COVID-19 multi-omics reveal organ-specific responses and biomarkers

Abstract

Objective

Post-COVID-19 syndrome is characterised by persistent immune dysfunction and multi-organ sequelae. This study aimed to characterise the systemic blood molecular landscape induced by SARS-CoV-2 infection and identify prognostic markers linked to skeletal muscle mass loss, a key driver of poor outcomes.

Methods

We enrolled 30 healthy controls and 307 COVID-19 patients, collecting 422 plasma samples for integrated proteomic and metabolomic profiling to investigate organ-specific molecular alterations in COVID-19.

Results

We comprehensively mapped the molecular landscape of COVID-19, encompassing immune, tissue-specific, and metabolic perturbations, and delineated their interactions. Focusing on organ-damage-related molecular patterns associated with disease progression and mortality, we found that skeletal muscle mass loss contributed to poor clinical outcomes of COVID-19 (p < 0.0001). Dysregulated arginine metabolism emerged as a key metabolic signature in fatal COVID-19 cases, with GLUL, GOT1, and citrulline showing significant correlation with skeletal muscle mass loss. Longitudinal analyses further revealed that reduced citrulline levels underlie the poor outcome of COVID-19 patients with muscle mass loss. These findings were robustly supported through multiple approaches: Mendelian randomization confirmed causal relationships between citrulline depletion, sarcopenia/fat-free mass loss, and COVID-19 mortality (p < 0.05), transcriptomic analyses of SARS-CoV-2-infected golden hamsters (GSE231910) provided additional support in enrichment of arginine biosynthesis (FDR < 0.05), and in vitro experiments further demonstrated that citrulline depletion promotes pro-inflammatory M1 macrophage polarisation — a key immunological feature of critical COVID-19. Leveraging these insights, we developed a skeletal muscle loss-specific prognostic prediction model for COVID-19 using GLUL, GOT1, and citrulline. This model effectively stratified patients into high- and low-risk groups (p = 0.035).

Conclusion

Our study advances the understanding of COVID-19-induced organ pathophysiology and provides a foundation for developing targeted therapeutic strategies for post-COVID sequelae.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12967-025-07459-2.

Introduction

The novel coronavirus disease 2019 (COVID-19) pandemic represents a global public health crisis of unprecedented scale. The pandemic has affected hundreds of millions worldwide, posing a substantial threat to public health and the economy. More seriously, the constant emergence of several variants of SARS-CoV-2 has led to increased transmission and immune escape of COVID-19, with the disease’s prevalence remaining uncertain and causing the recurrence of the epidemic [1, 2].

In post-COVID-19 condition (also known as long COVID), patients are still faced with problems of immune deficiencies and long-term sequelae, including fatigue, breathlessness, and cognitive dysfunction [3]. This condition often involves multi-organ dysfunction, affecting the liver, nervous system, heart, and skeletal muscle, leading to severe and protracted impairment of function [4]. One previous study [5] has revealed the pathophysiology of post-exertional malaise in patients with long COVID, including local and systemic metabolic disturbances, severe exercise-induced myopathy, and tissue infiltration of amyloid-containing deposits in skeletal muscles. Therefore, it is crucial to actively elucidate the molecular dynamics of the host and the multi-organ loss mechanisms triggered by SARS-CoV-2 to improve diagnostic and therapeutic strategies for patients with post-COVID sequelae.

Multi-omics is a valuable technique for deciphering the molecular landscape of diseases. Systematic screening of proteins, metabolites, and lipids during the progression of COVID-19 can reveal abnormal physiological regulatory processes in patients, including the complement system, cytokine release, and platelet degranulation [6]. Clinical proteomics enables the investigation of interactions between viral and host proteins, identifying the potential role of novel coronavirus proteins, and suggesting host factor processes and therapeutic interventions [7, 8]. Significant correlations have been revealed between metabolites and pro-inflammatory cytokines/chemokines in COVID-19 hosts, including IL-6, M-CSF, IL-1α, arginine, tryptophan, and purine metabolism [9]. However, specific molecular alterations and biomarkers associated with multi-organ tissue injuries have rarely been considered. These factors have limited our understanding of COVID-19 markers and biopathology, resulting in an inadequate range of treatment options.

To address this gap, this study collected 422 plasma samples through sequential time-point sampling from 30 healthy controls and 307 patients with COVID-19 during their admission for proteomics/metabolomics analyses. Clinical phenotypic and follow-up data were obtained. By integrating the analyses of the clinical phenotype, proteome, and metabolome, we explored the immune-metabolic landscape and interactions between organ tissues, immune cells, and blood molecules of COVID-19. We examined the distinct organ-specific molecular alterations of COVID-19 and revealed the relationships between skeletal muscle mass loss and clinical outcomes. Additionally, we investigated the effect of decreased citrulline levels on inflammatory macrophage polarisation and the clinical outcome of COVID-19 patients, which was further validated through a public data-based Mendelian randomisation (MR) analyses. Finally, GLUL, GOT1, and citrulline were identified as the prognostic markers specific to skeletal muscle loss of COVID-19, and a survival random forest model was developed for prognosis prediction.

Materials and methods

Ethics statement

The experimental protocol was established in accordance with the ethical guidelines of the Helsinki Declaration and was approved by the institutional review board and Ethics Committee (NCT05792865, 2023-Research-230). The requirement for informed written consent was waived because it is a retrospective study.

Patient enrolment

This study included 307 patients diagnosed with COVID-19 and 30 healthy volunteers. The inclusion criteria for patients with COVID-19 were: (1) Positive for SARS-CoV-2 via nasopharyngeal swabs with real-time RT-PCR; (2) age ≥ 18 years; and (3) at least one high-risk factor (age ≥ 60 years; history of diabetes, hypertension, cardiovascular disease, chronic lung disease, cerebral infarction, chronic liver, and kidney disease, cancer, smoking, or obesity). The exclusion criteria for patients with COVID-19 were: (1) death within 48 h of admission. (2) Pregnant or lactating women; (3) Participants in other interventional clinical trials.

Plasma sample collection

422 blood samples were obtained from 337 individuals admitted to Beijing Chao-yang Hospital, Capital Medical University, between December 1, 2022, and March 30, 2023. Among these, 337 baseline (first sampling) blood samples and 85 follow (sequential sampling) blood samples were collected using sequential time-point sampling. Blood samples were collected in ethylenediaminetetraacetic acid (EDTA) anticoagulation tubes and centrifuged at 1107 ×g for 10 min at 4 °C. The supernatant plasma was carefully aspirated, avoiding the buffy coat and platelet layer, and was immediately aliquoted to prevent metabolite degradation. All aliquots were then promptly stored at -80 °C until analyses.

Clinical definitions

Patients were classified into non-severe (NS), severe (S), and critical (C) groups according to disease severity. Severe disease was defined as the presence of at least one of these conditions: (1) percutaneous arterial oxygen saturation (SpO2) ≤ 93% at rest in the absence of oxygen inhalation; (2) respiratory distress, respiratory rate ≥ 30 rpm; (3) oxygenation index ≤ 300 mmHg; and pulmonary involvement > 50%. Critical disease was defined as the presence of at least one of these conditions: (1) the need for mechanical ventilation and/or pressor medications, (2) fulfilment of the diagnostic criteria for acute respiratory distress syndrome, and (3) fulfilment of the diagnostic criteria for septic shock. NS disease was defined as mild clinical symptoms that did not meet the criteria for a severe disease.

Patients were classified into the improved group and the worsened group according to the patient’s disease progression within 24-hour admission. Patients who met one of the following conditions were defined as a worsened group; otherwise, they were defined as an improved group: (1) Used vasoactive drugs within 24-hour admission; (2) sepsis/septic shock within 24-hour admission; (3) ARDS within 24-hour admission; (4) intubation within 24-hour admission.

Patients who died within 28-day hospitalisation were defined as the fatality group, those who survived to be discharged from the hospital within 28 days were defined as the survivor group, and those who still received treatments in the hospital after 28 days’ hospitalisation were not defined as either group.

LC-MS/MS-based proteomics analyses

The sample preparation procedures for proteomics included protein extraction, LysC/trypsin digestion, and peptide desalting. (1) Protein Extraction and LysC/Trypsin Digestion: Ten microlitres of plasma sample were aliquoted into a high-select top14 abundant protein depletion resin and incubated on a shaker for 30 min. The filtrate was collected by centrifugation after incubation, 200 µL of 8 M urea was added, and the mixture was centrifuged at 12,000 ×g for 10 min. DTT was added at a final concentration of 10 mM, and the mixture was incubated at 37 °C for 30 min. IAM was added at a final concentration of 20 mM, and the mixture was incubated in the dark for 30 min. Subsequently, 50 mM NH₄HCO₃ was added, and the mixture was centrifuged at 12,000 ×g for 10 min. Subsequently, 200 µL of 50 mM NH₄HCO₃ and 1 µg of LysC were added, and the mixture was incubated at 37 °C for 2 h. Finally, 1 µg of trypsin was added, and the mixture was incubated at 37 °C overnight. Finally, 10 µL of 10% TFA was added to terminate the peptide digestion reaction. (2) Peptide desalting: One hundred microlitres of methanol were added to the SoLAµ HRP plate and centrifuged at 600 ×g for one minute. One hundred microlitres of 80% acetonitrile containing 0.1% TFA were added and centrifuged at 1000 ×g for 1 min. Two hundred microlitres of 0.1% TFA were added and centrifuged at 1000 ×g for 1 min. Subsequently, the sample was transferred to a SoLAµ HRP plate and centrifuged at 1000 ×g for 2 min, and the sample loading procedure was repeated. Then, 200 µL of 0.1% THA was added for centrifugation at 1000 ×g for 2 min, 100 µL of 80% acetonitrile containing 0.1% TFA was added and centrifuged at 1000 ×g for 3 min, and the eluate was collected and concentrated in a concentrator at 40 °C. Finally, the remaining residues should be dissolved with 0.1% FA to a final peptide concentrate of 0.5 µg/µl, which is then ready for LC-MS analyses.

Peptide separation was conducted using a nano LC-Ultimate™ 3000 RSLC system (Thermo Fisher Scientific, USA) with an Acclaim PepMap C18 trap column (3 μm, 100 Å, 75 μm × 20 mm) (Thermo Fisher Scientific, USA) and analytical column (2 μm, 100 Å, 75 μm × 25 cm) (Thermo Fisher Scientific, USA). The mobile phase comprised a water solution containing 0.1% formic acid (A) and an 80% acetonitrile solution containing 0.1% formic acid (B). The gradient elution settings were as follows (flow rate = 400 nL/min): 0–4 min, 1–8% B; 4–85 min, 8–30% B; 85–90 min, 30–90% B; 90–91 min, 90–1% B; 91–95 min, 1–1% B. The temperature of the column oven was maintained at 55 °C. Data acquisition was conducted using a Q-Exactive HF-X Orbitrap high-resolution mass spectrometer (Thermo Fisher Scientific, USA). The primary parameters were spray voltage: 2.1 kV; capillary temperature: The temperature of the S-lens was 300 °C. The collision energy was set at 32% HCD, and the resolution was set to full MS 60,000@m/z 200 and MS/MS 30,000@m/z 200. The maximum injection time was set to 20 ms for full MS and 45 ms for MS/MS. The full MS scan range was set to m/z 350–1200, and the MS/MS scan range was set to 200–2000. The number of DIA windows was 50, and the isolation window was 17 Th.

Proteomics data pre-processing and peptide identification

Proteomics data pre-processing was conducted using DIA-NN v1.8.0 software. All raw data from the DIA experiment were imported into DIA-NN v 1.8.0 for targeted extraction [10, 11]. This was performed with the ultra-deep plasma spectral library by default to control the false discovery rate (FDR) at a level of less than 1% for peptide and protein levels, with the default parameters being used. Protein intensities were calculated using DIAN from the mean of the top three peptides. To reduce the analytical variations and ensure the reliability of the downstream analyses and results, the raw intensity data of proteins were normalised by variance stabilizing normalization (Vsn) using the vsn package [12] before statistical analyses.

LC-MS/MS-based metabolomics analyses

Aliquots of the samples were mixed evenly to prepare quality control (QC) samples for system pre-equilibration, system stability evaluation, and data quality control. A total of 50 µL of the plasma sample and 400 µL of methanol (with 10 µL of isotopic internal standard) were added to a 1.5 mL Eppendorf tube and incubated for 20 min. Subsequently, the supernatant was collected by centrifugation and dried in a centrifugal concentrator at a low temperature under reduced pressure. Finally, the dry residue was reconstituted with 200 µL of reconstituting solvent for LC-MS analyses.

Untargeted metabolomics analyses was conducted using Thermo Vanquish (Thermo Fisher Scientific) coupled with a Q-Exactive Plus Orbitrap high-resolution mass spectrometer (Thermo Fisher Scientific, USA). QC samples were inserted in an interval of ten test samples to monitor the stability of the instrument and normalise the variations during the run. Chromatographic separations were conducted using the Acquity™ BEH C18 column (1.7 μm, 2.1 × 100 mm, Waters, USA) and the Acquity™ BEH AMIDE column (1.7 μm, 2.1 × 100 mm, Waters, USA). The mobile phase for C18 column comprised a water solution containing 0.1% formic acid and 5 mM ammonium formate (A), and a methanol solution containing 5 mM ammonium formate (B). The gradient elution settings for C18 column were as follows (flow rate = 0.3 mL/min): 0–0.5 min, 2–2% B; 0.5–10 min, 2–98% B; 10–16 min, 98–98% B; 16–16.1 min, 98–2% B; 16.1–18 min, 2–2% B. The column oven temperature was 40 °C, and the injection volume was 5 µL. The mobile phase for the BEH AMIDE column comprised water containing 25 mM ammonium formate (adjust pH to 9.0 with ammonia water) (A) and acetonitrile (B). The gradient elution settings for the BEH AMIDE column were as follows (flow rate = 0.4 mL/min): 0–0.5 min, 95–95% B; 0.5–7 min, 95–65% B; 7–8 min, 65–40% B; 8–9 min, 40–40% B; 9–9.1 min, 40–95% B; 9.1–12 min, 95–95% B. The column oven temperature was 50 °C, and the injection volume was 5 µL.

Mass spectrometry data acquisition was conducted using the full scan plus DDA data acquisition mode in positive and negative ion modes. The ion source parameters were: Spray voltage: 4.0 kV(+) and 3.5 KV(-); capillary temperature, 320 °C; S-lens RF level, 50%; sheath gas flow, 45 arb; auxiliary gas flow rate, 10 arb; Aux gas heat temperature, 355 °C; collision energy, 40% HCD; fragmentation gas, ultra-pure nitrogen. The mass spectrometry parameters were scan range: m/z 70-1050; resolution setting, full MS 70,000 at m/z 200; MS/MS 17,500 at m/z 200; AGC target: full MS 1e6, MS/MS 1e5; Max IT full MS 100 ms, MS/MS 50 ms. Loop count: 10; isolation window: 1.5 m/z; (N) CE stepped: Nce, 20, 40, and 60; dynamic exclusion: 4.0 s.

Metabolomics data pre-processing and metabolite identification

Untargeted metabolomic data pre-processing was conducted using Compound Discoverer v.3.1. The raw data files were imported into the software to extract peak features. Then, a matrix containing the intensity, mass-charge ratio, and retention time of peaks was obtained. To reduce the analytical variations and ensure the reliability of the downstream analyses and results, the raw intensity data of metabolites were normalised by Vsn (3), and the batch effects were further removed by systematic error removal random forest method (SERRF) [13]. Subsequently, the peak features with mass-charge ratio, retention time, and associated MS/MS scans were employed to search databases (in-house standard database, NIST, mzCloud, and HMDB public spectra database) to assign formulas and structures. Finally, the metabolites were identified and annotated in accordance with the proposal of the Metabolomics Standardization Initiative [14], and additional identification confidence was provided for more concise and accurate annotation. In addition, all plasma metabolites exhibited high identification confidence levels (Levels 1 and 2). Level 1: Metabolites were matched with in-house standards by retention time, MS spectrum, and MS/MS spectrum after an in-house standard database search and comparison. Level 2: Metabolites were matched with the compounds by MS and MS/MS spectra after a public spectrum database (HMDB, NIST, and mzCloud databases) search and comparison.

Cell culture and treatment

RAW264.7 cells (CL-0190, Pricella, Wuhan, China) were cultured in Dulbecco’s Modified Eagle’s Medium (DMEM, C11965500BT, Gibco) supplemented with 10% heat-inactivated fetal bovine serum (FBS, ST30-2602, PAN-Biotech GmbH) at 37℃ in a humidified incubator under 5% CO₂. Cells were cultured to approximately 80% confluency and then subjected to no more than 20 cell passages. For experiments, cells were plated into 6-well plates at a density of 2.5 × 10⁵ cells per well. The treatment protocol was as follows: cells were pre-treated with the nitric oxide synthase inhibitor L-NAME (N5751, Sigma, 100 µM) or vehicle control for 2 h, followed by supplementation with citrulline (C7629, Sigma; 500 µM) or vehicle control for 30 min. Cells were subsequently stimulated with lipopolysaccharide (LPS; L2880, Sigma, 100 ng/mL) or vehicle control for 4 h to induce macrophage polarization.

RNA extraction and quantitative real-time PCR

Following 4 h of LPS stimulation, total RNA was isolated from RAW264.7 cells using TRIZOL reagent. Subsequently, 1 µg of the total RNA was reverse-transcribed into complementary DNA (cDNA) using the HiScript II Reverse Transcriptase kit (R223, Vazyme). Quantitative real-time PCR (qRT-PCR) was performed using SYBR qPCR Master Mix (Q331-03, Vazyme) in a 20 µL reaction volume containing 2 µL of cDNA template. The relative mRNA expression levels of target genes were calculated using the method, with β-actin serving as the internal control for normalization. The sequences of all primer pairs used in this study, which were synthesized by BGI, are provided in Supplementary Data Table S1.

Tissue-enhanced proteins

Tissue-enhanced proteins were defined as proteins encoded by genes that have an elevated expression (>= four fold in mRNA level) in the specific type of tissue compared with the average level in all other tissues [15]. To investigate the tissue specificity of the protein expressions, a total of 6,197 tissue-enhanced proteins were obtained from the Human Protein Atlas (HPA) database [16]. Then, 565 tissue-enhanced proteins were identified as detected in our proteomic dataset by searching the list of 6197 tissue-enhanced proteins. The number and proportion of differentially expressed tissue-enhanced proteins across various tissues were assessed to evaluate the extent of the damage to multiple organs.

Skeletal muscle mass evaluation

First, the appendicular muscle mass (AMS) of all COVID-19 patients was evaluated with an anthropometric prediction equation (PE) [17, 18]: AMS = 10.05 + 0.35 (weight) − 0.62 (body mass index, BMI) − 0.02 (age) + 5.10 (if male). Then, the skeletal muscle index (SMI) was calculated by adjusting ASM to height in meters squared [19]: SMI = ASM/height². Finally, low muscle mass was defined as values below a predetermined SMI cutoff value, which was 7.36 kg/m² in men and 5.81 kg/m² in women [20].

Mendelian randomisation (MR) analyses

A two-sample MR analyses was performed to explore the causal effects of sarcopenia/fat-free body mass, citrulline, and COVID-19 death, which was executed by ‘TwoSampleMR’ package [21]. The GWAS data for sarcopenia (GCST90007529: 256,523 Europeans; GCST90007531: 121,055 Europeans) and fat-free body mass (GCST90428121: 337,196 Europeans) were obtained from NHGRI-EBI GWAS catalog [22]. GWAS data for citrulline (GCST90200403: 8,157 Europeans; met-a-356: 7,773 Europeans) were obtained from the IEU OpenGWAS database [23] and NHGRI-EBI GWAS catalog [22]. GWAS data for COVID-19 death (UKBB_death_ALL_010421: 474 cases of COVID-19 death and 486,846 cases of normal population; UKBB_death_ALL_050921: 1,104 cases of COVID-19 death and 486,216 cases of normal population; UKBB_death_ALL_071420: 364 cases of COVID-19 death and 486,956 cases of normal population; UKBB_death_ALL_090820: 416 cases of COVID-19 death and 486,904 cases of normal population; UKBB_death_ALL_110320: 418 cases of COVID-19 death and 486,902 cases of normal population) was obtained from website (https://grasp.nhlbi.nih.gov/covid19GWASResults.aspx).

Five methods [inverse-variance-weighted (IVW), MR-Egger, weighted median, simple mode, and weighted mode] were used to evaluate and validate causal effects, and inverse-variance-weighted was the primary method used. Cochran’s Q statistic was calculated to evaluate heterogeneity. The MR-Egger intercept was performed to evaluate pleiotropy. Only a causal effect with an inverse-variance-weighted p-value of < 0.05 and without heterogeneity and pleiotropy (corresponding p-value of > 0.05) was identified as a significant effect.

Multi-omics factor analyses (MOFA)

Proteins and metabolites with FDR < 0.05 by ANOVA test followed by B-H adjustment between healthy controls (H), severe (S) patients, and critical (C) patients were selected for MOFA with the MOFA2 package (version 1.8.0). Besides, the expression data of those selected proteins and metabolites were Z-score-standardised before MOFA [24]. Multi-omics data were disassembled and reassembled into different factors, and the correlations between clinical variables and multi-omics factors were evaluated using Pearson correlation analyses.

Functional enrichment analyses

Protein functional analyses was conducted using the web-based platform Metascape (https://www.metascape.org) [25], which integrates various databases, including Gene Ontology Biological Processes, Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways, Reactome, and Wiki pathways. The significant protein-enriched pathways were identified as the pathways with FDR < 0.05. Metabolite functional analyses was conducted using MetaboAnalyst 6.0 [26], and the significant metabolite-enriched pathways were identified as the pathways with p < 0.05. Integrative pathway functional analyses was conducted using both differentially expressed proteins and metabolites with MetaboAnalyst 6.0, and significant integrative enriched pathways were identified as those with p < 0.05.

Gene set enrichment analyses

Gene set enrichment analyses was implemented using clusterProfiler package. The KEGG pathway analyses utilised the c2.all.v2024.1.Hs.symbols.gmt gene sets on the MsigDB database.

Immune infiltration analyses

The immune infiltration was executed with the CIBERSORT package [27], which is a deconvolution algorithm that can evaluate the proportion of 22 infiltrating lymphocyte subsets in a large number of samples.

Cox regression model and survival analyses

To identify proteins and metabolites associated with the 28-day prognosis of COVID-19 hospitalised patients, a multivariate Cox proportional hazards model analyses that adjusted for age, sex, BMI, and severity was implemented, and the proteins and metabolites with p < 0.05 were finally selected. Survival analyses was conducted using the Kaplan–Meier method, with proteins and metabolites initially stratified into high and low expression groups based on median levels. The significance threshold for the log-rank test was set at p < 0.05.

Dynamic trajectory analyses and network of proteins and metabolites

Firstly, proteins and metabolites with FDR < 0.05 by ANOVA test followed by B-H adjustment between healthy controls (H), severe (S) patients, and critical (C) patients were selected. Their expressions among each group were used for fuzzy c-means clustering, executed through the Mfuzz package (version 2.58.0) [28]. Then those proteins and metabolites were clustered into different groups. Next, those proteins and metabolites with a membership >0.5 or 0.25 were further categorised into several clusters (cluster 1, 2, 3, and 4) based on expression trends. Finally, the alteration trajectory of proteins and metabolites across H, S, and C groups was depicted with a heatmap and line plots (average ± SEM).

Finally, the Pearson correlation coefficients between proteins and metabolites of different clusters (cluster 1, 2, 3, and 4) were calculated with the Hmisc R package (version 4.7-2). The significant correlations between proteins and metabolites (p < 0.05 & |r| >0.5) were input into Cytoscape software (version 3.10.1) [29] to construct a correlation network for visualisation. The top 5 most connected proteins/metabolites (those with the top 5 most connections) were listed in the network.

Prognosis predictive model

GOT1, GLUL, and citrulline were identified as COVID-19 prognosis predictive markers associated with skeletal muscle loss, and their expression data were used for further model development. To develop a robust prognosis predictive model, 67 patients were randomly divided into a train set (fatality group, n = 48; survivor group, n = 197) and a test set (fatality group, n = 48; survivor group, n = 197). The survival random forest model was trained on the train set with the ‘randomForestSRC’ package [30], and the hyperparameters were optimised based on grid search with 5-fold cross-validation. The best model parameters were selected according to the C-index. The optimal model was established with the following parameters: n tree = 10, mtry = 1, node size = 5, nsplit = 1. The model displayed great predictive power (AUROC at 5-day = 0.94, 95% Cl: 0.88–1.00; AUROC at 15-day = 0.79, 95% Cl: 0.69–0.90; C-index = 0.70, 95% CI: 0.56–0.83) for predicting the survival outcomes of COVID-19 patients on the test dataset.

The risk score of each patient was estimated with the model, which could be used as an evaluation criterion to assess the survival outcomes of COVID-19. The best cutoff was 6.06, which was determined by using the surv_cutpoint function in the ‘survminer’ R package. Patients with a risk score greater than 6.06 would be stratified into a high-risk group, indicating a higher risk of poorer survival outcomes.

Statistical analyses

All data were subjected to analyses using R (version 4.2.2). The differential comparisons of continuous clinical variables were performed using the Student’s t-test or Wilcoxon rank-sum test, whereas the differential comparisons of categorical variables were performed using the chi-square test or Fisher’s exact test. One-way analyses of variance (ANOVA) followed by B-H adjustment was used to identify the differentially expressed variables across H, S, and C groups. Analyses of differential expression for proteomic and metabolomic data was conducted on log2-transformed data using the limma package (version 3.45.0) [31] with adjustments for age and sex. Thresholds for differential metabolites and proteins were set as: FDR < 0.05 and |log2(fold change)| >0.5. Paired Wilcoxon rank-sum test was conducted to identify the molecules altered over time in the worsened/improved group (p < 0.05). Pearson correlation coefficients were calculated between omics molecules and clinical variables using the Hmisc package. Heatmaps were constructed using the pheatmap package, bubble plots, box plots, and bar plots, and rose plots were implemented using the ggplot2 package, and a circos plot was constructed using the circliz package.

Results

Study design and sample characteristics

This study systematically analysed the immune and organ-specific molecular responses to COVID-19 using metabolomics and proteomics techniques. The overall workflow of this study is illustrated in Fig. 1A. Specifically, the cohort study included 5 non-severe (NS), 256 severe (S), and 46 critical (C) patients with COVID-19, as well as 30 healthy (HC) volunteers. The detailed participant information is presented in Supplementary Data Table S2 and Fig. 1B. Among the 256 severe patients, 39 were grouped into the improved group and 217 were in the worsened group according to their disease progression within 24 h of admission. Among the 307 patients with COVID-19, 48 patients died, 197 survived to be discharged from hospital, and the remaining 62 were still receiving treatments in the hospital within the 28-day follow-up period. A total of 422 blood samples were collected for proteomic and metabolomic analyses. Of these, 337 were collected at the initial sampling point (baseline) and 85 were collected at subsequent sampling points (follow-up). The inter-individual correlation and variability analyses demonstrated that the multi-omics data exhibited good reproducibility (Figure S1A-B). A total of 1651 proteins and 1687 metabolites were identified in the plasma samples (Figure S1C-D, Supplementary Data Table S3-6).

Fig. 1.

Overview of the schematic workflow and clinical parameters in the study. (A) Schematic summary of the study design. (B) Summary of omics measurements and clinical parameters of patients enrolled in the omics study. The y-axis displays COVID-19 patients, and the x-axis indicateds the number of days before and after admission (since disease onset)

Immune infiltration and multi-omics molecular panorama of COVID-19

To determine whether the pathophysiology of COVID-19 is associated with specific molecular changes, we compared proteomic and metabolic profiles of plasma samples from HC, NS, S, and C patients (Figure S2A). 180/271 proteins and 488/752 metabolites were significantly altered in S and C patients, respectively (Figure S2B), and a range of disorders in biological processes and pathways, including innate immune response, platelet activation, neutrophil degradation, arginine biosynthesis, and glycolysis/gluconeogenesis, were observed in patients (Figure S3A-B). Remarkably, significant correlations were observed between molecules and clinical parameters (Figure S2C). To interpret the correlations between molecular heterogeneity and clinical phenotype, we reduced the proteomic and metabolomic profiles into 12 factors (Figure S2D) by MOFA. Factor 7 was positively correlated with D-dimer levels, and factor 1 was positively correlated with death, 28-day prognosis, disease severity, creatinine level, and Charlson Comorbidity Index (CCI) (Figure S2E-F). These molecules were enriched in PPAR, ether lipid metabolism, and sphingolipid metabolism pathways.

We investigated the immune infiltration panorama of COVID-19 via the CIBERSORT algorithm. The ratio of 22 immune cells in the samples is shown in Fig. 2A. Compared with controls, COVID-19 patients had higher levels of B cells memory, T cells CD8, macrophages M0, M1, M2, dendritic cells activated, mast cells activated, eosinophils, and neutrophils; and had lower levels of B cells naïve and T cells CD4 naïve (Fig. 2B).

Fig. 2.

Immune infiltration and molecular panorama of COVID-19. (A) The ratio of 22 immune cells in each sample of healthy controls (H), severe COVID-19 (S), and critical COVID-19 (C) patients. (B) The proportion of immune cells in healthy controls and patients with severe and critical COVID-19 (*, p < 0.05; ns, not significant). (C) Pathway flowchart combining proteins and metabolites depicting key disturbed pathways in response to SARS-CoV-2 infection

Next, we depicted a molecular landscape of COVID-19 over disease progression with the differentially expressed proteins and metabolites (Fig. 2C). The expression levels of L-citrulline, aspartate, N-acetylornithine, and glutamate-ammonia ligase (GLUL) were reduced in patients with COVID-19. Conversely, urea and argininosuccinic acid levels were elevated. It has been demonstrated that citrulline depletion is necessary for the activation of pro-inflammatory macrophages and subsequent immune responses [32]. This phenomenon may be linked to inflammatory cytokines induced by SARS-CoV-2 [33]. The pentose phosphate pathway, glycolysis/gluconeogenesis, and nicotinate and nicotinamide metabolism were consistently upregulated at both the metabolite and protein levels. Studies have demonstrated that glycolysis and the pentose phosphate pathway can promote inflammation in immune cells [34]. The release of inflammatory cytokines in severe COVID-19 cases may be modulated by glycogen metabolism [35].

Dynamic molecular trajectory with the progression of COVID-19 disease

Four main clusters of COVID-19 disease progress evolutionary trajectories were identified that delineated key biological processes encompassing H, S, and C stages (Figure S4A). In response to SARS-COV-2 infection, molecules in cluster 1 (197 metabolites, 182 proteins) increased stepwise and molecules in cluster 2 (141 metabolites, 42 proteins) decreased stepwise. Only a minute amount of the molecules increased at the severe disease stage and quickly returned at the critical disease stage (cluster 3: 76 metabolites, 14 proteins) or decreased following severe stages and quickly returned at critical stages (cluster 4: 23 metabolites, 53 proteins). Enrichment pathways and correlation networks were generated for each cluster, highlighting potential regulators of biological processes and novel molecular functions through unexpected connections (Figure S4B-C).

Cluster 1 was enriched in molecules associated with thiamine metabolism, cysteine and methionine metabolism, pentose phosphate pathway, glutathione metabolism, and glycolysis or gluconeogenesis (Figure S4B), which demonstrated that the REDOX metabolism, protein biosynthesis and methylation, and central carbon metabolism (energy metabolism) are up-regulated along with COVID-19 progresses. As expected, we observed a sharp increase in plasma concentrations of carbohydrates (glucose, ribose, xylulose) and sulphur-containing amino acids (taurine, cystathione, cysteine, N-formyl-methionine), as shown in Figure S5A-C. 2-HG, cystatin C (CST3), N-acetyl galactosamine 4-sulphate (ASB), N, N-dimethyl guanosine (DMG), and vesicular integral-membrane protein VIP36 (LMAN2) were highlighted as the key regulators of Cluster 1 (Figure S4C). 2-HG is the reduced form of α-KG in TCA pathway, and the accumulation of 2-HG correlates with ROS production, abnormal methylation, and disruption of mitochondrial structure and function. LMAN2 plays a role as an intracellular lectin in the early secretory pathway. SARS-CoV-2 infection alters cellular RNA content, and our study found the characteristic elevations of specific modified nucleosides DMG in patients’ plasma during SARS-CoV-2 infection. DMG is a purine nucleoside and a primary degradation product of transport RNA; elevated circulating DMG levels might indicate cellular stress and be associated with the damage of tissue and immune cells upon SARS-CoV-2 infection [36]. CST3 is a small molecular protein that can freely pass through the glomerulus and can be used as a predictor of renal function [37]. The higher level of CST3 has been reported to be associated with high all-cause mortality and cardiovascular risk [38].

Cluster 2 was enriched in molecules associated with glycerophospholipid metabolism, alpha-linoleic acid metabolism, and biosynthesis of unsaturated fatty acids (Figure S4B). As expected, we observed a continuous decline trend in most free fatty acids, oxidative lipids, lysophosphocholines, lysophosphoethamines, and the related enzymes (Figure S5D-G). The correlation network in cluster 2 identified LPC(22:0), 6-HODE, LPC(18:0), 9,10-DHOME, and biotinidase (BTD) as the critical regulators (Figure S4C). Phospholipids are major components of cell membranes that can be metabolised to produce lysophospholipids and free fatty acids, and lower levels might reflect that viral replication depends on specific lipid components [39]. Pooled evidence suggested that polyunsaturated fatty acids control protein complex formation in lipid rafts associated with the function of two SARS-CoV-2 entry gateways [40]. 6-HODE and 9,10-DHOME were naturally occurring lipoxygenation products of linoleic acid. A previous study found a linoleic acid binding pocket in the locked structure of SARS-CoV-2 spike protein [41], providing a new strategy by supplementing linoleic acid for developing new drugs. Biotin is a vital cofactor for biotin-dependent enzymes for fatty acids, amino acids, and glucose metabolism. The low expression of BDT leads to the decreased activity of various carboxylases, resulting in mitochondrial function disorders, metabolic acidosis, organic aciduria, and a series of neurological and skin system damage [42].

Cluster 3 and cluster 4 comprise a small number of fluctuating metabolites and proteins, including butanone metabolism, glutathione metabolism, and nitrogen metabolism, which highlighted 3-hydroxybutyrate (3-HB) and anthrax toxin receptor 2 (ANTXR2) as the key regulators (Figure S4B-C). ANTXR2, a receptor protein related to cell adhesion, migration, and invasion, plays a protective role in vascular remodelling [43] and liver fibrosis [44]. 3-HB is an endogenous ketone molecule that functions as a ligand of hydroxy-carboxylic acid receptor 2 (HCAR2) and free fatty acid receptor 3 (FFAR3) metabolite in cells [45]. 3-HB alleviates the adipose cell’s lipolysis and protects immune cells from inflammation via activating HCAR2 receptor [46, 47], maintains glucose homeostasis and regulates inflammatory metabolism through HCAR2 [48]. A previous study [49] showed that the supplementation of 3-HB reduced glycolysis of CD4 + lymphocytes and improved respiratory chain function during COVID-19 infection, preventing weight loss, hypoxemia, and lung injury.

Multi-organ tissue injuries caused by SARS-COV-2

We characterised the organ tissue expression specificity of COVID-19 patients with tissue-enhanced protein analyses (Supplementary Data Table S7), in which the tissue-enhanced proteins were defined as proteins encoded by genes that have an elevated expression (>= four fold in mRNA level) in the specific type of tissue compared with the average level in all other tissues [16]. We observed an increase in the expression of tissue-enhanced proteins from multiple organs in severe and critical patients, including the brain, heart, liver, kidney, lymphoid, and skeletal muscle (Fig. 3A). This result indicated that a broad influence was caused by SARS-COV-2 virus on multiple organs. As shown in Fig. 3B, lung- and lymphoid-enhanced proteins were upregulated, with involvement in both adaptive and innate immune system processes (Figure S6A-B). Heart-enhanced proteins were primarily associated with ECM proteoglycans and defensive F8 secretion. Liver-enhanced proteins were predominantly linked to the metabolism of vitamins, cofactors, and interleukin-12 signalling. Ovary- or testis-enhanced proteins were linked to the post-translational protein phosphorylation and cell proliferation (Fig. 3B). Correlation analyses revealed intricate protein-protein interactions between multiple organs in severe COVID-19 patients (Figure S6C); however, such communications between multiple organs were disrupted in critical COVID-19 patients. These results further highlighted that injuries in multiple organs were caused in patients with critical COVID-19 disease.

Fig. 3.

Tissue-damage-related molecular alterations caused by COVID-19. (A) Rose plots indicating the proportions of differently expressed tissue-enhanced proteins (in colour) and non-differently expressed tissue-enhanced proteins (in grey) in severe (left) and critical (right) patients. (B) Heatmap of expression of tissue-enhanced proteins and their biological processes. (C) Sankey plots showing significant correlations (p < 0.05) between tissue-enhanced proteins from different organs, immune cells, and blood molecules of different pathways in severe (left) and critical (right) patients

Tissue, immune, and metabolic interactions of COVID-19

We analysed the correlation networks between tissue proteins, immune cells, and metabolic pathways in severe and critical COVID-19 patients (Fig. 3C). In our data, the coordination correlations between immune cells with heart-, brain-, liver-, lung-, and skeletal muscle-enhanced proteins were highly significant.

B cells naïve (BN) and T cells CD4 naïve (CD4) emerged as central regulatory factors of the regulatory network of severe COVID-19 (Fig. 3C). Activation of extrafollicular B cells has been identified as a dominant feature of severe and critical COVID-19 [50]. Highly activated and cytotoxic clusters of differentiation 4 + T lymphocyte responses might contribute to cell-mediated host tissue injury and COVID-19 progression [51]. We observed that B cells naïve were primarily associated with the expression of heart-, lung-, and lymphoid-enhanced proteins in severe COVID-19, which could be regulated by arginine biosynthesis and nicotinamide metabolism; T cells CD4 naïve were mainly related to the expression of heart- and liver-enhanced proteins, which could be regulated by glycolysis/gluconeogenesis, pentose phosphate pathway, and glutathione metabolism.

Macrophage M1 emerged as a central regulatory factor of the immune-metabolic network of critical COVID-19 (Fig. 3C). The systemic reaction resulting from massive macrophage activation plays the central role in SARS-CoV-2-associated complications [52]. In critical COVID-19 patients, macrophage M1 exerted influences on the heart, skeletal muscle, and brain, which could be regulated by arginine biosynthesis (N-argininosuccinate, N-acetyl-L-glutamate, and urea) and the pentose phosphate pathway (LDHA, MINPP1). These findings suggested that the coordinated correlation among organ tissues, immune systems, and metabolic molecules formed the foundation for host tissue injury and COVID-19 progression.

Skeletal muscle mass loss correlates to poor COVID-19 outcomes

To identify phenotypic molecules associated with the prognosis of COVID-19, we compared molecular profiles between the fatality group and the survivor group. A total of 203 proteins and 269 metabolites were altered in the fatality group (Figure S7A-B), in which 99 proteins and 94 metabolites were significantly correlated with the 28-day prognosis by Cox regression analyses (Supplementary Data Table S8). Further tissue-enhanced protein analyses revealed that the tissue-specific expressions of those differential proteins were obviously enriched in skeletal muscle (Fig. 4A). A similar phenomenon was presented in the comparison results of the worsened-improved group (Fig. 4B and S7C-D).

Fig. 4.

Skeletal muscle mass loss correlates to poor COVID-19 outcomes. (A) Rose plot indicating the number of death-related proteins (Cox analyses, p < 0.05; differential analyses, FDR < 0.05 & |log2(fold change)| >0.5) from different organ tissues. (B) Rose plot indicating the number of differentially expressed proteins from different organ tissues between the improved and worsened groups (FDR < 0.05 and |log2(fold change)| >0.5). (C) Box plots indicating the BMI level between survivor-fatality group and improved-worsened group. *, p < 0.05; **, p < 0.01. (D) The proportions of patients with muscle mass loss in the survivor, fatality, improved, and worsened groups. *, p < 0.05. (E) Kaplan–Meier curves predicting the 28-day hospitalised survival of COVID-19 patients stratified by status of muscle mass with a two-sided log-rank p-value. (F) Forest plot indicating Mendelian randomisation analyses results between sarcopenia/fat free body mass and COVID-19 death by inverse variance weighted method. *, p < 0.05

Given previous reports linking acute skeletal muscle loss to poor clinical outcomes in COVID-19 patients [53], we assessed skeletal muscle mass of patients using established tools like BMI [54] and the anthropometric prediction equation [17]. The results demonstrated lower levels of BMI and higher proportions of muscle mass loss in the fatality and worsened group (Fig. 4C-D). Critically, COVID-19 patients with muscle mass loss had significantly lower 28-day survival probability than those without muscle mass loss (p < 0.0001, Fig. 4E). As sarcopenia is characterised by low muscle mass and poor physical function, we performed the MR analyses between sarcopenia/fat-free body mass and COVID-19 death. The results further confirmed a significant causal relationship between the skeletal muscle mass loss and poor COVID-19 outcomes (Fig. 4F).

It is worth noting that we assessed the correlation between the anthropometric equation-derived indicators (SMI and muscle mass loss) and other key available clinical markers (serum creatinine and albumin) to validate the utility of the anthropometric prediction equation. The results (Figure S8) demonstrated a significant positive correlation between SMI and serum creatinine—a well-established surrogate for muscle mass [54, 55]; Conversely, muscle mass loss was significantly associated with lower serum albumin levels, a marker previously linked to future loss of appendicular skeletal muscle mass in older adults [56]. These correlations substantiate the utility of the equation-derived indicators as meaningful parameters in our study population, thereby strengthening the methodological rigor of our study.

The decreased citrulline underlies the poor outcome of COVID-19 patients with muscle mass loss

Significant disorders in arginine biosynthesis were observed in the blood samples of COVID-19 fatality group patients (Fig. 5A). Arginine metabolism is an important metabolic regulatory factor that regulates macrophages’ polarisation and inflammatory response. Previous studies have found that arginine metabolism disorder is an important metabolic characteristic of COVID-19 [9]. In our data, urea and N-acetylglutamate were up-regulated, and GLUL, GOT1, glutamate, citrulline, fumaric acid, and N-acetylornithine were down-regulated in the fatality group (Fig. 5A). To eliminate the potential influence of comorbidities, we conducted a sensitivity analyses on these key molecules, adjusting confounding factors for BMI, age, sex, and comorbidities (hypertension, heart failure, diabetes, kidney disease, liver disease, and psychiatric disease). The results (Supplementary Data Table S9) indicated that all molecules, except glutamate, exhibited significant alteration trends between the fatality group and the survivor group (FDR < 0.05). It is worth noting that those molecules show a significant correlation with clinical indicators such as BMI, SMI, WBC, and LymC (Fig. 5B). These results indicated that patients with COVID-19 exhibit skeletal muscle fibre atrophy and immune cell infiltration [57] via disturbance of arginine biosynthesis pathway. To corroborate these findings, we further analysed a skeletal muscle transcriptome dataset (GSE231910) [58] of golden hamsters infected with SARS-COV-2 for 60 days. The gene set enrichment analyses results showed that the arginine biosynthesis pathway was significantly enriched in the skeletal muscle tissues of golden hamsters infected with SARS-COV-2 (Figure S9A), which cross-validated our results.

Fig. 5.

(A) Metabolic pathway of arginine biosynthesis in the blood samples of the COVID-19 fatality group. (B) Correlation heatmap between clinical indicators and blood molecules of arginine biosynthesis pathway. *, p < 0.05; **, p < 0.01. (C) Kaplan–Meier curves predicting the 28-day hospitalised survival of COVID-19 patients stratified by expression levels of GLUL and citrulline with a two-sided log-rank p-value. (D) Real-time PCR analyses of M1 macrophage markers: Tnfα, Il1b, Il6, and Nos2 in RAW264.7 cells pre-treated with L-NAME (100 µM) or vehicle control for 2 h, followed by citrulline (Cit, 500 µM) supplement or vehicle control for 0.5 h, and followed by lipopolysaccharide (LPS) treatment (100 ng/mL) or vehicle control for 4 h. Bars represent mean ± SD, n = 9 (3 samples with 3 technical replicates). **p < 0.01, ***p < 0.001. (E) Boxplots indicating the expression level of citrulline in the survivor, fatality, improved, and worsened groups at baseline. ***, p < 0.001. (F) Boxplots indicating the expression level of citrulline in improved and worsened group patients at baseline and follow-up period. **, p < 0.01; ns, no significance

Survival analyses revealed that low expression of citrulline and GLUL levels indicated a poor survival rate among COVID-19 patients (Fig. 5C). Since citrulline is positively correlated with body weight, BMI, and SMI, there is a reasonable speculation that citrulline is a prognostic protective factor for COVID-19 patients with skeletal muscle mass loss. Therefore, we further conducted an MR analyses to investigate the causal relationship between citrulline and sarcopenia or body muscle mass. The results confirmed that low citrulline levels were significantly associated with a high probability of sarcopenia and low body muscle mass (Figure S9B). As previously established in Fig. 3C, macrophage M1 serves as a central hub in the immune-metabolic network of multi-organ damage in critical COVID-19. To investigate whether citrulline deprivation contributes to COVID-19 mortality in patients with muscle loss by promoting M1 polarisation, we conducted in vitro experiments using the RAW264.7 cell line. Our data demonstrated that LPS stimulation successfully induced a pro-inflammatory M1 phenotype, characterized by upregulated expression of Tnf, Il1b, Il6, and Nos2 (Fig. 5D) and downregulated expression of M2 markers Mrc1, Chil3, Il10, and Arg1 (Figure S10). Pharmacological inhibition of NOS2 with L-NAME, to mimic citrulline-deficient conditions, further enhanced LPS-driven M1 polarization. Conversely, citrulline supplementation attenuated this inflammatory response. Together, these findings provide mechanistic evidence that citrulline deficiency exacerbates macrophage-mediated inflammation, thereby supporting a causal pathway linking low citrulline to adverse outcomes in COVID-19 patients with muscle loss.

We also tracked the alterations of citrulline over the disease status in COVID-19 patients. The results showed that the citrulline level was first down-regulated in both the worsened and fatality groups at baseline (Fig. 5E). Then it was significantly elevated over time in the improved group but remained unchanged in the worsened group (Fig. 5F). These results indicated that the citrulline level was correlated with the disease progression of COVID-19, and the decreased level of citrulline underlies the poor outcome of COVID-19 patients with skeletal muscle mass loss. However, although GLUL is also a prognostic protective factor for COVID-19, given the positive correlation between GLUL and muscle mass loss, we speculated that the protective effect of GLUL level on the outcomes of COVID-19 might not originate from its effect on skeletal muscle.

Prognostic markers and predictive model of COVID-19 associated with skeletal muscle loss

The ability to predict the clinical outcomes of COVID-19 patients could inform the development of more effective medical strategies and enhance the survival rate. We identified GLUL, citrulline, and GOT1 as the critical prognostic biomarkers of COVID-19 by Cox analyses (Fig. 6A), which were specific to skeletal muscle loss (Fig. 5B). We successfully trained a survival random forest model on the train set (n = 47, Fig. 6B) and evaluated the model’s performance on the test set (n = 20) using a time-dependent receiver operating characteristic (ROC) curve. The area under the curve (AUC) at 5-day and 15-day was 0.94 and 0.79, respectively (Fig. 6C), demonstrating the model’s excellent predictive performance. We calculated the prognostic risk score for each COVID-19 patient, and evaluated the optimal cutoff value of the risk score as 6.06 using the ‘survminer’ algorithm (Fig. 6B). The patients were grouped into a high-risk group and a low-risk group according to the cutoff of 6.06. Survival analyses (Fig. 6D) revealed that the survival rate of the low-risk group was significantly higher than that of the high-risk group (p = 0.035).

Fig. 6.

Biomarkers predicting the prognosis of COVID-19 patients that associated with skeletal muscle mass. (A) Forest plot indicating the hazard ratio of GLUL and citrulline by Cox analyses. (B) Prognostic prediction of the train set (n = 47) and test set (n = 20) patients using the survival random forest model. The dotted line drawn at the cutoff value of 6.06 divided the patients into high- and low-risk groups. (C) The time-dependent receiver operating characteristic (ROC) curve of the survival random forest model on the test set (n = 20). (D) Kaplan–Meier curves showing the 28-day hospitalized survival of COVID-19 patients on test set (n = 20) stratified by prognostic risk scores (cutoff = 6.06) with a two-sided log-rank p-value

Discussion

Metabolic reprogramming represents a significant opportunity for viral replication, contingent upon the energy and biosynthetic precursors supplied by the host cell metabolic network [59]. A more profound understanding of how SARS-CoV-2 reprograms metabolic pathways in organ tissues may prove instrumental in developing more efficacious anti-COVID-19 therapeutic strategies. In this study, we utilised multi-omics data from 307 patients diagnosed with COVID-19 to identify the distinctive organ tissue-specific response and markers in COVID-19 associated with disease severity status, progression, and prognosis. Our multi-omics dataset encompasses several essential attributes crucial for the accurate characterisation of the dynamics of the host response (Fig. 1). These attributes include: (1) cross-sectional sampling with rich knowledge of clinical information, including disease severity, 28-day prognosis, and clinical laboratory indicators; (2) time-paired sampling of severe patients, which improves the understanding of the causal relationship between molecular expression changes and COVID-19 status.

In patients with severe and critical COVID-19, we observed alterations in the levels of immune cells (such as macrophages M0, M1, M2, dendritic cells activated, and mast cells activated) and in the biological process (such as neutrophil degradation, platelet degranulation, innate immune response, pentose phosphate pathway, and arginine biosynthesis) (Fig. 2). It is noteworthy that macrophage M1 was highly activated in the organ-immune-metabolic interaction network of critical patients, and it was correlated with the protein expressions specific to heart/skeletal muscle and the molecules of arginine biosynthesis (Fig. 3C). These results followed recent studies [60, 61] that SARS-CoV-2 has been demonstrated to induce immune disorders and cytokine storms, thereby activating clotting and complement pathways, resulting in multiple organ dysfunction. Elevated glucose levels have been associated with the accelerated progression of COVID-19 and a high mortality rate [62]. Glycolysis activates immune cells, leading to a pro-inflammatory response [63]. Phenylalanine has been identified as a marker of the severity of respiratory distress disease [64]. It has been demonstrated that inflammatory cytokines promote muscle breakdown and release phenylalanine for gluconeogenesis, thereby meeting the metabolic needs during infection [65]. SARS-CoV-2 impairs the metabolism and redox function of cellular glutathione [66]. Furthermore, studies have indicated that specific genetic variants, such as GSTP1 and GSTM3, may influence the susceptibility and severity of COVID-19 [67].

Our cohort exhibited profound alterations in the specific expressions of organ tissue proteins. The consequences of SARS-CoV-2 infection in multiple organ tissues have been well-documented [15]. Significant changes in the expressions of heart-, brain-, liver-, skeletal muscle-, tongue-, and lymphoid tissue-enhanced proteins were displayed in severe and critical COVID-19 patients (Fig. 3A). Consistent with recent studies [15, 68], the inflammatory effects in the lungs were related to the dysfunction of the adaptive immune system. Similarly, testis damage was related to cell proliferation, whereas brain damage was related to L1CAM interactions (Fig. 3B). Moreover, our findings underscore the distinct and critical damage patterns of skeletal muscle in COVID-19, with enrichment of skeletal muscle-specific proteins in the fatality and worsened group (Fig. 4A-B). The proportion of patients with skeletal muscle mass loss in the fatality group and worsened group obviously increased, and COVID-19 patients with muscle mass loss were more prone to poor outcomes (Fig. 4D-F). Skeletal muscle weakness, fatigue, pain, and injury are common symptoms and sequelae of COVID-19 [69]. It is plausible that SARS-CoV-2 interacts with ACE2 in skeletal muscles [70], direct SARS-CoV-2 viral infiltration into skeletal muscle or an aberrant immune system likely contributes to ICU- or COVID-19-acquired weakness [57]. Acute skeletal muscle loss in SARS-CoV-2 infection contributed to poor clinical outcomes in patients with COVID-19 [53].

Strikingly, our findings revealed the protective role of citrulline on skeletal muscle mass during the pathological improvement of severe COVID-19. The level of citrulline was initially down-regulated in both the fatality and worsened groups. Then it gradually increased over time in the improved group, which is different from that of the worsened group. Besides, the citrulline level was positively correlated with body weight, BMI, and SMI. Citrulline serves as a crucial regulator at the intersection of muscle and immune metabolism. Previous research has established its role in enhancing myofibrillar protein expression, modulating muscle energy metabolism [71], and promoting protein synthesis. Citrulline depletion has been shown to be essential for pro-inflammatory macrophage activation and immune responses [32]. Clinical studies further support that citrulline supplementation can improve body composition and physical function in older individuals with a low BMI, potentially preventing and managing sarcopenia and frailty [72, 73]. Our findings from multiple experimental approaches align with these studies, which further prove the improving effect of citrulline on the skeletal muscle of COVID-19 patients: MR confirmed a significant causal relationship between citrulline levels and sarcopenia/fat-free mass (Figure S9B); Transcriptomic analyses of golden hamsters infected with SARS-COV-2 revealed significant alterations in arginine biosynthesis pathways in the skeletal muscle (GSE231910, Figure S9A); In vitro experiments demonstrated that citrulline depletion exacerbates LPS-driven M1 macrophage polarization, while citrulline supplementation attenuates this inflammatory response (Fig. 5D and S10). To our knowledge, this is the first report to demonstrate that citrulline deficiency contributes to both skeletal muscle mass loss and increased mortality in COVID-19 patients, enhancing the understanding of the COVID-19 sequelae related to skeletal muscle mass loss.

Finally, we identified GLUL and citrulline as COVID-19 prognosis predictive markers associated with skeletal muscle loss, and developed a prognosis predictive model employing a machine-learning approach (Fig. 6). The capability of the model was fully evaluated on the test set with a time-dependent ROC curve. The model can effectively stratify patients into high-risk and low-risk categories according to the cutoff of 6.06. In contrast with low-risk patients, high-risk patients demonstrated poorer outcomes, suggesting the forceful predictive capability of the model. Through the model-guided patients’ stratification, COVID-19 patients could potentially benefit from more intensive monitoring, prompt intervention, and trials of therapeutic agents.

This study dissected the organ-specific molecular response and biomarkers of COVID-19, providing evidence for decoding the relationships between citrulline, skeletal muscle mass, and the clinical outcomes of COVID-19, which contributes to the prevention and treatment of patients with post-COVID sequelae. The limitations of this study are as follows: (1) The sample size of the proteomics dataset is small, and sampling time points are sparse. These factors may have influenced the accuracy of our results. In the future, a larger sample volume and additional sampling with more sequential time points should be included in the longitudinal cohort to validate the clinical utility of these biomarkers. (2) While the use of anthropometric equations is practical, it is less accurate compared to gold-standard techniques such as DEXA/MRI measurements. Future studies would benefit from incorporating direct imaging measures and functional assessments such as grip strength. (3) The CIBERSORT results suggested that various types of immune cells were infiltrating in COVID-19 patients. However, details of immune cell proportions were not further confirmed by flow cytometry. Further studies, such as flow cytometry, are warranted to validate the immune phenotypes of COVID-19.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

Conceptualization: Xian Ding, Ting Hu, Yi Zhang, and Zhuoling An; Data curation: Yi Zhang, Ting Hu, Haolong Liu, Rui Zhao; Formal analyses: Yi Zhang and Zhuoling An; Funding acquisition: Zhaohui Tong and Guozhi Jiang; Investigation: Xian Ding, Ting Hu, Yi Zhang, and Zhuoling An; Methodology: Haolong Liu, Xian Ding, Qiuhan Lu, Rui Zhao, Sicheng Huang, Yuwen Wu; Project administration: Zhuoling An and Guozhi Jiang; Resources: Zhuoling An; Supervision: Zhuoling An, Guozhi Jiang, and Zhaohui Tong; Visualization: Xian Ding, Qiuhan Lu, Sicheng Huang, Yuwen Wu; Roles/Writing - original draft: Xian Ding, Ting Hu, Qiuhan Lu, Sicheng Huang; Roles/Writing - review & editing: Xian Ding, Ting Hu, Zhuoling An. All authors reviewed the manuscript.

Funding

This research work was supported by the Ministry of Science and Technology of the People’s Republic of China, National Key Research and Development Program of China (Grant No. 2023YFC0872500 and 2021YFC2301305), the Shenzhen Science and Technology Program (ZDSYS20230626091203007).

Data availability

Data related to this study are available within the text, figures, and tables. All are available from the corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

The experimental protocol was established according to the ethical guidelines of the Helsinki Declaration and was approved by the institutional review board and Ethics Committee (NCT05792865, 2023-Research-230). The requirement for informed written consent was waived because it is a retrospective study.

Consent for publication

No individual information, images or videos were included in this study.

Competing interests

The authors declare that they have no competing interests or commercial affiliations.