Tissue, age, sex, and disease patterns of matrisome expression in GTEx transcriptome data

Abstract

The extracellular matrix (ECM) has historically been explored through proteomic methods. Whether or not global transcriptomics can yield meaningful information on the human matrisome is unknown. Gene expression data from 17,382 samples across 52 tissues, were obtained from the Genotype-Tissue Expression (GTEx) project. Additional datasets were obtained from The Cancer Genome Atlas (TCGA) program and the Gene Expression Omnibus for comparisons. Gene expression levels generally matched proteome-derived matrisome expression patterns. Further, matrisome gene expression properly clustered tissue types, with some matrisome genes including SERPIN family members having tissue-restricted expression patterns. Deeper analyses revealed 382 gene transcripts varied by age and 315 varied by sex in at least one tissue, with expression correlating with digitally imaged histologic tissue features. A comparison of TCGA tumor, TCGA adjacent normal and GTEx normal tissues demonstrated robustness of the GTEx samples as a generalized matrix control, while also determining a common primary tumor matrisome. Additionally, GTEx tissues served as a useful non-diseased control in a separate study of idiopathic pulmonary fibrosis (IPF) matrix changes, while identifying 22 matrix genes upregulated in IPF. Altogether, these findings indicate that the transcriptome, in general, and GTEx in particular, has value in understanding the state of organ ECM.

Figure 1

A transcriptomic and proteomic overview of the matrisome. (a) Four violin plots of 10,000 permutations of gene-protein Spearman’s correlations. The N of proteins per tissue is determined by how many protein/gene pairs were detected for each of the 4 matrisome classes. The black bars are the Spearman’s ρ correlation for the matrisome protein-gene pairs. (b) A heatmap correlation of matrisome proteins (top left) and gene mRNAs (bottom right) between tissues showing stronger correlations among proteins. (c) A Kendall’s τ correlation heatmap of each tissues’ median expression of matrisome genes. Multiple brain region datasets are collapsed into one tissue by using the median expression of each gene across these samples. GTEx abbreviations are used. (d) The median percent transcriptome of matrisome categories for each tissue. The colored bars each represent a different matrisome division.

Figure 2

Age-associated transcriptomic changes of the matrisome. (a) A plot indicating the 32 transcripts with the largest change in expression relative to age. The transcript with the smallest FDR-adjusted p-value per tissue is noted along with all other instances of that transcripts having a β estimate ≥ 0.15 in other tissues. Tissues with no significant findings (FDR-adjusted p > 0.05) were not included. (b) Representative images of the transverse colon with varying levels of multi-tissue ADIPOQ expression at two different powers. The left sample, GTEX-14C5O, had low levels of ADIPOQ and is notable for a thin submucosa (two-headed arrow) with few adipocytes. The right sample, GTEX-1KXAM, had high levels of ADIPOQ, with a notable submucosal area containing large numbers of adipocytes. The bars are 1 mm and 200 µm for the low and high power views respectively. (c) A scatterplot representing the percent of adipose tissue in the submucosa of the transverse colon for the top 6 and bottom 8 individuals ordered on body wide ADIPOQ expression. The datapoints (n = 2 to 6, median of 6, per individual) of multiple measured areas per sample are colored to represent the top (blue) or bottom (yellow) multi-tissue ADIPOQ expressors. Bars indicate the median level of submucosal adipocytes for each individual.

Figure 3

Sex based differences in the matrisome transcriptome. (a) A plot of sex DE genes. The transcript with the smallest p-value per tissue is noted along with all instances of that transcript having a significant β estimate in other tissues. Tissues with no significant findings were not included. (b) The sex-related genes that are DE in ADPSBQ and ADPVSC. Notably, ADPSBQ has 81 DE genes while ADPVSC has only 10. (c) The sex-related genes that are DE in skin not sun exposed (SKINNS) and skin sun exposed (SKINS). The size of the dot indicates the β estimate. The tissues are mostly congruent, having 17 shared genes with the same expression directionality. (d) A barplot comparing the Z score for the ADPSBQ ADP-Fib.1 gene cluster between males and females for both ADPSBQ and ADPVSC tissues. In ADPSBQ, which has more DE sex genes, cluster ADP-Fib.1 associates with the male sex. This cluster shares 11 genes with the DE gene list in (b). In ADPVSC, there is no sex difference. *=1.96e−05 (e) Adipose samples from two individuals showing differing levels of fibrosis. The top samples are from GTEX-IGN73 and the bottom samples are from GTEX-1K2DA. The left samples are ADPSBQ and the right samples are ADPVSC. The bar is 200 um. (f) The median percent adipose tissue from ADPSBQ (n = 34; 16F/18M) and ADPVSC n = 34; 16F/18M) samples, with the ADP-Fib.1 Z-score on the X axis. This data shows that the gene transcripts in the ADP-Fib.1 cluster positively correlate with adiposity and negatively correlate with fibrosis in ADPSBQ tissue.

Figure 4

Commonly upregulated and downregulated matrisome gene transcripts. (a) A histogram of transcript rank changes between combined TCGA normal samples (n = 110) and GTEx samples (n = 374) compared to TCGA cancer samples (n = 542) in lung. The histogram is colored on the decile of rank change. (b) A histogram of the sum of normal to cancer rank changes for gene transcripts across lung, breast, colon thyroid and prostate (n = 443). The histogram is colored on the decile of rank change. (c) A heatmap of the top and bottom rank-changed matrisome gene transcripts between normal tissue and cancer. The rank change order is split into deciles, (1—1st decile to 10—10th decile). NAs are represented by black boxes and indicate no expression in the normal or cancer samples.

Figure 5

Idiopathic Lung Fibrosis characterization using GTEx. (a) A heat map representing the centered normalized expression of matrisome genes in clusters LUNG-Fib.1 or LUNG-Fib.2 between lung samples. The x-axis of the lung samples indicates disease, batch, tobacco usage, sex, and age status of each. Age and being in batch Sivakumar—2 significantly correlate with IPF samples (logistic regression; p = 0.012; p = 0.029 respectively). (b) A violin-sina plot of the normalized score of gene expression cluster LUNG-Fib.1 and LUNG-Fib.2 comparing IPF, acute lung injury, ventilator injury, and normal lung samples from Sivakumar et al. The color of each data point indicates the batch.