Insights into Aging from the GTEx Tissue Collection using Omics Playground

The Genotype-Tissue Expression program (GTEx) is an extensive collection of gene expression profiles from 54 different tissue sites across more than 1000 individuals. This data can provide various insights on tissue-specific gene expression patterns. Due to its stratification across age-groups it has also been frequently used to analyse changes in gene expression associated with aging (e.g. Yang et al, 2015; Jia et al, 2018; Zeng et al, 2020) and develop age calculators (Ren and Kuan, 2020).

In this analysis, we aimed to replicate some of the previous findings by selecting 1753 samples from 26 tissue types (GTEx v8)  that were selected and divided into two age groups: ‘young’ (20-39 y.o.) and ‘old’ (60-79 y.o.) for each tissue type. The samples were batch corrected using the Omics Playground platform interface. We also combined all the tissues samples by age in an effort to identify gene expression profiles common across all groups. 

Clustering Analysis

The samples are clearly clustered by tissue type on a UMAP plot (Fig. 1). Forcing a grouping by age group did highlight some clusters with distinct gene expression. In particular, clusters S1 and S3 showed an upregulation in ‘old’ samples (Fig.2A). Annotation of these two clusters showed a correlation with various Hallmark genesets (Fig. 2B), the underlying mechanisms of which have been shown to be affected by aging in different tissues, including glycolysis (Ravera et al, 2019; Goyal et al, 2017 and Kanatsu-Shinohara et al, 2019), peroxisome activity (Ravera et al, 2019), sex hormones activity (Horstman et al, 2012), cholesterol homeostasis (Morgan et al, 2016), pancreatic endocrine function (De Tata, 2014) and drug (xenobiotic) metabolism (Kinirons and O’Mahony, 2004; Klotz, 2009; and Waring et al, 2017).

Figure 1.  UMAP plot showing the clustering of the samples by tissue type. Differences between tissues are more significant than differences between age groups.
Figure 2.  A: Heatmap of the samples clustered by age. B: annotation of the four gene clusters (S1-S4) obtained from the heatmap. Clusters S1 and S3 contain pathways affected by aging.

Differential Expression Analysis

Pairwise comparison of all the samples grouped by age groups was performed by intersecting three different analysis methods (EdgeR, DESEQ2 and limma) and selected rather permissive criteria (FDR<0.05, logFC>|0.2|) to take into account the heterogeneity of the tissues. A total of 176 genes were differentially expressed: 131 were upregulated and 45 downregulated (Fig. 3A). 

Among the top 10 most up-regulated genes (Fig. 3B) were several genes with previous correlations with the aging process. These included EDA2R, which has been associated with aging in the lungs (Jeong et al, 2020de Vries et al, 2017), alopecia (Prodi et al, 2008) and has been observed as a general marker of aging across tissues (Ren and Kuan, 2020).

SAA1 and SAA2-SAA4 are proteins associated with inflammatory responses and SAA1 overexpression is associated with various chronic inflammatory diseases, including age-related conditions such as Alzheimer, arteriosclerosis and rheumatoid arthritis (Genecards entry).

GDF15, a protein associated with inflammation, has been proposed as a biomarker for aging (Liu et al, 2021) with high expression levels linked to muscle weakness (Conte et al, 2020) and age-related monocyte immunosenescence (Pence et al, 2021). CDKN2A gene (involved in cellular senescence) expression levels have been shown to increase with age in rodents and regulation of activity has been shown to be correlated with extended lifespans (GenAge entry). 

DLK1 is  part of the DLK1-DIO3 imprinting control region and is involved in the regulation of cell growth and differentiation (Falix et al, 2012). Deregulation of the DLK1-DIO3 locus  has been linked to replicative senescence of adult adipose-derived stem cells (Garcia-Lopez et al, 2018).

LMO3 is an oncogene and contains a LIM-domain which can regulate gene transcription (GeneCards entry; Bach, 2000). Its expression increases with aging in adipose tissues and  overexpression is linked to cell growth and tumour growth (Digital Ageing Atlas entry). 

Finally there are three genes encoding haemoglobin subunits (HBA1, HBA2 and HBB). Haemoglobin is not exclusively found in erythrocytes and its expression in epithelial cells is considered essential for maintaining O2 homeostasis (Saha et al, 2014). While there is no direct correlation to aging, it is interesting to notice that aging itself is characterised by a decrease in oxygen supply to tissues (hypoxia) (Cataldi and Di Giulio, 2009) and that one of the responses to hypoxia is the upregulation of haemoglobin expression levels (Grek et al, 2011Emara et al, 2013).

Figure 3. Differential expression analysis of the combined tissue samples separated by age groups (young: 20-39 y.o., old: 60-79 y.o.). A: MA plot with significant differentially expressed genes highlighted in blue. B: Top 10 most upregulated genes in the pairwise comparison, old: light blue, young: dark blue. Upregulated genes have been previously identified as biomarkers of aging or linked to aging in various tissues.

While a comprehensive analysis of all of the individual tissue pairwise comparisons is beyond the scope of this short post, we took a look at the top genes for a selection of individual tissues. 

In the brain tissue, the most upregulated gene was the gene encoding the glial fibrillary acidic protein (GFAP) (Fig. 4A). This protein is expressed in numerous cells of the central nervous system (Wikipedia). GFAP expression is increased in the brains of aging mice (Kohama et al, 1995)  and also aging humans (Middeldorp and Mol, 2011). An increase in the amount of acidic isoforms of GFAP was also observed in Alzheimer’s disease patients (Korolainen et al, 2005). 

The most upregulated gene in the skin tissue was FAM38B (family with sequence similarity 83 member B) (Fig. 4B).  It functions as an oncogene that promotes epithelial cell transformation (Cipriano et al, 2012; Cipriano et al, 2013) and is overexpressed in the skeletal  muscles of older individuals (Tumasian et al, 2021), although no information is available on its expression in aging skin.

Finally, ALDOB (Aldolase B) was the most upregulated gene in the stomach tissue (Fig. 4C). This enzyme is preferentially expressed in the adult liver, kidney and intestine (GeneCards entry). Upregulation of ALDOB has been proposed as a prognostic biomarker for rectal cancer and has been found to be associated with metabolic syndrome in rats (Liu et al, 2011), driven by increased fructose levels. Interestingly, high levels of fructose intake have been shown to imimic several aspects of the aging process, such as increased inflammation and oxidative stress (Gatineau et al, 2017).

Figure 4. MA plot highlighting the most upregulated gene in three different tissues. A: brain, B: skin, C: stomach. Overexpression of these genes is linked to aging.

Drug Connectivity Map Analysis

There are several compounds that have been proposed as “anti-aging drugs”, including rapamycin, metformin and NAD+ precursors (Klimova et al, 2018). Using the “Drug Connectivity” module within the “Enrichment” tab of the platform, we looked for the modes of actions (MOA) of the most frequent compounds with anti-aging properties against the combined tissue comparison. Among the top 6 most enriched  inhibitory MOA, we observed two main categories: one related to microtubules and another related to the IGF1/PI3K/AKT/mTOR pathway (Fig. 5). 

The IGF1/PI3K/AKT/mTOR pathway plays a crucial role in various cellular functions with a direct effect on aging (Fig. 6). Caloric restriction, which has been shown to increase longevity, as well as several pharmaceutical interventions target this pathway and its inhibition is believed to be underlying the observed increases in lifespan (Lamming, 2014). Rapamycin is perhaps the most well known mTOR inhibitor that has been characterised as a universal anti-aging drug (Blagosklonny, 2019). Evidence for the effect of the inhibition of the other components of the pathway on longevity is also abundant (e.g. Heras-Sandoval, et al, 2011; Junilla et al, 2013; Noh et al, 2016; Chen et al, 2019).

Microtubules are polymers of tubulin and provide structure and shape to eukaryotic cells (Nogales, 2000). Abnormal microtubule regulation has been linked to age-related disorders and diseases (Raes, 1991; Saha and Slepecky, 2000; Apple and Chen, 2019), and microtubules often serve as targets for anti-cancer therapies (Kaur et al, 2014). Interestingly, microtubule polymerization and stability is also affected by PI3K/AKT activity (Fig. 7), indicating a potential interaction of PI3K/AKT inhibitors with it.

Figure 5. Most common MOA identified during analysis of the combined tissue gene expression profile against genesets from the Drug Connectivity Map L1000 database. Highlighted in red are inhibitors of microtubule activity, highlighted in green are inhibitors of the IGF-1/P13K/AKT/mTOR pathway. Literature strongly points to an anti-aging effect of such inhibitors.
Figure 6. The IGF1/PI3K/AKT/mTOR pathway orchestrates various cellular functions with a direct effect on senescence (source: Lamming, 2014).
Figure 7. Microtubule regulation in the cell. Note the involvement of PI3K and AKT in the regulation of microtubule polymerization and stability. Adapted from the CTS website.

Biomarker Discovery

We used the “Signature” panel of the platform to identify potential biomarkers for distinguishing aged samples across tissues. One of the possible decision trees indicated that the expression levels of two genes (namely SAA1 and LMO3) could be used to separate old from young samples (Fig. 8A). Among the list of potential biomarkers were also other genes that we described in the differential expression analysis section, namely DLK1, SAA2-SAA4 and HBA1 (Fig. 8B). Aging biomarkers can be used for various purposes, such as predicting the development of hyperglycemia (Lee et al, 2018) or identifying the biological age of a person (Sen et al, 2016). Most recently, inflammation-related markers have been proposed as a way to allow doctors to address age-related health issues (Sayed et al, 2021).

Figure 8. Potential Biomarkers for aging across all tissues. A: Decision tree based on the combined tissue samples. B: Boxplots of the expression levels of potential genetic biomarkers across age groups. A combination of these genes can be used to distinguish aged from young samples based on their expression levels.


Although this is just a preliminary look at the dataset, we could find that several Hallmark pathways associated with aging were showing distinct upregulation in ‘old’ samples. Differential Expression analysis of the combined tissues indicated the upregulation of several genes that had been previously described as markers of aging (e.g. EDA2R or GDF15). A cursory glance at the most upregulated genes in three tissue types (brain, skin and stomach) highlighted genes with a potential role in tissue-specific aging. GSEA with the Drug Connectivity map database indicated inhibitors of microtubule formation and activity and inhibitors of the IGF1/PI3K/AKT/mTOR pathway as potential “anti-aging” drugs, with the latter being a well studied anti-aging mechanism. Finally, SAA1 and LMO3 expression levels were proposed as potential biomarkers for aging across all tissues.

As a parting note, this analysis is mostly focused on identifying common patterns across all tissues. For anyone interested in a more in depth look, the data is now available for access under the code “gtex-aging-n40svaNnm” on our online platform.

Share this content