Age-associated increases in inter-individual gene expression variability across human tissues

This study analyzes nearly 1,000 individuals across 30 human tissues to reveal that aging is characterized not only by coordinated transcriptional changes but also by a significant increase in inter-individual gene expression variability driven by distinct biological pathways and shaped by local gene regulatory networks.

The Gist

Imagine your body as a massive, bustling city with billions of workers (cells) and millions of instructions (genes) telling them what to do. When you are young, this city runs like a well-oiled machine. The workers follow the instructions perfectly, and everyone in the city does roughly the same thing at the same time.

As the city ages, two things start to happen. Scientists have known about the first thing for a long time, but this paper shines a light on the second, which is just as important.

Here is the story of what happens as we age, explained simply:

1. The Old Way of Looking at Aging: "The Volume Knob"

For years, scientists studied aging by looking at the volume of the instructions. They asked: "Is the instruction to 'build muscle' turned up louder or turned down quieter as we get older?"

These are called Differentially Expressed Genes (DEGs). It's like noticing that the city's traffic lights have been set to "Red" more often than "Green" as the years go by. This is a coordinated, predictable change. The whole city shifts in the same direction.

2. The New Discovery: "The Static on the Radio"

This paper introduces a new concept called Differentially Variable Genes (DVGs). Instead of asking if the volume changed, the researchers asked: "Is the signal getting fuzzy?"

Imagine you are listening to a radio station.

• Young City: Every radio in the city hears the exact same clear song.
• Old City: The song is still playing, but now, every radio is hearing it slightly differently. One radio has a bit of static, another is slightly out of tune, and a third is skipping. The average song might be the same, but the consistency is gone.

The researchers found that as humans age, our genes become "noisy." Some people's cells follow the instructions perfectly, while others' cells get confused or act randomly. This "fuzziness" or variability accounts for a huge chunk of what happens when we age (about 15% of the total changes, and nearly 8% just due to getting older).

3. The "Stability Score" (The City's Quality Control)

To measure this, the scientists invented a new tool called the Gene Stability Score (GSS). Think of this as a "Quality Control Inspector" for the city.

• High Score: The gene is a reliable worker. It does the same job, every time, in every person. (Examples found: TBP, PUM1, TMEM199).
• Low Score: The gene is a "jittery" worker. Sometimes it works great, sometimes it barely works, and it varies wildly from person to person.

Why does this matter?
Scientists often use "reference genes" (like ACTB or GAPDH) as a baseline to measure other things, assuming they never change. This paper says: "Stop! Those aren't stable enough for aging studies!" They found that the old standards wobble with age. They recommend using the new "High Score" genes (like TBP) as the new reliable rulers for measuring aging.

4. The Network Effect: "The Neighborhood Watch"

The researchers looked at how genes talk to each other (the Gene Regulatory Network). They found that the "jittery" genes aren't random.

• If a gene is in a neighborhood where its neighbors are also "jittery," the whole block becomes unstable.
• It's like a rumor spreading in a chaotic neighborhood; the instability spreads from one gene to its neighbors.
• However, there are "Safe Zones" in the city. The researchers found a specific group of genes (centered around a gene called SP1) that act like a fortress. Even as the rest of the city gets noisy, this group stays rock-solid. These genes are often responsible for repairing DNA damage, suggesting that the body tries desperately to keep its repair crew stable even as everything else falls apart.

5. The Big Picture: Order vs. Chaos

The paper concludes that aging is a mix of two things:

1. The Coordinated Shift (DEGs): The whole city slowly changing its schedule (e.g., everyone working slower).
2. The Rising Chaos (DVGs): The city losing its synchronization. Some people's cells are confused, some are frantic, and some are calm. This "noise" is likely caused by random damage to our DNA and the environment piling up over time.

The Takeaway:
Aging isn't just about things slowing down; it's about things becoming unpredictable. The instructions are still there, but the signal is getting noisy. By understanding this "noise," we can better measure aging, find better ways to repair our cells, and perhaps one day, learn how to tune the radio back to a clear signal.

Technical Summary

1. Problem Statement

Aging is characterized by progressive physiological decline driven by genomic and epigenomic instability. While previous transcriptomic studies of aging have overwhelmingly focused on Differentially Expressed Genes (DEGs)—genes showing changes in mean expression levels—this approach overlooks a critical dimension: Differentially Variable Genes (DVGs). These are genes whose expression variability (stochasticity) changes with age, even if their mean expression remains stable.

Existing literature presents conflicting views on whether transcriptional variability increases with age, partly due to technical noise in single-cell RNA sequencing (scRNA-seq) and a lack of robust methods to quantify stability in bulk RNA-seq data. Furthermore, it remains unclear how inter-individual variability (between people) relates to intra-individual variability (cell-to-cell noise) and whether these patterns are driven by coordinated biological programs or stochastic events.

2. Methodology

The authors utilized a multi-faceted approach combining bulk RNA-seq, scRNA-seq, and gene regulatory network (GRN) analysis:

• Dataset:
  • Bulk RNA-seq: Genotype-Tissue Expression (GTEx) project v8, covering ~1,000 individuals across 30 tissue types.
  • scRNA-seq: 8 published datasets (Human Brain, Colon, Liver, Lung, Pancreas, Skeletal Muscle) processed via a standardized pipeline (Chromium 10x platform) to ensure consistency.
• Gene Stability Score (GSS) Development:
  • To quantify stability in bulk RNA-seq, the authors adapted algorithms originally designed for RT-qPCR reference gene selection: GeNorm, NormFinder, and the Coefficient of Variation (CV).
  • They transformed CPM (Counts Per Million) values into "pseudo-Ct" values to align with qPCR logic.
  • Principal Component Analysis (PCA) was performed on these three metrics to derive a unified score.
    • tsGSS: Tissue-specific Gene Stability Score.
    • gGSS: Global Gene Stability Score (aggregated across 30 tissues).
  • High GSS = Stable expression; Low GSS = High variability.
• Differential Analysis:
  • DEGs: Identified using DESeq2 (age modeled as a continuous covariate).
  • DVGs: Identified by comparing standard deviations between age/sex groups using Levene's test and Spearman correlation to assess progressive changes in variance.
• Network Analysis:
  • Utilized the TRRUST database (8,015 TF-target interactions) to analyze the Gene Regulatory Network (GRN).
  • Developed a Stability Similarity Metric to determine if genes with similar stability profiles cluster within local network neighborhoods.
• Contribution Modeling:
  • Used multiple linear regression (via the relaimpo R package) to quantify the relative contribution of age/sex-related DEGs and DVGs to the total expression variance (gGSS).

3. Key Contributions

1. Novel Metric (GSS): Established the first widely applicable metric to quantify gene expression stability in bulk RNA-seq datasets, bridging the gap between qPCR reference gene selection and transcriptomic aging studies.
2. Quantification of Variability: Demonstrated that DVGs account for a significant portion (~15% overall, 7.7% specifically for age) of inter-individual expression variability, a factor previously underappreciated compared to DEGs.
3. Cross-Scale Correlation: Provided evidence linking inter-individual variability (bulk tissue) with intra-individual cell-to-cell transcriptional noise (single-cell), suggesting a shared underlying mechanism.
4. Reference Gene Identification: Identified robust, age-invariant reference genes (e.g., TBP, PUM1, TMEM199) for human aging studies, correcting the assumption that traditional references like ACTB or GAPDH are stable across the lifespan.
5. Network Architecture Insight: Revealed that gene variability is not random but is structured by local GRN topology, with DVGs showing higher local coherence than DEGs.

4. Key Results

• Stability Metrics: The global GSS (gGSS) showed high consistency across tissues and validated against the Human Protein Atlas. The most stable genes (WDR20, SNW1, MED6) were involved in RNA processing and ubiquitination.
• DEGs vs. DVGs:
  • DEGs were enriched in core maintenance pathways (proteostasis, RNA processing), suggesting coordinated, directional shifts.
  • DVGs were enriched in signaling and homeostasis pathways (e.g., calcineurin-NFAT, amyloid-beta clearance), suggesting increased heterogeneity and stochasticity in these processes with age.
  • While some genes are both DEGs and DVGs, they largely represent distinct biological phenomena.
• Network Influence:
  • DVGs exhibited positive stability similarity (neighbors had similar stability), indicating that variability propagates through local regulatory motifs.
  • DEGs tended to be dissimilar from their local network neighbors, suggesting gene-specific regulatory responses.
  • Pathway enrichment analysis showed DEGs are highly enriched in known pathways, whereas DVGs showed limited pathway enrichment, supporting the "stochastic event" hypothesis.
• Inter- vs. Intra-individual Correlation:
  • There is a significant negative correlation between gGSS (inter-individual stability) and transcriptional noise (cell-to-cell CV).
  • Tissue Specificity: Age-related noise increases significantly in the brain (across all stability bins) and moderately in muscle (only in unstable genes), but not in the lung.
  • Cell-Type Specificity: In the brain, oligodendrocytes showed the strongest age-related noise increase, while neurons showed no significant change.
• The "SP1 Group": A subnetwork of stable genes regulated by stable transcription factors (centered on SP1) was identified. This group is highly resistant to age-associated noise and is enriched for DNA damage repair pathways.
• Variation Source: Surprisingly, genes that are both DEGs and DVGs (DEVGs) contributed more variation per gene than DVGs alone, indicating that mean shifts often accompany increased noise.

5. Significance

This study fundamentally shifts the understanding of the aging transcriptome from a purely "mean-shift" model to a dual-model involving both coordinated transcriptional programs (DEGs) and increased stochasticity (DVGs).

• Mechanistic Insight: The findings suggest that aging involves a breakdown in the buffering capacity of gene regulatory networks, leading to the propagation of stochastic genomic/epigenomic damage into transcriptional noise.
• Experimental Utility: The identification of TBP, PUM1, and TMEM199 as superior reference genes provides a critical tool for researchers conducting RT-qPCR on aging tissues, preventing artifacts caused by unstable controls.
• Therapeutic Implications: The discovery that DNA damage repair genes (specifically the SP1 network) remain stable despite aging suggests these pathways are robust targets for maintaining transcriptional fidelity. Conversely, the high variability in signaling pathways highlights potential points of failure in homeostasis during aging.

In conclusion, the paper establishes that aging is not just a change in what genes are expressed, but a fundamental increase in the unpredictability of gene expression across individuals and cells, driven by the interplay of stochastic damage and network architecture.