Longitudinal immune transcriptomic signatures are associated with carotid intima-media thickness over 18 years

This study demonstrates that peripheral blood mononuclear cell transcriptomic profiles are associated with long-term changes in carotid intima-media thickness, revealing that while cross-sectional immune signatures reflect immediate immune activation, longitudinal signatures linked to vascular remodeling are enriched for metabolic and redox pathways.

The Gist

Imagine your body is a bustling city, and your blood vessels are the main highways connecting all the neighborhoods. Over time, these highways can get clogged with "traffic jams" (plaque), leading to serious problems like heart attacks or strokes. This process is called atherosclerosis.

For years, doctors have tried to predict when these traffic jams will happen by looking at the usual suspects: high blood pressure, cholesterol levels, and smoking habits. But sometimes, two people with the exact same "traffic reports" have very different outcomes—one gets a massive jam, and the other's roads stay clear. Why?

This paper suggests the answer might be hidden in the immune system's "news feed."

Here is the story of the research, broken down simply:

1. The Long-Term Experiment (The 18-Year Movie)

The researchers studied a group of 148 healthy people (mostly children of factory workers, hence the "Barilla Offspring" name) starting in 2006. They took a snapshot of their health and their immune system's "mood" (gene activity) at that time.

Then, they waited 18 years. In 2024, they checked back on 101 of those same people. They didn't just ask how they felt; they took a high-definition look at their arteries (specifically the carotid arteries in the neck) to see how much "road thickening" (a sign of early plaque) had occurred.

2. The Detective Work (Reading the Immune System's Diary)

The scientists had a unique advantage: they had the "diary" (gene activity) from 18 years ago and the "road condition report" from today. They wanted to see if the diary entries from 2006 could predict the road damage in 2024.

They used three different detective strategies:

• Strategy A (The Snapshot): Did the immune system's mood in 2006 match the road damage in 2006?
• Strategy B (The Current Check): Did the immune system's mood in 2024 match the road damage in 2024?
• Strategy C (The Crystal Ball): Did the immune system's mood in 2006 predict the road damage in 2024?

3. The Big Discovery: Two Different Types of Clues

The study found that the immune system tells two different stories depending on when you look at it.

• The "Right Now" Story (Cross-Sectional): When they looked at the immune system and the arteries at the same time, the clues were all about active fighting. The immune cells were sounding alarms, calling for reinforcements, and preparing for battle. It was like seeing a city with sirens blaring and police cars speeding around right now. This made sense because the arteries were currently inflamed.
• The "Crystal Ball" Story (Longitudinal): This was the surprise. When they looked at the 2006 diary to predict the 2024 damage, the clues weren't about active fighting. Instead, they were about maintenance and structure. The genes pointed to issues with how cells build their walls, how they handle energy (metabolism), and how they manage stress.

The Analogy:
Think of a house.

• The "Right Now" clue is seeing smoke and fire trucks. It tells you the house is burning today.
• The "Crystal Ball" clue is finding that the foundation was built with weak bricks and the plumbing was corroded 18 years ago. You couldn't see the fire yet, but those structural flaws guaranteed the house would eventually collapse.

The study found that the structural flaws (metabolism and cell structure) in the immune system's genes 18 years ago were the real predictors of future artery damage.

4. The "Super-Connectors" (Hub Genes)

The researchers used a special computer method (better than the standard ones they tried) to find the most important genes. They found "Hub Genes"—think of these as the mayors of the immune system's city.

• Some mayors were in charge of calling the police (immune activation).
• Others were in charge of fixing the roads (cell structure and metabolism).

Even though the specific genes changed slightly over 18 years (because technology improved and people got older), the jobs these mayors were doing remained the same. The immune system's overall "personality" regarding vascular health stayed consistent.

5. Why This Matters

This is a game-changer for a few reasons:

1. Early Warning System: You don't have to wait for your arteries to get thick to know they are at risk. A simple blood test today could look at your immune system's "structural blueprint" and tell you if you are likely to have artery trouble in 20 years.
2. Beyond Cholesterol: It shows that the immune system isn't just a reaction to bad cholesterol; it's a driver of the problem itself. The way your immune cells handle stress and energy might be just as important as your diet.
3. Personalized Medicine: In the future, doctors might not just treat high blood pressure. They might prescribe treatments to "fix the foundation" of your immune system to prevent the roads from ever getting clogged in the first place.

The Bottom Line

Your blood cells carry a memory of your vascular health. Even before you see any signs of heart disease, your immune system's genes are whispering clues about how your arteries will age. By listening to these clues—specifically the ones about how your cells are built and how they handle energy—we might be able to keep our internal highways clear for much longer.

Technical Summary

1. Problem Statement

Cardiovascular disease (CVD) is driven by atherosclerosis, a chronic immunometabolic disorder. While Carotid Intima–Media Thickness (IMT) is a standard non-invasive surrogate for subclinical atherosclerosis, traditional risk factors fail to fully explain the heterogeneity in vascular remodeling trajectories.

• The Gap: There is a lack of robust longitudinal evidence linking systemic immune transcriptional states (measured in Peripheral Blood Mononuclear Cells, or PBMCs) to long-term structural vascular changes.
• The Challenge: Most existing studies are cross-sectional. Furthermore, integrating data across an 18-year period involves significant technological heterogeneity (transition from microarrays to RNA-seq) and methodological shifts in imaging, making direct longitudinal comparisons difficult.

2. Methodology

Study Design and Cohort

• Cohort: The "Barilla Offspring Study," a single-center observational cohort.
• Timeline: Baseline (T0, 2006–2007) and 18-year follow-up (T1, 2024).
• Participants: 148 healthy adults at baseline; 101 (68.2%) re-evaluated at follow-up.
• Phenotyping: Comprehensive clinical, metabolic, and vascular assessments (including carotid ultrasound for IMT) were performed at both time points. PBMCs were isolated for transcriptomic analysis.

Data Acquisition & Processing

• Transcriptomics:
  • T0: Microarray analysis (Agilent platform).
  • T1: High-depth RNA sequencing (Illumina NovaSeq, 2x150bp).
  • Normalization: Data were processed using limma (microarray) and STAR/TPM (RNA-seq).
  • Covariate Correction: To address the 18-year gap and technological shifts, an Ordinary Least Squares (OLS) regression was used to remove the effects of age and sex from both transcriptomic data and IMT measurements, creating "residualized matrices" for analysis.
• Vascular Phenotype: Mean IMT was used as the primary outcome to minimize the impact of focal plaques and measurement variability between the two time points.

Analytical Framework
Three distinct analytical configurations were modeled to predict Mean IMT:

1. Baseline Cross-Sectional: T0 Transcriptome $\rightarrow$ T0 IMT.
2. Follow-up Cross-Sectional: T1 Transcriptome $\rightarrow$ T1 IMT.
3. Longitudinal Prognostic: T0 Transcriptome $\rightarrow$ T1 IMT (predicting 18-year progression).

Machine Learning & Statistical Analysis

• Algorithm Comparison: Three regression methods were compared:
  1. LASSO (Least Absolute Shrinkage and Selection Operator).
  2. Random Forest Regression.
  3. Rank-Based Regression: An innovative method adapted from classification tasks, using a distance metric based on weighted enrichment scores (ES) to predict outcomes via neighbor averaging.
• Performance Metric: Ratio of Performance to Deviation (RPD) via 10-fold cross-validation.
• Functional Analysis:
  • Enrichment: STRING database for pathway analysis.
  • Semantic Similarity: Resnik's Information Content (IC) and Best-Match Average (BMA) algorithms to compare functional convergence between signatures despite low gene-level overlap.
  • Network Analysis: Construction of Protein-Protein Interaction (PPI) networks to identify "hub genes" (degree $\ge$ 3).

3. Key Contributions

1. Methodological Innovation: The study successfully navigated the technical heterogeneity of an 18-year longitudinal study (Microarray $\rightarrow$ RNA-seq) by employing covariate-residualized matrices and a rank-based regression approach. This allowed for the extraction of biological signals independent of age/sex and technological noise.
2. Algorithm Selection: The Rank-Based Regression method outperformed standard machine learning models (LASSO and Random Forest) across all configurations, achieving higher RPD values (1.26–1.38) and retaining sufficient gene complexity for functional enrichment.
3. Dual-Phase Discovery: The study distinguishes between the molecular drivers of current vascular state (cross-sectional) and future vascular remodeling propensity (prognostic), revealing distinct but converging biological mechanisms.

4. Key Results

Predictive Performance

• The Rank-Based method consistently yielded the highest RPD:
  • Baseline: 1.38
  • Follow-up: 1.26
  • Prognostic: 1.38
• This confirmed the method's superiority in extracting predictive signatures from high-dimensional, noisy omics data.

Transcriptomic Signatures & Pathways

• Cross-Sectional Signatures (Baseline & Follow-up):
  • Dominant Pathways: Immune activation, leukocyte proliferation, antigen presentation, and receptor-mediated signaling (T-cell receptor, MHC class II).
  • Hub Genes: CSF1R (monocyte/macrophage regulation), CD3E, CD86, HLA-DRA, ITGAM.
  • Interpretation: These signatures reflect active, concurrent inflammatory processes driving the current vascular state.
• Longitudinal Prognostic Signature (Baseline $\rightarrow$ Follow-up):
  • Dominant Pathways: Metabolic regulation (folate biosynthesis, D-xylose metabolism), redox processes, and cellular structural organization (MARVEL domain proteins, tight junctions).
  • Interpretation: Early metabolic stress and structural membrane integrity appear to be the primary drivers of future vascular remodeling, distinct from the acute inflammation seen in cross-sectional snapshots.
• Gene Overlap:
  • Direct gene overlap between signatures was minimal.
  • WNK4 was the only gene shared between Baseline and Follow-up signatures, suggesting a stable link between electrolyte/blood pressure regulation and IMT over nearly two decades.
  • Semantic Similarity: Despite low gene overlap, functional semantic similarity was high (Resnik-BMA: 4.26 for Baseline vs. Follow-up), indicating that the underlying biological processes (immunometabolism) remain conserved even as specific gene expression shifts.

Network Analysis

• Integrated networks showed a "well-mixed" topology, confirming that biomarkers from different time points and technologies integrate into shared functional modules rather than forming isolated clusters.
• Hub Genes (e.g., CSF1R, ITGAM, TJP1) acted as central biological drivers across all configurations.

5. Significance and Implications

• Mechanistic Insight: The study provides evidence that vascular remodeling is driven by a dual mechanism: active immune signaling (captured cross-sectionally) and underlying metabolic/structural dysregulation (captured longitudinally).
• Early Detection: PBMC transcriptomic profiles can serve as early, systemic readouts of vascular susceptibility, potentially identifying individuals at risk for adverse remodeling before clinical risk factors manifest.
• Clinical Translation: The findings suggest a shift from static risk factor evaluation to systems-level immune-metabolic profiling. This could refine cardiovascular risk stratification, particularly for individuals with "normal" conventional risk profiles but high immunometabolic risk.
• Robustness: The study demonstrates that longitudinal systems biology is feasible over decades, provided rigorous statistical methods are used to correct for technological drift and confounding variables.

Limitations:

• Moderate sample size (N=101 for longitudinal analysis) from a single center.
• Inability to perform direct absolute IMT comparisons due to ultrasound technology changes.
• PBMCs represent a composite of cell types; single-cell resolution is needed for finer granularity.

Conclusion:
Systemic immune transcriptional states are not merely reflections of current metabolic status but are persistent determinants of vascular aging. The study establishes a framework for using longitudinal immune transcriptomics to predict subclinical atherosclerosis progression, highlighting the critical role of immunometabolism in cardiovascular health.