We downloaded the RNA-Seq gene expression in human muscle biopsies of 119 older adults (GSE111017) from the Gene Expression Omnibus (GEO) (https://www.ncbi.nlm.nih.gov/geo/). This dataset includes three individual datasets: Hertfordshire Sarcopenia, Jamaica Sarcopenia, and Singapore Sarcopenia. Hertfordshire's Caucasian dataset includes four sarcopenia patients with low muscle strength and physical performance, five with low muscle mass strength, three with low physical performance, and 28 healthy individuals (aged 68–77 years). The inclusion criteria of the Hertfordshire Caucasian dataset were the availability of birth records detailing birth weight, and men were excluded if they had heart disease, myositis, neuromuscular conditions, or diabetes. The diagnosis of sarcopenia among Caucasians was based on the EWGSOP algorithm (1, 9). Jamaica's Afro-Caribbean dataset includes nine sarcopenia patients with low muscle strength and physical performance, five with low muscle mass strength, 11 with low physical performance, and 14 healthy individuals, and all participants were of at least three grandparents of African origin (aged 63–89 years) (9). The diagnosis of sarcopenia among Afro-Caribbeans was also based on the EWGSOP definition (1). Singapore's Chinese dataset includes 20 sarcopenia patients and 20 healthy individuals without information on low muscle mass strength and the physical performance of older adults (aged 65–79 years). Each Chinese participant gave written self-reported ethnicity, and the diagnosis of sarcopenia was based on the AWGSOP definition (10). Lean mass was measured using DXA scanning (9). We merged the three datasets using the “merge” function in the R package “base”, performed batch effects adjustment, and normalized the combined data for further analyses.

We used the single-sample gene-set enrichment analysis (ssGSEA) score (11) to evaluate the enrichment level of an immune signature in older adults based on the expression profiles of its marker genes. The ssGSEA score represents the enrichment score of a gene set in a sample based on the degree to which the gene set is coordinately upregulated/downregulated in the sample. We analyzed 30 immune signatures, including anti-inflammatory cytokines, B cells, CD4⁺ Regulatory T cells, CD8⁺ T cells, central memory T cells, cytolytic activity, cytolytic protein perforin, dendritic cells, effector memory T cell, effector T cell, effector Treg T cells, exhausted T cell, granulocyte, IGF1, immune-modulatory molecules, macrophages, MHC Class I, myokines, naive T cell, NK antimicrobial protein granulysin, NK encoded activating receptors, NK encoded inhibitory receptors, proinflammatory cytokines, resident memory T cell, resting Treg T cells, senescence-associated secretory phenotype (SASP), Th1 cells, Th2 cells, Type I IFN response, and Type II IFN response. The gene sets representing these immune signatures are listed in Supplementary material S1.

We identified differentially expressed genes (DEGs) between sarcopenia (S) and non-sarcopenia, low muscle mass (LMM)/non, and low physical performance (LMM_LP)/non using Student's t-test with a threshold of adjusted P-value [false discovery rate (FDR)] of <0.05 and fold change (FC) of mean expression levels of >1.3. Based on the DEGs, we identified KEGG (12) and GO (13) pathways differentially enriched between the identified groups using the “pathfindR” R package (14), with a threshold of FDR <0.05. The FDR was calculated by using the Bonferroni method (15).

We predicted sarcopenia (S) patients vs. non-sarcopenia control individuals based on the combined gene expression profiles. We performed 3-fold cross-validation (CV) in 33 sarcopenia patients and 86 control individuals. We trained the elastic net regression model classifier and predicted sarcopenia patients vs. non-sarcopenia control individuals. We used the CV to select the best λ with a fixed α parameter of 0.5. We reported the prediction performance (accuracy, specificity, balanced accuracy, and the Kappa statistic) in the 3-fold CV. We carried out the class prediction algorithm in the “caret” R package (16).

We used iRegulon software (17) to directly enable gene regulatory network mapping based on motif enrichment in the upregulated gene set of S patients. The library of motifs used was 10K [9,713 position weight matrices (PWMs)], track collection of 1,120 ChiP-seq tracks, the putative regulatory region of 20 kb centered around transcription start site (TSS), motif ranking database 20 kb centered around TSS (seven species), and track ranking database of 20 kb centered around TSS (ChiP-seq-derived). In addition, we used an enrichment score threshold of 3.0, a ROC threshold for AUC calculation of 0.03, and a rank threshold of 5,000. For transcription factor (TF) prediction, the maximum FDR on the motif similarity threshold was 0.001. The weighted gene coexpression network analysis (WGCNA) was used to identify the gene modules of coexpressed genes among the three sarcopenia stages (LMM, LMM_LP, and S). Based on the expression correlations between the hub genes in gene modules, we identified the enriched GO pathways that have significant correlations with specific traits. We analyzed the WGCNA using the R package “WGCNA” (18).

We used Student's t-tests (two-tailed) to compare two classes of samples. In comparisons among three classes of samples, we used ANOVA tests (two-tailed). In evaluating correlations between gene expression levels and immune signatures' enrichment levels and between expression correlations between two genes, we used Pearson's correlation coefficients (R). We employed FDR to adjust P-values in multiple tests. The FDR was calculated using the Benjamini–Hochberg methods. We used abbreviations of S, LMM, and LMM_LP to represent sarcopenia, low muscle mass, and low muscle mass with low physical performance but without sarcopenia, respectively.

We first identified the transcriptional signature of S patients compared to healthy individuals. This analysis revealed significant expression differences in sarcopenia patients in 459 unique protein-coding transcripts (419 upregulated and 40 downregulated genes, Supplementary material S1). In LMM patients, we found 74 unique protein-coding transcripts (60 upregulated and 14 downregulated genes, Supplementary material S1). In LMM_LP patients, we found 27 unique protein-coding transcripts (20 upregulated and seven downregulated genes, Supplementary material S1). There were five common genes among the three sets of DEGs, including AC092184.1 (uncategorized gene); FAM239B (family with sequence similarity 239 member B pseudogene); HSFX2 (heat shock transcription factor family, X-linked 2); PTCD2P1 (pentatricopeptide repeat domain 2 pseudogene 1); and SLURP1 (secreted LY6/PLAUR domain containing 1).

We analyzed pathway enrichment on significant genes to identify activated or repressed pathways in a given sample or group. We calculated the agglomerated z-score of each enriched pathway per sample, followed by unsupervised hierarchical clustering of the most significant KEGG and GO pathways between the pre-defined groups. We found that sarcopenia was associated with increased expression of genes related to the GnRH, neurotrophin, Rap1, Ras, and p53 signaling pathways (Figure 1A). In addition, we found that sarcopenia patients showed significant enrichment in RNA polymerase II-specific DNA-binding transcription factor binding and negative regulation of cyclin-dependent protein serine/threonine kinase activity pathways (Figure 1B). Furthermore, the significantly activated pathways in LMM patients included the VEGF, B-cell receptor, ErbB, and T-cell receptor signaling pathways (Figures 1C, D). Interestingly, LMM_LP patients showed lower enrichment scores in the B-cell receptor signaling, apoptosis, HIF-1 signaling, and the adaptive immune response pathways than LMM patients (Figures 1E, F). These results indicate that immune responses and biological and metabolic dysregulation occurred simultaneously with the patient's muscle mass and function deterioration.

To identify the fewest transcripts as diagnostic biomarkers for S patients, we used the elastic net regression to derive a discriminating model in the transcriptomic data from S patients and healthy controls. We trained this model on multiple random samples of the transcriptomic data using CV and ranked all the genes based on their regression coefficients (Figure 2A). Among deferentially expressed genes (DEGs) and the elastic net regression model, we found five common genes (TTC39DP, SLURP1, LCE1C, PTCD2P1, and OR7E109P) that discriminated between sarcopenia patients and healthy controls with a different range (Figure 2A). The relative expressions of these genes were generally upregulated or downregulated in sarcopenia patients than in healthy controls (Figure 2B). To alleviate sampling error, we performed 3-fold CV train/test sequences to obtain an average accuracy of 0.97, 95% CI of 0.8871 and 0.9995, kappa of 0.95, a sensitivity of 0.97, a specificity of 1.00, and a balanced accuracy of 0.98. Among these top-ranked genes, we found that SLURP1 and LCE1C were upregulated in S patients, while TTC39DP, OR7E109P, and PTCD2P1 were downregulated in them (Figure 2B). TTC39DP (tetratricopeptide repeat domain 39D, pseudogene) was the highest ranked among the five genes and discriminated between S patients and healthy controls with a regression coefficient of 95.99. SLURP1 and LCE1C discriminated between S patients and healthy controls with regression coefficients of 62.45 and 34.54, respectively (Figure 2A).

We further analyzed the expression differences of TTC39DP, SLURP1, and LCE1C in S, LMM, and LMM_LP among Caucasian, Afro-Caribbean, and Chinese descents (Figure 2C). SLURP1 and LCE1C showed the highest expression levels in sarcopenic Chinese descent. In addition, SLURP1 expression levels were higher in the S group of Caucasian descent than in the LMM and LMM_LP groups. SLURP1 expression levels were higher in the LMM_LP group of Afro-Caribbean descent than in the S and LMM groups. TTC39DP showed the maximum expression levels among the LMM_LP group of Caucasian and Afro-Caribbean descent. These results indicate that different expression levels of TTC39DP, SLURP1, and LCE1C among the pre-defined groups depend on the patient's ethnicity.

We analyzed 96 significantly upregulated genes under S status using iRegulon software. The analysis yields a top-scoring regulon containing GATA1, GATA2, and GATA3 as master regulators and nine predicted direct target genes, namely EHF, SLURP1, MYEOV, SERPINB2, SERPINA12, ARHGAP40, PCDH15, SULT2B1, and POSTN (Figure 3A). The predicted GATA1, GATA2, and GATA3 targets are likely functional targets associated with the different pathways, including signaling, response to stimulus, regulation of the biological process, multicellular organismal process, metabolic process, locomotion, localization, growth, developmental process, cellular process, biological regulation, and biological adhesion behavior (Figure 3B). Interestingly, two genes were associated with locomotion: POSTN and SLURP1. POSTN has been associated with metastatic outgrowth in melanoma (19), while the association of SLURP1 with locomotion remains limited.

Furthermore, we conducted further analysis to investigate biomarkers of different sarcopenia stages, according to different pathophysiologic mechanisms. These mechanisms include the neuromuscular junction, endocrine system, growth factor, muscle protein turnover, behavior-mediated pathways, inflammation-mediated pathways, and redox-related factors (20). However, due to the small sample size, significant results were found only in muscle protein turnover and neuromuscular junction pathways (Figure 3C). As expected, muscle protein turnover and neuromuscular junction pathways were higher and lower with severe sarcopenic patients than with LMM and LMM_LP, respectively. These results indicate the association of sarcopenia with the impairment of neurophysiological functions, alteration in the transduction of the action potentials of muscle, and structural alterations of the muscles.

In addition, WGCNA identified 34 significant gene modules differentially enriched between the S, LMM, and LMM_LP groups (Figure 3D). The 34 representative colors of these modules are indicated in salmon2, maroon, orangered1, sienna4, lightpink4, yellowgreen, paleturquoise, steelblue, lavenderblush2, lightslateblue, skyblue4, thistle1, mediumorchid, red, violet, coral3, mediumpurple3, yellow4, indianred4, plum, antiquewhite2, mediumpurple4, salmon4, mediumpurple2, lightcoral, brown2, saddle brown, lightgreen, darkmagenta, lightcyan, darkgreen, grey60, greenyellow, and lightsteelblue. The representative GO terms for these gene modules were olfactory receptor activity, response to ketamine, galanin receptor activity, galanin-activated signaling pathway, intrinsic component of nuclear outer membrane, cardiac left ventricular formation, chemokine (C-C motif) ligand 4 production, keratin filament, dihydrofolate reductase activity, regulation of endothelial cell activation, myoblast development, spermatogenesis exchange of chromosomal protein, negative regulation of high-density lipoprotein particle clearance, RISC complex, ATP:ADP antiporter activity, positive regulation of interleukin-17 secretion, CCR3 chemokine receptor binding, extrinsic component of synaptic vesicle membrane, intermediate filament, phytanate-CoA ligase activity, oxidative phosphorylation uncoupler activity, negative regulation of heart induction by canonical Wnt signaling, otic vesicle formation, regulation of transcription from RNA polymerase II promoter involved in forebrain neuron fate commitment, cornification, epidermal cell differentiation, neuron fate commitment, sensory perception, photoreceptor outer segment membrane, detection of chemical stimulus involved in sensory perception, keratinocyte differentiation, formation of quadruple SL/U4/U5/U6 snRNP, dynein complex, and positive regulation of cell growth, respectively.

The S group was associated with enriched pathways of intrinsic component of the nuclear outer membrane, cardiac left ventricle formation, chemokine (C-C motif) ligand 4 production, keratin filament, dihydrofolate reductase activity, regulation of endothelial cell activation, intermediate filament, phytanate-CoA ligase activity, oxidative phosphorylation uncoupler activity, negative regulation of heart induction by canonical Wnt signaling, otic vesicle formation, regulation of transcription from RNA polymerase II promoter involved in forebrain neuron fate commitment, cornification, epidermal cell differentiation, sensory perception, photoreceptor outer segment membrane, detection of chemical stimulus involved in sensory perception, formation of quadruple SL/U4/U5/U6 snRNP, dynein complex, and positive regulation of cell growth (Figure 3D). In contrast, the red RISC complex module was negatively associated with the S group. Moreover, the LMM and LMM_LP groups shared two common pathways with the S group, namely epidermal cell differentiation and the RISC complex. The RISC complex pathway was highly enriched in the LMM and LMM_LP groups, while the S group observed the opposite. Interfering RNAs (siRNAs) and microRNAs (miRNAs) act via RNA-induced silencing complexes (RISC) and can regulate gene transcripts negatively (21). RISC involves many biological processes, such as brain aging (21). It has been proven that the dysregulation of miRNA expression has been associated with reduced muscle plasticity (22), as observed in our results, and, as a consequence, will impair skeletal muscle adaptations to exercise (22).

We analyzed the association of different immune signatures with patients' muscle mass and function deterioration. First, we conducted a t-test analysis to identify the immune signatures (30 immune signatures) differentiating sarcopenia from healthy individuals. A total of 14 immune signatures significantly differentiated sarcopenia from healthy individuals (Figure 4A). Only proinflammatory cytokines were significantly upregulated in sarcopenia than in healthy controls. In contrast, Th1&Th2 cells, resident memory T cells, NK activating receptor, NK antimicrobial protein granulysin, granulocytes, exhausted T cells, effector memory T cells, cytolytic protein perforin, cytolytic activity, CD4⁺ T cells, B cells, and anti-inflammatory cytokines were less enriched in sarcopenia patients than in healthy controls. Second, we used the ANOVA test to compare immune signatures' expression levels among the S, LMM, and LMM_LP groups. In total, nine immune signatures significantly differentiate the three groups (Figure 4B). Type I IFN response and dendritic cells were higher in sarcopenia patients than in the other two groups. However, signatures of Th1&Th2 cells, resident memory T cells, granulocytes, CD4⁺ T cells, B cells, and anti-inflammatory cytokines declined with worsening muscle function and disease progression. These results strongly correlate immune system dysregulation with disease progression and muscle function deterioration.

Furthermore, we used the elastic net regression to derive a discriminating model of the 30 immune signatures based on patients with and without sarcopenia after training this model on multiple random samples of the immune signatures using CV. We ranked all the immune signatures based on their regression coefficients. We found that Th2 cells and granulocytes could discriminate between sarcopenia and healthy individuals with a regression coefficient of 100 and 77.9, respectively (Figure 4C). Figures 4A, B show that the Th2 cells were significantly downregulated in sarcopenia patients relative to healthy controls, LMM, and LMM_LP. These results indicate that sarcopenia patients have weaker immune responses than other pre-defined groups.

In addition, we analyzed the association of various immune signatures with the pre-defined top-ranked genes (TTC39DP, SLURP1, and LCE1C) that discriminated sarcopenia from healthy individuals. TTC39DP has shown a significant positive correlation with 11 immune signatures and one negative correlation with proinflammatory cytokines (Figure 5A). The highest positive correlations were observed in granulocytes and B cells. In contrast, SLURP1 and LCE1C significantly negatively correlated with immunomodulatory molecules and exhausted T cells (Figures 5B–E). Interestingly, TTC39DP expression levels were much lower in sarcopenia patients than in healthy controls, and its upregulation correlated with a better prognosis and stronger immune profile in older people. In contrast, SLURP1 and LCE1C expression levels were much higher in sarcopenia patients and associated with a worse prognosis and weaker immune profile.

We identified 103, 57, and 66 genes with strong positive expression correlations with TTC39DP, SLURP1, and LCE1C, respectively (R > 0.5). Interestingly, SLURP1 and LCE1C showed a strong positive expression correlation (R = 0.9, p < 0.001, Figure 6A). In contrast, no significant correlations existed between SLURP1 and LCE1C with TTC39DP (Figures 6B, C). Pathway enrichment analysis identified several significant KEGG pathways associated with SLURP1 and LCE1C expression, including Renin secretion, Ras signaling pathway, Rap1 signaling pathway, Parkinson's disease, inflammatory mediator regulation of TRP channels, GnRH signaling pathway, gastric acid secretion, cGMP-PKG signaling pathway, cAMP signaling pathway, calcium signaling pathway, p53 signaling pathway, neurotrophin signaling pathway, and C-type lectin receptor signaling pathway. These pathways were associated with the 57 and 66 genes strongly correlated with SLURP1 and LCE1C, respectively (Figures 6D, E, FDR < 0.05). However, pathway enrichment analysis with 103 correlated genes with TTC39DP did not show significantly enriched pathways.