An analysis of gene activity in intervertebral disc degeneration and osteoporosis has identified signals that point toward a possible shared inflammatory and tissue-supporting biology. The finding is a clue, not a clinical test: the work does not establish that either condition causes the other, or that any of the genes identified can diagnose or treat patients. The researchers combined separate public datasets and used computational analyses to search for genes that differed between disease and comparison groups. Their results offer a map of possible connections, but the evidence remains exploratory.
Two datasets, not one patient group
The material came from two very different sources. GSE124272 contained whole-blood samples from eight patients with magnetic resonance imaging-confirmed lumbar disc prolapse and eight healthy volunteers. GSE35956 contained bone marrow-derived mesenchymal stromal cells from five elderly people with primary osteoporosis and five middle-aged donors without osteoporosis. The datasets were analyzed separately and were not merged.
That design gives the analysis two disease-related comparisons, but not a matched human cohort with both conditions measured in the same people. It also means the disease signals were observed in whole blood in one dataset and bone marrow-derived cells in the other. Those differences make it difficult to tell whether a shared signal reflects a common process, the tissue being tested, or differences between the groups.
How the gene lists were built
Before testing for gene differences, the researchers inspected each processed GEO matrix separately, applied a log2 transformation when needed, normalized the distributions across samples, and mapped probes to official gene symbols. They also used principal-component analysis and hierarchical clustering to check the structure of the data. No samples were excluded. These quality-control steps describe how the data were handled, but they cannot make the datasets directly comparable.
To identify differences, the team used limma, a linear-model method, on each dataset. The analysis included empirical-Bayes moderation and Benjamini-Hochberg correction for multiple comparisons. Genes were counted as differentially expressed only when the absolute log2 fold change was at least 1 and the adjusted P value was below 0.05. In plain terms, the analysis required both a minimum expression change and an adjusted statistical signal.
The resulting lists were large in the disc-degeneration dataset: 1,874 genes were more highly expressed and 1,770 were less highly expressed in the disease group than in healthy volunteers. In the osteoporosis dataset, 437 genes were higher and 475 were lower than in the non-osteoporotic comparison group. These counts show how many genes passed the chosen filters; they do not, on their own, show that the genes drive disease.
Patterns beyond individual genes
Researchers then used gene-set enrichment analysis, or GSEA, to look for coordinated patterns among genes ranked by their moderated t statistics. Instead of focusing only on one gene at a time, this approach asks whether genes belonging to a named biological category tend to appear toward one end of the ranked list. The team ran each dataset separately with 1,000 permutations and required an adjusted P value below 0.05 and an absolute normalized enrichment score of at least 1.0 for a gene set to be called significant.
In the disc-degeneration data, reported negative signals included response to corticosterone, with a normalized enrichment score of -1.82 and a false-discovery-rate q value of 0.018, and glucosamine-containing compound metabolic process, with a score of -1.67 and q value of 0.031. In the osteoporosis data, amino sugar catabolic process showed negative enrichment, with a score of -1.74 and q value of 0.024. A positive signal for grooming behavior had a score of 1.59 and q value of 0.042, but the authors explicitly treated that result as exploratory.
What the findings can and cannot say
The biology suggested by these patterns should therefore be treated as a hypothesis about shared inflammation and extracellular-matrix processes, not as a demonstrated pathway. The underlying datasets were small and retrospective, and the first used whole blood from people with lumbar disc prolapse rather than disc tissue with histologically characterized degeneration. The osteoporosis comparison paired elderly patients with middle-aged donors, so age and clinical differences could affect the signal.
Because the datasets came from different tissue sources and microarray platforms and were not combined, the reported associations cannot establish causation or serve as validated biomarkers. The authors interpret the shared genes and related signatures as candidate clues for follow-up. Independent, adequately powered clinical cohorts, primary patient-derived cells and tissues, and gene-specific experiments would be needed to determine whether the candidates reproduce or play a causal role.
For now, the work is best read as a map for research rather than a tool for diagnosis, screening, prognosis or treatment. Its main contribution is to narrow a long list of gene-expression differences into biological signals that can be tested in better-matched human samples. Until that validation is done, the findings should not be used to guide individual medical decisions.
Paper data and sources
Original title: Integrated transcriptomic, machine learning, and network pharmacology analyses reveal shared inflammatory biomarkers between intervertebral disc degeneration and osteoporosis
Authors: Zheng M, Zhu G, Jiang H et al.
Journal/Repository: Frontiers in immunology
Status: Peer-reviewed
First online: 2026-10-19
DOI: Not available
Original paper · Full text