Splicing alterations in healthy aging and disease

Abstract

Alternative RNA splicing is a key step in gene expression that allows generation of numerous messenger RNA transcripts encoding proteins of varied functions from the same gene. It is thus a rich source of proteomic and functional diversity. Alterations in alternative RNA splicing are observed both during healthy aging and in a number of human diseases, several of which display premature aging phenotypes or increased incidence with age. Age-associated splicing alterations include differential splicing of genes associated with hallmarks of aging, as well as changes in the levels of core spliceosomal genes and regulatory splicing factors. Here, we review the current known links between alternative RNA splicing, its regulators, healthy biological aging, and diseases associated with aging or aging-like phenotypes.

1 |. INTRODUCTION

All organisms age, a process defined by a progressive decline of fitness and organ functions over time, ultimately leading to death (Kennedy et al., 2014; Kirkwood, 2005). Aging is associated with tissue deterioration and disorganization as a result of widespread genomic and cellular changes including the loss of stem cell renewal capabilities, genomic instability, metabolic alterations, altered cellular communication, loss of proteostasis, telomere attrition, cellular senescence, and epigenetic alterations (Lopez-Otin et al., 2013). Aging is measured on two separate clocks, a chronological age clock—the number of months or years since birth—and a biological age clock—the physiological or phenotypical age/status of cells, tissues, and organs. While both chronological and biological aging are progressive, not all organisms biologically age in the same way or at the same rate (Karasik et al., 2005; Kennedy et al., 2014; Kirkwood, 2005). Aging is the primary risk factor for major human pathologies, including cancer, diabetes, cardiovascular disorders, and neurodegenerative diseases, and therefore extensive research is being carried to understand the molecular basis of biological aging (Kennedy et al., 2014; Kirkwood, 2005). In the context of aging, much research has been targeted at understanding age-related changes in gene expression regulation via age-related genomic and epigenetic instability at the DNA level, transcription factor regulation, and age-related decline of proteasomal function at the protein level (Lopez-Otin et al., 2013; Niedernhofer et al., 2018). Recently, researchers have begun to uncover the significance of age-related gene regulation at the messenger RNA (mRNA) level, including through the regulation of alternative RNA splicing (Bhadra et al., 2020; Deschenes & Chabot, 2017; Lai et al., 2019; Latorre & Harries, 2017; Li et al., 2017; Stegeman & Weake, 2017). This review will explore the currently published links between alternative splicing (AS) and aging.

The central dogma states that genetic information stored in genomic DNA is transcribed to generate mRNA molecules, which are then translated into downstream, functional protein products. Expression of a gene (i.e., its functional output) is regulated via multiple biological processes at all three levels of this central dogma: at the DNA level via transcriptional regulation, at the RNA level via posttranscriptional regulation, and at the protein level via translational and posttranslational regulation. In the late 1970s, Rich Roberts and Phil Sharp discovered that, in contrast to simpler, bacterial genes, complex eukaryotic genes contain fragments of coding material (exons) interrupted by non-coding material (introns) (Berget et al., 1977; Chow et al., 1977). In the process of mRNA maturation, the introns of the pre-mRNA transcript are removed, and the exons are pieced together to generate a mature mRNA molecule, in a process known as constitutive RNA splicing. Metazoans have also adapted an mRNA editing process known as AS, that controls whether an exon (or intron) is included or skipped in the final mRNA transcript (Baralle & Giudice, 2017; Manning & Cooper, 2017). AS has been instrumental in evolution as it allows generation of diverse mRNA transcripts that can encode proteins of differing functions from the same gene (Nilsen & Graveley, 2010). AS can impact protein function, subcellular localization, stability, regulation, and structure. Over 95% of human genes undergo AS (Pan et al., 2008; Wang et al., 2008), and it is therefore a major contributor to proteomic and functional diversity. AS is highly regulated during healthy development and in disease. Alterations in the splicing regulatory machinery have been implicated in a variety of human diseases, including age-related pathologies such as progeria, neurodegenerative disorders, and cancer (Deschenes & Chabot, 2017; Scotti & Swanson, 2016; Urbanski et al., 2018).

2 |. AGE-RELATED ALTERATIONS IN SPLICING REGULATORY COMPONENTS

2.1 |. The splicing machinery

Alternative RNA splicing is a highly regulated process carried out by the core splicing machinery, called the spliceosome, along with regulatory splicing factors (SFs). The spliceosome, a very large (~3 MDa) ribonucleoprotein complex, functions as molecular scissors that recognize the intron-exon boundaries of genes defined by the 5′ splice donor, 3′ splice acceptor, and branch sites, and removes intronic regions from the pre-mRNA molecule (Wilkinson et al., 2020) (Figure 1a). The spliceosome contains over 300 components that undergo dynamic remodeling during the splicing reaction (Hegele et al., 2012). The core spliceosome is composed of several small nuclear RNA molecules that interact with “Sm” core proteins and additional proteins to form small nuclear ribonucleoprotein (snRNP) particles (U1, U2, U4, U5, and U6) (Shi, 2017). During the splicing reaction, first, the U1 snRNP binds, in an ATP-independent manner, the intronic 5′ splice site, including the conserved GU dinucleotide; whereas SF1, U2AF2, and U2AF1 bind to the branch point site, the polypyrimidine tract, and the conserved AG of the 3′ splice site, respectively (Figure 1a). Then, the U2 snRNP interacts, in an ATP-dependent manner, with the branch point site. This interaction is stabilized by the SF3a and SF3b protein complexes, as well as U2AF2 and U2AF1, and leads to the displacement of SF1 from the branch point site. Third, the preassembled U4/U6/U5 tri-snRNP is recruited to form a catalytically inactive complex, which then undergoes conformational changes, including release of U1 and U4, leading to spliceosome activation and formation of the catalytically active complex. The first catalytic step of splicing leads to lariat formation via cleavage of the 5′ splice site. The second catalytic step results in the joining of the two exons. Post-splicing, the spliceosome disassembles in an orderly manner, releasing the mature spliced mRNA, as well as the lariat bound by U2/U5/U6 (Bonnal et al., 2020; Shi, 2017; Wilkinson et al., 2020).

FIGURE 1.

The splicing machinery. (a) Graphical representation of the stepwise assembly of spliceosomal complexes on a pre-mRNA molecule and catalysis of the splicing reaction to generate mature spliced mRNA (5′SS, 5′ splice site; BPS, branch point site; Py-tract, polypyrimidine tract). (b) In addition to core spliceosomal components, regulatory splicing factors (SFs) acting as positive regulators (e.g., SRSF1 to SRFS12, TRA2α and TRA2β), or negative regulators (e.g., HNRNPA1, HNRNPH3, HNRNPM, HNRNPDL, etc.), can bind to exonic or intronic splicing enhancer (ESE or ISE) or silencer (ESS or ISS) sequences to fine tune splicing and promote exon inclusion or skipping. (c) Alternatively spliced sequences can be classified into the following patterns: cassette exon, alternative 5′ or 3′ splice site usage, inclusion of mutually exclusive exons, intron retention, alternative first or last exons. Alternative first exons arise as a result of both alternative promoter usage and alternative splicing, while alternative last exons arise as a result of both alternative polyA usage and alternative splicing

In metazoans, the 5′ and 3′ splice site consensus motifs are quite short and degenerate, that is, only two nucleotides are almost always invariant across splice sites, with the remaining nucleotides of the consensus motif are less conserved, making AS regulation possible (Schwartz et al., 2008). This regulation is carried out by regulatory SFs, a class of RNA-binding proteins that recognizes and binds auxiliary sites on the pre-mRNA, namely intronic or exonic splicing silencer or enhancer sequences, and promotes or represses splicing of that exon (Figure 1b). The serine-arginine rich (SR) proteins and heterogenous nuclear ribonucleoproteins (hnRNP) are two SF families that regulate AS by binding splicing enhancer or silencer sequences to activate or repress splicing respectively (Geuens et al., 2016; Howard & Sanford, 2015). SFs are critical for normal development and SF defects have been causatively implicated in several pathologies, including cardiac and liver dysfunction, brain abnormalities, diabetes, lupus, and cancer (Baralle & Giudice, 2017; Blech-Hermoni & Ladd, 2013; Colegrove-Otero et al., 2005; Dillman et al., 2013; Su et al., 2018; Torres-Padilla et al., 2001; Urbanski et al., 2018).

The core spliceosome, along with regulatory SFs, allows generation of differentially spliced mRNA isoforms from a given gene, in a given cell, at a given time. These spliced isoforms can include or lack additional exonic or intronic sequence(s) in their coding region or untranslated regions (Figure 1c), thus impacting the gene’s protein coding potential, its expression levels, or its posttranscriptional or posttranslational regulation. Whether an exon is included or skipped relies on numerous inputs, including the levels of SFs, which act in a dose-dependent manner (Geuens et al., 2016; Howard & Sanford, 2015). SF-levels and activity are therefore tightly controlled, at the epigenetic and transcriptional levels, posttranscriptionally via AS coupled to nonsense mediated mRNA decay (NMD), and posttranslationally by specific kinases (Geuens et al., 2016; Howard & Sanford, 2015). Many of these regulatory mechanisms are altered during aging and could therefore lead to changes in SF levels and activity in human and model organisms. Interestingly, components of the NMD mRNA surveillance machinery are implicated in the regulation of longevity in worms (Son et al., 2017), and may provide a link for RNA homeostasis with aging and longevity.

2.2 |. Changes in SF levels are detected with aging in multiple tissues

Aging studies profiling gene expression have highlighted age-related alterations in pathways and/or genes associated with RNA processing, including AS (Harries et al., 2011; Southworth et al., 2009). Changes in SF levels could lead to significant downstream AS changes, including changes in the levels of spliced isoforms implicated in aging hallmarks and age-related diseases. Age-related SF expression changes have been detected in multiple organisms (e.g., mice, rats, and humans) and tissues (e.g., blood, brain, muscle, skin and liver). While still speculative, it has been proposed that >50% of all age-related alterations in spliced isoforms could be linked with age-related changes in SF expression (Mazin et al., 2013; Meshorer & Soreq, 2002) (Figure 2).

FIGURE 2.

Age-related changes in splicing factor (SF) levels detected across multiple studies. Differential expression of genes or proteins implicated in AS regulation, including spliceosomal components, SR proteins, HNRNP proteins, or other regulatory factors, as measured by micro-array, RNA-seq, targeted QPCR, or proteomics, in tissues from young versus old human (h) or mouse (m). Blood samples are (i) human leukocytes (ages ranging from 30- to 104-years old) (Harries et al., 2011); (ii) human peripheral blood (lnCHIANTI study, 30–104-years old) (Holly et al., 2013); (iii) human peripheral blood (SAFHS study, 15–94-years old) (Holly et al., 2013); (iv) human peripheral blood (15–105-years old) (Peters et al., 2015); and (v) human whole blood (PIVUS study, comparing 70- vs. 80-years old) (Balliu et al., 2019). Brain samples are from (i) human pre-frontal cortex and cerebellum (0–98-years old) (Mazin et al., 2013); (ii) human whole brain (16–102-years old) (Tollervey, Curk, et al., 2011); (iii) mouse cortex (postnatal day 7 vs. 21 months) (Weyn-Vanhentenryck et al., 2018); (iv) mouse cerebral cortex (postnatal day 1 vs. day 56) (Kadota et al., 2020); and (v) mouse hippocampus (2 vs. 24 or 29 months) (Stilling et al., 2014). Liver samples are from (i) human liver (1–85-years old) (Chaturvedi et al., 2015); and (ii) mouse hepatocytes (postnatal day 1 vs. day 56) (Kadota et al., 2020) respectively. Mouse skin samples (4 vs. 18 or 28 months) are from Rodriguez et al. (2016). Heart samples are from mouse cardiomyocytes (postnatal day 1 vs. day 56) (Kadota et al., 2020). Skeletal muscle samples are from (i) mouse (4 vs. 18 or 28 months) (Rodriguez et al., 2016); and (ii) human (20–29 years vs. 65–71 years) (Welle et al., 2004); (iii) human (21–27 years vs. 67–75 years) (Welle et al., 2003); and (v) human (GESTALT study, 20–87 years) (Ubaida-Mohien et al., 2019)

2.2.1 |. Age-related SF changes in blood and hematopoietic lineages

Several longitudinal aging studies uncovered an enrichment for genes associated with splicing in human whole blood or specific blood cell types. Using 30–104-year-old human subjects from the InCHIANTI cohort, Harries et al. (2011) uncovered surprisingly few age-related transcripts in peripheral blood leukocytes as assayed by microarrays. However, gene set enrichment analysis identified mRNA splicing, polyadenylation, and other posttranscriptional event gene sets as negatively associated with age (Harries et al., 2011) (Figure 2). A complementary expression array study using the InCHIANTI cohort found that ~40% of the tested SF proteins were significantly associated with age, most of which were downregulated (Holly et al., 2013). Similar findings were described in the expression array analysis of the San Antonio Family Heart Study (SAFHS) cohort, comprised of human lymphocyte samples from 15- to 94-year-old subjects (Holly et al., 2013). Differentially expressed age-related transcripts shared between the InCHIANTI and SAFHS studies include spliceosomal components and SFs such as LSM2, LSM5, SF3B1, SRSF1, SRSF6, HNRNPD, HNRNPH3 (all downregulated), SUGP2 (upregulated), and HNRNPAB (different directions) (Holly et al., 2013). Additional analysis of peripheral blood samples of two follow-up visits of the InCHIANTI cohort found that the age-related differential expression of SFs HNRNPM and HNRNPA0 and spliceosomal component AKAP17A, are predictively associated with tests measuring cognitive decline in the same individuals (Lee, Pilling, et al., 2019). Interestingly, age-related expression changes in SRSF1, SRSF6, HNRNPD, HNRNPH3, LSM2, and LSM5, among other SF proteins, were identified in a more recent RNA-sequencing (RNA-seq) analysis of human peripheral blood (Peters et al., 2015). Moreover, the Prospective Investigation of Uppsala Seniors (PIVUS), a 10-year longitudinal aging study comparing whole blood samples of 70- and 80-year-old subjects, reported, using RNA-seq, a positive enrichment for genes associated with RNA metabolism and ribosomal proteins among the genes differentially expressed with age (Balliu et al., 2019). Changes in specific SF transcript levels were detected between the 80- and 70-year-old individuals (Figure 2). These alterations in SF levels could lead to alterations in their downstream target, and regulate, directly or indirectly, any of the ~500 introns differentially spliced with age detected in this study (Balliu et al., 2019). Interestingly, SF levels generally increased with age in this study, contrasting with the previously described transcriptomic studies. However, it is worth noting that the PIVUS study’s 10-year time scale is shorter than the previous aging studies, and the two PIVUS time points are 70 versus 80 years in age, which would both be considered as older age in previous studies. Therefore, it is possible that changes in SF expression may not be completely linear during aging, but also that the definitions of “young” and “old,” which are somewhat relative to the study design or population cohort, can influence the study’s findings.

2.2.2 |. Age-related SF changes in neuronal tissues and cell types

The brain exhibits unique and complex neuronal-specific AS patterns, as well as express a number of neuronal-specific or enriched SFs (Raj & Blencowe, 2015). Human brain samples from the temporal cortex, collected post-mortem from 16- to 102-year-old subjects, exhibited age-related decreased expression of SFs ESPR1, ESPR2, HNRNPK, SNRPB2, and SRSF2, increased expression of SLU7, as well changes in the expression of neuronally enriched SFs NOVA1 (decreased), PTBP1 (increased), and PTBP2 (decreased), which have been shown to regulate AS in neurons (Tollervey, Curk, et al., 2011) (Figure 2). Similarly, in the human cerebellum and prefrontal cortex, Mazin et al. (2013) not only reported age-related changes in expression of SFs (Figure 2), but also identified an enrichment in SF-binding motifs of PTBP1, PTBP2, hnRNPA1, hnRNPH1-3, hnRNPHF, TRA2β, SRSF2, and SRSF5, in the identified age-related spliced isoforms. Analogous to human brain tissues, mouse brain tissues also exhibit several expressional changes in a number of spliceosomal components and SFs. In murine brain early age-related gene expression studies suggested changes in several pre-mRNA processing genes, such as increases in Sf3a2, Ptb, Hprp22, and Hnrnph3, and decrease in Hprp16 (Lee et al., 1999; Meshorer & Soreq, 2002). In mouse cerebral cortex, the levels of several SFs changed in expression, predominantly decreasing with age, including Ptbp1 and Ptbp2, Nova2, and Rbfox2 (Kadota et al., 2020; Stilling et al., 2014; Weyn-Vanhentenryck et al., 2018) (Figure 2).

In Drosophila photoreceptors, another neuronal tissue, genes involved in both RNA processing were found to be both upregulated and downregulated with age. Specifically, two studies showed that over 30 homologs of human genes associated with splicing, including SAFB and SRSF2, were altered, mostly being downregulated with age in photoreceptors of young versus old fruit flies (Hall et al., 2017; Stegeman et al., 2018). Interestingly, many of these SFs were critical for both regulating AS of age-related differentially spliced genes, and for maintaining visual response in older animals, linking the downregulation of SF expression to an aging phenotype. Furthermore, single SF knock-down was not sufficient to recapitulate the age-related vision phenotypes or the age-related splicing profiles of photoreceptors; suggesting that the measured aging phenotypes were controlled by a combination of SFs or other proteins (Stegeman et al., 2018).

2.2.3 |. Age-related changes in epithelial tissues and muscles

Similar to blood and brain, enrichment analysis of differential gene expression in skin highlighted genes involved in AS and RNA processing as significantly increasing with age (Glass et al., 2013). For example, Wang, Wu, et al. (2018) demonstrate that PTBP1 expression decreases with age in human skin, but also that this change in SF expression leads to increased exon inclusion in age-related spliced isoforms which are direct targets of PTBP1. In addition, a proteomic study of human dermal fibroblasts identified age-related expression changes in four proteins involved in RNA splicing (i.e., SRSF9, DDX1, DDX3X, DHX15) (Waldera-Lupa et al., 2014). Moreover, Rodriguez et al. (2016) found altered expression of numerous genes, including SFs, with age in murine skin (mostly downregulated) as well as muscle (both up and downregulated), bone, thymus tissues (Figure 2). Similarly, in aging liver, an association study uncovered 88 RNA-binding proteins significantly altered with age in humans, including SFs SRSF4, TRA2β, SFSWAP, and hnRNPUL1 (Chaturvedi et al., 2015). This study did not report a common trend for SF alterations—some SFs were upregulated with age, others downregulated, whereas some initially increased but then decreased during human, mouse, or rat lifespan (Chaturvedi et al., 2015). Finally, an RNA-seq study of the mouse juvenile versus adult transcriptomes revealed that SFs, Srsf7, Ybx1, Hnrnpa0, Hnrnpa1, Hnrnpdl, Sfpq, and Srsf2, were more highly expressed in young cardiomyocytes, hepatocytes, and cerebral cortex (Kadota et al., 2020). Focusing on Srsf7, the authors demonstrated that this SF is critical for maintaining the juvenile identity of each of these cell types, and its knockdown increased the expression of senescence-associated genes and impaired mitochondrial functions (Kadota et al., 2020). Srsf7 regulates numerous age-dependent AS events in the liver, including isoforms of anabolism-associated genes Eif4a2 and Rbm7 (Kadota et al., 2020).

In contrast to blood and brain, studies in muscle have shown a pattern of upregulated SFs levels. In aged human skeletal muscle, multiple SR and HNRNP proteins, as well as RBFOX1, CELF1, CELF2, and MBLN1, are upregulated at the mRNA level (Welle et al., 2003, 2004); similarly, quantitative proteomic studies from human skeletal muscle reported large changes in SF proteins with age, including increased expression of SF3B1, SRSF1, HNRNPA1, HNRNPAB, and 53 additional spliceosomal proteins (Ubaida-Mohien et al., 2019) (Figure 2). Interestingly, a number of SFs have been previously implicated in the regulation of muscle-specific AS, including PTBP1, as well as RBFOX, CELF, or MBNL family members (Nakka et al., 2018). Furthermore, MBLN and CELF proteins are critical for both normal skeletal development, and mediate AS misregulation in muscular dystrophy (Brinegar et al., 2017; Kalsotra et al., 2008; Lee & Cooper, 2009; Lin et al., 2006). Acute upregulation of CELF1, CELF2, and MBLN1 has also been implicated in the response to toxin-mediated muscle injury (Orengo et al., 2011). Finally, defects in MBLN1 and RBFOX1 in muscular dystrophy result in mis-splicing of a number of AS exons (Klinck et al., 2014). Therefore, the upregulation of many of these transcripts and proteins with age may suggest that aged muscles activate regeneration-related pathways. The activation of these regenerative pathways may be a response to age-related muscle decline, potentially similar to the body’s attempt to compensate, albeit unsuccessfully, for muscle loss in muscular dystrophy patients (Yanay et al., 2020).

2.3 |. SFs associated with lifespan and longevity

Several studies have linked lifespan with changes in SF expression. In Caenorhabditis elegans, a screen for lifespan regulators identified multiple RNA-binding proteins that extended lifespan when inactivated post-developmentally, including the ortholog of human spliceosomal component PRPF38A (Curran & Ruvkun, 2007). Furthermore, knockdown of the C. elegans ortholog of human spliceosomal component SF1 was sufficient to abolish lifespan extension caused by dietary restriction in C. elegans, but did not shorten wild-type lifespan (Heintz et al., 2017). Mechanistically, SF1 modulates components of the TORC1 and AMPK pathways, two pathways associated with metabolism and aging (Heintz et al., 2017). Inhibition of additional orthologs of human spliceosomal genes SF3A2, SNRPD3, SNRNPB, HNRNPR, U2AF35, and U2AF65 significantly reduced lifespan of both wild-type and dietary restricted animals (Heintz et al., 2017). Another pro-longevity gene in worms is the homolog of human transcription elongation and splicing regulator TCERG1 (Ghazi et al., 2009; Sanchez-Hernandez et al., 2016). TCERG1 regulates splicing of BCL2L1, an apoptosis-regulating gene, by modulating the rate of RNA polymerase II transcription and triggers a switch towards the pro-apoptotic isoform (Montes et al., 2012). Finally, across several mouse strains with varying lifespans, expression of Hnrnpa1, Hnrnpa2b1, Hnrnpk, Hnrnpm, Sf3b1, Srsf3, and Tra2β in the spleen, or of Hnrnpa0, Hnrnpd, and Srsf3 in muscle tissues, correlated with lifespan (Lee et al., 2016). These observations in mice are further supported by findings in human and flies, in which HNRNPA2B1 and HNRNPA1 were also associated with parental longevity and general longevity respectively (Lee et al., 2016; Romano et al., 2014).

In summary, though there is much variability between studies, data from the blood, brain, liver, and skin suggest a pattern of SF downregulation with age, whereas SF transcripts are upregulated with age in skeletal muscle. Downregulation of SRSF1, SRSF2, SRSF5, SRSF6, TRA2B, SRRM1, HNRNPA0, HNRNPD, HNRNPH1, HNRNPH3, HNRNPHK, HNRNPHM, PTBP2, and RBFOX2 has been detected in at least five studies (Figure 2), highlighting a potential role in aging that needs to be further investigated. Yet, while several investigators have uncovered age-related changes in SFs, the mechanisms regulating age-related changes in SF levels are largely unknown. We can speculate that, like studies of SF misregulation in cancer, SF expression changes in aging are due to age-related changes in epigenetics, transcriptional changes and regulation, somatic mutations, signaling pathway alterations, and splicing patterns themselves (Dvinge et al., 2016; Urbanski et al., 2018). However, further investigation is necessary. Interestingly, it appears that no single SF is uniformly dysregulated across all aging gene expression studies. Similarly, the longevity studies do not point to a single or small number of conserved SFs that could be “master regulators” altered with age across all tissues and species. Instead, they suggest that aging is associated with deregulation of multiple SFs which in turn impact downstream spliced isoforms. This is likely due to many factors: (i) although most SFs are expressed ubiq-uitously, with the notable exception of most neuronal SFs, they do exhibit tissue-specificity in their targets. There is evidence for tissue specific regulation of gene expression and splicing (Wang, Wu, et al., 2018) and therefore SFs could display distinct functions across aged tissues; (ii) from a technical standpoint, this collection of studies utilized differing techniques to measure gene expression, used differential computational tools, and had different definitions of young and old ages; (iii) variability across studies may result from differential aging rate of tissues (i.e., Is a 60-year old heart as “old” as a 60-year old brain), as well as from cell composition of tissues used (i.e., some tissues are composed of more homogenous cell populations than others); (iv) many SFs themselves are alternatively spliced, often resulting in production of non-coding transcripts, a mechanism used to regulate SF protein levels (Lareau et al., 2007; Lareau & Brenner, 2015; Leclair et al., 2020). Therefore, while the transcript level could be unchanged or changing, this might not fully reflect the final protein levels. Actually, the idea that with increased age, protein expression, for a given gene, correlates less with its transcript expression has been shown in humans and monkeys—and potentially linked to posttranscriptional regulators including SFs, RNA binding proteins and miRNAs (Wei et al., 2015). Therefore, it is critical to further elucidate the molecular causes of these AS changes and their downstream consequences in the aging process. Whether and which of these SF alterations are the causes or consequences of aging remains to be determined. Expanding our understanding of how SF levels are regulated at the molecular level but also more holistically at the organism level will help answer these questions in the future.

3 |. AGE-RELATED ALTERATIONS IN SPLICED ISOFORMS

If the described changes in SFs with age are functionally significant, it might be hypothesized that they would lead to downstream changes in the isoforms spliced with age. Indeed, early studies in mice and rats first documented changes in mRNA processing with increased age that AS of fibronectin changes in aged rat tissues (Yannarell et al., 1977; Pagani et al., 1991). Since then, several transcriptomic studies have comprehensively characterized the AS repertoires in aged human and mouse tissues using RNA-seq or targeted microarrays. The resulting studies in human have supported the initial age-related splicing changes found in mouse and rat. As with the SF expression studies, experiments in humans and mice both support the hypothesis that changes in AS patterns occur with age—though individual AS events may be species-specific (Sieber et al., 2019) or not obviously conserved between species. In the following paragraphs, we will highlight the several studies that have identified age-related AS changes.

3.1 |. Age-related isoforms in blood and hematopoietic lineages

First, an early targeted analysis in peripheral human blood leukocytes identified 10 genes differentially spliced with age (Harries et al., 2011). A more recent longitudinal RNA-seq study of human whole blood from elderly females and males over 10 years uncovered robust AS changes in ~300 genes in 80 years old compared to 70 years old individuals (Balliu et al., 2019). The differentially spliced transcripts were implicated in RNA splicing, apoptosis, leukocyte, or circadian rhythm regulation (Balliu et al., 2019). Moreover, a study of human twins reported age-related AS changes, particularly in fat and skin, and to a lesser extend in blood and lymphoblastoid cell lines (Vinuela et al., 2018).

3.2 |. Age-related isoforms in neuronal tissues and cell types

Human brain tissues from the prefrontal cortex and cerebellum exhibit age-related changes in ~300 AS events, the majority of which are more included with age, as detected by RNA-seq (Mazin et al., 2013). These age-related AS events follow discrete patterns that could be linked to neural functions in distinct brain regions, and ~30% of these AS changes are predicted to affect the protein-coding portion of the transcript (Mazin et al., 2013). Another study of human temporal cortex detected, using a microarray approach, 1174 exons with age-dependent changes in AS, with an enrichment in transcripts related to metabolic processes and DNA repair (Tollervey, Curk, et al., 2011). Interestingly, these age-related AS changes correlated strongly with AS events detected in brains from subjects afflicted with neurodegenerative diseases (Tollervey, Curk, et al., 2011). In rat cerebral cortex, differential splicing patterns were observed in older animals, comparing isoforms in 6-, 12-, and 28 months old animals, and revealed that nonlinear isoform switching might be prevalent in brain aging, as it is in development (Wood et al., 2013). In the mouse hippocampus, age-related AS events occurred in genes related to neuronal function and, interestingly, transcripts that were differentially spliced had little overlap with genes that were differentially expressed (Stilling et al., 2014). Thus, the functional role of AS in aging may extend beyond determining transcript abundance (Balliu et al., 2019; Mazin et al., 2013; Rodriguez et al., 2016; Stilling et al., 2014; Tollervey, Curk, et al., 2011).

In the aging eye, 75 significant differentially spliced events were detected between young and old flies (Stegeman et al., 2018). Furthermore, RNA-seq using nuclear RNA isolated only from photoreceptor cells revealed 238 differentially spliced events, predominantly involved in visual function (Stegeman et al., 2018), therefore illustrating that tissue heterogeneity can mask critical changes in gene expression and splicing profiles in bulk RNA-seq.

3.3 |. Comparing age-related isoforms across tissues and species

Although AS changes with age was described above, comparative studies of samples from multiple tissues and species have not yet uncovered a common splicing signature or set of genes. This is exemplified in an extensive study of 48 distinct human tissues from >500 GTEX donors that found age-related AS events to be prevalent in most tissues but to be largely tissue-specific, with the skin and esophagus-mucosa having the highest number of AS events in old versus young individuals (Wang, Wu, et al., 2018). The age-related spliced transcripts detected across multiple tissues were involved in biological processes linked with aging, including ribosome function, mitochondrial function, DNA repair, DNA damage, and apoptosis. Mechanistically, age-related changes in AS profiles could be explained, at least in part, by concomitant age-related changes in the levels of upstream regulatory SFs. This study suggested that genome-wide AS profiles are a better predictor of biological age than gene or transcript expression profiles (Wang, Wu, et al., 2018). Similarly, in mice, analysis of skin, skeletal muscle, bone, thymus, and white adipose using exon expression microarrays revealed a number of age-related and tissue-specific AS events (Rodriguez et al., 2016). Interestingly, in this study, the number of differential AS events increased with chronological aging, with more differentially spliced events detected at 28 months than at 18 months compared to 3-month-old animals (Rodriguez et al., 2016). Gene ontology, pathway, and network analysis of age-related spliced transcripts across all five tissues highlighted pathways associated with posttranscriptional RNA processing, spliceosome, EIF2 translational control, and cancer (Rodriguez et al., 2016). Finally, a cross-species comparative analysis comparing AS changes in blood, brain, liver, and skin in young versus older humans, mice, and two fish species, revealed that overall less than 5% of transcribed genes are differentially spliced during aging (Sieber et al., 2019). Although most of the age-related differentially spliced transcripts are tissue- and species-specific, several spliced transcripts shared across tissues and species were associated with RNA processing, including splicing and translational control (Sieber et al., 2019).

3.4 |. Age-related intron retention

Increased intron retention with age is a pattern that has been highlighted in several studies and may represent an aging splicing signature, especially in the brain. A global increase in intron retention was detected in aging worms (Heintz et al., 2017), as well as Drosophila heads, where it was proposed to be linked with age-related changes in nucleosome occupancy (Adusumalli et al., 2019). Furthermore, several transcripts with age-related intron retention events detected in mouse and flies (Adusumalli et al., 2019; Lister et al., 2013; Stilling et al., 2014) were also conserved in the aging human brain (Adusumalli et al., 2019; Mazin et al., 2013). Moreover, an analysis of a dermal fibroblast dataset from young and old people highlighted intron retention as a significant age-related splicing event (Yao et al., 2020). However, additional studies argue that this is not the only significant trend in age-related AS; for example, in aging skeletal muscle, skipped exons were the most frequently detected AS event type, many of which were in genes implicated in mitochondrial processes and inflammation (Ubaida-Mohien et al., 2019). Also, when considering AS events in all GTEX tissues, the most prevalent AS event type across most tissues were cassette exons (both upregulated and downregulated), whereas intron retention events showed a bias towards being upregulated (Wang, Wu, et al., 2018).

In summary, taken together these studies indicate that AS patterns change with age. Like gene expression, it might not be surprising if AS events do not overlap between tissue types as a number of AS patterns are tissue specific (Barash et al., 2010; Wang et al., 2008; Xu et al., 2002). However, one might argue that aging does not lead to global AS dysregulation, but rather leads to discrete changes in certain types of AS events or altered AS of transcripts associated with specific pathways. To better understand which of these AS events might play a significant role in aging, further efforts need to be focused on elucidating which AS events result in proteins level changes, functional changes, and how those functional changes contribute to aging phenotypes. Furthermore, studies of the mechanistic link between SF expression and splicing of their direct targets, including profiling of binding sites for common SFs in age-related differentially spliced events, are also needed. Interestingly, a common theme of these aging studies is the differential splicing with age of genes implicated themselves in mRNA processing (Balliu et al., 2019; Rodriguez et al., 2016; Sieber et al., 2019; Wang, Wu, et al., 2018). Wang, Wu, et al. (2018) found mRNA processing and mRNA splicing to be among the top categories of age-related differentially spliced transcripts shared across human tissues. Also, age-related changes in exon usage in murine hippocampus were found to impact splicing of SF transcripts, such as Hnrnph1, HnrnpII, Hnrnpab, Srsf5, Srsf6, and Srsf11 (Deschenes & Chabot, 2017; Stilling et al., 2014). Furthermore, exon expression microarrays across multiple mouse tissues uncovered age-related differential splicing of Sf3b1, Sf3b2, Prpf3, and Prpf8 (Rodriguez et al., 2016). This needs to be further investigated as age-related splicing changes in SFs could change their protein expression and further disrupt AS patterns of their downstream targets.

4 |. HALLMARKS OF AGING AND SPLICING

Despite differences between tissues, aging leads to hallmark changes at the cellular and molecular levels, including genomic instability, epigenetic alterations, mitochondrial and metabolic dysfunction, telomere attrition, accumulation of senescent cells, loss of proteotasis, altered cell communication, and chronic low-grade inflammation termed inflammaging (Franceschi et al., 2018; Lopez-Otin et al., 2013). Recently, researchers have begun to uncover the significance of age-related gene regulation at the mRNA level. Below we review several of the links between splicing and known aging hallmarks (Figure 3 and Table 1).

FIGURE 3.

Age-related splicing alterations. Old versus young tissues and cells exhibit alterations in splicing that can be caused by mutations in, or changes in the levels of, splicing regulatory factors, the latter of which can occur due to copy number changes, or alterations in the epigenetic, transcriptional, posttranscriptional, or posttranslational regulation of SFs in response to signaling changes. These changes in SF levels lead to alterations in alternative splicing (AS) of their downstream targets, promoting events that follow one of the following patterns: cassette exon (CA), alternative 5′ or 3′ splice site selection (A5′SS or A3′SS), inclusion of mutually exclusive exons (MXE), or intron retention (IR). Many of the resulting AS isoforms occur in genes involved in cellular pathways associated with aging hallmarks

TABLE 1.

Isoforms associated with aging hallmarks

Gene namea	Isoform type and nameb	Aging hallmarks	Known SF regulators	References
TP53	Δ40p53 Δ133p53 p53β	Senescence	SRSF3	Fujita et al. (2009), Horikawa et al. (2014), Maier et al. (2004), Tang et al. (2013)
ING1	AFE ING1a/1b	Senescence	Unknown	Soliman et al. (2008), Vieyra et al. (2002)
ENG	ALE Exon 13/17	Senescence	SRSF1	Blanco and Bernabeu (2012)
EXOC7	CA Exon 7	Senescence	SRSF7	Georgilis et al. (2018)
SIRT1	CA Exon 8	Senescence	HUR, TIA, hnRNPA1	Lynch et al. (2010), Wang et al. (2016)
EXOC7	CA Exon 7	Senescence	PTBP1	Georgilis et al. (2018)
LMNA	A5′SS Exon 11	Telomere attrition Senescence	SRSF1, SRSF6	Lopez-Mejia et al. (2011)
hTERT	Multiple	Telomere attrition	NOVA1, PTBP1, SRSF11, hnRNPH2, hnRNPL	Listerman et al. (2013), Ludlow et al. (2018), Sayed et al. (2019)
RPS6KB1	ALE Exon 6a,c/7	Metabolism	SRSF1	Mogilevsky et al. (2018)
MKNK2	ALE Exon 14a/14b	Metabolism	SRSF1	Mogilevsky et al. (2018)
mTOR	RI Intron 5	Metabolism	Sam68	Huot et al. (2012)
IGF-I	A5′SS/ALE Exon 5	Metabolism	Unknown	Hameed et al. (2003)
BCL2L1	A5′SS Exon 2	Genomic instability Inflammaging	SRSF1, HNRNPA1, HNRNPA2/B1, PTBP1, RBM4, RBM5, RBM25, RBM10, PTBP1, Sam68	Shkreta et al. (2011)
NLRP3	CA Exon 5/7	Inflammaging	Unknown	Hoss et al. (2019), Latz and Duewell (2018)
IL-6R	CA Exon 9	Inflammaging	Unknown	Horiuchi et al. (1994), Lamas et al. (2013), Lust et al. (1992)
IL-32	CA IL-32γ	Inflammaging	Unknown	Heinhuis et al. (2011)
IL7R	CA Exon 6	Inflammaging	DDX39B	Galarza-Munoz et al. (2017)
CD45	CA exon 4-5-6	Inflammaging	HNRNPLL	Lynch and Weiss (2001), Motta-Mena et al. (2011), Oberdoerffer et al. (2008), Smith et al. (2013)
ILR15	CA Exon 2	Inflammaging	Unknown	Shakola et al. (2015)
IL-1RAP	ALE Exon 12/12b	Inflammaging	Unknown	Smith et al. (2009), Yoshida et al. (2012)
CADM1	CA Exon 8–9	Inflammaging	Unknown	Shakola et al. (2015)

^aEither human or mouse.

^bAS event types are indicated: cassette exon (CA), alternative 5′SS or 3′SS selection (A5′SS or A3′SS), inclusion of mutually exclusive exons (MXE), intron retention (IR), alternative first (AFE) or last (ALE) exon.

4.1 |. Metabolism, nutrient sensing, and splicing

Deregulated nutrient sensing is one of the key hallmarks of aging. Genetic polymorphisms or mutations that reduce the functions of the IGF-1 receptor, insulin receptor (IR), or downstream intracellular effectors such as AKT, mTOR, and FOXO, have all been linked to longevity in humans and model organisms (Barzilai et al., 2012; Fontana et al., 2010; Kenyon, 2010). Furthermore, caloric restriction can decrease the physiological manifestations of aging and increase lifespan or health span in all investigated eukaryote species (Balasubramanian et al., 2017). We review below examples of changes in spliced isoform or splicing regulators that have been implicated in cellular metabolism and aging (Bhadra et al., 2020).

4.1.1 |. Dietary restriction

During dietary restriction in both worms and mice, hundreds of genes involved in metabolism, RNA-regulatory processes, cellular signaling, and protein processing undergo changes in their AS patterns (Tabrez et al., 2017). In worms, those AS events are regulated at least in part by the ortholog of human HNRNPU, and its knockdown suppresses dietary restriction-mediated longevity (Tabrez et al., 2017). Caloric restriction also leads to expression changes in SF levels; co-expression analysis identified robust changes in gene modules implicated in RNA metabolism and AS regulation in mice, including Cpsf6, Srrn2, Luc7l2, SRm300, Prp44b, Prpf39, Sfpq, Sf3b1, and Sfrs18 (Swindell, 2009). Changes in the expression of SFs Hnrnpa1, Srsf1, Srsf6, Tra2β, or Sf1 were detected in at least one tissue in inbred mouse strains with variable lifespan responses to short-term or long-term caloric restriction (Lee, Mulvey, et al., 2019). Similarly, in monkeys, caloric restriction is linked to changes in numerous spliceosomal components in liver tissues both at the RNA and protein levels, including LSM2, LSM4, SRSF1, SRSF7, NCBP1, PHF5A, PRPF8, ACIN1, RBM8A, and DDX46 (Rhoads et al., 2018). These findings suggest that the AS machinery could be directly impacted by caloric restriction and possibly the enhanced longevity accompanying it.

4.1.2 |. Insulin and IGF-1 signaling pathways

The insulin and insulin-like growth factor (IGF) signaling pathway is critical for metabolic signaling and cellular growth, and has been linked with aging (van Heemst, 2010). Specifically, reduced IGF signaling has been associated with increased lifespan in flies, mice and worms (Altintas et al., 2016). Key players of the insulin/IGF signaling pathway undergo AS in mammals, including IGF-1, the ligand that binds the IGF-1 receptor, as well as the IR that is activated by insulin, IGF-I, or IGF-II (van Heemst, 2010). Splicing of IGF-1 produces at least three AS isoforms: IGF-1 Ea, IGF-1 Eb, or IGF-1 Ec (also known as MGF), which encode different precursor polypeptides with distinct posttranslational modification sites (Deschenes & Chabot, 2017). Interestingly, local expression of either IGF-1Ea or IGF-1Eb transgenes was protective against age-related loss of muscle mass and force in mice (Ascenzi et al., 2019). Furthermore, skipping or inclusion of IR exon 11 results in two isoforms, A or B, that differ in their ligand affinity in mammals (Belfiore et al., 2009). Alterations of the IR-A/IR-B ratio are associated with insulin resistance, aging, and increased proliferative activity of normal and neoplastic tissues and appears to sustain detrimental effects (Belfiore et al., 2017). Several SFs, including CUGBP, MBNL, SR proteins, hnRNPs, and Elav-like family members have been implicated in modulating this IR-A/IR-B balance (Belfiore et al., 2017). Similarly, worms and insects also express multiple IR transcript isoforms, likely generated through a combination of AS and the usage of alternative intronic polyadenylation sites. These short truncated IR isoforms lacks the intracellular tyrosine kinase domain and can act as decoy receptors and soak up substrate (Martinez et al., 2020; Vastermark et al., 2013). In worms, overexpression of these short IR isoforms extends lifespan; however, in this organism, IR isoforms are most physiologically relevant in the entry and exit of the dormant dauer state (Martinez et al., 2020). Continued research will reveal whether these isoforms are important in the context of physiological aging.

4.1.3 |. mTOR and AMPK signaling pathways

The mechanistic target of rapamycin (mTOR) kinase, a negative regulator of aging, is an essential component of two complexes: mTORC1 and mTORC2. mTORC1 signaling is activated in response to available nutrients, growth factors/insulin, ATP, oxygen, and amino acids in order to stimulate anabolic processes (e.g., protein and nucleotide synthesis) and inhibit catabolic process (e.g., autophagy and lysosome biogenesis) (Angarola & Ferguson, 2020; Liu & Sabatini, 2020). Genetic knockdown or pharmacological inhibition of mTORC1 signaling extends lifespan in yeast, worms, flies, and mammals—in part because mTOR inhibition decreases the energetic burden of translation, reduces harmful metabolic by-products, is associated with senescence, and promotes autophagy (Liu & Sabatini, 2020). In recent studies, mTOR has also been implicated in splicing and aging. In worms, dietary restriction leads to AS changes mediated by the ortholog of mammalian core spliceosomal factor SF1 and SF3A2, and SF1 overexpression is sufficient to promotes longevity (Heintz et al., 2017). Furthermore, SF1 knockdown suppresses the lifespan extension phenotype of several mTORC1 and AMPK worm mutants, placing SF1 downstream of mTORC1 signaling, with the exact link between splicing, longevity, and mTOR signaling still being unclear. A recent pair of studies uncovered that, even in yeast, canonical splicing patterns are altered in response to nutrient availability (Morgan et al., 2019; Parenteau et al., 2019). While the two studies did not agree entirely on the molecular mechanism of intron retention or their functional significance, both groups showed that introns accumulate in response to starvation in yeast. This increase in intronic sequences was dependent on mTORC1 and enhanced yeast survival during the stationary phase, likely by downregulating the splicing of ribosomal processing genes (Morgan et al., 2019; Parenteau et al., 2019).

Additional pieces of evidence directly link AS with the mTOR pathway. In human, the SF SRSF1 regulates AS of several targets associated with translational control, including S6K1 and MKNK2 (Anczukow et al., 2012; Karni et al., 2007; Mogilevsky et al., 2018) (Table 1). Moreover, SRSF1 interacts with both the phosphatase PP2A and with the protein kinase mTOR and promotes translation initiation by enhancing phosphorylation of 4E-BP1, a key component of the mTORC1 pathway (Michlewski et al., 2008). Another SR protein family member, SRSF7, impacts mTOR signaling (Kadota et al., 2020), and its protein level in the liver decreases with age in monkeys (Rhoads et al., 2018). In addition, mTORC1-regulated effector kinase, SRPK2, can control the expression of lipogenic enzymes by inducing efficient AS of their mRNAs through the activation of SR proteins and U1 (Lee et al., 2017). Finally, mTOR itself undergoes AS (Table 1), which is regulated by Sam68 (Huot et al., 2012), a SF that protects mice from age-related bone loss (Richard et al., 2005); the relevance of this AS event to aging is still to be determined.

The AMP-activated protein kinase (AMPK) is an energy sensor activated when energy levels are low in the cell and plays a key role in cell metabolic regulation. Opposite to mTOR, activated AMPK promotes catabolic processes and inhibits anabolic processes (Mihaylova & Shaw, 2011). Pharmacological activation of AMPK via metformin may increase lifespan in worms, flies, and mice (Salminen & Kaarniranta, 2012). Metformin treatment results in splicing alterations and the downregulation of SFs, including RBM3, though it is unclear whether this a direct effect of activated AMPK signaling (Laustriat et al., 2015). Also, a recent study identified SRSF1 as a substrate of AMPK (Matsumoto et al., 2020), further linking SFs and AMPK signaling.

Lastly, energy levels are also sensed by the NAD-dependent histone deacetylase sirtuin, SIRT1. SIRT1 deacetylates a variety of histones including at K16 of histone 4, K26 of histone 1, and K9 and K14 of histone 3, and its activation has been implicated in the regulation of aging phenotypes and lifespan extension in several studies (Canto et al., 2009; Canto & Auwerx, 2009; Deschenes & Chabot, 2017). SIRT1 and resveratrol, a plant compound that increases NAD levels and SIRT1 activity, are both being investigated in anti-aging studies, due to SIRT1’s complex roles in metabolism, inflammation, cellular senescence, and genome integrity (Lagouge et al., 2006; Lopez-Otin et al., 2013). Furthermore, stabilization of the SIRT1 mRNA can extend organismal lifespan and delay cellular senescence, and is regulated by SF hnRNPA1 (Wang et al., 2016). In addition, SIRT1 has also been linked with splicing and this link is discussed in more detail in Section 4.6.2.

4.1.4 |. Mitochondria dysfunction and reactive oxygen species

The accumulation of reactive oxygen species (ROS) due to progressive mitochondria dysfunction has been implicated in aging, although it remains to be determined whether it is a causal event or a consequence of the aging process (Back et al., 2012; Blagosklonny, 2008; Liguori et al., 2018). Both mitochondrial damage and/or ROS can lead to changes in AS patterns. For example, cells treated with paraquat, a compound that interferes with mitochondrial function causing energy deficit and oxidative stress, leads to dose- and time-dependent increase of several AS isoforms in apoptotic genes CASP3, CASP8, CASP9, or APAF1, transcription factor SIP1, or SF SMN (Maracchioni et al., 2007). Interestingly, these AS events correlated with ATP depletion, and compounds that generated ROS without directly damaging the mitochondria did not trigger AS changes, suggesting that an energy deficit triggers the signal that alters AS (Maracchioni et al., 2007). It is unclear if ROS can directly activate the splicing machinery, or rather elicits a downstream stress response that leads to changes in SF levels or activity.

In summary, these findings further support a potential, evolutionary conserved link between splicing and the role of dietary restriction and mTORC1 signaling in longevity. In mice, interventions targeting dietary restriction or mTORC1 signaling, such as rapamycin treatment, have been effective in extending lifespan (Harrison et al., 2009; Liu & Sabatini, 2020; Wu et al., 2013), but the role of splicing in these experiments has not yet been investigated comprehensively. However, in humans, the efficacity of dietary restriction or mTORC1 inhibition in modulating health span remains, for now, unclear (Liu & Sabatini, 2020). Future investigations of the links between splicing, longevity, and metabolism are needed to unravel the molecular basis of these differences between species.

4.2 |. Telomere attrition and splicing

Telomeres play a vital role in preserving the information in our genome by protecting it from nucleolytic degradation, recombination, repair, and interchromosomal fusion (Shay & Wright, 2019). With each cell division, as a normal cellular process, a small portion of telomeric DNA is lost. When telomere length reaches a critical limit, the cell undergoes replicative senescence and/or apoptosis. Telomere length may therefore serve as a biological clock to determine the lifespan of a cell and an organism. Accelerated telomere shortening is associated with early onset of many age-associated health problems, including coronary heart disease, diabetes, increased cancer risk, and osteoporosis, as well as with genetic disorders such as dyskeratosis congenita (Shay & Wright, 2019).

Telomerase activity, the ability to extend telomeres, is present in germline and certain hematopoietic cells, whereas somatic cells have low or undetectable levels of this activity and their telomeres undergo a progressive shortening with replication. Nearly all cancer cells up-regulate telomerase to re-elongate or maintain telomeres by de novo synthesis of telomere repeats on to chromosome ends (Shay & Wright, 2019). The reverse transcriptase component of telomerase, hTERT, undergoes AS to generate at least 22 distinct spliced isoforms, which differ in their activity—some having a dominant negative effect on telomerase activity—and in their expression in distinct cell types and during development (Wong et al., 2014). Part of the reactivation of telomerase in cancer cells involves AS of hTERT transcripts to produce full-length TERT, whereas many of the AS isoforms lack the domains encoding the telomerase activity. Splicing of hTERT is regulated by SFs hnRNPK, hnRNPD, SRSF11, hnRNPH2, hnRNPL, NOVA1, and PTBP1 (Kang et al., 2009; Listerman et al., 2013; Ludlow et al., 2018; Pont et al., 2012; Sayed et al., 2019). NOVA1 and PTBP1 act as enhancers of full-length hTERT splicing in neuronal cells, increase telomerase activity, and promote telomere maintenance in cancer cells (Ludlow et al., 2018; Sayed et al., 2019). Knockdown of these SF levels alters hTERT splicing, and therefore changes in SF levels, as detected during aging in several human and mouse tissues, could lead to changes in telomerase activity. Telomere length and telomerase activity have been linked with human healthy lifespan in centenarian studies (Tedone et al., 2019), yet hTERT AS patterns during aging have not yet been investigated.

Finally, telomere damage has been linked with other AS events. For example, progressive telomere damage in human fibroblasts leads to extensive changes in AS events in genes associated with cytoskeleton regulation, cell proliferation, or metabolism (Cao et al., 2011). Also, this progressive telomere damage leads to the production of progerin, a mutant lamin A protein generated through AS of the LMNA gene, and which has been implicated in premature aging in Hutchinson-Gilford progeria syndrome (HGPS) (Cao et al., 2011), and is further discussed in Section 5.1.

4.3 |. Senescence and splicing

Cellular senescence occurs when cells irreversibly arrest the cell cycle. Unlike cellular quiescence, cellular senescence is complex cellular state that involves cell morphological reconstruction, proteomic alterations, apoptotic resistance, and senescence-associated secretory phenotype (SASP) production (Munoz-Espin & Serrano, 2014). Senescence can be induced once a cell has reached its replicative potential or due to the application of a stressor (i.e., DNA damage, oxidative stress, and oncogenes). Throughout the lifespan of an organism, senescent cells accumulate in multiple tissues and organisms progressively lose the ability to clear them out (Jeyapalan et al., 2007). Thus, this accumulation of senescent cells has been deemed a critical hallmark of aging (Sikora et al., 2014). While senescence can be beneficial in deterring abnormal cell growth/cancer, the accumulation of senescent cells, and their pro-inflammatory effects, contributes to aging and age-related disease (van Deursen, 2014). Senescence induces an inflammatory SASP secretome that promotes carcinogenesis and age-related pathologies.

4.3.1 |. Senescence and spliced isoforms

Changes in AS have been detected during senescence (Deschenes & Chabot, 2017), and several spliced isoforms, such as TP53, ING1, and ENG, have been linked both with aging and senescence over the years (Table 1). The tumor suppressor protein p53, encoded by the TP53 gene, plays a critical role in regulating targets in several critical pathways including cell cycle arrest, DNA repair, apoptosis, and cellular senescence (Mijit et al., 2020). In response to DNA damage, the transcription factor p53 is phosphorylated, abrogating its repressive interaction with MDM2 and allowing p53 to promote expression of pro-senescence targets (Mijit et al., 2020). Splicing of TP53 generates multiple isoforms, including Δ40p53 (also called p44) a short isoform of p53 (p47 is the human ortholog), which is upregulated in aged mouse brains (Pehar et al., 2014) (Figure 4 and Table 1). Δ40p53 lacks the first transactivation domain of p53 but retains the second one due to retention of the i2 intron and use of the initiation codon in exon 4 (Ghosh et al., 2004). Its overexpression in mice causes a progeroid phenotype that resembles an accelerated form of aging, a suppression of cell proliferation, and increase in senescence (Maier et al., 2004). Δ40p53 leads to the onset of senescence by hyperactivating IGF signaling in mice (Maier et al., 2004). Typically, p53 decreases the expression of the IGF-1 receptor (Werner et al., 1996). However, overexpression of Δ40p53 in mice results in increases levels of the IGF-1 receptor and the downstream RAS-RAF-MEK-ERK pathway, resulting in induction of p21 and cell-cycle arrest (Deschenes & Chabot, 2017; Maier et al., 2004; Pehar et al., 2014). Similarly, two other TP53 isoforms, Δ133p53, which lacks both transactivations domains, and p53β, which lacks the oligomerization domain and the C-terminal basic region, regulate replicative senescence, but not oncogene-induced senescence, in normal human fibroblasts (Fujita et al., 2009; Horikawa et al., 2014; Tang et al., 2013). AS of p53β is regulated by the downregulation of SRSF3, an SF that triggers senescence when knocked down in fibroblasts (Tang et al., 2013).

FIGURE 4.

Splicing events associated with senescence. Senescence hallmarks include senescence-associated heterochromatin foci (SAHF), loss of nuclear laminin, increased reactive oxygen species (ROS), expression of senescence associated β-galactosidase (B-Gal), and senescence-associated secretory phenotype (SASP). Changes in splicing in TP53, EXOC7, ING1, or ENG gene have been reported to impact these senescence hallmarks. Schematic structures of AS isoforms, along with known SF regulators are shown. For each p53 isoform, the regions alternatively spliced are highlighted, and the protein domain encoded by each exon is indicated (DB, DNA binding domain; OD, oligomerization domain; TD, transactivation domain)

Another example of AS detected during senescence is the tumor suppressor and TP53 interacting partner, ING1, which can be spliced into two isoforms, INGa or INGb; increased ratio of the INGa to INGb isoform promotes senescence (Soliman et al., 2008; Vieyra et al., 2002) (Figure 4 and Table 1). Furthermore, AS of ENG to promote intron retention, a gene that encodes a transmembrane glycoprotein of the vascular endothelium, is altered in senescent endothelial cells (Blanco et al., 2008), and is regulated by SF SRSF1 (Blanco & Bernabeu, 2012) (Figure 4 and Table 1).

4.3.2 |. Senescence and SFs

Changes in SF levels have been detected during senescence. Replicative senescence in endothelial cells and fibroblasts is associated with changes in the expression of SF transcripts and results in downstream splicing changes (Holly et al., 2013). Altered senescence-induced expression patterns of key SF families suggested a loss of AS regulation complexity with senescence (Holly et al., 2013). Moreover, numerous RNA-binding proteins are downregulated upon senescence induction, and multiple differentially spliced events were shared between distinct pathways that induce senescence in human cells (Dong et al., 2018). Interestingly, many of the AS events occurred in genes associated with RNA splicing or the spliceosome itself. Combining differential gene expression data and binding motif data surrounding the differentially spliced genes, SFs SRSF1, SRSF7, QKI, RBFOX2, PTBP1, HNRNPK, HNRNPM, and HNRNPUL1, were predicted as key regulators of senescence-associated AS events (Dong et al., 2018). Similarly, senescent neuroblastoma cells were shown to suppress expression of SRSF7 (Kadota et al., 2020). Moreover, a recent study uncovered that, in vitro, decreased expression of U2AF1 levels during replicative senescence triggers intron retention in hundreds of transcripts, which in turn downregulates their gene expression (Yao et al., 2020). Interestingly, knockdown of one of the U2AF1 target genes, CNPE1, impacts senescence associated phenotypes, thus suggesting intron retention might directly contribute to senescence (Yao et al., 2020). Finally, several SFs, including PTBP1, have been shown to play a role in SASP production, for example by regulating AS of genes involved in intracellular trafficking, such as EXOC7 (Georgilis et al., 2018). While this example relates to oncogene-induced senescence (and not replicative senescence), changes in PTBP1 detected in aging tissues are evident (Georgilis et al., 2018) (Figure 4). Oncogenic SF SRSF1 has been associated with oncogene-induced senescence (Fregoso et al., 2013).

Moreover, several studies suggest that global dysregulation of regulatory SFs and spliceosomal components could trigger senescence. For example, a study in primary human fibroblasts using replicative and oxidative stress-induced senescence models detected changes in the expression of 58 genes associated with splicing regulation at the pre-senescence stage, prior to the detection of senescence-associated β-galactosidase activity (Kwon et al., 2021). Furthermore, homologs of resveratrol, a compound that can reverse senescent phenotypes and extend lifespans of model organisms, lead to increased SF levels (Latorre et al., 2017). In addition, inhibition of AKT and ERK signaling, two pathways implicated in senescence and lifespan, also increases SF expression indirectly, at least in part through the FOXO1 and ETV6 transcription factors (Latorre et al., 2019). Lastly, in contrast to human and mouse models described above, the naked mole rat, a unique aging model organism, shows little change in SF-levels with age (Lee et al., 2020). The naked mole rat is long-lived relative to its size, is resistant to both spontaneous and experimentally induced cancer, and exhibits an ameliorated stress response and decreased levels of senescence; therefore, the stable expression of SFs across ages might provide a link between AS and healthy aging, and could contribute to the exceptional lifespan and pro-longed health span of these animals (Lee et al., 2020). Therefore, modulating SF levels, either pharmacologically or genetically, could provide novel opportunities to delay or prevent senescence.

4.4 |. Genomic instability and splicing

Aging is associated with the accumulation of DNA mutations and increased genomic instability (Niedernhofer et al., 2018; Petr et al., 2020). Several lines of evidence correlate nuclear DNA damage with aging and suggest a causal link between DNA repair defects and accelerated aging in humans and mice (Niedernhofer et al., 2018). Interestingly, many genes implicated in the control of DNA repair are themselves alternatively spliced, including BRCA1, ATM, CHK2, or TP53 (Berge et al., 2010; Joruiz & Bourdon, 2016; Li, Harlan-Williams, Kumaraswamy, & Jensen, 2019). Reciprocally, genotoxic insults, including activation of classical DNA damage response kinases such as ATM, ATR, and DNA-PK, can alter gene expression responses, therefore modulating AS which, in turn, shapes the response to the damaging agent (Botto et al., 2020). Several SFs, including SR and HNRNP proteins, exhibit changes in their levels, AS profiles, phosphorylation or acetylation states, or subcellular distribution in response to DNA damage (Adamson et al., 2012; Busa et al., 2010; Ip et al., 2011; Leva et al., 2012; Magni et al., 2019; Matsuoka et al., 2007; Sakashita & Endo, 2010; Siam et al., 2019). The activation of the DNA damage response leads to changes in AS profiles in genes encoding components involved in DNA repair, cell-cycle control, and apoptosis (Dutertre et al., 2010; Paronetto et al., 2011; Paronetto et al., 2014). An example is the splicing of BCL2L1, a member of the Bcl-2 family, which generates two isoforms, BCL-xL and BCL-xS, that have opposing functions in apoptosis; the first prevents apoptosis whereas the latter promotes it (Boise et al., 1993). Splicing of the pro-apoptotic Bcl-xS isoform is regulated by SRSF10 in a manner dependent on the ATM and CHK2 DNA damage response pathway (Shkreta et al., 2011) (Table 1). Below we further discuss several key players in the DNA damage response pathway that have been linked both with splicing and aging, including BRCA1, ATM, PRMT5, and PARP1.

4.4.1 |. BRCA1

BRCA1 is a protein that maintains genomic stability, controls cell-cycle checkpoint activation, directs homologous recombination-mediated DNA double-strand break repair, and regulates transcription (Wu et al., 2010). BRCA1 germline mutations are associated with an increase in breast cancer risk in early adulthood until ages 30–40 years (Kuchenbaecker et al., 2017). Female BRCA carrier also have lower ovarian reserve, often experience premature menopause and increased ovarian aging (Ben-Aharon et al., 2018; Finch et al., 2013; Oktay et al., 2010; Rzepka-Gorska et al., 2006). Mice with homozygous Brca1 deletions display higher levels of cellular senescence and apoptosis and are embryonically lethal (Sharan et al., 1997; Xu et al., 2001); additionally Brca1^+/− p53^+/− mice display multiple premature aging phenotypes, such as osteoporosis, atrophy, kyphosis, decreased body weight, and increased tumor incidence (Cao et al., 2003). BRCA1 broadly functions as a scaffolding protein, facilitating the assembly of multiple and distinct multiprotein complexes, with various functions within the DNA damage response. However, following DNA damage, BRCA1 can also form a complex with mRNA splicing proteins via BCLAF1 (Savage et al., 2014). BCLAF1 promotes resistance to DNA damage and is required for efficient DNA repair and maintenance of genomic stability. The BRCA1/BCLAF1 interaction mediates the formation of a BRCA1-mRNA splicing complex, which drives the splicing of a subset of genes following DNA damage, including ATRIP, BACH1, and EXO1 (Savage et al., 2014).

4.4.2 |. ATM

Another factor involved in resolving double-stranded DNA breaks is ATM, a signaling kinase that acts upstream of p53. ATM activation is also a marker of cellular aging and a precursor to replicative senescence in primary human cells (Bakkenist et al., 2004). ATM and ATM-related DNA damage response proteins levels progressively decline with replicative senescence in human primary cells and with age in murine brain tissues (Qian et al., 2018). Treatment with chloroquine, a compound that increases ATM activity, delays aging and extends lifespan in C. elegans, as well as in a short-lived mouse model of progeria (Qian et al., 2018). Interestingly, the core spliceosome can be both a target and an effector of non-canonical ATM signaling (Tresini et al., 2015). Transcription-blocking DNA lesions promote chromatin displacement of late-stage spliceosomes and initiate a dependent positive feedback loop via ATM. Spliceosome displacement leads to R-loop formation triggered by pausing of RNA polymerase. In turn, R-loops activate ATM, which signals to impede spliceosome organization and augment ultraviolet irradiation-triggered AS at the genome-wide level (Tresini et al., 2015). The DNA-activated kinases ATM and ATR phosphorylate hundreds of proteins in response to ionizing radiation, including several hnRNP and SR proteins (Matsuoka et al., 2007). In addition, ATM knockdown regulates the levels of several SFs, including SRSF1, SRSF2, SRSF3, SRSF7, TRA2β, HNRNPD, HNRNPA1, and LSM14A (Holly et al., 2013).

4.4.3 |. PRMT5

Another protein linking double-strand break repair, aging, and splicing is the arginine N-methyltransferase 5 (PRMT5). This enzyme catalyzes the symmetrical arginine dimethylation of histones and non-histone substrates, including essential components of the spliceosome SmB, SmD1, and SmD3. PRMT5 depletion or inhibition induces DNA damage and genomic instability in a variety of tissues (Clarke et al., 2017; Hamard et al., 2018). Arginine methylation is required to maintain cells in a proliferation state and plays a key role in the maintenance of stem cells, cellular senescence, and premature aging (Blanc & Richard, 2017). The loss of PRMT5 leads to cellular senescence in glioblastoma and osteosarcoma models (Banasavadi-Siddegowda et al., 2017; Li et al., 2020). The loss of arginine methylation also leads to the depletion and exhaustion of stem cells in adulthood, and PRMT5 is required for normal adult hematopoiesis (Liu et al., 2015). Moreover, PRMT5 levels are regulated in an age-dependent manner, with a decline in testis, thymus, kidney, lung, and heart from 24-month-old as compared to 6-month-old rats (Hong et al., 2012). Conversely, PRMT5 levels are upregulated in cartilage from older patients with osteoarthritis, a whole-joint disease which occurs in 50% of the population ≥65 years of age (Dong et al., 2020). Mechanistically, PRMT5 depletion leads to aberrant splicing of the multifunctional histone-modifying and DNA-repair factor TIP60/KAT5, which in turn impairs homologous recombination (Hamard et al., 2018). This deficiency in homologous recombination creates a vulnerability in cells, as they increasingly rely on poly(ADP-ribose) polymerase (PARP) enzymes to repair their DNA; and PRMT5 and PARP inhibitors have synergistic effects on cancer cells (Hamard et al., 2018).

4.4.4 |. PARP1

Finally, PARPs are a family of DNA repair factors involved in the response to single-stranded DNA breaks and associated with tumor biology, oxidative stress, inflammatory, metabolic diseases, and mammalian longevity (Bai, 2015). First, poly(ADP-ribosyl)ation capacity in peripheral blood mononuclear cells of 13 mammalian species strongly correlates with their maximum lifespan, and declines with age in both humans and rodents (Grube & Burkle, 1992). Moreover, Parp1^−/− mice and cells are hypersensitive to DNA-damaging agents and display increased spontaneous genomic instability, and Parp1^−/− animals age moderately faster compared to wild-type control (Piskunova et al., 2008). Finally, treatment of human endothelial progenitor cells with a small molecule inhibitor of PARP1 leads to increased SIRT1 activity, decreased p53 acetylation, and attenuated production of stress-induced senescent cells (Zha et al., 2018). In addition to its role in aging, PARP1 has also been linked with splicing. First, PARP-1 binds to RNA, SFs, and chromatin, and might serve as a gene regulatory hub to facilitate co-transcriptional splicing, possibly by bridging chromatin to RNA and recruiting SFs (Krietsch et al., 2012; Matveeva et al., 2016; Melikishvili et al., 2017). Second, it can also influence AS decisions through the regulation of RNA polymerase II elongation (Matveeva et al., 2019). Third, PARP-1 binds to SF3B1, a member of the U2 spliceosomal complex, and both are found together at nucleosomes (Matveeva et al., 2016). Knockdown of PARP-1 decreases SF3B1 association with nucleosomes, implicating PARP1 in stabilizing the SF3B1-nucleosome complex. Depletion of PARP-1 or inhibition of its PARylation activity also results in changes in AS of a specific subset of genes, including genes involved in RNA and protein processing (Matveeva et al., 2016). Moreover, upon activation, PARP-1 generates poly(ADP-ribose) (PAR), which facilitates the recruitment of DNA repair factors. Interestingly, several SFs have been identified as PAR-binding proteins, including NONO, hnRNPA1, RBMX, and SRSF1, and are recruited at sites of DNA damage (Krietsch et al., 2012; Malanga et al., 2008). Upon PAR-dependent recruitment, SF NONO stimulates nonhomologous end joining and represses homologous recombination (Krietsch et al., 2012). Finally, SFs are themselves PARP-1 substrates and PARylation regulates the assembly-disassembly dynamics of RNP granules containing disease-related RNA-binding proteins, including hnRNPA1 and TDP-43 (Duan et al., 2019). hnRNPA1 can both be PARylated and bind to PARylated proteins or PAR. PARylation of hnRNPA1 controls its nucleocytoplasmic transport, whereas PAR-binding regulates its association with stress granules and phase separation (Duan et al., 2019). Given the links between DNA damage and aging, as well as between SFs and aging, one could imagine that age-related changes in SF levels could perturb the ability of PARP-1 to properly function and contribute to aging phenotypes.

4.5 |. Inflammation and splicing

Aging is accompanied by a decrease in adaptive immunity and the development of inflammaging, which contributes to the pathogenesis of age-related diseases (Franceschi et al., 2018; Ramos-Casals et al., 2003). This state of mild chronic inflammation, revealed by elevated levels of inflammatory biomarkers such as C-reactive protein and interleukin-6 (IL-6), is associated with and predictive of many aging phenotypes—changes in body composition, energy production and utilization, metabolic homeostasis, immune senescence, and neuronal health. Inflammation also leads to increased levels of ROS which can induce oxidative stress, a key component in chronic inflammation. Aged cells display increased levels of oxidant-damaged DNA, and the resulting oxidative stress leads to activation of pro-inflammatory pathways (Liguori et al., 2018).

4.5.1 |. Immune response and spliced isoforms

Although the effects of inflammaging on posttranscriptional gene regulation are not fully characterized yet, AS of immune signaling molecules is found altered in pathological conditions and in response to acute or chronic inflammation across multiple immune cell types (Martinez et al., 2012; Martinez & Lynch, 2013). Immune-activated AS events impact genes with known functions in immune cell biology, including cytokines and cell surface receptors, kinases, phosphatases, and adapter proteins (e.g., CD3, CD28, CD8, CTLA-4, MAP4K2, MAP3K7, MAP2K7, CD45, VAV1), transcription factors controlling inflammatory and antiviral responses (e.g., NFKB1 or STAT2), chromatin modifying enzymes (e.g., LEF1, ATF2, GATA3, RUNX1, EHMT2), or RNA binding proteins (Martinez et al., 2012; Martinez & Lynch, 2013; Rotival et al., 2019). In human monocytes, immune activation triggers a remodeling of the isoform repertoire, half of which correspond to switches between two functional protein-coding isoforms, suggesting AS-driven changes in downstream signaling potential (Rotival et al., 2019).

Several spliced molecules are of particular interest in the context of the immune response during aging. The inflammasome sensor NLRP3 is involved in the recognition of triggers that appear during physiological aging and during metabolic stress (Latz & Duewell, 2018). Human NLRP3, but not mouse Nlrp3, is regulated at the AS level to generate two major isoforms, the full-length variant and a variant lacking exon 5 that is unable to interact with NEK7 and hence loses its activity (Hoss et al., 2019) (Table 1). IL-6, an inflammatory marker elevated with aging, alters the splicing patterns of BCL2L1 promoting the pro-apoptotic BCL-xS isoform (Li et al., 2004). Both IL-6 and its receptor, IL-6R, are also regulated by AS. The soluble forms of IL-6R (sIL-6R) can be generated either by limited proteolysis or by AS of the receptor (Horiuchi et al., 1994; Lamas et al., 2013; Lust et al., 1992) (Table 1).

4.5.2 |. Chronic inflammation and spliced isoforms

Chronic inflammation and several AS isoforms with pro-inflammatory functions have been implicated in neurodegenerative and age-related diseases, including Parkinson’s disease, multiple sclerosis, and Alzheimer’s. An example in rheumatoid arthritis is IL-32, a proinflammatory cytokine that exists in six AS isoforms which differ in their potency to control inflammation, with IL-32γ being the most and IL-32β the least active isoform (Heinhuis et al., 2011) (Table 1).

Another chronic inflammatory autoimmune disorder, multiple sclerosis, has been associated with AS variants in the interleukin-7 receptor ILR7 that enhances skipping of exon 6 (Evsyukova et al., 2010; Gregory et al., 2007), increasing AS of the secreted form of the receptor sIL7R and its levels in plasma (Hoe et al., 2010; Lundstrom et al., 2013) (Table 1). ILR7 splicing is regulated by DDX39B, a SF which itself is also a risk allele for multiple sclerosis, and shows significant epistasis with the ILR7 risk variant (Galarza-Munoz et al., 2017). Elevated levels of sIL7R exacerbate disease severity in an experimental mouse model of multiple sclerosis, presumably by enhancing the activity or bioavailability of IL-7 (Lundstrom et al., 2013). Moreover, IL7R gene expression and its interacting partners have been associated with human healthy aging and familial longevity in the Leiden Longevity Study (Passtoors et al., 2012; Passtoors et al., 2015). First, relative to controls, “healthy agers,” that is, nonagenarian siblings and their offspring, exhibit reduced IL7R expression in blood. Second, higher IL7R gene expression levels, in both the nonagenarians and middle-aged individuals, associated with better prospective survival. These findings suggest that while gene expression levels of IL7R decrease with chronological age, higher IL7R expression levels could indicate a more “youthful profile” and thus better health (Passtoors et al., 2015). Further evidence that IL7 signaling may contribute to biological aging and longevity is that it is closely connected to mTOR signaling. For example, IL7 induces phosphorylation of the mTORC1 targets S6 and 4EBP1 (Brown et al., 2007), whereas mTORC2 suppresses IL7R gene expression by regulating FoxO1 phosphorylation (Lazorchak et al., 2010).

Finally, susceptibility to multiple sclerosis is also associated with the presence of a single nucleotide polymorphism that increases AS of exon 4 in CD45 at least in a subset of patients (Jacobsen et al., 2000). CD45 is a hematopoietic receptor-like tyrosine phosphatase that sets the threshold for signal transduction from immunoreceptor tyrosine-based activation motif-containing receptors, such as the T cell receptor. AS of exons 4, 5, and 6 produces distinct protein isoforms, which differ in their signaling potential (Shakola et al., 2015), and is regulated in a cell lineage- and stage-specific fashion (Ergun et al., 2013) by several SFs, including hnRNP proteins (Lynch & Weiss, 2001; Motta-Mena et al., 2011; Oberdoerffer et al., 2008; Smith et al., 2013) (Table 1).

4.5.3 |. Other spliced isoforms associated with inflammation

Other inflammation-related events include AS of: (i) interleukin-15 receptor, ILR15, a modulator of interleukin-15 signaling, and whose main isoform is localized in the nuclear membrane (Shakola et al., 2015), whereas deletion of exon 2 leads to a retention of the receptor in the endoplasmic reticulum, Golgi, and cytoplasmic vesicles (Dubois et al., 1999) (Table 1); (ii) interleukin-1 receptor accessory protein, IL-1RAP, an essential component of receptor complexes mediating immune responses to interleukin-1 family cytokines. IL-1RAP exists in two isoforms in the brain, IL-1RAcP and IL-1RAcPb, differing in the C-terminal region, and which play a role in the interplay between inflammation and neuronal survival (Smith et al., 2009; Yoshida et al., 2012) (Table 1); (iii) the cell adhesion molecule-1, CADM1, a member of the Ig superfamily, is spliced into four membrane-spanning isoforms that differ in the region upstream of the transmembrane domain, or a soluble isoform (Shakola et al., 2015) (Table 1). CADM1d, as the predominantly expressed isoform in mature cerebrum, combined with its role in enhancing nerve-mast cell interaction, could lead to exacerbated neurogenic inflammation (Shakola et al., 2015). Interestingly the Cadm1 gene is differentially methylated in old versus young hematopoietic stem cells (Sun et al., 2014).

4.6 |. Epigenetics and splicing

DNA methylation involves the addition of a methyl group on position 5 of cytosine by DNA methyltransferases to form 5-methylcytosines (Li & Zhang, 2014). In mammals, the major site for methylation is CpG dinucleotides; and CpG-rich regions, called CpG islands, frequently occur near promoters (Jones & Liang, 2009). Hypermethylated CpG islands in promoters are associated with gene silencing, whereas hypermethylated gene bodies increase gene expression (Li & Zhang, 2014). Strikingly, DNA methylation patterns change with age and age-related diseases (Ashapkin et al., 2017; Johnson et al., 2012; Jones et al., 2015). Aged tissues and cells exhibit increased genome-wide hypomethylation at non-CpG islands and promoter-specific hypermethylation (Fraga & Esteller, 2007; Maegawa et al., 2010). The Horvath epigenetic clock, an algorithm that measures age using DNA, utilizes the methylation status of 353 CpGs to predict the chronological age of humans across multiple tissues (Horvath, 2013). Interestingly, the CpGs used in the Horvath epigenetic clock are frequently related to splice sites, suggesting a link between epigenetics and splicing regulation during aging (Malousi et al., 2018). While it is unclear if cause or consequence, CpGs associated with a single mRNA transcript are more methylated than sites associate with many transcripts (Malousi et al., 2018). Further investigation is required to understand the connection between age-related changes in methylation and downstream changes in AS.

4.6.1 |. Methylation and splicing

Several studies support a strong link between splicing and methylation. First, reducing DNA methylation by either knocking down DNA methyltransferase proteins or treating with the DNA demethylation agent 5′-aza-2′-deoxycytidine alters AS patterns in honey bees, human cell lines, and mouse embryonic cells (Li-Byarlay et al., 2013; Maunakea et al., 2013; Yearim et al., 2015). Second, in multiple species and tissues, exonic sequences exhibit higher levels of methylation than surrounding intronic sequences (Chodavarapu et al., 2010; Gelfman et al., 2013; Khare et al., 2012; Lyko et al., 2010). Third, AS exons have lower levels of methylation than constitutively spliced exons (Choi et al., 2010; Gelfman et al., 2013); yet AS exons that are included tend to exhibit higher levels of methylation than those that are skipped (Flores et al., 2012; Maunakea et al., 2013; Song et al., 2017). However, the link between methylation level and exon inclusion is not linear, and methylation has been shown to promote both inclusion as well as skipping of AS exons (Wan et al., 2013; Yearim et al., 2015). The functional significance of methylation at constitutive exons has been questioned; indeed triple-knockout of Dnmt1/3a/3b in mouse embryonic stem cells had relatively little impact on splicing of constitutive exons, but a profound effect on AS exons (Yearim et al., 2015) Similarly, targeted demethylation using CRISPR-guided DNMT3A, a DNA methyltransferase, or TET1, a component of the demethylation system, alters AS of reporter minigenes in a position dependent manner, with exon-body methylation having a greater effect than intron methylation (Shayevitch et al., 2018). Therefore, methylation at exons might serve as a “fine-tuning” splicing element, and strong splice site consensus sequences at constitutive exons might be less susceptible to methylation induced splicing modulation (Yearim et al., 2015).

Though the underlying mechanisms of methylation-mediated splicing regulation are still debated, there are two working models of how DNA methylation might impact splicing: (i) methylation might recruit SFs to splice sites; or (ii) methylation may alter the kinetics of transcription, and due to the co-transcriptional nature of splicing this could in turn affect AS (Iannone & Valcarcel, 2013). In support of the first model, heterochromatin protein 1 (HP1), a family of proteins which interacts with H3K9me3 histone marks via their chromodomains, can also recruit SFs, including SRSF3 and several hnRNP proteins, to sites of methylation in mouse ES cells (Yearim et al., 2015). HP1 impacts AS of a subset of exons, silencing or enhancing exon recognition in a position-dependent manner (Yearim et al., 2015). In support of the second hypothesis, in the absence of DNA methylation, binding of the DNA-binding factor CTCF to CD45 exon 5 slows the elongation rate of RNA polymerase II, thus altering the kinetics of transcription, and promotes exon inclusion in human lymphocytes (Shukla et al., 2011). Conversely, methylation blocks CTCF binding and promotes skipping of the CD45 alternative exon 5. Similarly, depletion of the methyl-CpG-binding protein MeCP2 leads to AS misregulation in human fibroblasts and colorectal cancer cells (Maunakea et al., 2013). MeCP2 might AS by recruiting histone deacetylases to histones and change RNA polymerase II kinetics, thus promoting exon inclusion (Maunakea et al., 2013). MeCP2 was also found to regulate AS patterns in mouse neurons, and to interact with transcriptional regulators, SFs, and chromatin-related modifiers (Cheng et al., 2017). MeCP2 could therefore additionally impact AS by acting as a scaffolding protein that bridges splicing and chromatin regulators (Cheng et al., 2017).

4.6.2 |. Histone modifications and splicing

Histone modifications, both acetylation and methylation, are other epigenetic marks that change with age and are often associated with specific transcriptional states (Luco et al., 2010; McCauley & Dang, 2014; Michalak et al., 2019). With age, chromatin can take on a more “open” conformation due to increased histone acetylation and reduced histone levels at specific targets (McCauley & Dang, 2014). For example, age-related decline in expression of the histone deacetylase SIRT1 has been observed in yeast, mouse, and rat, and putatively linked to age-associated increases in H4K16 acetylation and altered transcriptional silencing (Braidy et al., 2011; Dang et al., 2009; Gong et al., 2014). In addition, inducible knockout of SIRT1 in neural stem cells resulted in 440 differentially expressed exons, with many of these AS changes occurring in genes related to cell cycle and DNA damage repair (Wang, Wang, et al., 2018). Further linking histone levels and splicing, knockdown of the stem-loop binding protein SLBP decreases the levels of H2B and H3 histones, increases transcription elongation rate, and alters AS patterns (Jimeno-Gonzalez et al., 2015). Moreover, increased levels of histone acetylation H3K9 in gene bodies lead to exon skipping in depolarizing neurons (Schor et al., 2009).

Furthermore, the enzymes that regulate the deposition of histone modifications, histone deacetylases, are themselves alternatively spliced. For example, SIRT1, the enzyme that regulates primary acetylated lysine 16 of histone 4 (H4K16ac), undergoes AS to generate an isoform lacking exon 8. This shorter isoform retains minimal deacetylase activity and exhibits distinct stress sensitivity, RNA/protein stability, and protein-protein interactions compared to full-length SIRT1 (Lynch et al., 2010). AS of SIRT1-ΔExon8 is regulated by p53, and can in return repress p53 expression (Lynch et al., 2010). This stress-sensitive splicing of SIRT1 might be influenced by a stress-induced downregulation of SRSF2 levels (Lynch et al., 2010). However, TIA1/TIAL1 and HuR have also been implicated in promoting exon 8 inclusion and skipping respectively (Deschenes & Chabot, 2017). Another isoform of SIRT1, lacking exons 2–9, is regulated at the splicing level in a p53-dependent manner, as well as at the translational level by CELF2 (Shah et al., 2012). This SIRT1-ΔExon2–9 isoform regulates p53 protein levels in non-stress conditions (Shah et al., 2012). The described SIRT1 spliced isoforms have diminished deacetylase activity, therefore suggesting that SIRT1 might exhibit a deacetylase-independent function, and its role in aging should be further investigated.

Finally, methylated histones located in exons near intronic regions are hypothesized to play a role in pre-mRNA splicing (Kolasinska-Zwierz et al., 2009; Wilhelm et al., 2011). Internal exons exhibit strong nucleosome-positioning signal, and these nucleosomes are enriched in trimethylated histone 3 lysine 36 (H3K36me3) marks (Andersson et al., 2009; Huff et al., 2010). In addition, proteins that bind the H3K36me3 mark can recruit SFs to the pre-mRNA, such as the chromatin-associated protein MRG15 which also binds PTB (Luco et al., 2010), or the short isoform of the chromatin-associated protein Psip1 (p52) which interacts with SRSF1 (Pradeepa et al., 2012), and can lead to downstream AS changes in target pre-mRNAs. Further linking histones, splicing and aging, the H3K36me3 mark plays a role in longevity in C. elegans, in which inactivation of methyltransferase met-1 decreased global H3K36me3 marks and shortened lifespan (Pu et al., 2015). The link between splicing and epigenetic regulation may be bidirectional, with histone methylation impacting the recruitment of SFs as described above, but also with AS patterns affecting the positioning of modifications of histones themselves (de Almeida et al., 2011; Kim et al., 2011).

5 |. SPLICING AND AGE-RELATED DISEASES

Alterations in AS have been causatively implicated in a number of human diseases, several of which display premature aging phenotypes or increase in incidence with age (Table 2).

TABLE 2.

Isoforms associated with age-related diseases

Gene name	Isoforma	Disease	Known regulators	References
BACE	A5′/3′SS Exon 3–4	Alzheimer′s	Unknown	Fisette et al. (2012), Mowrer and Wolfe (2008), Stilling et al. (2014), Zohar et al. (2005)
PFKP	IR multiple	Alzheimer′s	Unknown	Mitchelmore et al. (2004), Tollervey, Wang, et al. (2011)
NDRG2	IR multiple	Alzheimer′s	Unknown	Mitchelmore et al. (2004), Tollervey, Wang, et al. (2011)
APP	CA Exon 7/17	Alzheimer′s	RBFOX, ELAVL	Alam et al. (2014), Fragkouli et al. (2017)
PTK2B	IR multiple	Alzheimer′s	PTBP1, HNRNPC, CPSF7, ELAVL1	Raj et al. (2018)
PICALM	IR multiple CA exon 13	Alzheimer′s	PTBP1, HNRNPC, CPSF7, ELAVL1	Raj et al. (2018), Scheckel et al. (2016)
CLU	IR multiple	Alzheimer′s	PTBP1, HNRNPC, CPSF7, ELAVL1	Raj et al. (2018)
CD33	CA Exon 2	Alzheimer′s	PTBP1, SRSF1	van Bergeijk et al. (2019)
BIN1	CA Exon 12a	Alzheimer’s	ELAVL1, SRSF1, HNRNPA2/B1	Scheckel et al. (2016)
MAPT	CA Exons 2/3 Exon 10	Alzheimer′s frontotemporal dementia and parkinsonism linked to chromosome 17	SRSF1, SRSF2, SRSF3, SRSF3, SRSF6, SRFS7, SRFS9, SRSF11, TRA2B, RBM4, CELF2, CELF3, HNRNPG, HNRPE, PTBP2, RBFOX2	Liu and Gong (2008), Zhang, Yang, et al. (2020)
PSEN1	CA Exon 4	Alzheimer′s	Unknown	Braggin et al. (2019), De Jonghe et al. (1999), Love et al. (2015), Nornes et al. (2008), Sato et al. (1999)
PSEN2	CA Exon 5	Alzheimer′s	Unknown	Braggin et al. (2019), De Jonghe et al. (1999), Love et al. (2015), Nornes et al. (2008), Sato et al. (1999)
LMNA	CA Exon 11	Progeria	SRSF1, SRSF6	Lopez-Mejia et al. (2011)

^aAS event types are indicated: cassette exon (CA), alternative 5′SS or 3′SS selection (A5′SS or A3′SS), inclusion of mutually exclusive exons (MXE), intron retention (IR), alternative first (AFE) or last (ALE) exon.

5.1 |. Neurodegenerative diseases

The brain exhibits unique AS patterns which contribute to every step of nervous system development, including cell-fate decisions, neuronal migration, axon guidance, and synaptogenesis (Weyn-Vanhentenryck et al., 2018). Among age-related disorders, Alzheimer’s disease is characterized by progressive neurodegenerative phenotypes leading to memory loss and dementia, possibly caused by aggregation of amyloid beta in the brain (Kandalepas & Vassar, 2014; Zheng & Koo, 2011).

5.1.1 |. Spliced isoforms and their regulators associated with Alzheimer’s disease

AS isoforms have been identified in Alzheimer’s patients and model systems. For example, the BACE transcript encoding β-secretase, a key enzyme that degrades the amyloid precursor protein into amyloid beta, undergoes extensive AS, generating at least four distinct isoforms (Fisette et al., 2012; Mowrer & Wolfe, 2008). In addition, several studies reported differences in BACE isoforms when comparing young and old mice (Stilling et al., 2014; Zohar et al., 2005).

Furthermore, a comprehensive analysis of AS patterns in the dorsolateral prefrontal cortex of 450 subjects from two aging cohorts revealed that dysregulation of AS is a feature of Alzheimer’s disease and is, in some cases, genetically driven (Raj et al., 2018). The 84 significantly associated AS events include genes that have been previously shown to be associated with Alzheimer’s disease pathogenesis such as phosphofructokinase (PFKP), the α/β-hydrolase fold protein gene NDRG family member 2 (NDRG2), and the amyloid beta precursor protein (APP) (Mitchelmore et al., 2004; Tollervey, Wang, et al., 2011), as well as novel genes such as the Protein Tyrosine Kinase 2 Beta (PTK2B), or genes in known genome-wide association study loci such as PICALM, which codes for a clathrin assembly protein, or CLU, which encodes a secreted chaperone that prevents aggregation of non-native proteins (Raj et al., 2018). These splicing events were enriched in binding sites from 18 RNA binding proteins, including PTBP1, HNRNPC, CPSF7 and ELAVL1 (Raj et al., 2018). Notably, PTBP1 and SRSF1 can also regulate splicing of CD33, leading to an isoform associated with Alzheimer’s risk (van Bergeijk et al., 2019); whereas ELAVL1 has been previously linked with splicing regulation of BIN1 and PICALM in neuronal cells (Scheckel et al., 2016), and hnRNPC with translational regulation of APP mRNA (Borreca et al., 2016). These studies suggest that Alzheimer’s disease-related perturbations in AS are not simply owing to spliceosomal failure but rather that specific genes are reproducibly affected in a specific manner.

Finally, autosomal-dominant familial Alzheimer’s disease is caused by variants in presenilin 1 (PSEN1), presenilin 2 (PSEN2), and APP genes. Skipping of exons in PSEN1 or PSEN2 genes, which could lead to the expression of deleterious protein products, has been linked with the development of Alzheimer’s disease (Braggin et al., 2019; De Jonghe et al., 1999; Love et al., 2015; Nornes et al., 2008; Sato et al., 1999). APP exon 7 containing isoforms, which are regulated by SF members of the RBFOX protein family (Alam et al., 2014), are elevated in brain tissue and can activate the intracellular domain of APP as well as beta secretase and could contribute to the accumulation of toxic β-amyloid peptide (Belyaev et al., 2010).

5.1.2 |. Spliced isoforms of Tau

Splicing of Tau, a microtubule-associated protein encoded by the MAPT gene, has been linked with frontotemporal dementia and parkinsonism linked to chromosome 17, and can also contribute to Alzheimer’s disease (Hutton et al., 1998; Varani et al., 1999). AS modulates Tau function in normal brain by altering the domains of the protein, thereby influencing its localization, conformation and posttranslational modifications and hence its availability and affinity for microtubules and other ligands (Liu & Gong, 2008). Changes in the Tau protein isoform ratios lead to the formation of neurofibrillary tangle aggregates, characteristic of Alzheimer’s disease and other tauopathies (Chen et al., 2010).

Tau AS is regulated by a variety of SFs, including several SR proteins (Liu & Gong, 2008). Additionally, components of the U1 snRNP co-aggregate with Tau in neurofibrillary tangles in postmortem brain tissues from Alzheimer’s patients (Bai et al., 2013; Hales et al., 2014), Mapt transgenic mice (Maziuk et al., 2018; Vanderweyde et al., 2016), and in vitro (Bishof et al., 2018). Finally, a screen of candidate genes from Alzheimer’s-associated human genomic loci discovered that SmB, the fly ortholog of human SNRPN, modulates Tau-mediated neurotoxicity (Shulman et al., 2014). Follow-up studies in flies demonstrated that Tau expression triggers reductions in multiple core and U1-specific spliceosomal proteins, and genetic disruption of SF SmB, U1-70K, and U1A, enhances Tau-mediated neurodegeneration (Hsieh et al., 2019). Loss of function in SmB decreased survival, progressive locomotor impairment, and neuronal loss, independent of Tau toxicity (Hsieh et al., 2019). Interestingly, RNA-seq revealed similar profiles of AS errors in SmB mutant and Tau transgenic flies, as well as cryptic splicing errors in association with neurofibrillary tangle burden in human brains (Hsieh et al., 2019). Several other SFs have been implicated in neurodegenerative disorders, including TIA-1 that co-localizes with hyperphosphorylated Tau and aggravates Tau pathology (Vanderweyde et al., 2016), but also SFPQ which is recruited into TIA-1-positive stress granules after oxidative stress induction, and colocalizes with Tau (Younas et al., 2020). Hundreds of splicing quantitative trait loci have been identified in distinct human brain regions, including one that disrupts a putative binding site for RBFOX2 and increases inclusion of MAPT exon 3 in cerebellar tissues (Zhang, Yang, et al., 2020). Small molecules targeting AS of Tau are currently in development and can rescue disease-relevant AS of Tau pre-mRNA in a variety of cellular systems, including primary neurons (Chen et al., 2020).

5.2 |. Progeria and premature aging diseases

Progeria, or HGPS, is a premature aging disease associated with a high incidence of age-related disorders, such as car-diovascular impairment and atherosclerosis (Hisama et al., 2011; Merideth et al., 2008). One of the most common forms of HGPS exhibits a silent mutation in the LMNA gene encoding lamin A and C proteins. This mutation activates a cryptic splice site and leads to the production of a truncated spliced isoform, named progerin, which lacks 150 nucleotides and encodes a protein with impaired nuclear membrane functions (Hisama et al., 2011; Merideth et al., 2008). Progerin is not only expressed in HGPS patients, but also accumulates during normal aging (Ashapkin et al., 2019). AS of progerin is controlled by multiple SFs, including SRSF1 and SRSF6 (Lopez-Mejia et al., 2011). Splice-switching therapies that prevent or reverse pathogenic Lmna isoforms are currently being tested in HGPS models. Antisense morpholino-based therapy can reduce the accumulation of progerin and its associated nuclear defects, improving the phenotype of HGPS mouse models and substantially extending their life span (Osorio et al., 2011).

Splicing analysis of a mouse model of HGPS, containing an inducible LMNA mutation, revealed that the number of alternatively spliced genes increases with age (Rodriguez et al., 2016). Two-thirds of the age-related events were specific to this mouse model and not detected in wild-type aged animals (Rodriguez et al., 2016). Spliced genes in HGPS were overrepresented in relevant functions such as skin development, ECM-receptor interaction pathway, connective tissue disorders, and inflammatory diseases (Rodriguez et al., 2016).

5.3 |. Aging and cancer

Age is the greatest risk factor for developing cancer (de Magalhaes, 2013), and 60% of cancer patients are 65 or older. Most cancers can be considered an aging disease, although the shared mechanisms underpinning the two processes remain unclear. Several of the hallmarks of aging and cancer are shared, including genomic instability, telomere attrition, epigenetic changes, loss of proteostasis, decreased nutrient sensing, and altered metabolism, but also cellular senescence and stem cell function (Aunan et al., 2017). Gene expression analyses have been used to study cancer and aging, but only a few studies have investigated the relationship between gene expression changes in these two processes (Aramillo Irizar et al., 2018; Chatsirisupachai et al., 2019). Studies thus far suggest that the transcriptional alterations observed in degenerative aging diseases and cancer may affect overlapping sets of genes, but that they are regulated in opposite directions (Aramillo Irizar et al., 2018; Chatsirisupachai et al., 2019). Another feature common to both aging and cancer is the dysregulation of the splicing machinery (Urbanski et al., 2018). Tumor-associated alterations in RNA splicing result either from mutations in splicing-regulatory elements or from changes in components of the splicing machinery.

5.3.1 |. Mutations in spliceosomal genes

Recurrent somatic mutations in spliceosomal genes SF3B1, U2AF1, and SRSF2 occur in human tumors, most commonly in myelodysplastic syndromes (MDS), chronic myelomonocytic leukemia, acute myeloid leukemia, and chronic lymphocytic leukemia (Papaemmanuil et al., 2011; Yoshida et al., 2011). Mutations in each of these occur as heterozygous point mutations at specific residues, suggesting gain-of-function alterations, and are mutually exclusive with one another, presumably due to redundant effects and/or a limit on cellular tolerance of disrupted spliceosome function. Interestingly, the two most frequently mutated SFs are SF3B1, a core component of U2 snRNP involved in branch point site selection, and SRSF2, a serine/arginine-rich (SR) protein that acts both in alternative and constitutive splicing and interacts with U1 snRNP. Mutant SF3B1 is associated with decreased splicing fidelity and branchpoint recognition, leading to cryptic 3′ splice site selection (Carrocci et al., 2017; Darman et al., 2015; Tang et al., 2016). Conversely, mutations affecting SRSF2 alter its RNA binding preference in a sequence-specific manner, favoring splicing at C-rich sequences while having reduced affinity for G-rich sequences, and thereby alter the efficiency of exon inclusion (Kim et al., 2015; Zhang et al., 2015). Cells expressing mutant SF3B1 or SRSF2 display aberrant AS of mRNAs that promote NFκB signaling (Lee et al., 2018). Furthermore, SF3B1 and SRSF2 mutations elicit distinct effects on splicing and are synthetically lethal due to the cumulative impact on hematopoietic stem cell survival and quiescence (Lee et al., 2018).

The majority of patients with MDS are older than 55 years of age, and SF mutations are typically thought to be early events in MDS pathogenesis (Mossner et al., 2016; Pellagatti et al., 2016). Aging is associated with an accumulation of somatic mutations in normal dividing cells, including hematopoietic stem cells. Interestingly, large-scale genetic studies have revealed the prevalence of cancer-associated clonal SF mutations in blood of individuals without neoplasia, although at a lower frequency than mutations in epigenetic regulators TET2 and DNMT3A (Genovese et al., 2014; Jaiswal & Ebert, 2019; Xie et al., 2014). These findings suggest that the presence of SF mutations can result in clonal expansion in aging human bone marrow. Yet, hematopoietic stem cells from mice mutant for spliceosomal genes have a competitive disadvantage in vivo, unlike the clonal expansion observed in humans with these mutations (Jaiswal & Ebert, 2019).

5.3.2 |. Expression changes in splicing regulatory factors

In solid tumors, SFs exhibit frequent changes at the copy number or expression levels but are rarely mutated (Urbanski et al., 2018). SFs bind directly to pre-mRNA and regulate their downstream targets in a concentration-dependent manner (Long & Caceres, 2009); thus, changes in SF levels cause AS deregulation in tumors even in the absence of mutations. Changes in the expression of SFs from the SR protein family have been detected in several solid tumor types. For example, causal links have been identified between AS misregulation and breast cancer (Anczukow et al., 2012; Anczukow et al., 2015; Climente-Gonzalez et al., 2017; Hu et al., 2020; Koedoot et al., 2019; Sebestyen et al., 2016; Xu et al., 2014; Yoshida et al., 2015; Zhang, Zhang, et al., 2020). Moreover, several SFs are upregulated in breast tumors, and are sufficient to promote tumor initiation in breast cancer models (Anczukow et al., 2012; Anczukow et al., 2015; Karni et al., 2007; Park et al., 2019). A subset of these SF alterations are due to gene amplification (Karni et al., 2007) or to changes in their transcriptional regulators which include the MYC oncogene (Anczukow et al., 2012; Anczukow & Krainer, 2015; Das et al., 2012; Park et al., 2019). Two-thirds of all breast cancers occur above the age of 50 (Benz, 2008), making age the greatest risk factor for this disease in women. Aging is associated with the gradual accumulation of somatic mutations and of epigenetic, transcriptional, and posttranscriptional changes across cell and tissue types (Kenyon, 2010; Li, Shapiro, et al., 2019; Lopez-Otin et al., 2013; Pelissier Vatter et al., 2018). However, the mechanisms through which these aging-related molecular and cellular changes contribute to breast cancer development and risk are poorly understood (Fresques et al., 2019; LaBarge et al., 2016; Yau et al., 2007). Furthermore, how the expression or activity of SFs vary during mammary gland aging—and whether aging-related AS drives functional decline and leads to breast cancer—remains to be determined.

6 |. CONCLUSION AND PERSPECTIVE

AS, that is, the process that controls whether an exon or intron is included or skipped in a final mRNA transcript, is clearly altered during aging, across all species and all cell types examined thus far. Given the heterogeneity in AS profiles between tissues, and cell types, but also different aging rates between tissues, one can expect cell-type specific differences in the molecular links between AS and aging. The cumulative evidence that points to tissue- and species-specific heterogeneity in aging-associated AS patterns suggests that the role of AS events in aging may need to be considered on an event-by-event basis. Yet whether aging is a consequence or a driver, or both, of the changes in AS patterns observed in older organisms remains to be determined. A number of technical and conceptual challenges limit our ability to answer this question at the required resolution. First, published AS analyses have often been carried out with differing technologies used to detect AS patterns, or analyzed using distinct computational tools, thus rendering direct comparisons challenging. Second, it is unknown if age-related AS changes arise in a gradual or communitive fashion, or rather appear suddenly at a discrete time point triggered by a specific aging signal. Therefore, the age at which analysis is performed, and the number of timepoints considered, is likely important for detecting key AS patterns in studies with discrete timepoints. Third, model organisms differ in their AS profiles, with >90% of human genes undergoing AS, compared to ~25% in worms and ~65% in mice, rendering cross-species aging comparisons challenging. Moreover, even within the same species, distinct tissues differ in AS patterns during development, as well as with age. Finally, to date, most AS analyses have been carried out using bulk tissues and thus are not well suited for the detection of AS events in specific cell types, which may change in abundance within the tissue with aging. Thus, AS patterns in smaller cell populations may be masked by larger cell populations, and single-cell methods for AS analysis are needed to capture these changes. Finally, the accumulation of mutations that occur with age could impact AS patterns in aged individuals—either by directly impacting the SFs or the splice sites—and impact downstream AS patterns. Nevertheless, understanding the causes and consequences of age-related alterations, possibly at the single-cell level, will provide mechanistic insights into the molecular basis of aging, and might provide novel therapeutic opportunity for age-related diseases or for increased lifespan.