Scattered light reveals a chemical fingerprint of senescent cells

A microscope may be able to recognize a cell’s chemical state without dyeing it, destroying it or removing it from its surroundings. That matters because senescent cells—cells held out of the normal division cycle—do not all behave alike. Some patterns appear alongside repair; others accompany chronic dysfunction. A single molecular label can therefore reveal one part of the story while hiding the rest.

A study published on September 21 in Nature Aging approaches this identification problem by listening, in effect, to molecules ring. Its RamanOmics framework combines Raman microscopy, two forms of RNA measurement and machine learning. In sections of young and old mouse lung and skin, the researchers found a small lipid-linked feature in the Raman spectrum that repeatedly appeared with cells positive for p21, a commonly used senescence marker.

The result is an association, not a universal senescence detector. Understanding that distinction requires following four different measurements through the experiment.

What each layer contributes

Raman scattering is unusually useful here because molecules alter a tiny fraction of incoming light according to the ways their chemical bonds vibrate. The resulting spectrum acts like a chemical fingerprint. Unlike fluorescence microscopy, it does not require researchers to attach a glowing label to a chosen target. A review of Raman bioimaging describes both that label-free chemical sensitivity and a longstanding drawback: collecting the faint signal can be slow.

The researchers first used single-nucleus RNA sequencing to inventory 35,474 cells from mouse lung and 12,128 from skin. The arithmetic is reproducible:

35,474 lung nuclei + 12,128 skin nuclei = 47,602 nuclei

That is a substantial cellular inventory, but it should not be confused with 47,602 independent animals. Each young and old tissue group contained three mice, so biological replication remained much smaller than the cell count.

On tissue sections from the same sample blocks, the team collected Raman spectra across 600–1,800 inverse centimetres, a unit used to describe molecular vibrations. Raman imaging was followed by STARmap spatial transcriptomics, which targeted 890 genes. That sequence let the researchers compare a location’s chemical spectrum with nearby gene activity rather than trying to align unrelated pieces of tissue after the fact.

The time cost hiding inside the map

The study sampled the tissue every 3 micrometres and integrated the Raman signal for one second at each position. That sounds quick until the microscope must cover an area rather than a point.

For a one-square-millimetre map, the approximate acquisition burden is:

1,000,000 µm² ÷ (3 µm × 3 µm) ≈ 111,111 spectra

111,111 spectra × 1 second ÷ 3,600 ≈ 30.9 hours

That is about 30.9 hours of Raman acquisition alone. It excludes stage movement, focusing, setup, registration and data processing, so it is not the total analysis time. The calculation helps explain why the study identifies imaging area and speed as constraints: chemically mapping a large tissue section at this sampling interval is not yet a casual scan.

The narrow signal inside a broad chemical picture

Across lung and skin, p21-positive cells were associated with a lipid-linked Raman band around 1,131–1,135 inverse centimetres. The study also reported increased lipid-associated signals when senescent cells appeared in a mouse skin-wound model. The recurrence across two tissues and a repair setting makes the feature more interesting than a difference found in only one cell population.

But the light did not directly announce, “This cell is senescent.” It recorded molecular vibrations. The lipid interpretation came from the position of the spectral feature, while the senescence reference came from p21 RNA and the wider gene-expression data. Machine learning then combined leading Raman and molecular features into what the authors call a multimodal barcode.

That barcode is therefore broader than the Raman band itself. It summarizes agreement among chemical, genetic and spatial measurements. The accompanying Nature Aging commentary identifies this connection between label-free chemical imaging and spatial transcriptomics as the central advance: gene activity describes only one layer of cellular change, while Raman light provides a view of biochemical remodeling.

The tissue differences also resist a one-size-fits-all definition. In the study, senescent lung cells were associated with extracellular-matrix remodeling and transforming growth factor-β signaling. Skin cells instead showed stronger epidermal-differentiation programs. Young tissues carried more gene patterns consistent with repair, whereas old tissues carried more patterns associated with inflammation, survival and impaired metabolic or repair functions. Those are interpretations of coordinated gene programs, not proof that any one program caused the observed tissue state.

What would make the light trustworthy on its own?

A stand-alone Raman marker would face a harder test than this multimodal demonstration. A useful validation sequence would train a Raman-only classifier in advance, lock its decision rules, and then test it blindly on independent animals and tissues. It should be compared with several co-located senescence indicators—not p21 alone—and with cells that have stopped dividing for other reasons.

Researchers would also need to identify the chemistry behind the 1,131–1,135 band more directly, perhaps by tracing labelled molecules or comparing the same regions with lipid-sensitive mass spectrometry. Human samples would require their own validation. Success in mouse tissue sections does not establish detection inside living people, a diagnostic test or a way to choose treatment.

For now, the achievement is more precise and, in its own way, more delightful: faintly scattered light preserved a chemical view of cells in place, while RNA measurements supplied names and addresses. Their agreement reveals a promising fingerprint. It does not yet prove that the fingerprint belongs exclusively to senescence.