Transcriptome features of striated muscle aging and predictability of protein level changes

Abstract

We performed total RNA sequencing and multi-omics analysis comparing skeletal muscle and cardiac muscle in young adult (4 months) vs. early aging (20 months) mice to examine the molecular mechanisms of striated muscle aging. We observed that aging cardiac and skeletal muscles both invoke transcriptomic changes in innate immune system and mitochondria pathways but diverge in extracellular matrix processes. On an individual gene level, we identified 611 age-associated signatures in skeletal and cardiac muscles, including a number of myokine and cardiokine encoding genes. Because RNA and protein levels correlate only partially, we reason that differentially expressed transcripts that accurately reflect their protein counterparts will be more valuable proxies for proteomic changes and by extension physiological states. We applied a computational data analysis workflow to estimate which transcriptomic changes are more likely relevant to protein-level regulation using large proteogenomics data sets. We estimate about 48% of the aging-associated transcripts predict protein levels well (r ≥ 0.5). In parallel, a comparison of the identified aging-regulated genes with public human transcriptomics data showed that only 35-45% of the identified genes show an age-dependent expression in corresponding human tissues. Thus, integrating both RNA-protein correlation and human conservation across data sources, we nominate 134 prioritized aging striated muscle signatures that are predicted to correlate strongly with protein levels and that show age-dependent expression in humans. The results here reveal new details into how aging reshapes gene expression in striated muscles at the transcript and protein levels.

Figure 1:. Transcriptome changes in young adult mice vs. early aging mice.

A. Principal component analysis of normalized gene counts in the left ventricle (top) and quadriceps femoris (bottom) show separation by sex and age. B. Enriched pathways in aged vs. adult gene expression in two tissues. C-D. Normalized read counts showing sex and age expression among selected differentially expressed aging genes in (C) the heart and (D) the skeletal muscle (FDR <5% with the exception of Camk2b where FDR <10%).

Figure 2:. Shared aging signatures in aged cardiac and skeletal muscle.

A. Bar chart showing the adjusted P values (x-axis) and log2 fold-changes (log2FC; fill color) of the 20 aging associated genes identified in both tissues. B. Scatter plot showing a comparison of fold changes (aged vs. young adult) and a robust positive correlation (Spearman’s correlation coefficient ⍴ 0.65, P 0.0032) between the two tissues. X-axis: log2 fold-change in skeletal muscle; y-axis: log2 fold-change in the heart. Error bars: standard error of log2 fold-change; line: best-fit linear curve.

Figure 3:. Predictability of protein-level from across-sample transcript variance.

To estimate whether the quantified transcript changes might translate to the proteome, we considered the predictability of protein levels from their proxy transcripts on a gene-wise basis in large proteogenomics data sets. A. An elastic net is applied to 717 samples with matching transcriptomics and mass spectrometry data in the CPTAC collection. The average correlation (top) and R² values (bottom) between predicted and actual protein levels across samples in each of 10,693 genes are shown. B. Examples of an aging signature whose protein abundance across samples is well predicted by its proxy transcript (Anxa1) in matching samples and one that is poorly predicted (Uqcrc1). Each data point is one CPTAC sample. Brown: train set; blue: test set. C. Significantly enriched Reactome terms among transcripts that predict protein well (r ≥ 0.5) or poorly (r < 0.5). Size: protein count in pathway; fill color: adjusted P value; x-axis: gene annotation ratio. D. Scatter plot showing a robust correlation between the modeled gene-wise protein predictability here using CPTAC data with the Spearman’s correlation coefficient values between protein and RNA across 32 tissues in GTEx (r: 0.32, P: 2.6e–220). The modeled protein predictability values predict strong RNA-protein correlation (⍴ ≥ 0.5) in normal human GTEx tissues with an AUC of 0.69 (inset). E. Boxplot showing a breakdown of binned correlation values against GTEx correlation, the correlation plateaus at r ≥ 0.5 which may be due to potential overfitting or cross-sample biological differences.

Figure 4:. Correlation between RNA and protein levels from identical tissues.

A. Scatter plot showing the within-sample across-gene comparisons in the heart (left) and the skeletal muscle (right) for commonly quantified RNA and their proteins. Fill color: data frequency within bin. B. Scatter plot showing across-sample comparisons between RNA and proteins in the aging vs. young adult heart (left) and skeletal muscle (right). C. Log fold-change comparison at the RNA (x-axis) and protein (y-axis) level among commonly quantified proteins and transcripts with significant age-associated transcript level differences. Line: best-fit linear curves for the heart (red) and the skeletal muscle (blue). Error bars: standard errors of logFC.

Figure 5:. Conserved age-expression profiles of selected signatures in humans.

Box plots showing examples of GTEx v8 human normalized RNA expression levels across age groups in decadal brackets in four GTEx v8 human tissues (left to right) heart left ventricle, heart atrial appendage, skeletal muscle, and kidney cortex for A. Efemp1, B. Fkbp4, and C. Sod3. P values: ANOVA. Asterisks within plots denote Tukey’s post-hoc for individual group comparison. *: Tukey P < 0.05; **: P ≤ 0.01; ***: P ≤ 0.001; ****: P ≤ 0.0001.

Figure 6:. Prioritized age-associated signatures.

A. List of 134 prioritized aging signatures in the heart, skeletal muscle, and those common to both tissues. Top: paired bars represent −log10 P.adj in the heart and the skeletal muscles, respectively. B-D. The prioritized signatures had CPTAC RNA-protein correlation r ≥ 0.5 and ANOVA P ≤ 0.01 in GTEx v8 transcript expression against age groups in GTEx v8 heart left ventricle or skeletal muscle transcriptomes. X-axes correspond to panel A.

Figure 7. Non-coding RNA signatures in aging striated muscles.

A. Bar charts of differentially expressed annotated non-coding RNAs in the heart (top) and the skeletal muscle (bottom). Colors denote Gencode vM25 annotation gene biotype. B-C. Examples and genome tracks of two long non-coding RNAs Plet1os (B) and Foxo6os (C) that are differentially expressed in aging skeletal muscle.