voyAGEr profiles tissue-specific age-associated changes in gene expression and their differences between sexes.

For each of the 48 human tissues in GTEx, gene expression was linearly modelled in windows spanning 16 years centered in consecutive years of age, to estimate the effects thereon of Age, Sex and the interaction between them (Age&Sex), i.e., how the Sex effect changes with age, equivalent to how the Age effect differs between sexes (v. Materials and Methods). In each window (i.e., for each age in consecutive years), the percentage of genes with expression significantly altered by of each of those effects gives their respective transcriptomic impact (upper panels). voyAGEr thereby identifies the age periods at which major gene expression changes occur in each tissue. For example, in subcutaneous adipose tissue: major age-related transcriptional alterations are found at 30 and around 50 years of age (upper left panel), illustrated by the behavior of LMO3 (bottom left panel); major gene expression differences between males and females happen at around 60 years of age (upper center panel), as illustrated by GREB1 (bottom center panel); major differences between sexes in age-related gene expression alterations happen at around 50 years of age (upper right panel), as illustrated by ADAM33 (bottom right panel). Solid loess lines in the bottom panels (green for all donors, pink for females, blue for males). Gene expression (GE) in log2 of counts per million (logCPM).

Gene-centric analyses of expression alterations across age

A. Heatmap of MKI67 expression across tissues over age.

B. Heatmap of significance of Age-associated MKI67 expression alterations over age. p-values are for the moderated t-statistics of differential gene expression associated with the Age effect (v. the ShARP-LM approach in Materials and Methods). Notably, major transcriptional changes are observed in transverse colon (mid 20’s and after 55), breast (mid/late 40’s and early/mid 50’s), and liver (mid 50’s).

C. voyAGEr’s Gene tab interface. MKI67 expression in transverse colon is inspected. Donors’ information is shown in a table and the scatter plot can be interactively adjusted according to the donors’ condition of interest (Supplementary Figure 1B).

Tissue-specific assessment of gene expression changes across age

A. Heatmap of significance (FDR based on random permutations of age, v. Materials and Methods) over age of the proportion of genes with expression significantly altered with Age in the 48 analyzed tissues.

B. Exploration of gene expression changes across age in transverse colon.

a. Percentage of genes with significantly altered expression with Age over age. Three peaks at 27, 55 and 62 years old are noteworthy.

b. Clicking on the dot of a specific age (62.44 y.o. in plot a) gives access to the list of the most differentially expressed genes at that age, ordered by statistical significance of expression changes (p-value of moderated t-statistic).

c. Plot of expression of the chosen top gene in the table in b across age (bottom) in parallel with the significance of its expression alterations with Age. The expression of HOXA-AS3 significantly decreases at around 62 years old, concomitantly with the third peak of transcriptomic changes with Age.

Tissue-specific assessment of pathway expression changes across age

A. Heatmap of enrichment (above), given by GSEA’s normalized enrichment score (NES), of Reactome pathways in genes differentially expressed with Age over age. Pathways are classified into broad families (upper bar), whose associated biological processes and functions can be learnt from the word clouds of their most frequent terms (below). Only Reactome pathways found significantly associated (FDR ≤ 0.01) with gene expression alterations in at least one age window are included.

B. Enrichment, given by the significance of Fisher’s tests, of a user-provided gene set in genes differentially expressed (based on a user-defined p-value threshold) with Age over age. Here, the given gene set is composed of genes from Senequest 42 whose link with senescence is supported by at least 4 sources. The first and second peaks of Age-associated transcriptomic changes in the transverse colon appear to be enriched in senescence-related genes.

Tissue-specific assessment of age-associated progression of modules of co-expressed genes

A. Heatmap of eigengene expression for all the inferred modules of co-expressed genes in the heart’s left ventricle over age.

B. Scatter plot (above) of eigengene expression over age, in all left ventricle samples, for a selected module of 11 genes. The eigengene expression is derived from the 1st component of the single value decomposition of its genes’ expression. This module was analysed, with Fisher’s tests, for enrichment in cell types, based on markers from the literature (Skelly 84, Cui 83 and He 82, surnames of the first authors of the studies the markers are retrieved from) and found to be associated with fibroblasts, as revealed by the TreeMap below (where each rectangle’s area and darkness are proportional to the significance of its association with a cell type and its color linked to the markers’ source study).

C. Heatmap of association of the modules with five selected diseases, computed with the DOSE package 57.

voyAGEr’s Gene tab interface.

A. Heatmap of CDKN2A expression across tissues over age.

B. PCNA expression in transverse colon over age and its association with the donor’s history of pneumonia.

Sex-specific SALL1 (A) and KAL1 (B) expression alterations over age.

Scatter plots of SALL1 (A) and KAL1 (B) expression over age (above), where dots are colored based on the donor’s sex (male in blue and female in pink), are shown together with those of the significance of their difference between sexes (below).

Tissue-specific sex-differentiated expression.

A. Heatmap of significance over age, for each tissue, of the proportion of differentially expressed genes between sexes (left), in parallel with the number of sex-differentiated genes highlighted by Gershoni et al. (v. Figure 1 in 9) (right). Mammary is corroborated as the tissue with the most sex-differentiated gene expression. For tissues like adipose, skeletal muscle, skin, and heart, sex-related biases in expression mainly arise in late life.

B. Scatter plot relating, for each tissue (each dot), the number of sex-differentiated genes found by Gershoni et al. with the number of genes found by voyAGEr to be differentially expressed between sexes in at least 25 age intervals (left). This correlation is significant regardless of the considered minimum number of age intervals a gene must be differentially expressed in (right). Half visible dots in the left scatter plot represent tissues found to have no sex-differentiated genes by voyAGEr (the log scale would give them a minus infinite value on the Y-axis).

C. and D. Two examples of genes, MUCL1 (C) and NPPB (D), whose expression is known to be sex-differentiated in a tissue-specific manner (Figures S7 and 3 in 9, respectively). Plots of their sex-specific expression across age in parallel with the significance of their differences in expression alterations between sexes are shown for tissues reported in 9 as having sex-differentiated gene expression: breast and skin for MUCL1 and heart for NPPB. voyAGEr helps to refine the definition of the age periods at which those biases occur: late-life in breast and early-life in skin for MUCL1, and late-life in heart for NPPB.

Distributions of GTEx donors’ ages and its impact on the statistical power of detection of differential gene expression.

For thyroid as an example tissue, the sex-differentiated distribution of GTEx samples’ ages (v7) (above) are shown together with the percentage of differentially expression genes (% DE genes) over age (below) for the three variables (Age, Sex and Age&Sex), as well as their statistical significance, given by the ShARP-LM approach.

Principle of the ShARP-LM method.

ShARP-LM consists of fitting, for each tissue, linear models to the gene expression of samples from donors with ages within windows with a range of 16 years shifted through consecutive years of age, in a sliding window with window size = 16 and step size = 1 years of age. As samples at the ends of the dataset’s age range would be thereby involved in fewer linear models, we made the window size gradually increase from 11 to 16 years when starting from the “youngest” samples and decrease from 16 to 11 years when reaching the “oldest”. The shown example to illustrate the approach is the expression of LMO3 in subcutaneous adipose tissue over age, with 8 age windows are shown. Colored rectangles represent the age windows and colored lines the linear modelling of the gene expression over age in each of them.

Clustering of pathways from Reactome and KEGG, and level 3 Gene Ontology Biological Processes.

Based on their common genes (v. Materials and Methods), pathways from curated databases were clustered in 10 families.

Number of studied genes per tissue.

For each tissue, we kept the genes with at least 1 CPM of expression in at least 40% of the samples.