Immune aging in lupus: where did the extra 10 years come from?

Immune aging in lupus may be more pronounced than a patient's chronological age would suggest. Researchers have used machine learning to reconstruct immune-cell composition from gene expression data, finding a pattern in people with systemic lupus erythematosus that resembles an older immune system.

In the publicly available preprint of the study, the difference from the healthy age-related pattern was approximately ten years. This is a model-derived estimate of immune age, not proof that the disease makes the entire body age ten years faster.

The particularly interesting part is how the researchers obtained that estimate. They effectively reconstructed an immune-cell protein profile from data in which the proteins themselves had never been directly measured.

Predicting proteins that were never measured

Assessing the immune system requires more than counting white blood cells. T cells, for example, include many distinct populations: some have not yet encountered their corresponding antigen, others carry immunological memory, and still others have acquired highly differentiated cytotoxic functions.

Scientists distinguish these populations using proteins expressed on their surfaces. In laboratory immunophenotyping, antibodies are used to identify particular protein markers.

But there is another enormous source of biological information: single-cell RNA sequencing. It reveals which genes are active in individual cells without directly measuring the corresponding proteins.

This creates an important gap, since gene expression and protein abundance are related but not interchangeable.

Researchers at the Georgia Institute of Technology developed scEN, a machine-learning model designed to predict surface-protein abundance from single-cell gene expression data.

They trained it using bone marrow cells for which both RNA and surface-protein measurements were available. The model learned the relationships between the two kinds of data and then applied them to cells with RNA measurements alone.

The publicly available preprint describes testing the approach across several independent datasets. When a model trained on bone marrow data was transferred to peripheral blood cells, its mean correlation between predicted and measured protein abundance reached 0.798.

The team then asked whether these reconstructed protein profiles could reveal something more ambitious: the age of the immune system.

Nineteen cell populations as an immune-age signature

Aging changes not only the total number of immune cells but also the relative abundance of different populations.

Naive T cells, for example, generally become less abundant with age. Certain differentiated cytotoxic populations and other immune-cell subsets become more prominent.

Using the predicted surface markers, the researchers identified 19 immune-cell populations. They then developed a second model to estimate chronological age from the proportions of these populations.

The age predictor was trained on healthy individuals, learning how immune-cell composition typically changes over the adult lifespan.

According to the preprint, the model had a mean absolute individual prediction error of 9.76 years. That uncertainty matters: the model can capture population-level aging patterns, but it is not a precise biological-age test for an individual patient.

The researchers then applied their approach to previously collected single-cell data from people with systemic lupus erythematosus.

Why the immune system looks older in lupus

Systemic lupus erythematosus is an autoimmune disease in which the immune system attacks the body's own tissues. It involves complex immune dysregulation and can cause persistent inflammation.

In the lupus datasets, the model identified changes in immune-cell composition that resembled those normally associated with older age.

These included reduced naive T-cell populations, increases in certain differentiated cytotoxic cells, and changes in monocyte and natural killer cell populations.

The preprint reported an approximately ten-year upward shift in predicted immune age compared with the healthy reference pattern. The published journal abstract confirms the overall finding of an accelerated immune-aging signature, although it does not state the numerical estimate.

There is an important distinction between this observation and the broader interpretation.

The study does not establish that lupus causes the entire body to biologically age exactly ten years faster. The analysis used existing datasets, and protein levels in the lupus samples were computationally inferred rather than directly measured.

Disease activity, treatment and differences between study populations could also contribute to the apparent age shift.

Nevertheless, the result shows that a model trained on typical immune changes with age can identify a similar pattern of immune remodeling in an autoimmune disease.

The methodological contribution may be equally significant. Large collections of previously generated single-cell RNA data could potentially be reanalyzed to recover additional information about immune-cell populations without performing new protein measurements.

The lupus analysis illustrates why that matters: immune-cell composition may reveal biological changes that are invisible when looking at chronological age alone.