https://onlinelibrary.wiley.com/doi/full/10.1111/acel.70474?campaign=woletoc
chatGPT:
Here’s a concise summary, the main novelty, and a critique of the uploaded paper, An Extracellular Matrix Aging Clock Based on Circulating Matrisome Proteins Predicts Biological Aging and Disease.
Summary
The paper asks whether circulating extracellular matrix proteins, the “matrisome,” can be used as a focused aging biomarker rather than relying on broad, unconstrained plasma proteomic clocks. Using public SomaScan datasets, the authors show that many plasma matrisome proteins change with age and that the aggregate trajectory is not simply linear: core matrisome components show a U-shaped pattern, with a low point around midlife and a rise later in life. The authors argue this reflects age-related ECM remodeling rather than just generic plasma protein drift.
They then connect this aging signal to disease. In a large phenotype-association dataset from about 35,000 Icelandic individuals, matrisome proteins were associated with more disease phenotypes than non-ECM proteins, especially collagens, glycoproteins, proteoglycans, and ECM regulators. Disease-associated ECM signatures overlapped partly with age-associated ECM signatures, but disease signatures were broader and appeared slightly delayed relative to the aging signature, which the authors interpret as age-related ECM change preceding or contributing to disease-associated remodeling.
The core modeling result is an ECM aging clock built from plasma matrisome proteins. In the main human cohorts, ECM-only LASSO clocks predicted chronological age well, though usually slightly less well than unconstrained proteome-wide clocks. Across both discovery cohorts, the authors converge on a shared 14-protein ECM signature: PRL, PTN, CCL21, ADAMTS5, FMOD, WISP2, LAMC2, LUM, ADAMTS13, CTSV, CILP2, PLXNB2, ADAMTSL1, and C1QTNF3. Those 14 proteins outperformed random protein sets of the same size and gave age gaps that correlated strongly with broader proteomic age gaps.
The 14-protein model also transferred to external datasets. It correlated with age in plasma and, more weakly, in cerebrospinal fluid, but not in aqueous humor. In several disease datasets, cases tended to have a higher ECM age gap than controls, and case-control classifiers built from the ECM signature showed AUCs roughly from 0.70 to 0.94, depending on the dataset. Even a reduced 8-protein version retained moderate signal across biomaterials and diseases.
The paper also explores cross-species relevance. In the Lehallier human and mouse datasets, humans and mice both showed age-related matrisome change, but the exact protein-level age effects were poorly conserved. Human and mouse ECM clocks still tracked age within species, and in heterochronic parabiosis the old heterochronic mice showed lower ECM age gaps than old isochronic controls, while young heterochronic mice showed higher gaps than young isochronic controls. The authors interpret this as evidence that systemic rejuvenation and systemic stress are reflected in the ECM.
Finally, the paper tries to move from biomarker to intervention space. Using Mendelian-randomization-based prioritization and drug-target databases, the authors identify many druggable matrisome proteins, including a shortlist of 43 candidates implicated across multiple phenotypes and disease domains. About half of those were druggable, with 16 already having FDA-approved drugs.
What is novel
The main novelty is not “another aging clock,” but a domain-specific clock centered on the extracellular matrix. Most proteomic clocks are broad and hard to interpret biologically; this one is explicitly built from matrisome proteins, giving it a clearer mechanistic theme.
A second novelty is the claim that circulating ECM proteins have a distinct non-linear, U-shaped life-course trajectory, with core matrisome components behaving differently from associated matrisome proteins. That gives the paper a stronger biological story than a purely predictive model. The visual data on pages 3 and 5 are important here: page 3 shows the U-shaped core-matrisome trajectories, and page 5 shows the comparison between aging and disease signatures.
A third novelty is the attempt to connect three levels at once: aging biomarker discovery, disease association, and therapeutic prioritization. The study does not stop at age prediction; it asks whether ECM signals distinguish disease states and which ECM proteins might be druggable. That translational angle is broader than many clock papers.
A fourth novelty is the inclusion of cross-biofluid and cross-species testing, plus parabiosis. Even though the species overlap is weak at the protein level, the idea that species-specific ECM clocks track aging and respond to rejuvenation interventions is a useful conceptual contribution.
Critique
The strongest criticism is that this is still mainly a re-analysis of public datasets, not a prospective, purpose-built biomarker study. That is efficient and useful, but it means the paper inherits all the limitations of the source cohorts, including different age ranges, anticoagulants, ethnic composition, and platform coverage. The authors themselves note model coefficient differences across datasets, such as opposite signs for C1QTNF3, which suggests some instability in the specific clock weights.
Second, the clock is biologically more interpretable than an unconstrained proteomic clock, but it is usually not better at plain age prediction. In the main cohorts, unconstrained clocks performed slightly better. So the value of the ECM clock is interpretability and disease linkage, not raw prediction supremacy. That is a fair tradeoff, but it should be stated plainly.
Third, the U-shaped interpretation is plausible but not fully nailed down. The authors suggest that the later-life rise partly reflects overlap between aging and disease matreotypes, but the evidence is still inferential and cross-sectional. The paper does not truly separate “normal aging ECM remodeling” from “increasing occult disease burden with age.” In other words, the U-shape may be biologically real, but the mechanism behind it remains partly unresolved.
Fourth, the disease analyses are broad but somewhat shallow mechanistically. Showing that ECM proteins are associated with more diseases than non-ECM proteins is interesting, but a protein associated with many phenotypes can reflect pleiotropy, tissue damage, fibrosis, inflammation, leakage, or assay architecture rather than a clean causal role in aging. The Mendelian randomization step helps prioritization, but it still does not prove that modifying those proteins will improve aging biology.
Fifth, the cross-species story is mixed. The paper’s conceptual conclusion is that ECM aging is conserved, but the actual protein-level overlap between human and mouse is limited. The authors do acknowledge that a human-based platform used in mice may partly explain this. Still, it weakens any strong claim that the same ECM proteins are universal aging markers across mammals.
Sixth, the parabiosis results are interesting but should be interpreted carefully. The old heterochronic mice looking “younger” and young heterochronic mice looking “older” fits classic parabiosis logic, but the sample is modest, and the paper notes that the downstream protein associations driving these effects did not survive multiple-testing correction. That makes the mechanistic interpretation exploratory rather than definitive.
Seventh, there is a general biomarker issue: the paper moves back and forth between chronological age prediction, biological age, disease discrimination, and rejuvenation response. Those are related but not identical endpoints. The ECM clock likely captures a mixture of structural remodeling, inflammation, fibrosis, tissue turnover, and disease burden. That mixture may be useful clinically, but it means the biological meaning of “higher ECM age” is still somewhat composite rather than precise.
Bottom line
I think this is a strong and genuinely interesting paper. Its best contribution is showing that the ECM is not just one contributor among many in proteomic aging clocks, but may be a coherent biomarker axis in its own right. The work is less convincing as proof of causality than as proof of signal: circulating matrisome proteins carry real information about age, disease state, and intervention response. The clock is probably most valuable as an interpretable, biologically themed complement to broader proteomic clocks, rather than a replacement for them.
If you want, I can also turn this into a more detailed section-by-section critique or a claim-by-claim table.
The paper identifies a core set of “key” extracellular matrix (ECM / matrisome) proteins that repeatedly drive the aging clock signal.
1) The main 14 “core” proteins (most important result)
Across both human cohorts, 14 proteins consistently appear in the ECM aging clocks and are treated as the central signature:
- PRL (prolactin)
- PTN (pleiotrophin)
- CCL21
- ADAMTS5
- FMOD (fibromodulin)
- WISP2
- LAMC2
- LUM (lumican)
- ADAMTS13
- CTSV
- CILP2
- PLXNB2
- ADAMTSL1
- C1QTNF3
These are the most robust predictors of biological age in the study:
- They were repeatedly selected by machine learning models
- They generalised across datasets
- They outperformed random protein sets of the same size
Conceptually: this is the “ECM aging signature”.
2) Functional grouping (what kinds of proteins these are)
These 14 proteins are not random—they cluster into ECM-relevant functions:
ECM remodeling enzymes
- ADAMTS5, ADAMTS13, ADAMTSL1, CTSV → Proteases / regulators of ECM turnover
Structural ECM components
- FMOD, LUM, LAMC2, CILP2 → Collagen organisation, matrix structure, elasticity
ECM-associated signaling / secreted factors
- PRL, PTN, CCL21, WISP2, C1QTNF3 → Cytokines / growth factors / signaling molecules interacting with ECM
ECM-affiliated / guidance proteins
- PLXNB2 → Cell–matrix signaling, tissue organisation
This mix is important:
It captures both structure (collagens/proteoglycans) and remodeling + signaling, which fits the idea that ECM aging = dynamic remodeling, not just accumulation of damage.
3) Broader “frequently selected” proteins (larger set)
Beyond the 14, the authors also note ~23 proteins frequently selected in models, including:
- GDF10, GDF3, MSTN, IFNA7, IL19
- TIMP4, CST6
- EMILIN3
- plus several of the core 14 above
These reinforce that:
- secreted factors and ECM regulators dominate the aging signal
4) Cross-species conserved signals (much weaker overlap)
Between humans and mice, only a small subset overlaps, including:
- GDF15
- PLXNB2
- SPARC, IGFBP2, WISP1
- chemokines like CXCL12 / CXCL13
Important point:
- The process (ECM remodeling) is conserved
- But the specific proteins are largely species-specific
5) Proteins strongly linked to disease
The study also highlights ECM proteins heavily associated with disease (not identical to the clock set), e.g.:
- PLXNB2, SEMA6B, CILP2, ADAMTSL2 (mental disease)
- INHBC, LEP, IGFBP1, IL1RN (digestive disease)
Overlap exists, but:
- disease signature is broader than aging signature
Bottom line
The key proteins in this paper are the 14 matrisome proteins above, which together form a minimal ECM-based aging clock.
Mechanistically, they point to:
- Active ECM remodeling (ADAMTS, CTSV)
- Structural matrix integrity (LUM, FMOD, LAMC2)
- ECM-linked signaling (WISP2, PTN, CCL21)
The big takeaway is not any single protein, but that:
Aging signal in blood is strongly encoded in ECM turnover + ECM-associated signaling.
If you want, I can break these down one-by-one mechanistically (e.g., how LUM vs ADAMTS5 vs WISP2 change with age and what pathways they map to).
The paper itself does not make splicing a main theme. It identifies the 14 ECM-clock proteins as aging/disease markers, but it does not analyze splice isoforms, splice-factor control, or splice-switches for those proteins directly.
The cleanest answer is that the links to splicing are mostly of three kinds.
First, some of the clock genes are themselves known to have alternative transcript variants. The clearest example is LAMC2, where NCBI and the original gene paper note two transcript variants generated by alternative splicing of the 3′ terminal exon. PTN, FMOD, and PRL are also annotated as having multiple transcript variants, with PTN specifically noted to use alternative promoters plus alternative splicing.
Second, for some of these proteins, the splicing link is more indirect but biologically important: they act in ECM pathways whose substrates or receptors are strongly splice-dependent. A good example is ADAMTS5. ADAMTS5 is an ECM-remodeling protease in the paper’s clock and disease discussion, but a major part of its biology is cleavage of proteoglycan substrates such as versican and aggrecan, both of which have well-known splice isoforms. So the effect of ADAMTS5 can depend on which ECM splice isoforms are present, even if ADAMTS5 itself is not primarily famous as a splicing-regulated gene.
Third, some of the signal may sit one step away from the ligand, in receptors or pathway members whose isoforms alter how the ECM-clock proteins act. The best example here is PRL: prolactin itself has multiple transcript variants, but the stronger functional splicing story is actually the prolactin receptor (PRLR), which has alternatively spliced membrane-bound and soluble isoforms that can change prolactin signaling output. That means PRL-related aging signals could be modulated by splicing even if the plasma PRL measurement is just one analyte.
For the 14 proteins specifically, I would separate them into strong, plausible, and weak/unclear splicing links.
Stronger direct evidence
LAMC2: definite alternative splicing with distinct isoforms.
PTN: multiple transcript variants via alternative promoters and splicing.
FMOD: multiple transcript variants are annotated.
PRL: multiple transcript variants are annotated, though they encode the same protein.
Mainly indirect ECM/splice-context links
ADAMTS5: acts on ECM substrates with important splice isoforms, especially versican/aggrecan.
PLXNB2: not a classic splicing paper example here, but disease studies report pathogenic splicing changes in PLXNB2, which at least shows the gene is sensitive to splice disruption.
Weaker or currently unclear from the sources I checked
CCL21, ADAMTS13, CTSV, CILP2, ADAMTSL1, C1QTNF3, WISP2, LUM, CILP2: there may be transcript variation for some of these, but I did not find equally solid primary-source support quickly enough to treat them as established splice-driven examples. For CILP2, the paper discusses it as a disease-associated ECM protein, but not through splicing.
So the practical takeaway is this:
These proteins are not mainly spliceosome proteins. Their relationship to splicing is mostly that:
-
some of their genes have alternative isoforms,
-
they sit in ECM networks where isoform choice matters a lot,
-
and in a few cases receptor/pathway splicing probably changes their functional meaning.
That makes splicing a plausible upstream regulator of the ECM clock, but this paper does not prove that the age signal is being driven by splice changes rather than by transcription, secretion, proteolysis, tissue leakage, inflammation, or disease burden.
If you want, I can turn the 14 proteins into a small table with columns for “direct splice isoform evidence,” “indirect ECM-splicing link,” and “likely relevance to aging.”