repair (Shamanna et al., 2016). These two drugs, as well as daunorubicin which targets TOP2A, may therefore worsen health status. Cinchonine, which targets CYP2D6 was similar to pro-longevity drug profile in terms of the genes up-regulated by the pro-longevity drugs but show opposite profile for the down-regulated genes. The functions that are associated with the genes downregulated by the drugs are autophagy or immune function related categories. Most of the known pro-longevity drugs are suggested to function through inhibition of PI3K / mTOR pathways, favouring autophagy. It appears that cinchonine would not function in the same way. However, considering that it targets CYP2D6, which was shown to have a role in lifespan regulation in C. elegans, it is possible that this drug has a distinct mechanism to modulate ageing. Mann et al. previously suggested that expression of CYP2D6 increases with age in the human brain and is lower in Parkinson's disease (Mann et al., 2012). Considering that they suggest this protein might be important to inactivate neurotoxins, inhibiting this protein using cinchonine might function in the same direction and exacerbate ageing by down-regulating one of the cellular responses.
Novel candidates that are not in the GenAge or the DrugAge databases can offer new targets and mechanisms to modulate ageing. Thioridazine and trifluoperazine are serotonin and dopamine receptor antagonists used for the management of psychoses, including schizophrenia. Thioridazine is withdrawn from the market due to its side effects related to cardiac arrhythmias. Ye et al. screened a library of compounds for lifespan extension in C. elegans and identified a couple of drugs targeting serotonin and dopamine receptor antagonists including thioridazine hydrochloride, which extends lifespan by 31% in C. elegans (Ye, Linton, Schork, Buck, & Petrascheck, 2014). Thus, it is likely that these drugs also have antiageing effects. Emetine is the principal alkaloid of the ipecac root. It is a eukaryotic protein translation inhibitor. A recent study investigated the effect of protein translation inhibition on cellular senescence. They suggest that cytoplasmic protein accumulation is an important cause of the cellular senescence and mild protein translation inhibition can prevent senescence induction in normal and tumour-derived human cells (Takauji et al., 2016). Although both this information and our results suggest that emetine can help alleviating the ageing, Takauji et al. did test the effect of emetine on senescence and could not detect any significant result. Atropine oxide is predicted to target muscarinic acetylcholine receptors ("CHEMBL2146145," n.d.), which are suggested to be important for various brain functions as well as pathologies such as Alzheimer's and Parkinson's diseases (Langmead, Watson, & Reavill, 2008). The information regarding the effect of atropine oxide on different muscarinic acetylcholine receptors, however, is limited to make a conclusion whether this drug could be beneficial or damaging for the human brain ageing. Securinine is a GABA(A) receptor antagonist. GABA receptors are started to gain attention as potential targets for neurodegenerative diseases (Rissman, De Blas, & Armstrong, 2007). The drugs tested and shown to have an impact on cognitive abilities so far, however, are mainly GABA(A) agonists or GABA(B) antagonists (Li et al., 2016). Considering that GABA(A) subunit expression levels show a decrease with age (TableS2), and securinine is an antagonist, it is possible that it acts in the same direction as ageing and exacerbates it. Rifabutin shows high similarity to the pro-longevity drug profile and clusters together with levothyroxine sodium and geldanamycin, which are known pro-longevity drugs ( Figure  S9, cluster 2). Rifabutin is an antibiotic but it is reported to also target BCL6 (Evans et al., 2014). BCL6 gene is not in GenAge databases, however, there are studies linking this gene to ageing using human gene expression data (Glass et al., 2013) and through its role in cell proliferation and senescence, regulated by miR-127 (Chen, Wang, Guo, Xie, & Cong, 2013). Thus, it is possible that rifabutin helps to reduce the effect of damaging changes induced by ageing, through targeting BCL6.

The column labels written in bold indicates the drugs in the DrugAge database. Annotation columns show the up-or down-regulation of each category in ageing and prolongevity drug profile. The small table shows the number of probe-sets in each category we defined to reflect pro-longevity drug profile
For each category (up or down in both ageing and the pro-longevity drug profile), we calculated the percent similarity between each drug we identified and the pro-longevity drug profile. Particularly, we calculated the percentage of the consistent expression changes induced by each drug that are in the same direction as the pro-longevity drug profile (see Methods). We generated hierarchical clustering of the drugs based on the percent similarities ( Figure S9). Most of the drugs showed a similar profile for the probe-sets down-regulated by the pro-longevity drugs (except for cinchonine). The drugs with negative and positive CMap scores clustered separately, but this separation is mainly driven by just a subset of the genes -the ones up-regulated by the prolongevity drugs. Thus, it is still possible that drugs with both negative and positive CMap scores can alleviate as well as worsen ageing, through different mechanisms of action. Overall, we divided the hierarchical tree into seven clusters to explain the trends in each group. Cluster 1 had three drugs; atropine oxide, GW-8510 and tretinoin. These drugs showed a strong similarity for the probe-sets down-regulated by the prolongevity drugs, however, they also down-regulate the probe-sets up-regulated by the pro-longevity drugs. Cluster 2, which included two DrugAge drugs; levothyroxine sodium and geldanamycin, as well as a novel candidate; rifabutin, showed relatively lower, but above 50% similarity, for the probe-sets down-regulated according to the pro-longevity drug profile. These drugs were also similar to the pro-longevity drug profile in terms of the up-regulated probe-sets. Cluster 3 was similar to cluster 1 but the dissimilarity in terms of the up-regulated probe-sets was less pronounced. Cluster 4 consisted of only one drug showing a quite distinct profile compared to the other drugs; cinchonine. It was similar to the pro-longevity drugs in terms of the up-regulated probe-sets but it showed an opposite profile for the down-regulated probe-sets.
Cluster 5 was also a one-drug cluster with wortmannin, which is a DrugAge drug. Except the probe-sets up-regulated in ageing and down-regulated by the drugs, wortmannin showed similar profile to the pro-longevity drug profile. Cluster 6 was the group with known pro-longevity drugs except for vorinostat. As expected, the percent similarities were the highest in this cluster. Cluster 7 also showed quite a remarkable resemblance to the pro-longevity drug profile, except for the probe-sets that were down-regulated in ageing but reversed by the pro-longevity drugs, which were neither similar nor dissimilar having 50% similarity.   Figure S11: a) Heatmap showing the pairwise correlation coefficients among GTEx datasets, corresponding 17 major and 35 minor tissue types. The intensity of the colours on the heatmap shows the magnitude of the correlation coefficient. b and d) The same as (a) but using only the genes in brain ageing signature compiled using microarray (a) and GTEx brain dataset (b). c)Heatmap showing the proportion of the changes in the same direction with ageing signature compiled using microarray signature. The colour shows the similarity (red), and dissimilarity (blue) based on the majority of change.