The study, led by Associate Professor Jacques Behmoaras and first author Dr. Harry Park from the Centre of Biomedical Data Science, utilized single-cell RNA sequencing data to map the genetic activity of immune cells across the entire human lifespan. The cohort included individuals aged 19 to 97, with a diverse demographic composition that included a significant number of participants of Asian ethnicity. This large-scale integration of publicly available research datasets allowed the team to create a comprehensive immune atlas, offering a high-resolution view of how the immune system evolves over time. The analysis focused on the transcriptomic profiles of various immune cell types, particularly T cells, which are central to coordinating the body’s defense mechanisms against pathogens.

The researchers discovered that the fundamental machinery responsible for cellular function, specifically the genes involved in the synthesis of RNA and proteins, experiences its most pronounced decline during two specific windows: around age 40 and again after age 60. These periods represent critical inflection points where the immune system undergoes rapid reorganization. Rather than a steady erosion of capacity, the data suggests that the immune system undergoes abrupt changes in gene expression patterns. This nonlinear trajectory implies that the body’s immune resilience is not simply lost bit by bit, but rather that specific functional capacities are compromised at distinct life stages, potentially leaving individuals more vulnerable to specific types of immune dysregulation during these periods.

T cells emerged as the primary drivers of these nonlinear changes. The study found that T cells, a type of white blood cell essential for fighting infections, showed the most significant alterations in gene activity during the first wave around age 40. This early shift likely reflects the central role T cells play in coordinating the immune response and their heightened sensitivity to the initial effects of aging. Interestingly, the specific subtypes of T cells that undergo profound changes at age 40 are different from those that drive the second wave of changes after age 60. This distinction suggests that T cells have specialized roles at different stages of life, and that the aging process affects these specialized functions in a staggered manner.

One of the most significant findings of the study is the identification of sex-specific differences in immune aging trajectories. While both men and women experience the two major waves of immune aging, the specific patterns of gene activity and the types of T cells involved differ between the sexes. These differences may help explain why women are more likely than men to develop autoimmune diseases, such as lupus and rheumatoid arthritis, despite generally living longer. In Singapore alone, more than 600,000 people, or about 11 percent of the population, suffer from autoimmune conditions. The study suggests that the immune system’s response to aging is not uniform, and that sex-specific biological pathways may predispose individuals to different types of immune-related disorders at different stages of life.

Professor Antonio Bertoletti, a co-author of the study and a leader in the Emerging Infectious Diseases Signature Research Programme at Duke-NUS, emphasized the importance of understanding these nonlinear changes. He noted that previous evidence had suggested T cells are among the immune cells most affected by aging, but the new study provides a detailed map of how and when these changes occur. Understanding the specific biological pathways that drive these shifts could help explain why older adults are more susceptible to infections and inflammatory conditions. The findings suggest that interventions aimed at boosting immune function may need to be timed to coincide with these critical windows of vulnerability to be most effective.

To further explore the implications of these findings, the scientists developed artificial intelligence models capable of predicting biological age based on the immune cell data. Biological age is an estimation of the ‘true age’ of the body, which can differ from chronological age. By using these AI models, researchers can identify individuals whose immune systems are aging faster or slower than their peers. This tool could be used to tailor preventive strategies and treatments based on an individual’s biological age and sex, moving away from a one-size-fits-all approach to aging. The ability to predict biological age from immune cell profiles opens up new possibilities for personalized medicine, where treatments can be adjusted to match the specific needs of a patient’s immune system.

The study was conducted in collaboration with several Signature Research Programmes at Duke-NUS, including those in Cardiovascular and Metabolic Disorders, Cancer and Stem Cell Biology, and Emerging Infectious Diseases. Partners from the NUS Yong Loo Lin School of Medicine also contributed to the research. Professor Sheemei Lok, Duke-NUS’ Interim Vice-Dean for Research, highlighted the significance of the findings for the field of aging research. She stated that the study provides a more detailed map of how the human immune system changes across the lifespan, and importantly, how those trajectories differ between men and women. This knowledge is fundamental for developing precise approaches to maintaining health as people grow older, rather than relying on generalized strategies that may not account for individual biological differences.

The research was supported by the National Research Foundation, Singapore, and the National Medical Research Council, underscoring the importance of this work in the context of Singapore’s healthcare landscape. The findings have broad implications for global health, as the patterns of immune aging observed in this diverse cohort are likely to be relevant to other populations. By identifying the specific windows of immune vulnerability and the sex-specific differences in these trajectories, the study provides a foundation for future research into targeted interventions. Understanding the nonlinear nature of immune aging could lead to the development of therapies that are more effective at preventing or treating age-related immune disorders, ultimately improving the quality of life for older adults.

As the global population ages, the need for a deeper understanding of the biological mechanisms underlying age-related diseases becomes increasingly urgent. This study represents a significant step forward in that effort, providing a new perspective on how the immune system changes over time. By revealing the distinct waves of immune aging and the sex-specific differences in these trajectories, the researchers have opened up new avenues for investigation and intervention. The findings suggest that the future of aging research lies in a more nuanced, personalized approach that takes into account the complex, nonlinear dynamics of the immune system. This work not only advances our scientific understanding but also holds promise for improving clinical outcomes for patients suffering from immune-related diseases.

Subject of Research: Nonlinear trajectories of sex-specific immune aging at the single-cell level

Article Title: Immune ageing accelerates around 40 and 60, with key differences between men and women

Article References: Immune ageing accelerates around 40 and 60, with key differences between men and women. (n.d.). Original publication

Image Credits: AI Generated

DOI: Not provided

Keywords: Immune Aging, Single-Cell RNA Sequencing, T Cells, Autoimmune Diseases, Sex-Specific Differences, Biological Age, Genome Activity, Duke-NUS Medical School, Nature Communications, Nonlinear Aging, Personalized Medicine, Immune Atlas

Expression quantitative trait loci are genomic positions, typically single nucleotide variants, that correlate with differences in the activity of nearby genes. In classical eQTL studies, which aggregate gene expression measurements across whole tissues, these associations reveal only the average effect of a variant across a mixture of cell types. That averaging is a serious limitation for immunology, because peripheral blood contains dozens of specialized cell populations—T cells, natural killer cells, monocytes, B cells, and many rarer intermediates—each with its own regulatory grammar. A variant that strongly boosts a gene in natural killer cells may be invisible in a bulk measurement dominated by monocytes. By pairing each donor’s genome with single-cell transcriptomes, the researchers could test the association between genotype and expression separately within each cell population, resolving effects that bulk approaches blur away.

The technical pipeline behind the study was substantial. Whole-genome sequencing data were processed through established variant-calling workflows, with germline small variants identified using the Strelka2 caller, following best practices endorsed by the Global Alliance for Genomics and Health for benchmarking variant calls. Coverage calculations were performed with Mosdepth, and sequencing quality was assessed against community standards to ensure that the genetic data underpinning the eQTL analysis were reliable. On the transcriptomic side, single-cell RNA-seq data from the 230,000 PBMCs were integrated across donors using the Harmony method, a widely adopted algorithm that corrects for batch effects and donor-specific technical noise while preserving genuine biological differences between cell types. Cells were then classified into 18 populations spanning the major branches of the immune system, providing the cellular resolution required for cell-type-specific association testing.

With genotypes and cell-type-resolved expression in hand, the team carried out cis-eQTL mapping using Matrix eQTL, a computational framework designed to perform the enormous matrix operations required for fast association testing across millions of variants and thousands of genes. The result was a catalog of 1,233,644 cis-eQTLs distributed across the 18 cell types. Importantly, this was not simply a numbers exercise. The researchers subjected their findings to a series of validation and interpretive analyses designed to ask whether the variants they detected showed the hallmarks of genuine regulatory elements, and, more provocatively, what evolutionary forces have shaped them.

One of the most intriguing findings to emerge from these secondary analyses concerns evolutionary conservation. When the team examined the genomic regions harboring their strongest eQTLs, they found that the most statistically significant associations tended to sit in less conserved regions of the genome—stretches of DNA that have diverged relatively rapidly between species. These variants were also concentrated in the regulatory regions of more divergent genes. This pattern suggests that immune gene regulation is an evolutionary hotspot, consistent with the well-documented observation that genes involved in host defense are frequent targets of positive selection. Rapid turnover of regulatory elements may allow populations to adapt to changing pathogen landscapes, but it may also help explain why immune-related variants are a rich source of susceptibility to chronic inflammatory and autoimmune disease in modern humans. The finding carries a double edge: the same regulatory flexibility that enabled adaptation to ancient pathogens may predispose contemporary genomes to misfire.

To move from statistical association to mechanistic understanding, the researchers turned to deep learning models of cis-regulatory sequence. Neural networks trained on genomic data can learn the relationship between DNA sequence and regulatory function, predicting how transcription factors bind to specific sequence contexts and how single-base changes alter those interactions. Drawing on approaches pioneered by tools such as DeepSEA, Basset, and the Enformer-style sequence models developed in recent years, the team applied these computational models to their eQTL catalog to ask, for each variant, which transcription factor binding sites are disrupted and in which cellular context that disruption matters. This step transformed the analysis from a list of correlated positions into a functional hypothesis-generating resource: each eQTL could now be annotated with a predicted mechanism of action grounded in sequence-level regulatory biology.

The functional analysis focused on genes with well-established roles in immunity, including NKG7, members of the HLA family, MIF, and MS4A1. NKG7 encodes a protein essential for the cytotoxic function of natural killer cells and CD8 T cells, involved in the trafficking of lytic granules that deliver the killing blow to infected or malignant target cells; variants affecting its expression could plausibly modulate antiviral and antitumor immunity. HLA genes, which encode the human leukocyte antigen molecules that present viral peptides to T cells, are among the most polymorphic loci in the human genome and have been repeatedly implicated in COVID-19 susceptibility and severity. MIF, the macrophage migration inhibitory factor, is a potent inflammatory mediator whose circulating levels correlate with severe COVID-19 pneumonia. MS4A1, better known as CD20, defines B cells and is the target of widely used monoclonal antibody therapies. By tracing how variants influence the expression of these genes through disrupted transcription factor binding—including factors such as those involved in myeloid and lymphoid differentiation—the study connected genetic variation to plausible cellular mechanisms.

The disease context of the study is explicit in its framing. Prior work by overlapping teams had used single-cell transcriptomics to identify immune cell signatures associated with severe Delta-variant COVID-19, and large consortia had shown that severe disease is marked by a dysregulated myeloid cell compartment. The GWAS Catalog lists numerous loci associated with COVID-19 outcomes, but translating those associations into functional biology has remained a central challenge for the field. The new eQTL resource provides a bridge: because it assigns regulatory variants to specific immune cell types and annotates their predicted effects on gene expression, it can be used to prioritize candidate causal variants among the many statistical associations emerging from disease-genetic studies. The authors also point toward broader applications, noting parallels with single-cell eQTL studies in brain, autoimmune disease, and other contexts that have revealed how cell-type-specific regulation shapes genetic risk.

The scale of the cellular census deserves emphasis. Thirty donors may sound modest compared with the million-person cohorts of cardiovascular genetics, but single-cell eQTL studies trade sample size for resolution: every donor contributes tens of thousands of individual cells, each measured across the whole transcriptome. The 230,000-cell dataset allowed the researchers to detect eQTLs not only in abundant populations like CD4 T cells and monocytes but in rarer states such as intermediate monocytes and specific lymphocyte subsets, where regulatory effects would be hopelessly diluted in bulk analysis. The identification of 18 cell-type-resolved regulatory landscapes from this relatively small cohort illustrates the power of the approach, and the authors’ analytical strategy—conservation analysis, transcription factor modeling, and pathway interrogation—provides a template that larger consortia are likely to follow as single-cell eQTL mapping matures.

Methodologically, the study also underscores how much of modern genomics depends on careful assembly of open computational tools. Beyond the core machinery of variant calling and eQTL mapping, the researchers drew on annotation frameworks for regulatory elements derived from more than a thousand epigenomic datasets, packages for gene ontology and pathway enrichment, tools for predicting DNA shape features that influence transcription factor binding, and simulation frameworks for testing the significance of overlaps between genomic intervals. This layered infrastructure allowed a single research group to integrate population genetics, transcriptomics, chromatin biology, and machine learning into a coherent narrative about how sequence variation becomes functional variation in the immune system.

The limitations of the study are those inherent to its design. A cohort of 30 individuals restricts statistical power for detecting rarer variants and weaker regulatory effects, and the donor population limits generalizability across ancestries—an important consideration given that eQTL effects and linkage patterns differ among populations. The deep learning predictions, while mechanistically informative, remain computational hypotheses that would need experimental validation through reporter assays or CRISPR-based perturbation of individual variants. Nevertheless, the study’s value lies in the framework it establishes: a complete chain from genome sequence, through cell-type-resolved expression, to predicted transcription factor mechanism, anchored to a disease of global significance.

As single-cell sequencing costs continue to fall and paired genotyping-transcriptomics cohorts grow, resources of this kind are expected to expand rapidly in scale and diversity. What this study demonstrates is that the payoff of such investment is not merely a longer list of associations, but a progressively sharper picture of the regulatory code that governs human immunity—and with it, new opportunities to understand, predict, and ultimately intervene in diseases where the immune system holds the balance between recovery and catastrophe. For COVID-19, whose genetic architecture continues to be dissected years after the pandemic’s peak, that sharper picture may help explain at last why the same virus produces a mild illness in one person and a life-threatening one in another.