Introduction

Aging population is a huge challenge faced by all countries around the world. Given the rapid growth of the global aging population, researchers are interested in identifying treatments that would delay the physiological, metabolic, and functional decline that gradually occurs in various systems, organs, and tissues of the body as they age. Additionally, it is well known that aging is closely related to a variety of complex diseases including partial cancer, Alzheimer’s disease, Parkinson’s disease, type 2 diabetes, multiple cardiovascular diseases, and neurodegenerative diseases etc.1,2,3,4,5,6. While understanding the biological basis of the aging process is a major scientific challenge that will require integration of molecular, cellular, genetic and physiological approaches7. We hope that we can use model organisms instead of humans to do some research on diseases and drugs, and ultimately achieve the purpose of delaying aging and reducing the occurrence of diseases related to aging. However, it is not clear whether the aging research done on mice is effective on humans. Therefore, in this paper we compared the aging mechanism of human and mouse on multiple tissues at the level of gene expression.

With the advent of various high-throughput sequencing technologies, such as RNA-seq8, the development and improvement of the novel gene expression databases has made it possible to define aging processes by analyzing the transcriptional differences between the young and old. The Genotype-Tissue Expression (GTEx) Portal (https://www.gtexportal.org/home/)9,10 is a resource database generated from an analysis of RNA sequencing data from 1641 samples across 43 tissues from 175 individuals whose ages range from 20 to 79. This provides a data basis for us to study the relationship between gene expression and aging in human tissues.To elucidate the aging differences between humanand mouse at the molecular level, we systematically assessed the relationship between human and mouse aging by comparing age-related gene expression. We hope that this study, along with future work, will help researchers to justify the utilization of model organisms in research on aging and aging related diseases.

An early study of aging between species was performed by McCarroll et al. comparing transcriptional changes among C. elegans, D. melanogaster, Saccharomyces cerevisiae and Homosapiens, and showed that most of the changes associated with aging are species-specific, aging between C. elegans and D. melanogaster is highly conserved11. Khaitovich et al. analyzed gene expression in various brain regions of human and chimpanzees, and clarified that human and chimpanzee have significant differences in aging12. Zahn et al. provided the AGEMAP gene expression database and explored similar age-regulated genes and gene sets in different species: M. musculus, H. sapiens, D. melanogaster, and C. elegans, and it was eventually found that there was no overall correlation between mouse and human aging- related expression changes, similarity was only found in several specific gene sets13. And Yang et al. also showed that the aging genes were significantly different between human and mouse14. The results were not ideal since the datasets they used with small size of samples or with poor data quality.

In this work, we studied age-related genes in 18 tissues of human and mouse (see Fig. 1). We applied Deseq2 to perform differential expression analysis on the young and the old samples of 15 human tissues collected from GTEx database, and compared the results with DEGs of 15 mouse tissues studied by Wang et al. with CD algorithm15 from Gene Expression Omnibus(GEO) data16 (see Table 1), we also compared the DEGs of 3 pairs of human and mouse tissues studied by Wang et al. from GEO data15 (see Table 2). Then we performed functional analysis on these DEGs. Furthermore, we compared the aging DEGs in 18 tissues crosswise for human and mouse respectively, especially contrasted the four tissues associated with the brain. Since human and mouse DEGs are obtained by different algorithms, we applied CD and DESeq2 to analyze the DEGs of human Adipose_Subcutaneous respectively in order to compare the two algorithms.

Figure 1
figure 1

Workflow of the aging project.

Full size image
Table 1 The 15 tissues of human and mouse from GTEx database and GEO database respectively.
Full size table
Table 2 The 3 tissues of human and mouse from GEO database.
Full size table

Results

DEGs between young and old human samples from GTEx

GTEx Portal is a resource database generated from an analysis of RNA sequencing data of 1641 samples across 43 tissues from 175 individuals, built to help researchers study the relationship between genetic variation and gene expression in human tissues. In this paper, we used 15 human tissues RNA-seq datasets in GTEx for differential analysis (see Table 1). There are many methods for differential expression analysis of RNA-Seq data so far17,18,19,20,21,22,23. Anders et al. have proved that DESeq is the most conservative method among edgeR, DESeq, ShrinkSeq, NBPSeq, TSPM, voom + limma, vst + limma, baySeq, EBSeq and SAMseq20. But Love et al. clarified that DESeq2 is better than DESeq24. So, in this paper, for 15 human tissues from GTEx9, we used edgeR25,26, DESeq27 and DESeq224 to call differential genes in the young and old samples, and we call these DEGs as “age-related genes”. We found that DESeq2 is more sensitive than the other two methods and the number of age-related genes obtained by DESeq2 is the largest.

We summarized the number of age-related genes in 15 human tissues in Table 3. And these genes can be found in the Supplementary Dataset 1.

Table 3 Overview of differential expression analysis in 15 human and mouse tissues.
Full size table

DEGs between the young and the old samples from GEO data

GEO28,29,30 is a database provided by the National Center for Biotechnology Information (NCBI). In the study, in order to compare gene expression differences in young and old mice, we downloaded age-related genes expression profiles of multiple tissues in mouse from the GEO database29. Since these data are microarray data31, we used limma algorithm to perform differential expression analysis. However, the numbers of age-related genes were smaller than those derived by Wang et al.15, so we directly used the results of the DEGs they obtained. For the corresponding GSE (Series) information of each tissue, see Supplementary Table S1 and Supplementary Dataset 2 shows the detailed summary of DEGs in 15 mouse tissues (matching the tissues obtained from GTEx) from GEO database obtained by Wang Z et al.

Moreover, we also summarized the age-related genes of three pairs of human and mouse tissues (see Table 2) that are matched exactly from GEO database. Table 4 provides the numbers of age-related genes integrated in brain, retinal_periphery and hematopoietic_ stem_cell of human and mouse respectively. For a more detailed summary of age-related genes, see Supplementary Dataset 2.

Table 4 Overview of differential expression analysis in 3 human and mouse tissues from GEO database
Full size table

Comparison of human and mouse homologous age-related genes

To compare gene expression across mouse and human fairly, we restricted our genes in both species to homologous genes, or genes that are at least 80% similar in both species. Most homologous genes have the same or similar biological functions, and the regulatory pathways are similar. Homologous genes were selected using HOM_MouseHuman Sequence.rpt from MGI Data and Statistical Reports (http://www.informatics.jax.org/downloads/reports/index.html). More detailed information on these homologous genes can be found in Supplementary Dataset 3.

In column 6 of Table 3 and column 3 of Table 4, we showed homologous age-related genes in 18 human tissues, the numbers of which range from 1 to 6078. The numbers of homologous age-related genes in 18 mouse tissues range from 493 to 5215, as shown in column 9 of Table 3 and column 6 of Table 4.

The comparative analysis of human and mouse homologous age-related genes was mainly carried out from three perspectives:

Quantifying the overlap of human and mouse homologous age-related genes

The overlap of homologous age-related genes of 18 human and mouse tissues can be seen in column 10 of Table 3 and column 7 of Table 4 respectively, and the numbers of which range from 0 to 820. In kidney and small intestine, there aren’t overlapping homologous age-related genes between human and mouse.

The Fisher’s exact test

To get a statistically demonstration, we performed fisher’s exact test on homologous age-related genes of human and mouse 18 tissues. For example, in terms of human Liver and mouse liver, we used the total homologous genes of human and mouse as the background (14212), and made fisher’s exact test on aging genes of human Liver (108) and aging genes of mouse liver (4756) (Table S2). In Tables 3 and 4, we show the p-values of 18 pairs of tissues obtained by fisher’s exact test, and their adjusted p-values. We define tissues with adjusted p-value < 0.05 as tissues that are significantly correlated in human and mouse. There are 9 pairs of tissues that are significantly correlated, and the three pairs of tissues from GEO database are more similar. Also, we note that the three pairs of tissues data are all microarray data, and the same algorithm was used to analyze the DEGs.

Enriched functions of homologous age-related genes

In this section, we performed gene functional analysis with David32 on homologous age-related genes obtained from 18 pairs of human and mouse tissues, and adjusted enrichment p-values using a Benjamini-Hochberg procedure. Corrected p-values were considered significant if pBen < 0.05. We showed the top 10 enriched terms for every pair of tissues of human and mouse in Table S3, and detailed results can be found in Supplementary Dataset 4. The number of overlapping GO and KEGG terms33,34 in 18 human and mouse tissues ranges from 0 to 68 (see Table S4).

As shown in Table S3, the functional enrichment analysis revealed that homologous aging-related genes were significantly enriched in GO:0031012~extra cellular matrix between human Heart_Atrial_Appendage and mouse heart. And DR Sell et al. have proved that the extra cellular matrix undergoes progressive changes during senescence35. We also see that GO:0005615~extracellular space is enriched between human spleen and mouse spleen.

We also found that Phosphoprotein was the term of the homologous age-related genes enriched in Ovary, Brain_Cerebellum, Adipose_Visceral_ (Omentum), Lung, Heart_Left_Ventricle, Artery_Aorta, Muscle_Skeletal, Brain_Cortex, Brain_Hippocampus, brain and Adipose_ Subcutaneous significantly between human and mouse. And Kahn A et al. have declared that changes in cellular expression of phosphoprotein are linked to insulin resistance, tumor cell invasion, and cellular senescence36,37. And homologous age-related genes relating to the cytoplasm were significantly enriched in Ovary, Adipose_Visceral_(Omentum), Lung, Heart_ Left_Ventricle, Artery_Aorta, Muscle_Skeletal, Brain_Cortex and Brain_Hippocampus between human and mouse. Dou Z et al. have discovered that the cytoplasmic chromatin-cGAS -STING pathway promotes the senescence-associated secretory phenotype in primary human cells and in mouse38.

Crosswise comparison of homologous age-related genes between tissues

Here, we carried out pair wise comparison of homologous age-related genes of 18 tissues in human and in mouse separately. A more detailed summary of overlapping genes and fisher’s exact test p-values can be found in Supplementary Dataset 5.

When analyzing human homologous age-related genes, for Adipose_Visceralis, as an example, the tissue with the biggest overlap of homologous age-related genes is lung. Inomata et al. have found an association between the visceral adipose tissue level and lung function39. And excessive abdominal visceral fat contributes to increase plasma IL-6, which, in turn, is strongly associated with all-caused and cause-specific mortality in older persons with obstructive lung disease40,41. We also found that in the comparison of human 18 tissues, the two tissues with the highest number of overlapping DEGs are Muscle_Skeletal and Lung. This is consistent with the findings of Serres et al. who found that impaired skeletal muscle endurance in patients with chronic obstructive pulmonary disease was associated with altered lung function and reduction in associated physical activity42. Furthermore, the p-value obtained by fisher’s exact test indicates that the tissue most correlated with Adipose_Subcutaneous is Muscle_Skeletal (2.793932e-55), and Brain_Cortex is significantly correlated with Brain_Hippocampus (8.349845e-220).

In terms of 18 mouse tissues, for neocortex, the tissue with the biggest number of overlapping homologous age-related genes is Hippocampus, the overlapping number is 849 and the p-value of fisher’s exact test is 1.169441e-199. This result is consistent with human.

Comparison of homologous age-related genes in human Brain_Cerebellum, Brain_Cortex, Brain_Hippocampus and brain (from GEO)

Here, we did a more in-depth study of the four tissues associated with human brain: Brain_Cerebellum, Brain_Cortex, Brain_Hippocampus and brain (from GEO). 39 homologous age-related genes are overlapped in these four tissues (see Table S5). Biological interpretation of these DEGs was carried out using ClueGO v2.5.143 in Cytoscape44, we reserved the terms with p-value < 0.05 (see Fig. 2), and got 56 overlapping terms (see Table S6).

Figure 2
figure 2

Functional groups in ClueGO Overview. ClueGO analysis of DEGs in Brain_Cerebellum, Brain_Cortex, Brain_Hippocampus and brain from human donors. Overview chart with functional groups including specific terms for DEGs. The percentage of genes per term is shown in each group.

Full size image

Comparison of homologous age-related genes in mouse cerebellum, neocortex, hippocampus and brain

Similarly, we made a further comparison of the four tissues associated with mouse brain: the cerebellum, neocortex, hippocampus and brain. There are just 8 overlapping age-related DEGs among these four tissues (see Table S5). As the studying process of human brain, the results of mouse brain biological interpretation are in Fig. 3, and there is no overlapping terms among these four tissues in mouse.

Figure 3
figure 3

Functionally grouped networks on cerebellum, neocortex, hippocampus and brain for mouse. Functionally grouped network with terms as nodes linked based on their kappa score level (≥ 0.4), where only the label of the most significant term per group is shown. Each node in the figure represents a term, and the node size represents the term enrichment significance. Functionally related groups partially overlap. The connection between the nodes reflects the correlation between the terms, and the color of the node reflects the enrichment classification of the node.

Full size image

It is worth noting that GFAP appears in both human and mouse overlapping DEGs list. Middeldorp et al. have already proved that the astrocytic cytoskeleton protein GFAP plays role in many processes in the brain, and they discussed the versatility of the GFAP cytoskeletal network from gene to function with a focus on astrocytes during human brain development, aging and disease45. Furthermore, GFAP in Cerebrospinal Fluid (CSF) serves as a potential biomarker of Alexander disease that is comparable between mouse models and human patients46.

Comparison of CD and Deseq2 methods

In order to compare the two methods of CD and Deseq2, we performed differential expression analysis on young and old samples of human Adipose_Subcutaneous tissue using CD and Deseq2 methods respectively. We found overlapping 637 out of the top 2000 DEGs in both CD and Deseq2. That is 32% of the top DEGs were identified using both methods.

Discussion

In the comparison of age-related genes in multiple tissues of human and mouse, we used GTEx data and more sensitive algorithms than the previous studies, and we found 9 pairs of tissues were significantly correlated between human and mouse on aging. The results were similar to those of Zahn13 and Yang14.

By functional enrichment analysis of DEGs, we have found some terms related to aging, such as GO:0031012~extracellular matrix35, Phosphoprotein36,37, Cytoplasm38, Cell cycle, Cell division, ATP-binding and GO:0005515~protein binding et al.

When we performed a crosswise comparison of 18 tissues in human and mouse respectively, we found that the human Brain_Cortex aging is significantly associated with Brain_Hippocampus aging, which was also found in mouse. Next, we focused on comparing four brain-related tissues in human and mouse, and found a gene–GFAP–related to aging in both human and mouse.

Since human and mouse DEGs are obtained by different algorithms, it is necessary to parallel the two methods over the same dataset to make sense of the impact of technical error. So we applied CD and Deseq2 to analyze the DEGs of human Adipose_Subcutaneous respectively. Also, because we only focused on the overlapping of aging genes in human and mouse, we were not positioned to identify human-specific gene expression changes related to aging. More research is needed to find human specific pathways and mechanisms that contribute longer lifespan in human47.

Materials and Methods

Data collection

We downloaded human multi-tissue gene expression data from the Genotype-Tissue Expression (GTEx) Portal (https://www.gtexportal.org/home/). And two age-related differential expression gene data from Enrichr (http://amp.pharm.mssm.edu/Enrichr/#stats). These two datasets are Aging_Perturbations_from_GEO_down and Aging_Perturbations_ from_GEO_up which are obtained by applying CD algorithm48 to the GEO data (https://www.ncbi.nlm.nih.gov/geo/) to analyze the age-related genes. Comparisons between human and mouse DEGs were based on homologous genes which used HOM_ MouseHumanSequence.rpt obtained from MGI Data and Statistical Reports (http://www.informatics.jax.org/downloads/reports/index.html).

Matching of tissues

We matched 15 tissues between GTEx data and GEO data, and then compared the DEGs related to aging between human and mouse. In addition, in terms of the GEO data itself, we found three additional human and mouse tissues which are matched. So we collected 15 human tissues from GTEx data, 3 human tissues from GEO data, and 18 mouse tissues corresponding to human tissues from GEO data (see Table 1 and Table 2).

Data pre-processing

We restricted GTEx RNA-seq tissue-wide expression data to individuals who were 30 or under (young), and 65 or over (old), and removed genes that had either 0 or 1 read in minimal pre-filtering.

Differential gene expression analysis

We applied Deseq2 to identify age-related genes in humans24,49. Deseq2 algorithm has two requirements of inputting data: (1). Deseq2 requires that the input data be a matrix of integers, and (2). the matrix is not standardized. It is worth noting that Deseq2 has its own strategy for calculating the scaling factors. For data visualization purposes, we log transformed our data, and added a pseudo count to avoid undefined values. Deseq2 provides two types of transformation methods for count data: regularized-logarithm transformation (rlog24) and variance stabilizing transformation (VST27). Both transformations produce transformed data on the log2 scale which has been normalized with respect to library size or other normalization factors24. Usually, rlog is used when the data set is less than 30, VST is used for large data sets, and the most appropriate one is automatically selected during the Deseq2 analysis process (Fig. S1). Then, we used the negative binomial distribution to calculate the statistical significance (p-values) among all genes across datasets50, and FDR corrected using the Benjamini-Hochberg method51,52,53. Genes were considered differentially expressed if their adjusted p-value < 0.05.

For GEO data, DEGs are obtained by the CD algorithm48. In this paper, we directly used the DEGs on GEO data obtained by Wang et al.15.

The Fisher’s exact test

For each pair of tissues, the statistical significance of the difference between human aging genes and the mouse aging genes was assessed by fisher’s exact test54,55,56. P-values were corrected for multiple-hypothesis testing using Benjamini-Hochberg correction51, with a significance threshold of adjusted p-value < 0.05.

Gene function enrichment analysis

In this paper, DEGs were annotated by David tools (V6.7) (DAVID; http://david.abcc.cifcrf.gov/)32,57 and ClueGO v2.5.143 in Cytoscape44. In these two analyses, we adopted the threshold p-value < 0.05.