Source-linked AI summary
Post-transcriptional regulation across human tissues
Alexander Franks, Edoardo Airoldi, Nikolai Slavov
TL;DR
The paper asks how transcriptional and post-transcriptional regulation shape human tissue protein levels. It finds that scaled mRNA levels predict overall protein levels well, but protein levels for individual genes across tissues are poorly predicted, implicating tissue-specific post-transcriptional regulation.
Problem
Human tissue identity depends on protein levels, but the relative contributions of transcriptional and post-transcriptional regulation remain contested.
Method
The study compares scaled mRNA levels with protein levels across human tissues and examines protein-to-mRNA ratios.
Results
Overall protein levels are well predicted by scaled mRNA levels, whereas individual genes’ protein levels across tissues are poorly predicted by mRNA levels.
Takeaways & Limitations
Tissue-specific proteomes involve substantial post-transcriptional regulation, so mRNA fold-changes may not reliably estimate protein fold-changes between cell types.
Takeaways & Limitations
The data cannot accurately quantify the contributions of different regulatory factors because of collection and handling effects.
Abstract
from arXiv · showhide
Transcriptional and post-transcriptional regulation shape tissue-type-specific proteomes, but their relative contributions remain contested. Estimates of the factors determining protein levels in human tissues do not distinguish between (i) the factors determining the variability between the abundances of different proteins, i.e., mean-level-variability and, (ii) the factors determining the physiological variability of the same protein across different tissue types, i.e., across-tissues variability. We sought to estimate the contribution of transcript levels to these two orthogonal sources of variability, and found that scaled mRNA levels can account for most of the mean-level-variability but not necessarily for across-tissues variability. The reliable quantification of the latter estimate is limited by substantial measurement noise. However, protein-to-mRNA ratios exhibit substantial across-tissues variability that is functionally concerted and reproducible across different datasets, suggesting extensive post-transcriptional regulation. These results caution against estimating protein fold-changes from mRNA fold-changes between different cell-types, and highlight the contribution of post-transcriptional regulation to shaping tissue-type-specific proteomes.
Author Summary
The paper separates variation between different proteins from variation within the same protein across tissues. It finds that mRNA levels predict overall protein abundance but poorly predict a given protein’s tissue-specific levels, implicating post-transcriptional regulation.
- The study asks whether human tissue protein levels are set mainly by corresponding mRNA levels or by post-transcriptional mechanisms.
- For an individual gene, protein levels across tissues are poorly predicted by its mRNA levels, suggesting tissue-specific post-transcriptional regulation.
- Overall protein levels are well predicted by scaled mRNA levels.
- These apparently contradictory findings reflect two sides of Simpson’s paradox.
Introduction
The introduction distinguishes mean-level variability between proteins from across-tissues variability within proteins, because genome-wide mRNA–protein correlations conflate biologically distinct sources of variation. The paper therefore separately evaluates transcriptional and post-transcriptional contributions across human tissues.
- Absolute mRNA–protein correlations mix variation between proteins, variation across conditions or cell-types, and measurement or technological error.
- Mean-level variability describes differences between the average abundances of different proteins and their corresponding mRNAs.
- Across-tissues variability describes variation within the same protein across different tissue-types, cell-types, or physiological conditions.
- The study separately quantifies transcriptional and post-transcriptional contributions to both variability sources across human tissues.
- Much mean-level protein variability is explained by mRNA levels, whereas across-tissues protein variability is poorly explained by mRNA levels.
- Some unexplained across-tissues variance is reproducible across datasets, supporting post-transcriptional regulation as a significant contributor to tissue-specific proteomes.
Results
The analysis separates mean-level variability between proteins from across-tissues variability within proteins, showing that aggregated mRNA–protein correlations can obscure tissue-specific regulation. Relative protein-to-mRNA ratios instead reveal substantial, functionally coordinated across-tissues post-transcriptional regulation, although measurement noise limits precise attribution.
- Separating variability sources: Scaled mRNA explains a large fraction of total protein variance, but this conflates between-protein mean-level variability with within-protein across-tissues variability.The conflation can produce large positive aggregate correlations even when within-gene tissue trends are near zero.
- Separating variability sources: Across-tissues correlations between scaled mRNA and protein levels are not representative of corresponding within-gene mRNA–protein correlations across tissues.Across datasets, low or negative within-gene correlations coexist with very high conflated correlations.
- Separating variability sources: Across-tissues variability spans roughly 2–10-fold, far below the 10^3–10^4-fold abundance range across proteins, yet it defines tissue-type biological identity.The analysis therefore focuses on factors regulating across-tissues protein variability.
- Reliability and regulation: Measurement noise from sample collection and assay error substantially constrains estimates of transcriptional versus post-transcriptional contributions.Noise is largely study-dependent in both mRNA and protein measurements, while relative same-gene quantification can minimize systematic biases.
- Reliability and regulation: Approximately 50% of across-tissues protein variance is consistent with mRNA-linked transcriptional regulation and approximately 50% with post-transcriptional regulation.The average across-tissues mRNA–protein correlation was 0.29, corresponding to R2 = 0.08 before reliability correction.
- Coordinated post-transcriptional regulation: Relative protein-to-mRNA ratios vary across tissues in functionally enriched and reproducible patterns, including higher ribosomal-protein ratios in kidney than stomach.rPTR correlations are reproducible for most tissues, with reported values of ρ = 0.7–0.8 for many tissues and weaker values for others.
Discussion
Across human tissues, mRNA levels explain much of the differences between protein abundances but poorly explain variability in the same protein across tissues. Reproducible, functionally concerted protein-to-mRNA ratios instead support substantial, context-dependent post-transcriptional regulation, although measurement noise limits quantitative attribution.
- Highly abundant proteins have highly abundant mRNAs, indicating transcription sets much of the overall protein abundance range.
- Across-tissues protein variability is poorly explained by mRNA variability, implicating post-transcriptional regulation.
- Measurement noise and systematic sample-collection and handling biases limit accurate quantification of regulatory contributions across tissues.
- Merged mRNA isoforms and protein proteoforms limit distinct isoform-level quantification in peptide- and short-read-based approaches.
- Strong enrichment of protein-to-mRNA ratios within functional gene sets demonstrates functionally concerted post-transcriptional regulation.
- Kidneys show high protein-to-mRNA ratios for energy-production gene sets, consistent with post-transcriptional regulation supporting their high energy demands.
- The relative contributions of transcriptional and post-transcriptional regulation vary substantially with the tissues compared.
- Tissue-type-specific proteomes may involve more post-transcriptional regulation than stimulation that preserves cell identity.
Methods
The study combines large-scale RNA-seq and shotgun mass-spectrometry measurements across twelve human tissues, with normalization, reliability correction, and gene-set analyses used to assess mRNA–protein relationships and tissue-specific protein-to-mRNA ratios.
- Data: RNA-seq and shotgun mass-spectrometry datasets measured 6,104 genes across twelve human tissues, while approximately 8% of mRNA and 40% of protein measurements were missing.A targeted validation dataset contained 33 proteins across five tissues.
- Normalization: Measurements were median-aligned across tissues to correct multiplicative differences, using the first tissue as an arbitrary baseline and applying the same normalization to mRNA and protein.Median alignment was chosen because medians are more robust to outliers than means.
- Scaling mRNA levels: After normalization, log PTR ratios were defined as normalized log protein minus log mRNA, with residual ratios representing tissue-specific post-transcriptional regulation and measurement noise.Scaled mRNA was defined by adjusting mRNA with median PTR ratios to estimate mean protein levels.
Tables
The tables and figures document dataset coverage, reliability, tissue-specific rPTR variation, and the limits of using scaled mRNA to explain across-tissues protein variability.
- Dataset construction: The consensus dataset has the highest proteome coverage and best agreement with the validation dataset.It was constructed by merging datasets [20] and [21].
- Variance explained: Scaled mRNA improves overall protein-level correlation but does not indicate across-tissues predictive accuracy.Figure 1 reports RT = 0.89 after scaling, while individual-gene across-tissues correlations can remain RP ≈0.
- Measurement reliability: Reliability estimates are obtained independently within and across studies for relative mRNA and protein measurements.Protein reliability uses non-overlapping peptide subsets, while mRNA reliability compares measurements across subjects and datasets.
- Functional rPTR variation: Functional gene sets show significant tissue-specific rPTR variation, including higher kidney rPTR for ribosomal, NADH dehydrogenase, and respiratory proteins.Kidney effects and contrasting stomach effects are reported at FDR < 1%, with estimates summarized on a log10 scale and reproduced across studies.
- Consensus validation: The consensus protein dataset averages noise across source datasets and provides more reliable tissue-proteome quantification.Figure 4 compares correlations with two mRNA datasets and mean squared error against targeted validation measurements.
- Reproducibility: Relative rPTR estimates for GO terms and genes reproduce across independent datasets with confidence intervals reported in the supplemental tables.Table S1 covers GO terms, and Table S2 covers genes.