Conserved aging-related signatures of senescence and inflammation in different tissues and species

Gene expression discriminates among tissues more strongly than ages

To evaluate the homogeneity of the investigated transcriptomic data, we performed t-SNE dimensionality reduction of the RNA-Seq libraries based on the measured expression strengths of our 464 preselected senescence- and inflammation-associated genes (see Figure 1). We observed a homogeneous clustering of the samples with respect to species and tissue, with almost no segregation of the three different ages within these clusters. This observation was already made and reported in similar studies based on different sets of genes and indicates that aging is a relative subtle process, at least on the transcriptional level [9]. Nevertheless, separations between the youngest and oldest time point can be observed within the skin clusters of all four species but to a much weaker extent in the human samples. This observation might suggest a pronounced and conserved difference in the activity of senescence and inflammation processes during skin aging of evolutionarily distinct species. This observation and the fact that the skin is also relatively easily accessible makes it especially interesting for interspecies comparisons of age-related senescence and inflammation. However, it remains to be further investigated whether this observation can be made in similar transcriptomic studies.

Figure 1. A t-distributed stochastic neighbor embedding (t-SNE) of the analyzed RNA-Seq libraries. All RNA-Seq samples were clustered based on the expression patterns of the selected senescence and inflammation related genes, utilizing the t-SNE approach. All tissues form distinctive species-specific and non-overlapping clusters with exception to very few single outliers. Additionally, larger species-independent tissue clusters were drawn to improve the visualization of the data. However, the three different time points did not generally separate in independent clusters of their own. A weak segregation can only be observed among the mature and old-aged skin samples of all four species. For more information, see Supplementary Data 10.

Another observation is that although the different tissue samples form species-specific clusters, they are still well separated from each other, i.e., forming tissue-specific clusters, with the exception to the Homo sapiens and Mus musculus blood samples. This observation hints that the similarities of the inflammaging process among the same tissues of different species prevails over any systematic effect within one individual.

Expression variance is more controlled in long-lived individuals

It is proposed that individuals that have reached a comparatively high age have somehow counteracted the effects of inflammaging by unknown anti-inflammation processes [5, 6]. As one approach to validate this hypothesis, we analyzed how the mean variance in gene expression changes during aging. Interestingly, we found a significantly lower relative standard deviation of the old-age time points compared with the aged time points in almost all species and tissues and even lower than the mature time points in selected cases (see Figure 2 and Supplementary Data 4). This observation indicates that the selected genes associated with inflammation and senescence have a more stable expression, suggesting that both age-driving processes are in fact more controlled in the long-lived individuals. Similar observations were recently made in brains of long-lived mice [10]. Exceptions are the skin samples of all species other than humans, which show the opposite pattern of a significantly increased variance in gene expression in the old-age time point. The skin is the most diverse of the tissues studied in this work, with mice and humans having fur and hair, whereas fish have scales. In addition, it is one of the most exposed tissues and therefore faces a potentially higher diversity of factors that affect ageing in skin [11]. As a result, skin cells are subjected to more and different stresses, and the rate of senescence is increased compared with other tissues, which also affects the rate of inflammation within the skin tissues due to the pro-inflammatory secretory phenotype that senescent cells can develop [12]. In contrast, human skin showed a decreased variance in gene expression, indicating less pronounced or at least tighter controlled inflammatory and senescence processes. Such a distinct behavior could be the result of various intrinsic or extrinsic physiological factors like the thinning of the epidermis, the degeneration of the extra cellular matrix, or a decline of the microvasculature due to a reduced angiogenic capacity [24]. Besides, additional factors like external applications of antioxidants on skin might also play role in it [11].

Figure 2. Change of relative standard deviations of gene expression in all four species with age. For each investigated species and tissue, the measured variance in transcript expression of the preselected senescence- and inflammation-related genes is displayed for every time point (M – mature, O – old, OA – old-aged). The upper and lower bounds of the box plots represent the respective 2.5% percentiles. All displayed differences in the mean variance of gene expression among the aged and old-aged time points and almost all other age comparisons are significant (*: p-Value ≤ 0.01, **: p-Value ≤ 0.001, ***: p-Value ≤ 0.0001,) within each tissue and species, individually. A general decrease of variance in gene expression can be observed in the old-age time points, except for the skin samples of Mus musculus, Danio rerio and Nothobranchius furzeri. This indicates a tighter control of inflammaging processes in long-lived individuals, reducing negative effects and helped them to reach the high age. For detailed information, see Supplementary Data 4.

The only other observed exception is the brain of Nothobranchius furzeri, which displayed a strongly increased gene expression variance at the old-age time point compared with both younger time points. This pattern is similar to that in fish skin and is already reflected in the sample clustering as mentioned above (see Figure 1). This observation could indicate a weaker protection of the Nothobranchius furzeri brain against inflammaging, but it remains to be further studied.

To verify that the more controlled gene expression observed in long-lived individuals is not biased due to our preselected genes, we repeated the same analysis with all expressed genes in every species and tissue. We confirmed our observation with all tissues showing the same relative variance changes as before, in principal. The overall variance within each age group was decreased, which was expected because many more genes were included in the analysis that show no change in expression during aging (see Supplementary Data 4). Nevertheless, in order to confirm that long-lived individuals really benefited from controlled gene expression, further investigations, such as longitudinal studies, are necessary.

Oxidative stress response tends to be the prevailing process towards old age

To examine whether certain biological functions predominantly drive the inflammaging processes in a tissue- or species-dependent manner, we analyzed the identified differentially expressed genes (DEGs) with respect to their molecular function. We found that most of the DEGs within the brain and liver samples belong to immune and inflammatory response processes, whereas changes in the expression of senescence-related genes are predominately related to skin aging (see Figure 3, Supplementary Data 6 and Supplementary Data 11). Only few DEGs could be identified in blood age comparisons for humans and mice, making it difficult to interpret these results in a meaningful way.

Figure 3. Differentially expressed genes during tissue aging and associated biological processes. The upper part shows the number of identified differentially expressed genes (DEGs) within all three age comparisons (early aging, late aging, longevity) in every investigated species and tissue. The lower part shows for every comparison the dominant biological processes as determined by the majority of the annotated functions of the respective DEGs. Note that in some comparisons only few DEGs could be identified and hence may only poorly reflect the underlying age-related processes (n.d. – no data). For detailed information, see Supplementary Data 6 and Supplementary Data 11.

It is a known and common observation of aged skin that it is more susceptible to infections, physical damage and reduced epidermal barrier integrity as well as other age-related deficiencies [14]. Epidermal stem cells maintain the tissue’s homeostasis and loss of those stem cells due to premature senescence is the main cause of aging within the skin [15]. Our data confirm this observation, showing mainly senescence-related DEGs within the skin of all four species already in the early and the late aging comparisons. However, whether the source of epidermal stem cell senescence is primarily intrinsic or extrinsic factors is controversially discussed [16, 17]. Apoptosis appears to play a more important role during aging of the liver than in any of the other investigated tissues. This is a confirmed observation, because liver homeostasis is mainly regulated through apoptotic processes and many liver dysfunctions and diseases are related to apoptosis [18, 19].

Most interestingly, we observed that DEGs associated with the oxidative stress response and oxidative processes occur predominantly within the late age and longevity comparisons (i.e., comparisons with the long-lived individuals) of the brain, liver and skin samples. This observation suggests that those individuals that grew older than their species average survival age have either a generally more active or better regulated response mechanism against reactive oxygen species (ROS). Free radicals such as ROS and their great impact on the aging process are much discussed [8, 20]. During respiration of oxygen, free radicals are produced as a harmful byproduct that can oxidize macromolecules such as DNA or proteins and damage them in this process. The main point of aging theories involving oxidative stress is that with age, the rate of harmful ROS accelerates because ROS first target the mitochondria (because they are mainly generated in that location), leading to an accumulation of damage in these cell organelles. The more degenerated the mitochondria become, the higher the rate of released ROS becomes, injuring genomes and membranes of neighboring cells and, as a final after-effect, resulting in their senescence or apoptosis and therefore aging [21]. Different antioxidant mechanisms exist within the cells to prevent the release of free radicals and protect the cells from such deleterious consequences [22]. It was observed that long-lived species appear to have higher protection against ROS than short-lived species [23, 24]. Our data suggest that the same is true for long-lived individuals within one species, confirming the importance of oxidative stress response for longevity.

Conserved aging expression signatures across tissues and species

To identify a more precise signature of the abovementioned observations (Figure 3), we analyzed the expression patterns of our preselected gene set in greater detail. We were particularly interested in genes that show a constant increase or decrease towards old age that could also be consistently observed between the investigated tissues and species.

We found 16 genes that were consistently higher expressed and 10 that were consistently lower expressed during aging in at least three of the four species and tissues, totaling 26 genes that show a conserved expression pattern with aging (see Figure 4). Certain of these genes have already been reported in the context of aging, and we observed them to be the most conserved genes from our preselected set among species and tissues. Hence, we discuss several of these genes based on their biological function in additional detail. More senescence- and inflammation-related genes were found to have conserved expression patterns but to a lesser extent and can be found in Supplementary Data 7.

Figure 4. Heatmap representation of potentially conserved senescence- and inflammation-related genes. Genes are represented as mouse orthologues (a complete list of gene orthologues can be found in Supplementary Data 8). Numbers indicate log₂ fold changes between two compared ages, where a positive value indicates an upregulation (blue), and a negative value downregulation (red) of the respective gene with aging. All significant changes in gene expression are indicated in bold. For detailed information, see Supplementary Data 7.

Immune/inflammatory response

The gene marco encodes for a scavenger receptor (Marco – macrophage receptor with collagenous structure), which is typically found on macrophages but also other immune cells [25, 26], and acts in the innate immune system by binding and clearing pathogens and initiating an inflammatory response [27, 28]. We observed marco to be constantly upregulated (mostly by an increase of two-fold or more) towards old-age in all four species and within the blood, liver and skin. Although Marco does not cause inflammation directly, it is important for the activation of other receptors, such as members of the toll-like family [28]. These receptors subsequently activate NF-κB, inducing an inflammatory response. Interestingly, Marco is associated with Alzheimer’s disease, because it is also present on microglia and a decreased response of these cells has been observed after binding amyloid beta peptides [29].

The coagulation factor II receptor (PAR1) is encoded by the gene f2r, which similar to marco shows persistent upregulation with age in blood, liver or skin within all four species. The major function of PAR1 is mediation between coagulation and inflammation activity, and therefore it has a key role in inflammatory response activation [30, 31]. In addition, PAR1 activity is implicated in aging-associated cardiovascular diseases [32].

Four more genes (cd40, sh2d1b1, ptafr, adam8) coding for membranous receptors have a common upregulated signature through several aging tissues and species. These genes have all been shown to promote pro-inflammatory signaling, either directly by binding platelet-activating factors (ptafr [33]), activating antigen presenting cells (cd40 [34]), and acting as a regulator of antigen receptor signal transduction (sh2d1b1 [35]) or more indirectly by releasing and degrading other cell surface receptors of leukocytes (adam8 [36]).

Observation of a generally higher expression of ceruloplasmin encoded by the gene cp in all species and tissues with age (except Nothobranchius furzeri brain) is not unexpected due to its role as an acute-phase protein. These reactants are increasingly released in the blood plasma during an inflammatory response and support the innate immune response [37]. Thus, increased cp expression is most likely a result of the chronic inflammatory nature of aged tissues.

Lef1 is one of the few immune-related genes that showed a decreased expression with age in the blood and skin of human and mice respectively and the brains of both examined fish. The main function of the LEF1 (lymphoid enhancer-binding factor 1) protein is to enhance expression of the T-cell receptor alpha chain [38]. However, a variety of different interaction partners are known, linking LEF1 as an important regulator within the WNT and TGF pathways and implicating a role in apoptosis and cell proliferation of leukocytes [39, 40].

Oxidative stress response

As previously mentioned, ROS and oxidative stress are widely believed to be the driving force behind the cellular aging process [8]. The source of oxidative stress is likely an imbalanced activity of the respiratory chain and antioxidant processes. We observed a constant age-depended upregulation of the genes cyba, cybb, cyp26b1 and nrros in various combinations of the investigated tissues and species. The first two genes encode for the light and heavy chains of the Cytochrome b-245 protein, a superoxide-producing subunit of the NADPH oxidase [41]. By producing and releasing ROS, phagocytes use Cytochrome b-245, mainly as an antimicrobial strategy during an infection. However, the role of Cytochrome b-245 during aging is still debated [42, 43], and it is associated with several degenerative diseases due to uncontrolled production of ROS [44, 45]. Cyp26b1 encodes another cytochrome protein, Cytochrome P450, which is involved in many metabolic reactions, especially the oxidation of NADPH [46]. Higher expression of cyp26b1 during aged time points is suggested to aid in degradation of toxic substances that have accumulated with age [47]. The genes nrros and gpx8 both code for proteins that belong to antioxidant mechanisms controlling the production of ROS. NRROS directly interacts with Cytochrome b-245 and mediates its degradation, limiting the rate of ROS production of the associated NADPH oxidase complex [48]. Similarly, the enzyme GPX8 protects cells from ROS-induced damage by its peroxidase activity by reducing free hydrogen peroxide to water [49]. However, whereas gene expression of nrros rises with age, most likely to regulate the increased Cytochrome b-245 activity, the translation of gpx8 is downregulated in many tissues and species. We observed the opposite regulation during aging only in the human skin and mouse liver samples. Nevertheless, gpx8 serves as an interesting potential therapeutic target and it was previously shown in the Caenorhabditis elegans model that the deletion of several GPx family members is the cause of an accelerated aging process that results in shorter life-span [50].

Senescence and apoptosis

Most of the 26 identified genes with conserved expression patterns in aging, are strongly related to the cell cycle control or apoptotic processes. Both replicative senescence and apoptosis are immediate causes for the loss of somatic and stem cells with age and can be similarly triggered by chronic inflammation and oxidative stress [51, 52].

We found four of these genes to encode direct inducers of senescence: ccnf (Cyclin-F) [53], scrib [54]. cdkn2b (p15^INK4b) [55] and rassf4 encoding an inhibitor of Cyclin D1 [56]. We found all of these genes to be expressed in a senescence-promoting fashion towards aging (rassf4 and cdkn2b upregulated; scrib and ccnf downregulated) in the examined tissues and species. Additionally, selected genes interacting with cell cycle regulators also showed persistent expression towards old age, namely, dab2 [57], lgals1 [58], and brat1 [59]. The proteins CDCA8 and Survivin encoded by cdca8 and birc5, respectively, are both associated with apoptosis. By inhibiting caspase activity, Survivin is a repressor of apoptosis, whereas CDCA8 directly interacts and stabilizes Survivin [60, 61]. However, both proteins show converse directions of regulation during aging between the same tissues and within individual species, resembling rather weakly conserved expression signatures.

Potential common tissue-specific marker genes of aging

We also focused on senescence- and inflammation-related genes that change their expression during aging consistently and similarly in a tissue-specific manner. These genes should present conserved age-related changes and might be the driving factors of tissue-specific aspects of the inflammaging process, probably closely linked to the specialized function of the respective tissue. In a first comparison of the identified DEGs of the different species, we observed only a few common genes that occur as differentially expressed in all species of a single tissue (see Figure 5 and Supplementary Data 9). However, the blood and liver samples revealed no DEGs to be common in all of the respective species, and two were identified within the brain (cybb, cd68) and another two within the skin (Gpx7, Tnfrsf25). Cd68 is a member of the cell surface receptors of macrophages and other monocytes and is a well-known marker for macrophage activation. The Cytochrome b-245 subunit encoding gene cybb and its potentially harmful role during aging were already introduced in the previous section. Tnfrsf25 (a member of the TNF receptor superfamily) stimulates the proliferation of T cells or can initiate apoptotic signaling, leading either to survival or cell death, and is therefore an important regulator of T cell development [62]. Gpx7 encodes for another glutathione peroxidase and has the same antioxidant function as its above described homolog gpx8.

Figure 5. Common inflammation- and senescence-related genes that are significantly changed with age. Venn diagrams showing the overlap of the identified differentially expressed genes among the four investigated tissues. Only few genes are commonly differentially expressed among all species of any of the four tissue comparisons. For detailed information, see Supplementary Data 9.

However, none of the four genes exhibits features that serve as a tissue-specific marker because they do not display a consistent up- or downregulation during aging within each investigated species or share the same expression pattern with other tissues. Therefore, we applied a similar but more stringent filtering process to reveal genes displaying the characteristics of potential tissue-specific markers. With this approach, we identified several genes that showed a constant age-related transcriptional increase exclusively in the liver or skin samples of the investigated species.

Liver

Four constantly upregulated genes were identified in the liver samples of Mus musculus, Danio rerio and Nothobranchius furzeri: jag2, anax1, ralb, and sfrb. The protein encoded by jag2 (Jagged-2) is a known activating ligand of Notch2, which is mainly involved in many different developmental processes regulating cell fate decisions [63]. Most interestingly, jag2 overexpression is not only shown to play a critical role in the formation of plasma cell myeloma and progression of other tumors [64, 65] but also induces the secretion of interleukin-6 [66]. Interleukin-6 is a potent stimulator of immune and inflammatory responses and acts in the development and progression of many age-associated diseases, such as Alzheimer’s disease [67], atherosclerosis [68], diabetes [69] and various cancers, making it an attractive therapeutic target [70, 71]. Because the liver shows strong signs of inflammation during aging (Figure 3), overexpression of jag2 might be one conserved driving factor of inflammatory processes within hepatic cells and could be of interest as an additional therapeutic target. In contrast, Annexin A1, the protein encoded by anax1, has distinct anti-inflammatory and protective properties, by inhibiting NF-κB signal transduction and counterregulating pro-inflammatory signals in a variety of immune cells [72, 73]. The age-related upregulation of Annexin A1 in the liver could be a component of a tissue-specific mechanism attempting to cope with the chronic state of inflammation. SFRP2 (encoded by sfrp2) has an oncogenic character and is associated with cancer formation by acting as an inhibitor of the canonical WNT/βcatenin pathway [74, 75]. The last of the four conserved liver-specific upregulated genes, ralb, encodes for one of two Ral protein paralogs (Ral-B) and acts as a major modulator of a multitude of cellular processes [76]. Although many functions are shared between the paralogs Ral-A and Ral-B, the latter is more specifically involved in the activation of apoptosis [77], which we observed to occur more predominantly during liver aging (Figure 3). Additionally, Ral-B has clinical significance because it promotes tumor progression of several cancers and activation of the innate immune response [76, 78] and hence might contribute to inflammatory stress in the liver.

Skin

If including all four species, we did not observe any gene displaying the characteristics of a potential conserved marker gene for skin aging. Considering the diverse nature of skin among the studied species we decided to examine the skin in two separate groups of, fishes and mammals.

Between human and mouse, only pecam1 was identified to be consistently upregulated towards old age. The gene encodes the protein Pecam-1, also known as CD31, which is a common immunohistochemistry marker used to evaluate tumor growth and has a major role in the removal of old neutrophils, which could explain its higher expression with age. However, this gene is already known as a biomarker of inflammatory processes [79].

The two fishes Danio rerio and Nothobranchius furzeri share 6 genes that showed a persistent increased expression with aging exclusively in their skin and their encoded proteins are all involved in mitotic progression of the cell cycle: ckap2 [80], cdca8 [61], aurkc [81], cep55 [82], mis12 [83] and spc25 [84]. This observation might indicate that fish skin cells are not as affected by senescence as the skin of humans and mice, which is in line with the observation that the skin of zebrafish shows a high regenerative capacity even at higher age [85].

High-throughput transcriptomic data acquisition

All RNA-Seq libraries involved in this study originate from the JenAge consortium (http://www.jenage.de/) and were first published by Irizar et al [1]. These libraries are accessible at NCBI’s Gene Expression Omnibus (Homo sapiens: GSE75337, GSE103232; Mus musculus: GSE75192, GSE78130; Danio rerio: GSE74244 and Nothobranchius furzeri: GSE52462, GSE66712).

High-throughput transcriptomic data processing

For each species we investigated up to four different tissues at three different ages: young mature (M), aged (A) and old-aged (OA) (see Figure 6).

Figure 6. Overview of the analyzed high-throughput transcriptomic data. For each of the four investigated species, up to four different tissues were sampled at three different ages: one mature (M) time point, one aged (A) time point and one old-aged (OA) time point. Three different comparisons were made to reveal significant differences in expression of genes during early aging (M vs. A), late aging (M vs. OA) and longevity (O vs. OA). For more details, see Supplementary Data 1.

First, Prinseq (v0.20.3) [92] was used to trim and filter the RNA-Seq libraries, by clipping reads on either ends to achieve a minimum base quality score of 20 and discarding all reads with a length of less than 15 nt or containing more than two ambiguous N bases. Second, quality-trimmed RNA-Seq libraries were mapped onto the according recent genomes (as from Ensembl release version 92 [93] and [94, 9]) using TopHat2 (v2.1.1) [95] under consideration of the default parameters. This procedure allowed the identification of spliced reads, and the mapping of single reads to multiple best-fit locations. Third, read counting was performed by Featurecounts (v1.5.3) [96], with the counts normalized to transcripts per million (TPM) [97] to discard all genes with a value of ≤ 1 in every sample from subsequent expression analyses. Finally, differentially expressed genes were identified via the DESeq2 (v1.10.0) [98] Bioconductor package by comparing the three different time points of each species and tissue. False discovery rate adjustment of the resulting gene’s p-values was performed according to previous work [99]. Details of all DEG results, together with the raw and normalized count values are given in detail in Supplementary Data 1.

Collection of genes of interest and t-SNE dimensionality reduction

An initial list of 769 genes relevant in the context of senescence and inflammaging was compiled from a detailed literature search (see Supplementary Data 2). From those 769 genes, we removed all genes that were either missing an ortholog in any of the four species or were not differentially expressed (FDR adjusted p-value < 0.05) at least once in any species and tissue, ending with the final list of 464 genes (see Supplementary Data 3). Based on the measured expression of these 464 genes the t-SNE algorithm [100] was used for dimensionality reduction in order to investigate the clustering behavior of the analyzed RNA-Seq libraries (see Figure 1).

Gene expression coefficient of variation analysis

To estimate the general gene expression variation among the different time points of each species and tissue, we calculated the coefficient of variation for every expressed gene (i.e., a TPM value greater than 1) as the ratio of its standard deviation to its mean. The significance of the variation differences among time points was determined using the two-sided t-test. We repeated this analysis using only the 464 genes of interest (for details, Supplementary Data 4).

Gene function annotation and expression pattern analysis

The exact biological functions of our preselected gene set of interest were obtained from the functional annotation database David (version 6.8) [101] (Supplementary Data 5). For every age comparison (Figure 6) we determined the dominant biological processes by the majority of the annotated functions of the respective DEGs (for details, see Supplementary Data 11). To identify common signatures of senescence- or inflammation-related processes, we filtered for genes that were differentially expressed at least once within one of the age comparisons and showed the same direction of up- or downregulation by a minimum threshold of 10% in every other age comparison. This rather low effect-size threshold can be justified, because it was previously shown that expression changes related to aging tend to be more subtle [102]. This process was conducted for every species and tissue individually. Based on this gene subset, we further searched for common expression patterns of single genes among the different tissues and species. When searching for tissue-specific signatures, we applied the same filtering strategy but raised the minimum threshold of expression changes to 25%.

Additional information

All relevant data can be found within the paper and its supporting information files. The online supplement is integrated into the open science framework (https://osf.io/) and available under https://osf.io/kzq5y/ or https://doi.org/10.17605/OSF.IO/KZQ5Y. As soon as this manuscript gets published, the Supplementary Data will be made publicly available and the corresponding link and doi will be provided