SUMMARY
Telomerase is best known for its role in the maintenance of telomere length and its implications for ageing and cancer. The mechanisms, kinetics and tissue-specificity underlying the protective or deleterious mechanisms of telomerase, however, remain largely unknown. Here, we sought to determine the telomerase-dependent and - independent transcriptomic changes with ageing, in the gut and brain, as examples of high and low proliferative tissues, respectively. We hypothesised this could shed light on common telomerase-dependent and -independent therapeutic targets aimed at preventing or ameliorating age-associated dysfunction in both tissues. For this, we used the zebrafish model which, similarly to humans, depends on telomerase for health- and lifespan. We performed whole tissue RNA sequencing of gut and brain, in naturally aged zebrafish alongside prematurely aged telomerase null mutants (tert-/-), throughout their lifespan. Our study highlights stem cell exhaustion as the first main hallmark of ageing to be de-regulated in WT zebrafish gut and brain. Towards the end of life, altered intercellular communication becomes the main hallmark of ageing de-regulated in both gut and brain, and this is accelerated in both tissues, in the absence of telomerase. Finally, we identify 7 key gene changes common between the gut and brain at the early stages of ageing, highlighting potential early intervention therapeutic targets for preventing age-associated dysfunction in both tissues.
1 INTRODUCTION
Ageing is the strongest risk factor for chronic diseases. How and why this is the case remain important questions in the field, especially as key research has shown that targeting common hallmarks of ageing, such as cellular senescence (Baker et al., 2016; Baker et al., 2011), can have a positive impact across multiple tissues and ameliorate several chronic diseases of ageing at the same time. There are well-known key hallmarks of ageing, such as genomic instability, telomere attrition, epigenetic alterations, loss of proteostasis, de-regulated nutrient sensing, mitochondrial dysfunction, cellular senescence, stem cell exhaustion and altered intercellular communication(Lemoine, 2021; López-Otín, Blasco, Partridge, Serrano, & Kroemer, 2013). However, a major challenge in ageing research is to identify where and when these potentially pathological changes start and when the tipping point between homeostasis and loss of function takes place (Rando & Wyss-Coray, 2021). Additionally, several lines of evidence suggest there may be specific tissues where age-related changes start earlier, potentially influencing others (de Jong, Gonzalez-Navajas, & Jansen, 2016; Rera, Azizi, & Walker, 2013). One example is the gut, which has been suggested to be a trigger for multiple organ failure (Cardoso et al., 2008). Evidence suggests that the kinetics of ageing can vary dramatically between cells, tissues (Shokhirev & Johnson, 2021; M. J. Zhang, Pisco, Darmanis, & Zou, 2021) and individuals, and that this is influenced not only by intrinsic but also extrinsic factors, recently discussed elsewhere (Rando & Wyss-Coray, 2021). These considerations are of particular importance when the aim is to understand how changes in ageing lead to disease and how, when and where to intervene. This is crucial in order to shift towards a more preventive form of medicine, which is a current global ambition (Rudnicka et al., 2020), aiming to match the dramatic increase of lifespan we have experienced in the past century, with an equivalent increase in years of healthy living, i.e, healthspan (England, 2017).
Tissue-specific transcriptomics analysis over the lifecourse can offer important insights into the downstream molecular mechanisms potentially driving the pathology of ageing. Significant research is being dedicated to these approaches in different animal models, including in mice (Schaum et al., 2020; Tabula Muris, 2020; M. J. Zhang et al., 2021). Different animal models may offer different insights into the mechanisms of ageing, and some models may be better suited to explore the role of specific human hallmarks of ageing. The role of telomere attrition in natural ageing can be considered a hallmark of ageing that may benefit from additional and complementary models, beyond the mouse (Forsyth, Wright, & Shay, 2002; Gomes et al., 2011; Sullivan et al., 2021). Once such model is the zebrafish that, like humans, age and die in a telomerase-dependent manner (Anchelin et al., 2013; Carneiro, de Castro, & Ferreira, 2016; Madalena C Carneiro et al., 2016; Henriques, Carneiro, Tenente, Jacinto, & Ferreira, 2013; Henriques & Ferreira, 2012). Restricted telomerase expression and function are key determinants of natural ageing in humans, underpinning multiple age-related diseases (Blackburn, Epel, & Lin, 2015). However, the role and the dynamics of telomerase-dependent changes that may contribute to tissue-specific ageing are still poorly understood. This is partially due to the fact that telomerase appears to have multiple functions in the cell, that go beyond the maintenance of telomere length, recently reviewed elsewhere (Segal-Bendirdjian & Geli, 2019).
Telomerase is best known for its telomere-dependent function (i.e. canonical functions), acting as a reverse transcriptase, maintaining telomere length through its catalytic domain (TERT protein) and RNA template (TERC) (Greider & Blackburn, 1985). Telomeres are (TTAGGG)n DNA repeats that together with a complex of proteins (known as Shelterin) create a “cap-like” structure at the end of linear chromosomes (de Lange, 2004), preventing the ends of linear chromosomes from being recognised as deleterious DNA double strand breaks (Ferreira, Miller, & Cooper, 2004). However, in humans, due to time- and cell-specific-limited telomerase expression, telomeres shorten with ageing, leading to proliferative exhaustion and replicative senescence (Bodnar, 1998; d’Adda di Fagagna et al., 2003). Importantly, there is accumulation of cellular senescence with ageing in humans (Dimri et al., 1995) and senescence has been linked to several age-associated diseases (Ovadya & Krizhanovsky, 2014). Additionally, short telomeres themselves can lead to de-regulated gene expression, particularly in genes near the chromosome ends, due to loss of the “telomere positioning effect” (TPE), which is known to regulate gene expression of genes at least up to 10MB away from the chromosome ends (Robin et al., 2014).
Growing evidence now suggests that telomerase also has activity independent of its action at telomeres, known as non-canonical (Goodman & Jain, 2011; Romaniuk et al., 2018; Segal-Bendirdjian & Geli, 2019; Sung, Ali, & Lee, 2014). In the nucleus, these non-canonical functions include transcriptional regulation of genes involved in inflammation, including nuclear factor kappa B (NFkB) and tumour necrosis factor alpha (TNFα) (Deacon & Knox, 2018; Ghosh et al., 2012; Mattiussi, Tilman, Lenglez, & Decottignies, 2012), as well as genes involved in cell proliferation (Choi et al., 2008; Sarin et al., 2005) and cell survival (Cao, Li, Deb, & Liu, 2002; Rahman, Latonen, & Wiman, 2005). Telomerase can also translocate to the mitochondria, where it has been shown to play a protective role against DNA damage and oxidative stress (Ahmed et al., 2008; Haendeler et al., 2009).
As tissues with high cellular turnover present accelerated telomere erosion (Bodnar, 1998; H. W. Lee et al., 1998), it is reasonable to think that telomerase functions are likely to primarily affect highly proliferative tissues. Accordingly, premature accumulation of critically short telomeres has been identified in high proliferative tissues such as the gut, in tert-deficient animal models. Nonetheless, the role of telomerase and telomeres is not restricted to highly proliferative tissues. In the brain, considered a predominately post-mitotic tissue, telomerase has been shown to have a protective role against excitotoxicity (Eitan et al., 2012), oxidative stress (Spilsbury, Miwa, Attems, & Saretzki, 2015), and neuronal death (J. Lee et al., 2010), all involved in neurodegenerative diseases. Studies in late-generation telomerase-deficient mice have suggested that limited telomerase expression is associated with premature accumulation of senescence-associated markers in different cell populations including Purkinje neurons, cortical neurons and microglia (De Felice et al., 2014; Jaskelioff et al., 2011; Jurk et al., 2012; Raj et al., 2015). Telomerase is therefore a promising target to promote healthy ageing in multiple tissues and so the identification of mechanisms driving telomerase-dependent ageing could enable the identification of targeted therapies to improve healthspan.
In this study, we aimed to determine the telomerase-dependent and -independent transcriptomic changes and their kinetics occurring during ageing, in both brain and gut within the same individuals. We hypothesised that this would allow us to identify key age-associated genes and pathways that become prematurely de-regulated in both or either tissue, providing key insights into the early stages of ageing in these tissues and likely interactions. We further hypothesised that this may highlight potential common telomerase-dependent and -independent therapeutic targets for early intervention aimed at preventing age-associated dysfunction in both tissues. To address these questions, we performed RNA sequencing in whole tissues (gut and brain) of WT fish (2, 9, 22 and 35 months of age) alongside telomerase mutant fish (tert-/-) (2, 9 and 22 months of age). tert-/- zebrafish, extensively characterised elsewhere (Anchelin et al., 2013; Madalena C Carneiro et al., 2016; Henriques et al., 2013), display no telomerase activity and have significantly shorter telomeres from birth, consequently ageing and dying prematurely. Ageing is usually described as a time-dependent change in tissue homeostasis, that increases the probability of disease and death (Hayflick, 2007). However, whether the genes and pathways driving or accompanying these time-dependent changes are also consistently changing in a time-specific manner, remains unresolved (Rando & Wyss-Coray, 2021). We therefore decided to combine a time-series analysis (STEM), which allowed the identification of genes and pathways that are consistently up or down-regulated over-time, with the more traditional differential gene expression (DEGs) analysis between young and old animals. This combined strategy allowed the identification of genes that change in a monotonic, time-dependent manner (STEM), versus genes that change at specific stages of life (DEGs).
We show that although the gut and brain have distinct transcriptomic signatures of ageing, both tissues display hallmarks of ageing as early as 9 months in the WT zebrafish. Importantly, telomerase depletion accelerates the appearance of such hallmarks in both gut and brain. In particular, we identify stem cell exhaustion as the common principal hallmark of ageing at the early stages of ageing, in both tissues. Further, we identify altered intercellular communication, in which immunity and inflammation play a central role, as the main telomerase-dependent hallmark of ageing common between the gut and brain. Finally, we conclude that the gut displays telomerase-dependent hallmarks of ageing at an earlier age than the brain and that these include changes in several key genes that have also been included in the GenAge database, a benchmark curated database that included genes involved in ageing across different organisms, including in humans (Tacutu et al., 2018). Finally, we identify 7 key gene changes common between the gut and brain at the early stages of ageing, namely ccnb1, kif2c, serpinh1a, temem37, si:ch211-5k11.8, cfap45 and eif4ebp3l.
2 RESULTS
2.1 Identification of monotonic, time-dependent gene signatures and process changes with ageing in the zebrafish gut and brain
In order to identify telomerase-dependent and -independent transcriptional signatures of ageing in the zebrafish gut and brain, we performed RNA-Sequencing of whole tissues, throughout the lifespan of WT and telomerase-deficient (tert-/-) fish (Fig. 1 and source data). While tert-/- fish have a lifespan of c. 12-20 months, WT fish typically die between c. 36-42 months of age (Madalena C Carneiro et al., 2016). The data shown here include 4 age-groups of WT (2, 9, 22 and >30 months), corresponding to young, adult, median lifespan and old. As the tert-/-fish have a shorter lifespan compared with their WT siblings, the data include 3 age-groups of telomerase-deficient fish (2, 9, and 22 months), which correspond to young, medium lifespan and old. Each group has a sample size of 3 animals and the brains and guts from within the same groups of animals were used (Fig 1A). The reads were aligned to the latest zebrafish genome build GRCz11 (Lawson et al., 2020) and resulted in uniquely mapped read percentages ranging from 92.1% to 94.4%, which is a readout of good quality (Lawson et al., 2020). All samples had at least 10 million uniquely mapped reads, except G11 which had around 8 million. Further quality control, using a principal component analysis (PCA) including all the samples, revealed that one of the gut samples clustered with the brain samples, and not with the gut samples (Fig. 1B). This was considered to be a technical error and this sample (G7) was therefore excluded from further analysis. To analyse the overall impact of the genotype and age on transcriptomic regulation, we then performed a PCA in the samples from the gut and brain, separately. We further observed that some samples cluster per age, but there are some genotypes that separate quite distinctly, despite being of the same age (Fig 1C and D). As an example, the WT and tert-/- gut samples are quite distinct, and the tert-/- 2-month samples cluster closely to the WT at 9 months than the WT at 2 months (Fig 1C), providing the first hint of an acceleration of the ageing transcriptomic profile in the tert-/-. We found that there was higher variability in the gut than in the brain, between samples within each age group (Fig 1C and D). A summary of the number of significant differentially expressed genes (DEGs) in all samples is represented in Fig. 1E. How these DEGs relate to which other, how many overlap and how many are in common or accelerated in the absence of telomerase (tert-/-), will be explored later in the manuscript.
To identify genes and pathways that are consistently up- or down-regulated in a time-dependent manner in natural ageing in the zebrafish gut and brain, we grouped the genes into temporal expression profiles, by time-series analysis using Short Time-series Expression Miner (STEM) software (Ernst & Bar-Joseph, 2006). To determine whether the temporal profiles were associated with specific biological processes and pathways, pathway over-representation analysis (ORA) were performed for the genes assigned to the significant STEM profiles. Enrichments of GO Biological Process (GOBP), GO Molecular Function (GOMF), GO Cellular Compartment (GOCC), Kyoto Encyclopedia of Genes and Genomes (KEGG) terms and REACTOME pathway terms were therefore analysed for each profile. Time-series analysis of WT gut identified 9 different profiles, with 2 of them containing up-regulated genes (profiles 7 and 21. Total of 523 genes), and 7 containing down-regulated genes (profiles 32, 31, 23, 9, 12, 22 and 34. Total of 11,594 genes) (Fig. 2A1 and source data). Interestingly, in the tert-/-gut, all the profiles identified by time-series analysis contain up-regulated genes (profiles 8, 6, 15 13, 12 and 11. Total of 10,317 genes) (Fig. 2A2 and source data). Enrichment analysis showed that the profiles containing up-regulated genes are associated with immune response, while profiles containing down-regulated genes are largely associated with proliferation, cellular response to DNA damage, and DNA damage repair, in both WT and tert-/- gut (Fig. 2 A1.1, 1.2 and A2.1). To help contextualise our analysis, we performed a further classification of enriched processes according to the hallmarks of ageing, which have been previously identified (Lemoine, 2021; López-Otín et al., 2013). This classification further strengthened the observation that the up-regulated profiles include genes mostly involved in altered intercellular communication, in which immunity and inflammation play a key role, whereas the down-regulated profiles identify stem cell exhaustion and genomic instability as the main hallmarks of ageing, to which the genes affecting proliferation, DNA damage and repair are likely to contribute (Fig. 2 A1.1, 1.2 and A2.1 and source data).
In the WT brain, we identified 9 temporal profiles, 6 of them including up-regulated genes (profiles 42, 29, 40, 30, 21 and 48. Total of 7,230 genes), and 3 including down-regulated genes (profiles 1, 12 and 26. Total of 561 genes) (Fig. 2B1 and source data). In the tert-/-brain, time-series analysis revealed 6 different profiles. Profiles 4, 0 and 3 containing down-regulated genes (total of 5,374 genes) and profiles 15, 12 and 11 containing up-regulated genes (total of 1,155 genes) (Fig. 2B2 and source data). As in the gut, up-regulated profiles reveal genes mostly involved in immune regulation and inflammation and down-regulated profiles are mostly involved in cell cycle, genome stability and DNA damage responses, in both genotypes This is further highlighted when placed into context by the analysis based on the hallmarks of ageing, where up-regulated profiles identify altered intercellular communication whereas the down-regulated ones identify stem cell exhaustion as the main hallmarks affected (Fig 2 B1.1, 1.2; B2.1, 2.22 and source data).
In summary, STEM analysis and enrichment pathways in both gut and brain ageing show a general trend towards up-regulation of genes involved in immune response and down-regulation of genes involved in cell cycle, DNA damage and repair. This general trend is recapitulated in the absence of telomerase. These mechanisms are all known contributors to the well-described altered intercellular communication, genome stability and stem cell exhaustion hallmarks of ageing, respectively (Lemoine, 2021; López-Otín et al., 2013).
2.2 Comparing the hallmarks of ageing over time, in WT and tert-/- zebrafish gut and brain
More than just identifying the signatures of natural WT ageing in the gut and brain and identifying telomerase-dependent and -independent changes, we wanted to understand if there were particular changes occurring before others, and how their kinetics compared between the gut and the brain. In order to contextualise our analysis in the light of the well-described hallmarks of ageing, (Lemoine, 2021; López-Otín et al., 2013), we used the main hallmarks of ageing as headers in which we could group the different changes in processes that were enriched from both the STEM profiles and all DEGs. This allowed us to compare the effects of age, genotype and tissue on the evolution of the key hallmarks of ageing. When we combine all the gene changes (up- and down-regulated) and associated biological processes affected (Fig. 3 and source data), we observe distinct tissue-specific signatures of ageing, namely in the gut (Fig. 3A) and brain (Fig. 3B), particularly when considering the time-series analysis on its own (STEM) (Fig 3. A1 and B1). Whereas ageing in the WT gut seems to be predominantly affected by the de-regulated nutrient sensing initially, the brain displays stem cell exhaustion as the main hallmark affected at the early age of 9 months (Fig. 3A1, B1, respectively). When we combine the different analysis, though (Fig. 3A3 and B3), both gut and brain have stem cell loss as the main hallmark of ageing associated with the enriched processes found at 9 months of age, highlighting this as a potential contributor to the early stages of ageing in both tissues. Towards the end of life, both gut and brain have altered intercellular communication as the main hallmark of ageing identified. Importantly, independently of the analysis, the tert-/-zebrafish show accelerated hallmarks of ageing. In specific, a tert-/-2-month-old gut profile is very similar to a WT 35 gut month profile (Fig. 3A). In the brain, at 2 months of age the tert-/-mutant also displays some of the hallmarks of ageing that also become altered in WT ageing at later ages, particularly altered intercellular communication, but the tert-/- brain becomes much more similar to the aged WT from the age of 9 months onwards (Fig. 3B). This suggests that hallmarks of ageing accelerated in the absence of telomerase are developing earlier in the gut than in the brain.
Finally, when we look at the overall number of gene expression changes, i.e, not just the ones associated with the hallmarks of ageing, we observe that there is a general increase in the number of changes in gene expression with ageing (Fig 3 A4 and B4). However, this increase does not appear to be linear. In specific, in the gut, the number of DEGs is fairly low until 9 months of age, after which there seems to be an inflexion point and the number of DEGs increases up to 5-fold in the oldest WT (>35 months) and 3 fold in the oldest tert-/- (22 months). In the brain, there seems to be a more gradual, consistent increase over time, both in WT and tert-/-. This increase in number of DEGs with ageing may be a consequence of the known de-regulation of gene expression with ageing due to de-repression of heterochromatin and/or changes in epigenetic markers.
2.3 When does a tert-/-resemble an aged WT the most?
It is of note that, even though the tert-/-accelerates hallmarks of ageing in the gut and brain, the set of genes identified are not necessarily the same as in WT ageing. In specific, looking at the genes at the early stages of ageing, the most significant hallmark of ageing shared between the 2 months old tert-/- and WT 9 months gut is nutrient sensing, and only 2 up-regulated genes are in common, namely lipca (Hepatic triacylglycerol lipase precursor) and pla1a (phospholipase 1a). Interestingly, hepatic lipase has been involved in atherogenesis (Santamarina-Fojo, Gonzalez-Navarro, Freeman, Wagner, & Nong, 2004), and age-related macular degeneration (Neale et al., 2010), diseases that have been hypothesised to have a parallel response to tissue injury induced by multiple factors, including impaired immune responses, and oxidative stress (Neale et al., 2010). In line with the general de-regulated inflammatory response we see in the gut, with ageing (Figs. 2 and 3 and source data), phospholipase 1a has been reported to be up-regulated in inflamed gut tissue of Crohn’s disease patients (Hong et al., 2017) and has a complex role in the regulation of immunity and inflammation (recently reviewed in (Zhao, Hasse, & Bourgoin, 2021)). In the brain, there are no genes in common between the 2 months old tert-/- and WT 9 months associated with altered intercellular communication or genomic instability, the 2 main hallmarks of ageing shared between the genotypes at the early stages of ageing. Following this type of analysis, we could then ask at what age does the tert-/- best mimic naturally aged WT, at the level of gene expression from the shared hallmarks of ageing and these “other” genes we found to be associated with ageing but not obviously associated with the described hallmarks of ageing. We therefore analysed the overlap between all DEGs identified in the old WT (35 months) and the tert-/-at the different aged (2, 9 and 22 months), using Venn diagrams created using the Venny 2.1 online platform (Oliveros, 2007-2015) to ask this question. We observe that the 2 months old tert-/- has the most genes shared with the 35 months old WT gut (59 genes in common) (Fig. 4 A1 and Supp. Fig. 1) whereas it is the tert-/- at 22 months that has the most genes shared with the 35 months old brain (112 genes in common) (Fig. 4 B1 and Supp. Fig.1). This similarity between the 2 month and 22 months tert-/-, with the gut and brain, respectively, is also apparent when we just look at the pattern of the main hallmarks of ageing accelerated in the tert-/- (Fig. 3, 4 and Graphical Abstract). Together, these data suggest that the gut is displaying telomerase-dependent hallmarks of ageing at an earlier age than the brain, consistent with what would be expected for a high versus low proliferative tissue. Of relevance, In the brain, of the genes de-regulated in the tert-/- 22 month that are in common with the WT aged at 35 months ccna2 (cyclin a2), cdk6 (cyclin-dependent kinase 6), chek1 (checkpoint kinase 1), mad2l1(Mitotic spindle assembly checkpoint protein mad2a), tacc3 (Transforming, acidic coiled-coil containing protein 3), top2a (DNA topoisomerase ii alpha) and mcm2 (DNA helicase, MCM2 minichromosome maintenance deficient 2) have also been included in the Human Ageing Resources databases (Tacutu et al., 2018) and are mostly located within the same cluster (green), identified using k-means clustering in STRING analysis (Szklarczyk et al., 2021) (Fig 4 and source data). Of the proteins encoded by these genes, chek1 (Poehlmann et al., 2011), cdk6 (Morris, Hepburn, & Wynford-Thomas, 2002), mad2l1 (Lentini, Barra, Schillaci, & Di Leonardo, 2012), tacc3 (Schmidt et al., 2010)have been reported to be involved in cellular senescence,. Additionally, Top2a has also been involved in neuron proliferation (Watt & Hickson, 1994) and mcm2 depletion in mice leads to decreased proliferation in various tissue stem cell progenitors (Pruitt, Bailey, & Freeland, 2007).
2.4 Analysis of telomerase (tert)-dependent gene changes of old age
Once we identified what WT ageing looked like at the level of time-dependent gene expression changes over the life-course and the main biological processes affected, we sought to determine how much of these were likely to be telomerase-dependent. If a gene expression change or a biological process alteration is accelerated in the absence of telomerase (i.e becomes significant at an earlier age in the tert-/-), we consider it to be telomerase dependent, as has been described before (Madalena C Carneiro et al., 2016; Henriques et al., 2013). Conversely, if none of these pre-requisites are met, we consider the gene/process alteration to be telomerase-independent. With this in mind we performed Venn diagrams (Oliveros, 2007-2015) to identify the telomerase-dependent significant gene alterations of old age, i.e, DEGs present in the WT at 35months old, when compared with the WT young control. For this, we used the DEGs identified in the STEM profiles (i.e., genes that change monotonically, in a time-dependent manner), combined with the more traditional DEG analysis, which include all gene changes, whether they change consistently in a time-dependent manner across the lifecourse or not (called “all DEGs” from hereafter) (Fig. 5 and source data). From these analyses, we identified 50 significant DEGs (out of 491; c. 10%) (present in old age (WT 35 months) that are prematurely de-regulated in the tert-/- gut (Fig. 5 A) and 100 genes (out of 428; c.23%) to be prematurely de-regulated in the tert-/- brain (Fig. 5 B). Importantly, most of these genes are directly or indirectly involved in known hallmarks of ageing (Lemoine, 2021; López-Otín et al., 2013), namely altered intercellular communication (including immunity, inflammation, extra-cellular matrix), genome stability (including DNA replication and repair), stem cell exhaustion and mitochondrial dysfunction, as was also evident in Fig 3 and now further detailed below. In the gut, telomerase-dependent gene expression changes with ageing include up-regulation of genes involved in immune response, such as tlr18 (Toll-like receptor 18 precursor), sytl1 (Synaptotagmin-like protein 1) and down-regulation of genes involved in metabolism, such as cyp8b1 (Bertaggia et al., 2017) (Cytochrome P450, family 8, subfamily B, polypeptide 1) and igfbp2a (Insulin-like growth factor-binding protein 2). Within the down-regulated genes, there are other well-known genes such as sox6 (involved in stem cell function), vav3b (involved in wound healing), (Fig 5 and source data). Additionally, we identify a number of non-annotated genes, which may represent novel tert-dependent genes, that would require further investigation. Importantly, amongst these tert-dependent genes of “old age”, there are a number of homolog or closely related genes which have been previously identified in ageing datasets (Tacutu et al., 2018) such as several igfbps (insulin-like growth factor binding proteins), igf (insulin growth factor), mapks (mitogen-activated protein kinases), tlr3 (toll-like receptor 3) and nrg1 (neuroregulin 1).
In the brain, tert-dependent ageing gene expression changes include mostly down-regulation of genes involved in cell cycle or neurogenesis, such as aurkb (aurora kinase B), chek1 (checkpoint kinase 1), ccnb1 (cyclin b1), cdk2 (cyclin-dependent kinase 2) and neurod4 (neuronal differentiation 4), dld (delta d), nog1 (noggin 1), respectively, as well as genome stability and DNA repair, such as rad54l (rad54 like), mcm5 (minichromosome maintenance complex component 5) and smc4 (structural maintenance of chromosomes 4). Up-regulated genes are mostly involved in immune response or inflammation, such as cxcl18b (chemokine (C-X-C motif) ligand 18b), mhc2a (major histocompatibility complex class II integral membrane alpha chain gene), socs1a (suppressor of cytokine signaling 1a), irf8 (interferon regulatory factor 8) and csf1b (colony stimulating factor 1b (macrophage)). Of note, amongst these tert-dependent DEGs of old age, we identified dre-mir-29b-1, which encodes for a regulatory micro RNA 29 (mir29), widely described to be up-regulated in ageing, in response to DNA damage, alongside P53 (Ugalde et al., 2011) and can act as a protective response, limiting excessive iron-exposure and damage in neurons (Ripa et al., 2017). As for the gut, there are a number of homolog or closely related genes which have been previously identified in ageing (Tacutu et al., 2018), such as chek1, mad2l1 (MAD2 mitotic arrest deficient-like 1 (yeast)), dl3 (delta like 3, nog (noggin), ifnb1 (interferon beta), socs2 (suppressor of cytokine signaling 2), amongst many others, which can be found in the GeneAge database.
Since it is known that short telomeres themselves can lead to de-regulated gene expression, particularly in genes near the chromosome ends due to loss of the “telomere positioning effect” (TPE) (Robin et al., 2014), we proceeded to map the genes identified to be de-regulated prematurely in the absence of telomerase to their chromosome location, with the aim of determining whether they located to within up to 10MB of either of the telomeric ends (Fig. 5C1). We found that whereas most tert-dependent gene changes of old age do not locate to the end of chromosomes, in both gut and brain (Fig. 5C2 and C3). However, there are significantly more tert-dependent genes located at the end of chromosome in the gut than in the brain, (Fig. 5 C4). This is consistent with gut being a more proliferative tissue, where telomeres are likely to shorten more, which would be particularly exacerbated in the absence of tert, leading to a higher probability of TPE contributing to gene transcription alterations.
2.5 Transcriptional changes in common between the gut and brain at the early and late stages of ageing
Even though the aged phenotype is something usually observed late in life, the underlying molecular and cellular mechanism driving these changes can, arguably, start from the moment you are born (Gladyshev, 2021). To understand what significant transcriptional changes are taking place in the gut and brain and, in particular, in common between the gut and brain, we compared all DEGs and STEM DEGs in these tissues at the earliest time point we detect significant changes (9 months) and at the late stages of ageing, i.e, at the latest time point analysed of 35 months (Fig 6, Supp Fig 3 and source data). We identified 7 gene changes in common between WT gut and brain at the early stages of ageing (9 months of age). In specific, we identified 5 down-regulated genes, namely ccnb1, kif2c, serpinh1a, temem37 and si:ch211-5k11.8, and 2 up-regulated genes, namely cfap45 and eif4ebp3l. Of these, ccnb1 (G2/mitotic-specific cyclin-B1) and kif2c (Kinesin-like protein) are both proteins involved in cell-cycle progression. Whereas Ccnb1 has been reported to be involved in normal stem-cell/progenitor maintenance in the brain (Hagey et al., 2020) and gut (Basak et al., 2014); kif2c can act as a DNA damage repair protein, and, accordingly, its depletion leads to significant accumulation of DNA damage (Zhu et al., 2020). STRING analysis suggests these proteins are likely to be co-expressed in a variety of organisms including zebrafish and humans, highlighting potential functional links (Fig 6 B and source data). The downregulated elf4ebp3l gene (Eukaryotic translation initiation factor 4E-binding protein 3-like) encodes a repressor of translation and is inhibited in response to TORC1 (mammalian target of rapamycin complex 1) (Boutouja, Stiehm, & Platta, 2019).
At “late stages” of ageing, we identified 23 gene changes in common between the gut and brain. Of these, STRING analysis identified 2 main protein network interactions, namely a link between cd59 and cd99l2 and link between il2rb and b2m. In specific, cd59 and cd99l, which our data show to be downregulated in old age, in both gut and brain, have both been cited in the literature as markers of new-born neurons and oligodendrocyte progenitor cells, respectively, in a single-cell transcriptomic analysis of the adult zebrafish brain (Lange et al., 2020). In the gut, cd59, also known as protectin, has been shown to be downregulated in the gut epithelium of ulcerative colitis and Chron’s disease patients and thought to render the tissue susceptible to inflammatory damage (Scheinin et al., 1999). cd99 has been shown to be a key molecule in modulating monocyte migration through endothelial junctions (Schenkel, Mamdouh, Chen, Liebman, & Muller, 2002) and monocyte differentiation to macrophages is known to be triggered by endothelial transmigration (Gerhardt & Ley, 2015; Li et al., 2020), including in the brain via migration through the blood-brain barrier (Ifergan et al., 2007). Il2rb and b2m, which our data show to be up-regulated in both gut and brain at the “late stages” of ageing, are key molecules involved in adaptive and innate immune function. Whilst the IL2R beta is important for T-cell mitogenic response to IL-2, the b2m (Beta-2-microglobulin) protein is a component of the major histocompatibility complex I (MHCI), involved in antigen presentation. Intriguingly, b2m has been shown to be present in a soluble free-form, and has been found to be systemically increased with ageing in humans and in individuals with neurodegenerative diseases (Smith et al., 2015). Importantly, heterochronic parabiosis experiments between young and old mice suggest that increased b2m with ageing leads to cognitive impairment and neurodegeneration, and has hence been identified as a systemic pro-ageing factor (Smith et al., 2015). Additionally, increased IL2 receptor expression has been shown to contribute to CD4 differentiation and exhaustion of their naïve pool and therefore ability to respond to infection with ageing, in humans (Pekalski et al., 2013; H. Zhang, Weyand, & Goronzy, 2021).
3 DISCUSSION
In this study, we used RNA sequencing analysis to determine the kinetics of telomerase-dependent and -independent transcriptomic changes occurring during WT ageing in brain and gut tissues, using the zebrafish as a model. We hypothesised that this would allow us to identify key age-associated genes and pathways that become prematurely de-regulated in both or either tissue, providing key insights into the early stages of ageing in each tissue and highlight potential interactions.
3.1 STEM versus all DEGs analysis
Ageing can be described as a time-dependent change in tissue homeostasis, that increases the probability of disease and death (Hayflick, 2007). Whether the genes and pathways driving these time-dependent changes are also consistently changing in a time-specific manner, remains unresolved. To account for both possibilities, we performed a time-series analysis (STEM) and then combined this with the more traditional differential gene expression (DEGs) analysis between young and old animals. Even though the gene changes identified with the STEM analysis were also picked up by the traditional DEGs analysis, the STEM analysis provided a much tighter, restrictive view of the transcriptomic signatures of ageing. In the gut, particularly, the main hallmarks of ageing identified using the STEM analysis are quite different from the ones using the traditional all DEGs. In particular, STEM analysis identifies mitochondrial dysfunction and de-regulated nutrient sensing as the main hallmarks of WT ageing at the earlier stages of ageing in the gut (WT 9 months), whereas all DEGs analysis identified stem cell dysfunction as the principal hallmark de-regulated at that age. The significance of these differences is difficult to judge, but it can suggest that changes in genes affecting mitochondrial function and nutrient sensing have a mostly monotonic trajectory in gut ageing, whereas the ones involved in stem cell maintenance can have different behaviours at different times throughout life. Nevertheless, this difference was not observed in the WT brain or in the tert-/-ageing, where STEM and all DEGs analysis led to very similar conclusions regarding the identity or kinetics of the main hallmarks of ageing affected. Importantly, the kinetics of gene changes and processes identified in our data match very well to the phenotypes of ageing previously reported in the tert-/- and WT ageing. In particular, in the gut, impaired cell proliferation in the gut is one of the first tert-dependent phenotypes of ageing identified, followed by cellular senescence and inflammation later in life (Madalena C Carneiro et al., 2016; Henriques et al., 2013). Moreover, the recently reported tert-dependent changes in macrophage immune activation and increased gut permeability (Pam S. Ellis, 2022) are consistent with key tert-dependent gene changes of old age identified here. An example of such genes are cd99, potentially involved in macrophage differentiation via trans-endothelial migration (Gerhardt & Ley, 2015; Li et al., 2020; Schenkel et al., 2002); and cldn5a (claudin 5), involved in tight-junctions (Lu, Ding, Lu, & Chen, 2013).
3.2 Main hallmarks of ageing
A simplistic prediction of how the kinetics of the hallmarks of ageing would behave over the lifecourse could be that, at early ages, we would detect more changes affecting the primary hallmarks of ageing, i.e, the “causes of damage”, namely genomic instability, telomere attrition, epigenetic alterations and loss of proteostasis. At the last stages, one could predict that we would detect more significant changes in the integrative hallmarks, i.e, the “culprits of the phenotype”, namely stem cell exhaustion and altered intercellular communication, of which inflammation is a key component, as described in (López-Otín et al., 2013). However, either separately or combined, neither STEM nor all DEGs analysis show this. Our combined analyses show that most of the gene changes occurring at the early stages of WT ageing are observed from 9 months and are mostly involved in stem cell maintenance, in both gut and brain. One potential explanation for this observation, is that our qualitative analysis was not able to distinguish between such hallmarks or is under-estimating the primary hallmarks. Another explanation comes from the relatively low sample size used for each time-point and the heterogeneity of individuals physiology amongst each population. As recently showed in (Dambroise et al., 2016), zebrafish age following the two-phase model first proposed flies (Tricoire & Rera, 2015), based on the age-related intestinal permeability assessed using the Smurf assay they previously described (Rera, Clark, & Walker, 2012). Moreover, we have recently shown that gut permeability with ageing is accelerated in the absence of telomerase (tert-/-) (Pam S. Ellis, 2022). Following this model and considering the longevity curve from the population we sampled, the proportion of Smurfs might have been approximately, <10% at 2 month, 25% at 9 month, 50% at 22 months and >80% at 35 months. The chances to have selected at least 1 Smurf by accident are approximately 27% at 2 months, 57% at 9, 86% at 22 and 99% at 35%. This could contribute to the heterogeneity or potential bias towards having more of a specific ageing signature over another (smurf versus non-smiurf).
At first glance, it is surprising that telomere dysfunction is not picked up as a main hallmark of ageing in the telomerase mutant model. However, this may simply be the reflection of the number of genes that have been directly implicated in each of these hallmarks. In specific, there are a lot less genes that one would identify as directly involved in telomere dysfunction, when compared to stem cell dysfunction, for example. The main culprits for telomere dysfunction would be telomerase and the shelterin components, whereas for stem cell dysfunction, all the cell cycle and DNA damage repair proteins can play a role. For example, chek1, fxr1 and daxx, which are all de-regulated in the old WT gut and brain (chek1) (see source data), have all been identified as potential regulators of telomeres (Nersisyan et al., 2019). Yet, because that is not the main function one would attribute to such genes, these would have been missed as part of the telomere dysfunction hallmark of ageing. Additionally, it is not possible to distinguish from just looking at lists of DEGs if such gene was de-regulated due to telomere dysfunction in the first place, or if its dysfunction will affect telomere function indirectly. These are some of the considerations that highlight the complexity of ageing and the non-linearity of how the hallmarks of ageing may drive ageing as well as each other and it is important to have them in mind when interpreting our analyses. Nevertheless, at the later stages of WT ageing (>35 months), altered intercellular communication, a previously described “culprit of the phenotype” is indeed the most significant hallmark of ageing detected in common between the gut and brain. Finally, in terms of the progression hallmarks of ageing, different hallmarks may play more prominent roles at specific times of life, and may be replaced by others at other times, explaining why we don’t necessarily always see a linear accumulation of the different hallmarks of ageing over-time. This is particularly evident in the gut, where stem cell exhaustion is the main hallmark identified by the combined STEM and ALL DEGs analysis, and it is barely represented at the later stages of life. In the brain, the progression seems to be more linear, though, and most hallmarks present at the early time point of 9 months remain until old age, when other hallmarks are further added.
3.3 Telomerase- and/or telomere-dependent changes with ageing
If a gene expression changes or a biological process alteration is accelerated in the absence of telomerase (i.e becomes significant at an earlier age in the tert-/-), we considered it to be telomerase dependent. If we take a step back and look at the processes and, in turn, hallmarks of ageing affected in the absence of telomerase, it becomes clear that the tert-/- is indeed accelerating many of these changes. It is particularly evident when we look at the pie charts depicting the different hallmarks of ageing affected at each time point in WT versus tert-/-. Here, it is striking how a 2-month-old tert-/- gut resembles a WT gut at the old age of 35 months, and how a 9-and 22-month-old tert-/- brain resembles an old WT brain at 35 months. This is further highlighted by the further comparison we performed, where we asked at which age does the tert-/- share more gene expression changes with the WT old (>35 months). In this analysis, we show that the 2 months old tert-/- has the most gene expression changes shared with the 35 months old WT gut, whereas it is the tert-/- at 22 months that has the most gene expression changes shared with the 35 months old WT brain. This suggests that the gut is displaying telomerase-dependent hallmarks of ageing at an earlier age than the brain, which is consistent with what would be expected for a high versus low proliferative tissue. Accordingly, when looking at the specific gene changes accelerated in the absence of tert, i.e, we identify significantly more tert-dependent genes located near the ends of chromosomes in the gut than in the brain, and therefore more liable to have been altered due to telomere shortening. In the future, it would be insightful to test how many of the tert-dependent gene changes occur due to non-canonical functions of telomerase involved in transcriptional regulation.
Nevertheless, tert-dependent gene changes in both tissues are still a minority, serving as a reminder that “all roads lead to Rome”, and one should exercise caution when trying to identify genes influencing the natural process of ageing using premature ageing models. It is not necessarily the same genes influencing the processes of ageing in these models, even if the processes and phenotypes are accelerated. We should also have this in mind when we compare the sets of genes identified in this study with those previously identified as implicated in human ageing, and not necessarily be surprised if only a small percentage of those are shared. One could argue that it is more important that the processes are shared. Nevertheless, we do identify many gene changes shared between zebrafish and human ageing databases as highlighted throughout the results’ section.
3.4 Which genes to focus on if we were to target age-associated dysfunction in the gut and brain
As a final step in our analysis, we wanted to identify common gene expression changes between the gut and the brain at the earliest stages of ageing, in our case, from the age of 9 months. We hypothesised that this may highlight potential common therapeutic targets for early intervention to prevent age-associated dysfunction in both tissues. We identified 8 significant DEGs in common between the gut and brain at 9 months of age, 5 down-regulated (ccnb1, kif2c, serpinh1a, si:ch211-5k11.8 and tmem37) and 2 up-regulated (cfap45 and eif4ebp3l). We could then hypothesise that restoring expression of these genes to youthful levels would have a positive impact on delaying or ameliorating the development of ageing phenotypes in both these tissues at the same time. Of note, we identified that, amongst these, ccnb1 and eif4ebp3l were telomerase-dependent changes in the brain. If so, one could hypothesise that telomerase re-activation in the brain should restore these genes to young WT levels and potentially contribute to amelioration of ccnb1- and eif4ebp3l-dependent ageing phenotypes in the brain.
4 CONCLUSIONS
We provide the first systematic analysis of transcriptomic changes throughout the lifespan of zebrafish in the gut and brain of the same group of individuals, leading to the identification of key genes and processes likely involved in driving the ageing process in these tissues. Many of these genes have previously identified in human ageing databases and many of them are potentially new genes of ageing, which will have to be experimentally tested in relevant model organisms. This analysis and the open access availability of its source and raw data should provide a key steppingstone and framework supporting future work for understanding the ageing process and using zebrafish for studying ageing.
5 MATERIALS AND METHODS
5.1 Zebrafish husbandry, genotypes and ages
Zebrafish were maintained at the standard conditions of 27-28ºC, in a 14:10 hour light-dark cycle, and fed twice a day with Artemia (live rotifers) and Sparus (dry food). All the experiments were performed in the University of Sheffield. All animal work was approved by local animal review boards, including the Local Ethical Review Committee at the University of Sheffield (performed according to the protocols of Project Licence 70/8681).
Two strains of adult zebrafish (Danio rerio) were used for these studies: wild-type (WT; AB strain) and tert-/- (tertAB/hu3430). Wild-type fish were obtained from the Zebrafish International Resource Center (ZIRC). The telomerase mutant line tertAB/hu3430 was generated by N-ethyl-nitrosourea mutagenesis (Utrecht University, Netherlands(Wienholds et al., 2003)). It has a T→A point mutation in the tert gene and is available at the ZFIN repository, ZFIN ID: ZDB-GENO-100412–50, from ZIRC. The fish used in this study are direct descendants of the ones used previously29,30, by which point it had been subsequently outcrossed five times with WT AB for clearing of potential background mutations derived from the random ENU mutagenesis from which this line was originated. The terthu3430/hu3430 homozygous mutant is referred to in the paper as tert−/− and was obtained by incrossing our tertAB/hu3430 strain. Genotyping was performed by PCR of the tert gene13,14. In order to study age-related phenotypes in the zebrafish gut and brain, we used >30 months old fish for what we consider old in WT (in the last 25-30% of their lifespan), and we considered the tert-/- old fish at the equivalent age (>22 months), which approximately corresponds to the last 25-30% of their lifespan. In specific, ‘Old’ was defined as the age at which the majority of the fish present age-associated phenotypes, such as cachexia, loss of body mass and curvature of the spine. These phenotypes develop close to the time of death and are observed at >30 months of age in WT and at >22 months in tert-/-(Madalena C Carneiro et al., 2016; Henriques et al., 2013).
5.2 Tissue dissection and RNA extraction
Animals from various age-groups were used for RNA-Sequencing (WT at 2, 9, 22 and >30 months of age; and tert-/- at 2, 9 and 22 months of age). Three biological replicates were used per group. Fish were culled by overdose of MS-222, followed by confirmation of death. Then, the whole tissues were rapidly dissected in cold PBS (Sigma-Aldrich), transferred to a microcentrifuge tube containing 100 μl of Trizol (Thermo Fisher Scientific), snap-frozen in dry ice and stored at -80°. To isolate the RNA, extra 50 μl of Trizol was added to each sample, and the tissue was homogenized with a mechanical homogenizer (VWR International) and a 1.5 pestle (Kimble Chase, Vineland, NJ, USA). After 5 min incubation at room temperature (RT), 30 μl of chloroform (1:5, VWR International) was added and the samples were incubated for further 3 min at RT before centrifuged at 13,000g, for 30 min, at 4°C. Isopropanol (Thermo Fisher Scientific) was then added to the aqueous phase of the solution, and the resultant mix was incubated for 10 min at RT, before centrifuged (13,000g for 15 min at 4°C). Finally, the pellet was twice washed in 250 μl of ice cold 75% ethanol and left to air-dry, before resuspended in 14 μl of nuclease-free water. RNA integrity was assessed with the bioanalyzer Agilent 2100 Bioanalyzer (Agilent, Santa Clara, CA, USA). All the samples had a RNA integrity number (RIN) ≥9.
5.3 RNA-Sequencing
The RNA-Sequencing was performed at the Genomics and Sequencing facility of Sheffield Institute for Translational Neuroscience (SITraN). Library preparation was performed following the Illumina methodology. mRNA was extracted from 500 ng of RNA with oligo-dT beads, capturing poly(A) tails, using the NEBNext Poly(A) mRNA Magnetic Isolation Module (New England Biolabs Inc). cDNA libraries were made with the NEBNext® Ultra™ RNA Library Prep Kit for Illumina, following the manufacturer’s instructions (New England Biolabs Inc). The samples were then sequenced on an Illumina HiSeq SQ, using a high output run and sequencing by synthesis V3 chemistry on a single 100 bp run. The data was imported into a FASTQ file format in order to perform the analysis.
5.4 RNA-Sequencing analysis
5.4.1 Data processing
Quality control was performed using MultiQC version 1.9. Cutadapt version 3.0 was used for trimming the first 13 bases from the reads to remove poor quality base pairs in the reads. Read alignment was performed as follows. Single-end reads were aligned against the reference genome Danio_rerio.GRCz11.dna. primary_assembly.fa using STAR. A bespoke alignment index was built using annotation file Danio_rerio.GRCz11.103.gtf and an expected read length of 88 bp. Ht-seq count was run in non-stranded mode to obtain per gene read counts.
5.4.2 Differential expression
To identify signatures of ageing, WT gut and brain samples were subjected separately to DESEq2 analysis, comparing the time points 9, 22 and >30 months with the time-point of 2 months. Then, tert-/- samples at 2, 9 and 22 months were contrasted with WT at 2 months, in order to identify telomerase-dependent ageing processes.
5.4.3 Time-series analysis
Short Time-series Expression Miner (STEM) software was used to assign genes to predefined expression profiles genes based on their expression across the time points. For this, the 2 months’ time-point was used as the zero time point for the analysis. The maximum number of model profiles was set to 50; the maximum unit change in model profiles between time points was set to 2; and the minimum absolute log ratio of expression was set to 1.0, with change based on maximum − minimum. Significance level of the model profiles was set to 0.05 with Bonferroni correction. Minimum correlation for profile clustering was set to 0.7. The statistically significant temporal profiles were visualised as line plots using ggplot2. Median expression fold change values of the genes in each profile were shown on the plots with a thicker line.
5.4.4 Enrichment analysis of temporal profiles from STEM
Enrichment analysis was performed using all the differentially expressed genes (DEGs) and using the genes identified in the STEM analysis, separately. Gee-set over-representation analysis (ORA) of GO Biological Process (GOBP), GO Molecular Function (GOMF), and GO Cellular Compartment (GOCC) terms were performed using the enrichGO function of clusterProfiler package version 3.18.0. Minimum and maximum gene set sizes were set to 10 and 500, respectively. Results were simplified using the simplify function of clusterProfiler with the similarity cut-off set to 0.7 and minimum adjusted p-value used as the feature for selecting the representative terms. Enrichment results with adjusted p-values below 0.05 and at least 3 core enrichment genes were considered significantly enriched. ORA of KEGG and REACTOME pathway terms were performed using the enrichKEGG and enrichPathway functions of clusterProfiler. Minimum and maximum gene set sizes were set to 10 and 500, respectively. The same significance criteria for the enrichments were used as for the GO term enrichments. Results of the enrichment results were visualised as barplots or as pie charts. Barplots were made using the ggbarplot function of R package (R studio version 2021.09.2), ggpubr version 0.4.0, showing a maximum of five pathways with adjusted p-value below 0.05 per pathway category. Pie charts were made using Prism GraphPad version 9.0.
5.4.5 Protein-protein interaction network analysis
The search tool for retrieval of interacting genes (STRING) (https://string-db.org) database, which integrates both known and predicted PPIs, was used to predict functional interactions of proteins(Szklarczyk et al., 2021). Active interaction sources, including text mining, experiments, databases, and co-expression as well as species limited to “Danio rerio” and an interaction score > 0.7 (high confidence) were applied to construct the PPI networks. The network was further clustered using k-means clustering to a specific number of up to 3 clusters.
5.4.6 Venn Diagram analysis
Venn Diagram analysis was performed using the online tool Venny 2.1.0 – BioinfoGP (https://bioinfogp.cnb.csic.es/tools/venny/)(Oliveros, 2007-2015).
5.4.7 Statistical analysis
A chi-square test was used in Fig 5 to compare between the relative chromosome location of telomerase-dependent genes in the gut and brain, using raw data values, even though it is the % that is represented in the graphs, to ensure accurate analysis. P value <0.05 was considered significant and denoted with a *, whereas ns was used to denote non-significant differences.
AUTHOR CONTRIBUTIONS
RRM and CMH conceived and designed this work; RRM performed additional RNA sequencing beyond that provided by Genevia services; RRM, MR and CMH analysed and interpreted RNA sequencing results and co-wrote the manuscript; CMH designed the figures with input from co-authors.
COMPETING INTERESTS
The authors declare no competing interests
DATA AVAILABILITY STATEMENT
The RNA sequencing data from this experiment were deposited in gene expression omnibus (GEO) and will be made available when in peer review.
SUPPORTING INFORMATION
Supporting information in the form of figures, tables and source data can be found online in the Supportive information section at the end of the article
FIGURE LEGENDS
ACKNOWLEDGEMENTS
The authors would like to acknowledge Genevia Technologies Oy for the main bulk of RNA sequencing analysis service. We are very grateful to Joao Ribeiro for valuable advice regarding the many functionalities of Microsoft Excel, which greatly facilitated data analysis. This work was generously funded by a University of Sheffield PhD studentship to RRM, a Sheffield University Vice Chancellor’s Research Fellowship and a Wellcome Trust/Royal Society Sir Henry Dale Fellowship (UNS35121) to CMH.