From Collapse to Active Self‐Repair: Integrative Multi‐Omics and Machine Learning Analysis Map the Hepatic Metabolic Adaption in Response to Simulated Spaceflight Stress

ABSTRACT

Long‐term spaceflight poses substantial challenges to human physiology, with the liver being highly susceptible due to its central metabolic role. To determine whether hepatic alterations represent transient stress or sustained remodeling, we performed an integrated multi‐omics analysis in a rat model simulating chronic space radiation and microgravity. Herein, we applied an integrated multi‐omics and AI‐driven analytical framework combining histopathology, cytokine and miRNA profiling, proteomics, metabolomics and Western blot validation. After 21 days of simulated space conditions, rats exhibited significant hepatic atrophy, histopathological injury, and metabolic dysfunction resembling a NAFLD‐like phenotype, accompanied by multi‐omics signatures of impaired oxidative phosphorylation, disrupted TCA cycle activity, altered lipid‐metabolic regulation, and inflammatory remodeling. During a 14‐day recovery phase, hepatic atrophy and histological lesions were incompletely improved, with omics changes suggesting partial restoration of mitochondrial related energy metabolism, PPAR associated lipid regulation, and fatty acid β‐oxidation. Machine learning‐based proteomics identified a panel of energy‐related and lipid‐metabolic proteins that robustly distinguished injury from recovery states. External validation with NASA GeneLab transcriptomic datasets supported the suppression of extracellular matrix programs and structural repair during injury. Together, these findings organize the hepatic response to simulated spaceflight into (1) AMPK/PPAR‐γ/PGC‐1α‐centered energy‐related lipid/mitochondrial regulation, (2) ACSM5/CRAT‐associated fatty‐acid utilization and carnitine‐shuttle remodeling, and (3) TGF‐β/IGF1‐related structural repair and anabolic signaling. This study provides a comprehensive organ‐level overview for understanding hepatic adaptation to extreme spaceflight environments and identifies potential targets for mitigating astronaut health risks during long‐duration missions.

This study, using multi‐omics and machine learning methods, revealed the effects of simulated spaceflight on the liver: the process from acute injury to active recovery, but with the presence of some irreversible molecular scarring. This may increase astronauts' long‐term risk of developing non‐alcoholic steatohepatitis (NAFLD) and fibrosis, pathologies commonly found on Earth.

1. Introduction

The capacity of complex organisms to maintain systemic homeostasis under extreme environmental stress represents a fundamental frontier in biological and medical research. Long‐term spaceflight imposes multifaceted challenges to human physiology, characterized by the simultaneous exposure to microgravity, chronic low‐dose cosmic radiation, circadian rhythm disruption, and confinement [1, 2, 3]. These stressors collectively perturb multiple organ systems, including cardiovascular, musculoskeletal, immune, and crucially, metabolic networks [3, 4]. Prior studies, such as the NASA Twins Study, have partially revealed the systemic nature of these alterations, showing significant metabolic, immune, and microbiome disruptions that underscore the profound effect of the space environment on human biology [4, 5, 6].

In mammals, the liver functions as the central metabolic regulator, coordinating a wide range of processes critical for systemic homeostasis—from energy balance, lipid and amino acid metabolism to detoxification, and immune modulation [7, 8]. Given its central role, the liver acts as both a sentinel and a target in stress responses, readily reflecting systemic disturbances caused by environmental challenges. Evidence from both astronauts and animal models indicates that spaceflight induces significant hepatic perturbations, encompassing mitochondrial respiratory impairment, redox imbalance, dysregulated lipid metabolism reminiscent of non‐alcoholic fatty liver disease (NAFLD), and altered immune surveillance [9, 10]. However, a critical limitation of previous investigations is that these studies focused on isolated stress factors or single time points, leaving the subsequent recovery process upon return to Earth largely unexplored.

A fundamental yet unresolved question is whether the hepatic response to spaceflight constitutes a transient, reversible adaptation or a persistent metabolic reprogramming. This knowledge gap directly impedes the rational design of effective countermeasures for protecting astronaut health during future long‐term missions to the Moon and Mars. To address this issue, a longitudinal systems‐level analysis is necessary to delineate the complete adaptive pathway, from initial injury through adaptive remodeling to post‐flight recovery.

In this study, we employed an integrated multi‐omics strategy with AI‐driven analysis in a composite rat model that simulated the combined chronic space radiation and microgravity, followed by a ground‐based recovery period. We combined histopathological assessment, cytokine/microRNA profiling, and high‐resolution 4D‐Smart‐DIA proteomics with untargeted metabolomics, enhanced by co‐expression network analysis and machine learning‐based data analysis. By mapping the core molecular signatures of injury, reversible adaptation and persistent maladaptation, and cross‐validating these findings with authentic NASA GeneLab data, this study aims to establish the boundaries of hepatic resilience and identify the molecular basis of long‐term spaceflight risk.

Using this integrated framework, we summarized the hepatic response to simulated spaceflight across three major aspects [1]. Energy‐related lipid/mitochondrial regulation was supported by AMPK suppression, dynamic PPAR‐γ/PGC‐1α remodeling, and mitochondrial‐related omics signatures [2]. Fatty‐acid utilization and carnitine‐shuttle remodeling was evidenced by ACSM5/CRAT‐associated protein changes and acylcarnitine‐related metabolic alterations [3]. Structural repair and anabolic signaling was supported by TGF‐β/IGF1 dysregulation and extracellular matrix‐associated changes in the NASA GeneLab comparison. These findings provide an organ‐level overview and indicate that hepatic recovery after simulated spaceflight was partial and accompanied by persistent molecular remodeling.

2. Material and Methods

2.1. Construction of a Rat Model to Simulate Combined Radiation and Weightlessness Exposure

Sprague–Dawley (SD) wild‐type male rats (weight: 200 ± 20 g, age: 5–6 weeks) were supplied by SiPeiFu Biotechnology Co. Ltd. (RRID: RGD_70508). Upon arrival, rats were maintained in sterile housing conditions at 25°C, with a 12 h light/dark cycle and rats had free access to food and water. All methods were carried out in accordance with relevant guidelines and regulations and approved by the Animal Care and Utilization Committee of Beijing Institute of Technology (Beijing, China) (Approval No. [SYXK‐BIT‐20220114001]). The sample size was determined based on previous similar studies [11, 12] to ensure sufficient statistical power for detecting proteomic and metabolomic changes.

After 3 days of adaptive feeding, a total of 12 healthy male rats were randomly divided into three groups using a random number table, the control group (CON = 4), the group exposed to simulated one‐time proton radiation and 21‐day weightlessness to mimic the long‐term space flight environment (SCE = 4), the recovery group of the simulated model (REC = 4). All rats included in the respective experiments finished the study, that is, there was no attrition. No animals or data points included in the corresponding analyses were excluded from the analysis.

One time whole‐body proton irradiation was used to simulate space radiation condition [13]. Rat box irradiation (one SD rat in a transparent acrylic box) was used to ensure that the experimental rats were kept in a stable position during the whole irradiation period. The energy of proton beam was selected as 100 MeV, the energy transfer density was selected as 0.8 keV/μm, and the dose rate of 0.5 Gy/min was used to irradiate the whole body of each rat at the Center for Innovative Applications of Anti‐radiation Technology of the National Institute of Atomic Energy Agency. The cumulative dose of irradiation was 1 Gy.

To simulate the long term microgravity condition in space, the 30° tail lifting method was used after one time proton irradiation for 21 days [14]. The tails of all the rats were hung up, the front legs were on the ground, and the hind legs were suspended to make sure the rats were at an angle of about 30° to the ground. The one‐time proton irradiated rats were stably acclimated for 30 min before being transported to the animal room for tail suspension to prevent stress reactions.

For the recovery group, after one‐time proton whole body irradiation and 21 days tail suspension, rats were returned to a normal feeding environment for 14 days.

2.2. Histopathological Analysis

The rats were euthanized by CO₂ inhalation followed by decapitation to ensure minimal pain. The rat livers were excised and completely immersed in 4% paraformaldehyde (Beijing Solarbio Science & Technology Co. Ltd., Beijing, China) for histopathological analysis. The liver tissue was fixed for 24 h, dehydrated, and embedded in paraffin. Standard histological processing was performed. Tissues were dehydrated through a graded ethanol series (SCRC, 100092683, Shanghai, China) using a dehydrator (DIAPATH, Donatello, Montichiari, Italy), cleared in xylene (SCRC, 10023418, Shanghai, China) or Environmentally Friendly Dewaxing Transparent Liquid (Servicebio, G1128‐1L, Wuhan, China), and embedded in paraffin using an embedding machine (Wuhan Junjie, JB‐P5, Wuhan, China). Paraffin blocks were sectioned at 4 μm thickness with a rotary microtome (Shanghai Leica, RM2016, Shanghai, China), mounted on adhesive slides (Servicebio, G6012‐1, Wuhan, China), and air‐dried. For H&E staining, sections were deparaffinized and rehydrated, then stained with hematoxylin (Servicebio, G1076, Wuhan, China) for 5 min, differentiated, blued, and counterstained with eosin for 5 s. Dehydration was performed through graded ethanol and normal butanol (SCRC, 100052190, Shanghai, China), followed by xylene clearance and mounting with neutral gum (SCRC, 10004160, Shanghai, China) and cover glasses (Citotest, 10212432C, Jiangsu, China). Processed sections were examined under an upright optical microscope (Nikon, ECLIPSE E100, Tokyo, Japan) equipped with an imaging system (Nikon, DS‐U3, Tokyo, Japan) to assess cell morphology, inflammatory infiltration, and other pathological features. Histopathological scoring and assessment were performed by two independent observers blinded to the experimental groups.

2.3. Measurement of Inflammatory Factors IL‐1β and TNF‐α in the Liver Tissue

The contents of TNF‐α and IL‐1β in rat liver homogenate were measured by ELISA. Liver tissue of each rat was homogenized. The levels of TNF‐α and IL‐1β in the rat liver tissue were determined by the commercial enzyme‐linked immunosorbent assay (ELISA) kits following the instructions of the ELISA kits. Catalog number: YX‐091201R from Shanghai Guduo Biotechnology Co. Ltd. for IL‐1β, Catalog number: YX‐201406R from Shanghai Guduo Biotechnology Co. Ltd. for TNF‐α.

2.4. RT‐qPCR Analysis

Total RNA was extracted from liver tissue using the reagent kit (catalog number R0024, Beyotime Biotechnology, Beijing, China), following the kit's instructions. The resulting RNA samples were subsequently reverse transcribed into complementary DNA using the reagent kit (model M‐MLV, Beijing Solarbio Science & Technology Co. Ltd.). The RT‐qPCR was performed by real‐time PCR (IQ5, Bio‐Rad) utilizing β‐actin as a housekeeping gene. The expression of each gene was normalized to that of β‐actin. The relative expression levels (fold change) of each target gene were normalized to that of β‐actin mRNA levels in each sample using the 2^−∆∆CT method. Each experiment was conducted with three biological replicates to guarantee independent measurements. The forward primer sequence (F) and the reverse primer sequence (R) were shown in Table S1.

2.5. 4D‐Smart‐DIA Quantitative Proteomics Analysis

2.5.1. Tissue Sample Extraction

Liver tissue samples were individually pulverized in liquid nitrogen. The resulting powder was lysed using SDT buffer (containing 100 mM NaCl, 8 M urea and 1% protease inhibitors) supplemented with 1/100 volume DTT, followed by ultrasonication (model S‐250D from Branson Ultrasonics Co., Danbury, CT, USA) on ice for 5 min. Lysates were incubated at 95°C for 8–15 min, immediately cooled on ice for 2 min, and centrifuged at 12000× g for 10 min at 4°C. The supernatant was collected and alkylated with iodoacetamide (IAM) for 1 h at 25°C in the dark. Proteins were precipitated by thoroughly vortexing the alkylated lysate with 4 volumes of pre‐cooled acetone, followed by incubation at −20°C for a minimum of 2 h. After centrifugation (12,000× g, 15 min, 4°C), the pellet was collected, washed once with 1 mL cold acetone, and subsequently dissolved completely in Dissolution Buffer (DB buffer).

Protein concentration was determined using a Bradford assay kit (model P0006 from Beyotime Biotechnology, Beijing, China). A standard curve was generated using bovine serum albumin (BSA) solutions ranging from 0 to 0.5 g/L. Both BSA standards and diluted sample solutions (20 μL final volume per well) were loaded in triplicate onto a 96‐well plate. Following the rapid addition of 180 μL G250 dye reagent to each well, the plate was incubated at 25°C for 5 min. Absorbance was measured at 595 nm, and sample protein concentrations were calculated based on the standard curve. For quality assessment, 20 μg of each protein sample was resolved by SDS‐PAGE using a 12% polyacrylamide gel. Electrophoresis was performed at 120 V through the stacking gel (20 min) and 150 V through the resolving gel (50 min). Proteins were visualized by staining with Coomassie Brilliant Blue R‐250 reagent and destained until bands were clearly defined.

Protein samples were adjusted to a final volume of 100 μL with DB lysis buffer (8 M urea, 100 mM triethylammonium bicarbonate [TEAB], pH 8.5). Samples were digested with trypsin at 37°C for 4 h, followed by a second addition of trypsin and overnight digestion. The digestion was terminated by acidification with formic acid to pH less than 3. Acidified samples were centrifuged at 12000× g for 5 min at 25°C. The supernatant was loaded onto a C18 solid‐phase extraction column. Peptides were washed three times with washing buffer (0.1% formic acid, 3% acetonitrile). Peptides were eluted with elution buffer (0.1% formic acid, 70% acetonitrile). Eluted peptides were collected and lyophilized.

2.5.2. LC–MS Condition

Lyophilized peptides were reconstituted in 10 μL of mobile phase A (0.1% formic acid in water). After centrifugation at 14000× g for 20 min at 4°C, 200 ng of peptide (based on pre‐digestion quantification) was injected. Chromatographic separation was performed on a Vanquish Neo UHPLC system (Thermo Scientific) equipped with a C18 pre‐column (Acclaim PepMap 100, 5 mm × 300 μm, 5 μm) maintained at 50°C and an analytical column (PepMap Neo UHPLC C18, 150 μm × 15 cm, 2 μm). Mobile phases consisted of A (0.1% formic acid in water) and B (0.1% formic acid in 80% acetonitrile).

The elution gradient begins at time zero (0 min) with a flow rate of 2.5 μL/min and a mobile phase composition of 96% A and 4% B. At 0.2 min, the flow rate decreases to 1.3 μL/min while maintaining the same mobile phase ratio (96% A/4% B). At 0.3 min, the flow rate further decreases to 0.8 μL/min as the composition shifts to 92% A and 8% B, remaining stable at these conditions until 0.5 min. Between 0.5 and 14.2 min, the flow rate holds at 0.8 μL/min while the mobile phase gradually changes to 77.5% A/22.5% B at the 14.2 min mark. By 21.1 min, still at 0.8 μL/min, the composition reaches 65% A/35% B. Subsequently, at 21.5 min, the flow rate increases to 2.5 μL/min concurrent with a shift to 45% A/55% B and initiation of a Column Wash step. At 21.9 min, the mobile phase rapidly transitions to 1% A/99% B while maintaining the 2.5 μL/min flow rate, and this condition persists until the run stops at 22.6 min.

Eluting peptides were analyzed using a Thermo Orbitrap Astral mass spectrometer (Waltham, MA, USA) equipped with an Easy‐Spray (ESI) ion source. The key mass spectrometry parameters were as follows: spray voltage—2.0 kV; ion transfer tube temperature—290°C; full MS1 scan range—m/z 380–980; MS1 resolution—240 000 (at m/z 200); MS1 automatic gain control (AGC) target—500%; Data‐Independent Acquisition (DIA) mode with 300 variable windows (2 Th width); normalized collision energy (NCE)—25%; MS2 scan range—m/z 150–2000; MS2 resolution (Astral analyzer)—80 000; maximum MS2 injection time—3 ms. (.raw) files were generated.

2.5.3. Proteomic Data Analysis

2.5.3.1. Quality Control

The mass spectrometry data files were processed using Proteome Discoverer software (version 2.1, RRID: SCR_014477) for database searching. Searches were performed against the UniProtKB database restricted to the species Rattus norvegicus (RRID: SCR_004426). A false discovery rate (FDR) threshold set at less than 1% is applied for peptide‐spectrum matches (PSMs), peptides, and proteins. Proteins with all values missing in any group (CON/SCE/REC) were removed. Proteins exhibiting at least 3 missing values in at least 2 groups were excluded.

2.5.3.2. Differential Expression Analysis for Proteomic

Differential expression analysis of proteomic data was performed using UniProt IDs as unique identifiers derived from database searching. Homogeneity of variance between groups was evaluated using Levene's test with an alpha level of 0.01. Proteins exhibiting a Levene's p‐value exceeding 0.01 were considered to have equal variances. To compare the differences between two different groups, an independent samples t‐test was conducted. A two‐tailed test was employed to determine whether there was a significant difference between the group means (in either direction). A dual‐threshold criterion defined statistically significant differential expression: (1) a protein expression fold change threshold of 1.2, and (2) a p‐value threshold of 0.05. Proteins were classified as differentially expressed only when meeting both criteria simultaneously.

2.5.3.3. Pathway Enrichment Analysis

Functional enrichment analysis was conducted using the STRING platform (version 12.0, RRID: SCR_005223). This analysis utilized the STRING API and Python (version 3.10.16, RRID: SCR_008394). The species was specified as Rattus norvegicus , identified by NCBI taxon ID 10116. The STRING enrichment method was employed with default parameter settings. For proteins intended for enrichment analysis, UniProt IDs were converted to gene names using the online UniProt ID mapping tool (https://www.uniprot.org/id‐mapping, accessed on 20 December 2024, RRID: SCR_002380). Then, the STRING ID mapping tool was employed to convert gene names or protein descriptions into standardized STRING query items. For a given protein list, the STRING API returns enrichment results including Gene Ontology (biological processes, molecular functions, cellular components) and KEGG pathways (RRID: SCR_012773). Pathway enrichment significance was determined using a false discovery rate threshold of 0.05. Data visualization was performed using Python (version 3.10.18) with the Matplotlib (RRID: SCR_008624) and Seaborn libraries.

2.6. Weighted Gene Correlation Network Analysis (WGCNA)

Weighted gene correlation network analysis (WGCNA) was performed using the OmicVerse package (version 1.7.1) in Python (version 3.10.16). WGCNA was used to identify highly correlated protein modules related to the experimental conditions (CON/SCE/REC). Following data preprocessing, the top 3000 most variable proteins based on median absolute deviation were selected for network construction. A signed adjacency matrix was generated using a soft‐thresholding power determined via scale‐free topology analysis (power range from 1 to 20), followed by calculation of a topological overlap matrix to minimize noise. Dynamic hybrid tree‐cutting with deepSplit of 2 and pamRespectsDendro of False identified co‐expression modules, whose eigengenes were computed for downstream analysis. Module‐trait associations and p‐value were calculated by default parameters. Module with p‐value less than 0.05 was identified as the key module.

2.7. Four Machine Learning Models for Identifying Key Proteins

To identify the key proteins that distinguish between the injury and recovery groups under the simulated composite space environment, four machine learning models were implemented using scikit‐learn (version 1.5.2, RRID: SCR_002577) in Python (version 3.10.16, RRID:SCR_024202). Therefore, the computational task of the machine learning model was formulated as a binary classification task. All models used fixed random states (the random seed for all models was set to 42) and balanced class weights.

LASSO (Least absolute shrinkage and selection operator) was applied for sparse feature selection using L1 penalty. The model was configured with inverse regularization strength C = 30 000 (minimal regularization), “liblinear” solver, and maximum iterations = 10 000. Proteins with non‐zero coefficients were considered selected.

Random Forest was used for non‐linear feature importance assessment. 1000 trees with “sqrt” mode “max_features”, Gini impurity splitting criterion, and no depth restriction are included. Feature importance was calculated as mean Gini impurity reduction across all trees.

Linear Support Vector Machine (SVM) with L2 penalty was implemented for maximum‐margin separation. The regularization parameter C was set as 0.1. The regularization parameter controlled margin strictness, with a linear kernel for interpretable coefficients. Feature weights were extracted from the primal hyperplane coefficients.

Elastic Net logistic regression combined L1 and L2 penalties (L1_ratio was set as 0.01) using “saga” solver. The regularization parameter C was set as 0.1. Max_iter was set as 10 000. This balanced feature selection (sparsity) and coefficient stability (correlation handling).

The absolute values of signed coefficients in LASSO, 0–1 scaled Gini importance scores in Random Forest, signed hyperplane coefficients in SVM, and signed coefficients in Elastic Net are defined as coefficient or importance. Proteins were ranked by the absolute coefficient or importance values per model‐task combination. Top 30 proteins are selected for each model.

2.8. Protein–Protein‐Interaction Network Analysis

The protein–protein‐interaction network was constructed using the STRING database (version 12.0). The analysis was restricted to the target species Rattus norvegicus (taxonomic identifier: 10116). The full STRING network was retrieved, encompassing both functional and physical protein associations. A minimum interaction confidence score threshold of 0.300 was applied.

2.9. Metabolomic Analysis

Non‐targeted metabolomic analysis was performed at Beijing Novogene Co. Ltd. Approximately 100 mg of each liver tissue sample that had been thoroughly ground in liquid nitrogen was used. To each sample, 500 μL of pre‐cooled 80% (v/v) methanol in water as the extraction solvent was added. The mixture was vortexed to mix thoroughly and then placed in an ice bath for 5 min. The sample was then centrifuged at 15000 g for 20 min at 4°C to precipitate cell debris and insoluble matter. The supernatant was carefully aspirated, transferred to a new centrifuge tube, and mass spectrometry‐grade pure water was added to dilute the methanol concentration to a final concentration of 53% (v/v) to match the subsequent liquid chromatography conditions. To remove possible precipitation, the diluted sample was centrifuged again at 15000 g for 20 min at 4°C. Finally, the clear supernatant was collected and transferred to an autosampler sample vial for subsequent LC–MS analysis.

The LC–MS analysis was performed using a Vanquish UHPLC system coupled to a Q Exactive HF‐X mass spectrometer (Thermo Fisher Scientific, Germany). Chromatographic separation was achieved on a Hypersil Gold C18 column (100 × 2.1 mm, 1.9 μm; Thermo Fisher Scientific, USA). The column oven was maintained at 40°C. The mobile phase consisted of solvent A (0.1% formic acid in water) and solvent B (methanol), delivered at a flow rate of 0.2 mL/min. The gradient elution program was as follows: 0–1.5 min, 2% B; 1.5–3 min, a linear gradient from 2% to 85% B; 3–10 min, a linear gradient from 85% to 100% B. The composition was then returned to 2% B at 10.1 min and held for 1.9 min for column re‐equilibration, with a total run time of 12 min.

The raw file (.raw) obtained by mass spectrometry detection was imported into Compound Discoverer 3.3 (CD3.3, RRID: SCR_014477) software for spectral processing and database search. The high‐quality mzCloud database constructed by the standard was combined with the mzVault and MassList databases to match and identify the molecular characteristic peaks (RRID: SCR_014669). This process enabled the qualitative and quantitative analysis of the metabolites, followed by rigorous data quality control. Multivariate statistical analysis of the metabolites was performed, including principal component analysis (PCA). The biological significance of metabolites was explained through functional analysis such as metabolic pathways. Differential metabolites were screened according to the standards of FC > 1.2 or FC < 0.83 and p‐value < 0.05. Data visualization was performed using Python (version 3.10.18) with the Matplotlib and Seaborn libraries.

2.10. Integrative Analysis With NASA Authentic Spaceflight Data

The transcriptomics dataset GLDS‐137 from the NASA GeneLab database was used (https://osdr.nasa.gov/bio/repo/data/studies/OSD‐137). We performed differential expression analysis between the spaceflight group (FLT) and the ground‐based baseline control group (BSL) in this dataset using a bioinformatics pipeline identical to our animal model data analysis. Briefly, raw data were quality‐controlled, aligned using STAR (RRID: SCR_004463), and differentially expressed genes were identified using DESeq2 (RRID: SCR_000154). The selection criteria (p < 0.05 and FC > 1.2) were consistent with our in‐house proteomic analysis. Finally, we compared the list of significantly differential expressed genes obtained from the GLDS‐137 dataset with our own list of identified differential expressed proteins to identify overlapped core response molecules for subsequent analysis.

2.11. Statistic Methods

Quantitative data from body and liver weight measurements, cytokine ELISAs, RT‐qPCR analyses, and other targeted experimental assays are presented as mean ± SEM. Statistical analyses for these targeted assays were performed using GraphPad Prism software (version 9.3.0, RRID: SCR_002798). For comparisons among the CON, SCE, and REC groups, data normality and homogeneity of variance were assessed using the Shapiro–Wilk test and Levene's test, respectively. When the assumptions for parametric analysis were met, one‐way analysis of variance (ANOVA) followed by Tukey's post hoc test was used to determine inter‐group differences. Statistical significance was defined as p < 0.05. For proteomic and metabolomic datasets, pairwise differential feature screening was performed as an exploratory discovery‐level analysis using predefined fold‐change and nominal p‐value thresholds. Proteins or metabolites with a fold change > 1.2 or < 0.83 and a nominal p‐value < 0.05 were considered differentially altered for downstream visualization, integration, and biological interpretation. These feature‐level p values were used for exploratory screening rather than confirmatory multiple‐testing‐adjusted inference. For functional enrichment and pathway analyses, FDR‐adjusted enrichment statistics were used where implemented by the corresponding software, database, or analysis platform. WGCNA, machine‐learning feature selection, and NASA GeneLab comparison analyses were used as complementary discovery and integration approaches. Details of each omics workflow are provided in the relevant Methods sections.

2.12. Western Blot

Western blotting was performed to provide targeted orthogonal validation for key proteins identified from the RT‐qPCR and proteomics analyses. Total protein was extracted from rat liver tissues, separated by SDS‐PAGE, and transferred onto PVDF membranes. After blocking with 5% non‐fat milk, the membranes were incubated with primary antibodies against AMPK (Selleck Chemicals, F0269), PPAR‐γ (MedChemExpress, HY‐P80872), PGC‐1α (MedChemExpress, HY‐P80783), TGF‐β1 (Selleck Chemicals, F1624), ACSM5 (Proteintech Group, 67 334–1‐Ig), CRAT (Yeasen Biotechnology, P106973), IGF1 (Proteintech Group, 28530‐1‐AP, RRID:AB_2881164). β‐actin (Proteintech Group, 66009‐1‐Ig) was used as the internal loading control. Membranes were then incubated with HRP‐conjugated Affinipure Goat Anti‐Rabbit IgG (Proteintech Group, SA00001‐2) or HRP‐conjugated Affinipure Goat Anti‐Mouse IgG (Proteintech Group, SA00001‐1), as appropriate. Bands were visualized using enhanced chemiluminescence.

3. Results

3.1. Simulated Spaceflight Conditions Induce Hepatic Atrophy and Histological Damage With Partial Reversibility Upon Recovery

Studies have indicated that spaceflight and simulated space environments induce substantial physiological stress, resulting in systemic and organ‐specific damage, with the liver being particularly affected. The combined effects of microgravity and radiation have been shown to disrupt hepatic metabolism, intensify oxidative stress, and impair hepatocyte integrity. These pathological changes underscore the liver's high susceptibility to space stressors and underscore its pivotal role in systemic adaptive responses. However, the actual severity of hepatic atrophy and the mechanisms controlling its recovery—key determinants of long‐term astronaut health—are still largely unknown.

To assess the physiological impact of long‐term spaceflight, rats underwent one time proton irradiation followed by 21 days of simulated microgravity (SCE), and a 14‐day recovery period (REC). The SCE exposure induced a systemic stress response, resulting in significant reductions in both total body weight and liver weight. The SCE group exhibited a 26.7% lower body weight than controls (CON) (287.7 g vs. 392.7 g; p < 0.01). After 14 days of recovery, body weight partially recovered to 303.0 g but remained 22.8% below control levels (Figure 1A).

FIGURE 1.

Simulated spaceflight induces hepatic atrophy and histological injury, followed by partial recovery. (A–C) Quantification of (A) body weight, (B) liver weight, and (C) liver‐to‐body weight ratio in control (CON), 21‐day simulated space condition exposure (SCE), and 14‐day recovery (REC) groups. Data are presented as mean ± SEM (n = 3 per group). Statistical significance was determined by one‐way ANOVA followed by Tukey's post hoc test. *p < 0.05 and **p < 0.01 vs. CON; #p < 0.05 and ##p < 0.01 vs. SCE. (D–I) Representative H&E‐stained liver sections from CON, SCE, and REC groups (n = 3 rats per group), (D–F) show low‐magnification images, and (G–I) show the corresponding high‐magnification images of the boxed regions. The CON group shows normal hepatic architecture, whereas the SCE group shows disrupted hepatic cords, hepatocellular swelling (red arrow), and inflammatory cell infiltration (black arrow). The REC group shows partial histological recovery with improved lobular organization.

Liver weight similarly decreased by 35.9% in SCE versus CON (9.0 g vs. 14.1 g; p < 0.01), but recovered to 92.5% of the CON baseline after REC (44.5% increase vs. SCE; p < 0.05) (Figure 1B). The liver‐to‐body weight ratio declined from 3.52% (CON) to 3.14% (SCE), then rebounded to 4.31% in REC—a 37.3% increase over SCE and 22.4% above CON (p < 0.01) (Figure 1C). These findings indicated that recovery not only reversed hepatic atrophy but also induced a transient phase of hyperplastic or hypertrophic liver growth.

Histopathological analysis using H&E staining revealed substantial liver injury under SCE conditions. Control (CON) livers showed normal lobular architecture with well‐organized hepatic cords (Figure 1D,G), whereas SCE livers displayed severe hepatocellular swelling, cytoplasmic rarefaction (red arrow), and disorganized cords, accompanied by focal inflammatory infiltrates in portal areas (black arrow) (Figure 1E,H). These changes were largely reversed after recovery (REC), with restored hepatic architecture, repopulated hepatocytes, and markedly reduced edema and inflammation (Figure 1F,I).

These findings revealed that simulated spaceflight triggered severe but partially reversible hepatic atrophy and pathology. The pronounced compensatory recovery observed underscores the liver's remarkable regenerative plasticity and provides a foundation for elucidating the molecular mechanisms of stress‐induced hepatic remodeling.

3.2. Simulated Spaceflight Stress Triggers Dynamic Hepatic Metabolic Dysregulation

To elucidate the hepatic response to simulated spaceflight, we first examined key regulators related to energy‐related lipid/mitochondrial regulation and repair‐associated signaling across the 21‐day exposure (SCE) and 14‐day recovery (REC) phases. The relative mRNA expression levels of TGF‐β, AMPK, PPAR‐γ, and PGC‐1α were quantified across CON, SCE, and REC groups.

During the 21‐day SCE period, relative AMPK mRNA expression was significantly downregulated by 24.1% (1.01 in SCE vs. 1.33 in CON; p < 0.01; Figure 2A). In contrast, PPAR‐γ and PGC‐1α expression increased markedly by 376% (1.00 in SCE vs. 0.21 in CON; p < 0.01; Figure 2B) and 59.1% (1.05 in SCE vs. 0.66 in CON; p < 0.01; Figure 2C), respectively. Concurrently, TGF‐β mRNA levels decreased by 46.9% (1.02 in SCE vs. 1.92 in CON; p < 0.05; Figure 2D).

FIGURE 2.

Simulated spaceflight triggers dynamic hepatic metabolic, inflammatory, and microRNA regulatory changes. (A–D) Relative mRNA levels of key metabolic and repair‐associated regulators, including (A) AMPK, (B) PPAR‐γ, (C) PGC‐1α, and (D) TGF‐β, in CON, SCE, and REC groups, as determined by RT‐qPCR. (E–H) Representative Western blot validation of the corresponding proteins, including (E) AMPK, (F) PPAR‐γ, (G) PGC‐1α, and (H) TGF‐β, with β‐Actin used as the loading control. (I–J) Hepatic levels of inflammatory cytokines, including (I) TNF‐α and (J) IL‐1β, as measured by ELISA. (K–M) Relative hepatic levels of key microRNAs, including (K) miR‐34a, (L) miR‐20b, and (M) miR‐130b, as determined by RT‐qPCR. Data in A–D and K–M are presented as mean ± SEM (n = 4 per group), and data in I–J are presented as mean ± SEM (n = 3 per group). Western blot images are representative of 3 biological replicates per group. Statistical significance was determined by one‐way ANOVA followed by Tukey's post hoc test. *p < 0.05 and **p < 0.01 vs. CON; #p < 0.05 and ##p < 0.01 vs. SCE.

In the recovery phase, AMPK expression further decreased to 0.89, reflecting an 11.9% reduction relative to SCE (p < 0.05 vs. SCE; Figure 2A). PPAR‐γ expression declined to 0.79, a 21.0% decrease from SCE yet still remained higher than the CON baseline (p < 0.01; Figure 2B). In contrast, PGC‐1α expression increased to 1.58, representing a 50.5% increase over SCE and a 139.4% rise compared with CON (p < 0.01; Figure 2C). TGF‐β expression partially rebounded to 1.15 during recovery, but remained below the CON baseline (Figure 2D).

To further determine whether these transcriptional changes were reflected at the protein level, Western blot analysis was performed for the four proteins. AMPK protein abundance was reduced after simulated spaceflight exposure and remained lower during recovery, consistent with sustained impairment of energy‐sensing signaling (Figure 2E). PPAR‐γ protein increased in the SCE group and declined during recovery, whereas PGC‐1α remained elevated after recovery, indicating continued lipid/mitochondrial regulatory remodeling (Figure 2F,G). TGF‐β protein abundance was decreased after SCE and did not fully return to the CON level during recovery, supporting incomplete restoration of repair‐associated signaling (Figure 2H).

We next assessed inflammatory cytokines in liver tissue. During SCE, hepatic TNF‐α decreased by 8.5% (302.84 to 277.07 pg/mL; p < 0.05; Figure 2I), while IL‐1β showed a non‐significant decline from 15.68 to 13.99 pg/mL (Figure 2J). After recovery, TNF‐α partially returned toward the control level, whereas IL‐1β rebounded to 17.59 pg/mL, exceeding both SCE and CON levels (Figure 2J). Three key hepatic lipid metabolism‐associated miRNAs are also quantified. All three miRNAs showed marked upregulation during SCE. The levels of miR‐34a, miR‐20b, and miR‐130b increased substantially in the SCE group compared with CON (Figure 2K–M). Notably, these inductions were rapidly and substantially reversed during recovery, with miR‐34a, miR‐20b, and miR‐130b all decreasing toward or below control levels (Figure 2K–M).

3.3. Analysis of Hepatic Dysfunction Mechanism Based on Proteomics Analysis

To characterize the hepatic proteomic response to simulated spaceflight, we analyzed protein expression patterns across control (CON), 21‐day simulated spaceflight condition exposed (SCE), and 14‐day recovered (REC) groups. Principal component analysis (PCA) delineated three distinct, non‐overlapping clusters, indicating clear separation among the experimental groups (Figure 3A). Using a fold‐change threshold of 1.2 and a p‐value threshold of 0.05, differential expression analysis between SCE–CON and REC–SCE comparisons revealed extensive proteomic alterations, as visualized by volcano plots (Figure 3C,D). A total of 392 up‐regulated and 206 down‐regulated differentially expressed proteins (DEPs) were identified in the SCE group compared to the CON group, highlighting a massive initial stress response (Figure 3C). Analysis comparing REC–SCE identified 344 up‐regulated and 564 down‐regulated DEPs (Figure 3D).

FIGURE 3.

Proteomic analysis of the liver tissue. (A) Principal component analysis (PCA) of the hepatic proteome across control (CON, n = 4), simulated spaceflight‐condition exposed (SCE, n = 4), and recovery (REC, n = 3) groups. Ellipses represent 95% confidence intervals. (B) Venn diagram illustrating the overlap of up‐regulated and down‐regulated proteins between the SCE vs. CON and REC vs. SCE comparisons. (C, D) Volcano plots illustration DEP for (C) SCE vs. CON and (D) REC vs. SCE. Significantly up‐regulated (red) and down‐regulated (blue) proteins are highlighted based on a threshold of p < 0.05 and FC = 1.2. Numbers of proteins in each category are indicated in the legends. All panels were generated from CON (n = 4), SCE (n = 4), and REC (n = 3) samples.

To further elucidate the dynamics of proteomic remodeling, we categorized proteins according to their expression trajectories (Figure 3B). A prominent group of 122 proteins displayed a “reversed” expression pattern—either downregulated during exposure and upregulated during recovery (25 proteins), or conversely, upregulated during exposure and downregulated during recovery (97 proteins)—suggesting their involvement in adaptive recovery processes. In contrast, smaller subsets of proteins remained either persistently upregulated (8 proteins) or downregulated (9 proteins), indicating a sustained molecular imprint of spaceflight exposure.

3.3.1. Uncovering Mechanisms of Simulated Spaceflight‐Induced Hepatic Injury by Proteomics

To elucidate the systemic impact of simulated spaceflight, we performed functional enrichment analysis on the differentially expressed proteins (DEPs) from the SCE versus CON comparison (Figure 4A,B). KEGG pathway enrichment revealed that the simulated composite environment of space radiation and microgravity induced liver damage dominantly through disruption of oxidative phosphorylation, thermogenesis, retrograde endocannabinoid signaling, and metabolism of branched‐chain amino acids and sphingolipids, alongside activation of pathways linked to non‐alcoholic fatty liver disease and neurodegenerative disorders (Figure 4C,G).

FIGURE 4.

Proteomic analysis of the hepatic tissue under simulated space exposure (SCE). (A, B) Heatmaps of (A) upregulated and (B) downregulated differentially expressed proteins (DEPs) between the SCE and control (CON) groups. (C) Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis of DEPs. (D–F) Gene ontology (GO) enrichment analysis of DEPs, showing the top enriched terms for (D) biological process, (E) cellular component, and (F) molecular function. Bars in (C–F) are ranked by −log₁₀ (p‐value), and the number of proteins for each term is indicated. (G) The UpSet plot showing the intersections of proteins among the enriched KEGG pathways. The bar chart on the left indicates the total number of DEPs per pathway, while the top bar chart shows the size of each intersection. Sample size annotation: The analyses were based on DEPs from the SCE vs. CON comparison (CON n = 4, SCE n = 4).

Gene Ontology (GO) enrichment analysis was performed to characterize the functions of DEPs comparing SCE versus CON. Key biological processes (Figure 4D) included metabolic process, mitochondrion organization, organic substance metabolic process, mitochondrial respiratory chain complex assembly, cellular respiration, electron transport chain, oxoacid metabolic process, carboxylic acid metabolic process, and oxidative phosphorylation.

Cellular component enrichment (Figure 4E) implicated that DEPs were enriched in mitochondrion, cytoplasm, intracellular membrane‐bounded organelle, and protein‐containing complex, revealing mitochondrion as the focal point of the molecular dysfunction. Mitochondrion related cellular component included mitochondrial protein‐containing complex, mitochondrial inner membrane, mitochondrial envelope, mitochondrial membrane, mitochondrial respirasome, mitochondrial respiratory chain complex I, and mitochondrial ribosome.

Molecular function enrichment (Figure 4F) was mainly enriched for catalytic activity, oxidoreductase activity, ion binding, electron transfer activity, NADH dehydrogenase activity, transferase activity, and oxidoreduction‐driven transmembrane transporter activity. Together, these results depicted a multilayered cellular crisis initiated by severe mitochondrial damage, which propagated to disrupt core metabolism and activate pathways resembling neurodegenerative and NAFLD‐associated stress responses. The intersections of proteins among the enriched KEGG pathways are shown in Figure 4G.

3.3.2. Cross‐Platform Analysis of Proteomics and NASA GeneLab Transcriptomics

To benchmark our findings against authentic spaceflight conditions, we compared our proteomic data with a publicly available NASA GeneLab transcriptomic dataset from space‐flown mice (GLDS‐137). Cross‐species analysis identified 41 consistently dysregulated proteins/genes, confirming that our ground‐based model faithfully recapitulates key aspects of authentic spaceflight‐induced hepatic pathology (Figure 5A).

FIGURE 5.

Integrative analysis with NASA GeneLab data defines a conserved signature of spaceflight‐induced hepatic injury. (A) Heatmap displaying 41 significantly dysregulated DEPs consistently altered in both the simulated spaceflight model (SCE vs. CON) and authentic NASA GeneLab flight data (Flight vs. Ground). (B) Protein–protein interaction (PPI) network of the 41 signature proteins. Nodes are colored by functional classification. Edge thickness represents interaction confidence (confidence score > 0.25). Sample IDs are directly shown for Figure 5A.

Functional and network analysis (Figure 5B) revealed that the largest module consisted of a densely interconnected network of down‐regulated proteins, overwhelmingly enriched in extracellular matrix (ECM) components. Profound suppression of key structural proteins, including Col1a2, Col3a1, and Fbn1, along with the anabolic regulator Igf1, reflected a systemic anabolic suppression and provided a mechanistic basis for the observed hepatic atrophy. This conserved ECM downregulation might suggest a fundamental adaptive trade‐off, whereby energetically costly structural maintenance was compromised to prioritize immediate survival needs.

In contrast, the second module comprised a network of up‐regulated proteins that functionally validated our findings of a severe intracellular stress response. This module included proteins implicated in oxidative stress management (e.g., Ppox, Vkorc1l1) and mitochondrial calcium regulation (Mcu), offering cross‐platform evidence for the mitochondrial dysfunction identified as the epicenter of spaceflight‐induced injury. Furthermore, the dysregulated proteins involved proteostasis and lipid metabolism, such as Cers6 (sphingolipid metabolism) and the down‐regulation of the protective adipokine Adipoq.

3.3.3. Analysis of Hepatic Self‐Restorative Recovery Mechanism Based on Proteomics

To investigate the molecular drivers of hepatic recovery, we analyzed 122 proteins exhibiting reversed expression dynamics between the exposure (SCE) and recovery (REC) phases—comprising proteins either down‐regulated during SCE and up‐regulated during REC, or vice versa. Functional enrichment analysis of this subset revealed a coordinated reprogramming aimed at restoring hepatic homeostasis after combined radiation and microgravity stress. Heatmap visualization clearly illustrates the expression reversal for proteins down‐regulated during exposure and up‐regulated during recovery (Figure 6A), as well as those showing the opposite trend (Figure 6B), providing direct evidence of dynamic molecular reprogramming at the protein level.

FIGURE 6.

Functional analysis of proteins with reversed expression profiles during injury and recovery. (A, B) Circular heatmaps identifying two distinct profiles of reversibly regulated proteins. (A) injury‐upregulated/recovery‐downregulated and (B) injury‐downregulated/recovery‐upregulated proteins. (C) KEGG pathway enrichment analysis. (D–F) Gene Ontology (GO) enrichment analysis for (D) Biological Process, (E) Cellular Component, and (F) Molecular Function. For panels C–F, bars are ranked by significance, and protein counts are shown. Sample size annotation: Proteins were selected from the reversed‐expression subset identified across SCE–CON and REC–SCE comparisons using CON (n = 4), SCE (n = 4), and REC (n = 3) samples.

KEGG pathway analysis further demonstrated that these dynamically regulated proteins are significantly involved in oxidative phosphorylation, thermogenesis, PPAR signaling, retrograde endocannabinoid signaling, ribosomal pathways, and processes associated with non‐alcoholic fatty liver disease and neurodegenerative disorders (Figure 6C).

Critically, this proteomic reprogramming was functionally aligned with the transcriptional changes observed earlier. The substantial up‐regulation of PPAR‐γ mRNA (Figure 2B) during exposure and its sustained elevation throughout recovery closely mirrored the dynamic reversal of its downstream targets in the proteome. For example, the suppression of lipogenic enzymes and the rebound of fatty acid β‐oxidation proteins within the “reversed” protein set are consistent with a PPAR‐mediated metabolic shift. Together, these multi‐omics findings establish the PPAR signaling axis as a central regulator of the hepatic recovery process.

GO enrichment analysis of the DEPs highlighted mitochondrial functional restoration as a central theme of the recovery phase. Key enriched biological processes were primarily related to energy metabolism and mitochondrial biogenesis, including mitochondrial respiratory chain complex I assembly, mitochondrion organization, cellular metabolic process, electron transport chain, metabolic process, respiratory electron transport chain, generation of precursor metabolites and energy, energy derivation by oxidation of organic compounds, mitochondrial ATP synthesis coupled with electron transport (Figure 6D).

GO analysis also revealed mitochondrion as the central organelle in the molecular recovery process. During the recovery phase, DEPs were significantly enriched in mitochondrial‐related cellular components, including mitochondrial respiratory chain complex I, mitochondrial protein‐containing complex, mitochondrial inner membrane, mitochondrial envelope, mitochondrial matrix, and mitochondrial ribosome (Figure 6E).

Molecular functions were dominated by NADH dehydrogenase (ubiquinone) activity, oxidoreductase activity, electron transfer activity, structural constituent of ribosome, catalytic activity, and primary active transmembrane transporter activity (Figure 6F). Collectively, these findings indicate that recovery from spaceflight stress is driven by a coordinated reactivation of mitochondrial energy production and ribosomal biogenesis.

3.3.4. Persistent Proteomic Alterations Reveal Incomplete Hepatic Repair and Long‐Term Pathogenic Risk

To identify persistent molecular alterations induced by combined radiation and microgravity, we analyzed proteins that remained dysregulated throughout both the exposure (SCE) and recovery (REC) phases. Heatmaps illustrated the expression patterns of eight persistently up‐regulated proteins including Eif4ebp2, Zbed5, Gpt, Slc25a22, Hsdl2, Crat, Acsm5, Ech1 (Figure 7A; Table 1) and nine consistently down‐regulated proteins (Rig1, Derl2, Amy1, Igf1, Lgalsl, Uba7, Cpamd8, Mug5, Mug1; Figure 7B; Table 2).

FIGURE 7.

Heatmaps displaying the expression profiles of proteins that were (A) persistently up‐regulated or (B) persistently down‐regulated across CON, SCE, and REC groups. Each row corresponds to an individual protein, and each column corresponds to a biological replicate. The color scale indicates the z‐score normalized expression, with red denoting higher abundance and blue denoting lower abundance relative to the mean expression across all samples. Hierarchical clustering was performed on the rows (proteins). Sample IDs are directly shown.

TABLE 1.

Sustained up‐regulated proteins across both exposure (SCE vs. CON) and recovery (REC vs. SCE) phases.

UniProt	Gene	FC(SCE:CON)	p(SCE:CON)	FC(REC:SCE)	p(REC:SCE)
P25409	Gpt	1.38	0.014	1.37	0.038
A0A0G2K5L2	Slc25a22	1.33	0.001	1.23	< 0.001
Q6AYT9	Acsm5	2.91	0.030	2.68	0.016
A0A8I6A0I3	Eif4ebp2	1.37	0.045	1.85	0.041
A0A8I5ZUC6	Hsdl2	1.49	0.006	1.39	0.034
A0A8L2QF26	Ech1	1.32	0.033	1.84	0.011
A0A8L2RAJ5	Crat	1.33	0.033	1.92	0.012
D3ZTR5	Zbed5	1.26	0.038	1.44	0.042

TABLE 2.

Sustained down‐regulated proteins across both exposure (SCE vs. CON) and recovery (REC vs. SCE) phases.

UniProt	Gene	FC(SCE:CON)	p(SCE:CON)	FC(REC:SCE)	p(REC:SCE)
A0A0G2K735	Uba7	0.77	0.033	0.72	0.002
A0A8I6GIB4	Derl2	0.79	0.001	0.78	0.022
A0A0G2JX40	Igf1	0.65	0.032	0.53	0.039
A0A8I5ZU41	Amy1	0.79	0.015	0.72	0.014
A0A0G2JUP5	Mug1	0.63	0.045	0.73	0.025
A0A8I6ABY4	Lgalsl	0.77	0.018	0.63	0.006
D4ADT5	Rig1	0.80	< 0.001	0.81	< 0.001
A0A8I6AMC9	Cpamd8	0.69	0.011	0.66	0.049
A0A8I5Y1T7	Mug5	0.59	0.015	0.59	0.019

The persistently up‐regulated proteins reflected a state of chronic metabolic stress accompanied by compensatory adaptation. These up‐regulated molecules are primarily involved in enhancing mitochondrial fatty acid β‐oxidation (e.g., Crat, Acsm5, Ech1) [15, 16], modulating amino acid metabolism (Slc25a22), and promoting stress‐adaptive responses. This expression profile reflected a long‐term, adaptive shift toward fatty acid catabolism, likely to compensate for the persistent mitochondrial dysfunction identified in our earlier analyses [17, 18]. The continued elevation of the hepatic stress marker Gpt further supported the notion of ongoing metabolic challenge [19, 20]. Notably, Eif4ebp2, a key inhibitor of Cap‐dependent translation downstream of mTOR, remained elevated. This conversely, the persistently down‐regulated proteins might indicate a sustained suppression of cellular defense. This group of proteins was associated with impaired growth signaling (Igf1) [21, 22], protein quality control (Derl2 in endoplasmic reticulum‐associated degradation, Erad, Uba7 in ISGylation) [23, 24], immune response (Rig1, Lgalsl) [25, 26], and protease inhibition (Mug1).

To further validate these persistent proteomic alterations, Western blot analysis was performed for three representative proteins. ACSM5 and CRAT remained elevated in both SCE and REC groups relative to CON, supporting sustained fatty‐acid utilization and carnitine‐shuttle‐associated remodeling. In contrast, IGF1 remained reduced after simulated spaceflight exposure and did not fully recover during the recovery phase, indicating incomplete restoration of anabolic and repair‐related signaling (Figure S4A–C).

3.3.5. WGCNA Network Analysis of Hepatic Proteome Identifies co‐Expression Modules Associated With Spaceflight‐Induced Injury and Recovery

To elucidate the systems‐level organization of the hepatic proteome beyond individual protein changes, we performed weighted gene co‐expression network analysis (WGCNA) independently on the stress (SCE) and recovery (REC) phases. This strategy enabled the identification of distinct modules of co‐expressed proteins that define the molecular state during simulated spaceflight exposure and subsequent recovery.

The initial phase of a WGCNA is hierarchical clustering and soft threshold search. For the CON versus SCE comparison, a scale‐free network (soft power β = 5) was constructed, grouping the proteome into 21 distinct co‐expression modules. Several of these modules showed strong association with the stress condition (Figure 8A–C). Notably, the Gold (409 proteins), Silver (297 proteins), Yellowgreen (192 proteins), Forestgreen (90 proteins), and Lavender (68 proteins) modules were highly positively correlated with the SCE group (|r| > 0.7, p < 0.05). To functionally annotate these stress‐induced networks, we cross‐referenced them with the 208 proteins enriched from the KEGG pathways during the exposure phase (Figure 8D). This revealed a significant overlap, confirming that these WGCNA modules represented the core biological machinery driving the pathological response to spaceflight.

FIGURE 8.

WGCNA identifies protein co‐expression modules associated with injury and recovery phases. (A, E) Determination of the optimal soft‐thresholding power (β) for network construction in the (A) injury and (E) recovery phases, based on scale‐free topology fit (left) and mean connectivity (right). (B, F) Sample clustering dendrograms for the (B) injury and (F) recovery datasets. (C, G) Module‐trait correlation heatmaps identifying modules significantly associated with the (C) injury and (G) recovery states. Color intensity corresponds to the correlation strength (red: Positive, blue: Negative). (D, H) UpSet plots illustrating the overlap between proteins in key modules and those in KEGG pathways for the (D) injury and (H) recovery analyses. The injury network was constructed from CON and SCE samples, and the recovery network from SCE and REC samples. For the samples in WGCNA, sample IDs are directly shown in (B) and (F).

Next, we investigated how this network architecture was rewired during recovery phase by performing a second WGCNA comparing the SCE and REC groups. This analysis also yielded a robust network structure (soft power β = 6; Figure 8E,F) with 20 distinct modules (Figure 8G). A large module, Silver group (749 proteins), was strongly positively correlated with the SCE state. The Sienna (165 proteins), Lightgreen (256 proteins), and Forestgreen (102 proteins) modules were also highly correlated with the REC state (|r| > 0.8, p < 0.05), collectively forming a robust recovery‐activated network. Integration of this recovery network with the dynamically “reversed” proteins from KEGG analysis provided compelling functional insights (Figure 8H).

3.3.6. Identification of Key Proteins Using Four Machine Learning Models

To identify the most important molecular determinants of the hepatic injury phase, we applied a convergent machine learning strategy to a set of 132 candidate proteins derived from KEGG‐enriched and WGCNA stress‐associated modules. Four machine learning models, LASSO, Random Forest, SVM, and Elastic Net (Figure 9A), were employed as feature importance rankers to further identify the key proteins related to simulated space environment stress as well as hepatic recovery. Given the sample size, these models were utilized for consistent feature selection across the entire dataset rather than for training predictive classifiers.

FIGURE 9.

Multi‐model ensemble analysis identifies core protein signatures for hepatic injury. (A) Feature importance ranking of proteins across four independent models: (a) LASSO logistic regression, (b) random forest, (c) support vector machine (linear kernel), and (d) elastic net regression. (B) Protein–protein interaction (PPI) network constructed from the intersecting features of at least two models, revealing a mitochondria‐ and lipid‐centric core module. (C) Comparative feature importance analysis across machine learning models. Green dots indicate top‐ranked proteins, with darker shading highlighting recurrent selection. (D–H) Representative expression patterns and classification performance of the top features (Asah1, Cers6, Derl1, Ndufaf6, and Pah). Left panels: Boxplots showing protein expression across control (CON), spaceflight condition (SCE), and recovery (REC) groups. Right panels: ROC curves demonstrating the discriminatory power of each protein in classifying CON vs. SCE and SCE vs. REC states. Sample size annotation: N = 4 in CON group, n = 4 in SCE group, n = 3 in REC group.

This integrated analysis identified a core hepatic injury signature comprising 19 proteins, each selected by at least two algorithms in the CON vs. SCE comparison. A high‐confidence subset of five proteins—Asah1, Cers6, Derl1, Ndufaf6, and Pah—was consistently prioritized by three or more models (Figure 9B). Protein–protein interaction (PPI) network analysis of this signature revealed a highly interconnected functional landscape, primarily centered on lipid signaling, proteostasis, and mitochondrial pathways (Figure 9C).

Expression analysis confirmed significant upregulation of Asah1, Derl1, and Pah during the injury phase. Furthermore, individual protein‐based receiver operating characteristic (ROC) curve analysis demonstrated the strong discriminatory power of all five core signature proteins, with near‐perfect area under the curve (AUC) values ranging from 0.94 to 1.00 for distinguishing control (CON) from spaceflight‐exposed (SCE) samples (Figure 9D–H).

Having defined the molecular signature of the initial injury, we next sought to identify the key proteins of the subsequent recovery process. We applied the same multi‐model machine learning pipeline to a set of 39 proteins that were reversely regulated in CON versus SCE and SCE versus REC comparison, training the algorithms to distinguish the recovery state (Figure 10A).

FIGURE 10.

A machine learning‐derived signature reveals a mitochondria‐centric hub driving hepatic recovery. (A) Top 20 predictive proteins identified by four independent machine learning models (a. LASSO, b. Random Forest, c. SVM, and d. Elastic Net). (B) PPI network of the 18 high‐confidence proteins selected by at least three models, revealing a tightly connected mitochondrial Complex I module. (C) Dot matrix for the 39 candidate proteins across samples in SCE and REC. Green dots denote top‐20 feature proteins. (D–I) Expression levels and highly discriminatory performance of the six core proteins (Adipoq, Ndufs4, Pdxk, Ndufs7, Ndufa12, Ndufb3) selected by all four models for distinguishing CON, SCE, and REC states. Sample size annotation: N = 4 in CON group, n = 4 in SCE group, n = 3 in REC group.

This integrated approach identified the recovery signature of 18 key proteins (Table S2) that were selected as top features by at least three of the four models (Figure 10C). PPI network analysis revealed a tightly interconnected core module (Figure 10B). Several top‐ranked proteins were related to mitochondrial respiratory Complex I and energy metabolism. This data‐driven finding supported the involvement of Complex I‐related energy metabolism during recovery.

Further analysis identified a smaller subset of six recurrent features selected across all four models. Individual ROC of the signature proteins confirmed their statistical discriminatory performance, with each protein exhibiting a perfect or near‐perfect AUC of 1.00 for classifying both the stress (CON vs. SCE) and recovery (SCE vs. REC) transitions (Figure 10D–I).

3.4. Analysis of Hepatic Dysfunction Mechanism Based on Metabolism Analysis

To delineate the hepatic metabolic trajectory under simulated spaceflight stress and subsequent recovery, we conducted untargeted metabolomic analysis. Simulated spaceflight exposure (SCE) triggered a dramatic metabolic shift, with 56 up‐regulated and 121 down‐regulated metabolites identified relative to controls (CON) (Figure 11A). The subsequent 14‐day recovery period (REC) prompted further extensive remodeling, with 81 up‐regulated and 79 down‐regulated metabolites relative to the SCE group (Figure 11B). A Venn diagram illustrated the dynamic regulation of metabolites across conditions, identifying distinct expression patterns—including a subset of 23 metabolites that displayed a complete reversal in expression trends between phases (Figure 11C). Heatmaps provided a global visualization of the metabolites significantly altered during the exposure phase (Figure 11D,E) and those dynamically regulated throughout the entire process (Figure 11F).

FIGURE 11.

Analysis of hepatic dysfunction mechanism based on metabolism analysis. (A, B) Volcano plots showing differential expressed metabolites in the (A) SCE vs. CON and (B) REC vs. SCE comparisons. Significantly up‐regulated (orange) and down‐regulated (blue) metabolites are highlighted. (C) Venn diagram illustrating the overlap between up‐ and down‐regulated metabolites across the exposure and recovery phases. (D, E) Heatmaps showing z‐score normalized expression of metabolites that were significantly (D) down‐regulated or (E) up‐regulated in the SCE vs. CON comparison. (F) Heatmap showing the z‐score normalized expression of the 23 metabolites that exhibited a reversed expression pattern between the exposure and recovery phases. (G) KEGG pathway analysis of metabolites significantly altered during the exposure phase (SCE vs. CON). (H) KEGG pathway analysis of the 23 “reversed” metabolites, highlighting pathways central to the recovery process. (I) Violin plots showing the expression levels of representative metabolites that were persistently dysregulated across CON, SCE, and REC groups. Data are presented as mean ± SEM. Statistical significance was determined by one‐way ANOVA with Tukey's post hoc test. *p < 0.05, **p < 0.01 vs. CON; #p < 0.05 vs. SCE. Metabolomic analyses were performed using CON (n = 4), SCE (n = 4), and REC (n = 3) samples.

KEGG pathway analysis of differentially regulated metabolites during the injury phase revealed a multi‐system disruption consistent with our proteomic findings (Figure 11G). A prominent cluster of enriched pathways involved neuroendocrine signaling, including “GABAergic synapse”—driven by accumulated γ‐aminobutyric acid (GABA)—and “retrograde endocannabinoid signaling”. This emergent neuroendocrine‐like signature aligned with neurodegenerative pathways identified in proteomic data, together indicating systemic perturbation of liver–brain cross‐talk. Functionally, this systemic disruption was characterized by a severe redox control failure, as shown by enrichment of “glutathione metabolism” (involving L‐glutamic acid) and, notably, the “ferroptosis” pathway. The depletion of glutathione precursors such as L‐cysteine offered a direct metabolic connection to mitochondrial dysfunction observed at the proteomic level.

We next sought to identify drivers of metabolic recovery by analyzing 23 metabolites whose expression was dynamically reversed between the exposure and recovery phases. KEGG pathway analysis of these metabolites revealed a highly coordinated repair program that strongly aligned with the PPAR‐mediated recovery signature identified in our proteomic data (Figure 11H). The most significantly enriched pathways were predominantly associated with membrane integrity, including “glycerophospholipid metabolism” and “choline metabolism”, driven by the rebound of key structural lipids such as phosphatidylcholine (PC 40:7). These findings indicated that active restoration of cellular and organellar membranes—critically compromised during the initial mitochondrial injury—constituted a central objective of the recovery process. This structural repair program was accompanied by recalibration of systemic stress signaling, as reflected in the enrichment of “cortisol synthesis and secretion” pathway, marked by normalized levels of cortisol precursors such as cortodoxone.

Finally, we identified a persistent “molecular scar” reflecting incomplete metabolic recovery, which functionally corroborated the sustained proteomic alterations (Figure 11I; Table 3). Most notably, carnitine continued to accumulate significantly throughout the recovery phase. This offered compelling metabolic evidence for a lasting bottleneck in mitochondrial fatty acid β‐oxidation, consistent with the sustained up‐regulation of proteins such as Crat and Ech1. In contrast, key signaling lipids including prostaglandin B1 remained suppressed, indicating prolonged disruption of inflammatory signaling pathways. Together, these lasting metabolic alterations functionally validated that the liver established a new, altered homeostatic state—one marked by the molecular scars previously identified at the proteomic level.

TABLE 3.

Fold change and p‐values for persistently dysregulated metabolites.

Name	FC(SCE:CON)	p(SCE:CON)	FC(REC:SCE)	p(REC:SCE)
Prostaglandin B1	0.631	< 0.001	0.714	0.001
Pleuromutilin	0.526	0.009	0.686	0.008
11‐Ketoetiocholanolone	0.826	0.003	0.821	0.009
DL‐Carnitine	1.701	0.031	3.508	0.001
L‐	1.402	0.013	2.422	< 0.001

3.5. Integrative Omics Analysis With NASA GeneLab Data

To further validate the mechanistic underpinnings of hepatic adaptation, we performed an integrated analysis combining our proteomic and metabolomic datasets with the NASA GeneLab transcriptomic dataset (GLDS‐137). This multi‐layer concordance analysis focused on three convergent axes: (1) energy‐related lipid/mitochondrial regulation, (2) fatty‐acid utilization and carnitine‐shuttle remodeling, and (3) structural repair/anabolic signaling.

3.5.1. Energy‐Related Lipid/Mitochondrial Regulation

The integrated analysis first supported a major disturbance in energy metabolism and mitochondrial regulation during the injury phase. Cross‐layer concordance was observed in Oxidative Phosphorylation and TCA Cycle pathways. In the NASA GeneLab dataset, Ndufs and Ndufa family transcripts were broadly downregulated, consistent with the alteration of mitochondrial complex I‐related proteins in our proteomic data. This energy‐related change was further reflected in the metabolome, where TCA cycle intermediates, including L‐Malate and alpha‐Ketoglutarate, were reduced. Together with AMPK suppression and dynamic PPAR‐γ/PGC‐1α changes observed in our experimental model, these findings support disrupted energy sensing and lipid/mitochondrial regulatory remodeling under simulated spaceflight stress.

3.5.2. Fatty‐Acid Utilization and Carnitine‐Shuttle Remodeling

The second component involved altered fatty‐acid utilization. NASA transcriptomic data showed suppression of the Hadha/Hadhb and Cpt1a axis, which was consistent with changes in fatty‐acid β‐oxidation‐related proteins in our proteomic dataset. At the metabolite level, the accumulation of L‐Palmitoylcarnitine and DL‐Carnitin, together with increased Palmitic acid, suggested an imbalance in fatty‐acid transport and mitochondrial lipid handling. This was further supported by the persistent alteration of ACSM5 and CRAT, two proteins linked to fatty‐acid activation and carnitine‐shuttle‐associated metabolism. Redox‐related changes, including dysregulation of Glutathione S‐transferases (GSTs) and altered GSH/GSSG balance, were interpreted as accompanying stress responses rather than separate main mechanisms.

3.5.3. Structural Repair and Anabolic Signaling

The third component involved structural repair and anabolic signaling. NASA GeneLab comparison supported suppression of extracellular matrix‐associated programs during spaceflight, consistent with changes involving structural maintenance and repair capacity. In our model, this structural/anabolic component was also reflected by altered TGF‐β and IGF1, together with partial histological recovery after the recovery period. During recovery, PPAR‐associated signals and the machine‐learning‐prioritized protein Adipoq were linked to improved energy/lipid regulation, while replenishment of membrane‐related metabolites such as Phosphatidylcholine and Phosphoethanolamine suggested partial restoration of cellular and organellar membrane integrity. However, the persistence of selected protein and metabolite alterations indicated that structural and anabolic recovery remained incomplete.

4. Discussion

In the long‐term spaceflight missions, alterations in hepatic hemodynamics due to fluid redistribution, coupled with the metabolic demands of coping with systemic inflammation and oxidative stress generated by radiation and microgravity, can significantly impact astronauts' liver function [27, 28]. By combining proton radiation with head‐down tilt unloading for 21 days, following a 14‐day recovery phase, we have established a rat model to simulate long‐term spaceflight condition. The choice of a 1 Gy proton dose and acute exposure (0.5 Gy/min) was determined based on NASA's projections for deep‐space missions, where the total effective dose for a 30‐months Mars mission is estimated at about 1.1 Sv [29]. Since protons make up about 90% of space radiation, this 1 Gy “stress test” models both the total mission impact and the risks of high‐flux Solar Particle Events. This model allows for the identification of the “vulnerability hubs” and peak damage states in hepatic metabolism, which might otherwise be masked by the slow adaptive kinetics of chronic exposure. By establishing this high‐dose baseline, we can map the liver's regenerative potential and the transition from injury to active self‐repair during the recovery phase. This study presented a comprehensive longitudinal analysis of hepatic response to simulated spaceflight, integrating multi‐omic analysis with functional validation.

In the current studies, the simulated space environment induced significant hepatic atrophy and histopathological damage, consistent with previous reports of spaceflight‐associated organ atrophy [30]. The reduction in absolute liver weight and disorganization of hepatic cords underscored the severity of environmental stress. Notably, the recovery phase was marked not only by mass restoration but also by increased liver‐to‐body weight ratio, suggesting a hyperplastic or hypertrophic regenerative response. This rebound highlighted the liver's remarkable plasticity and aligns with clinical observations of post‐stress regenerative growth [31, 32, 33].

At the molecular level, the regulatory patterns we observed recapitulated established mechanisms in metabolic and hepatic diseases. Suppression of AMPK had been widely observed in NAFLD and obesity, where impaired energy sensing exacerbated lipid accumulation and insulin resistance [34]. Similarly, dysregulated TGF‐β signaling, a hallmark of hepatic fibrosis and cirrhosis, promoted extracellular matrix deposition and progressive liver dysfunction [35]. In addition, upregulation of PPAR‐γ and PGC‐1α aligned with protective adaptations seen in metabolic syndrome and mitochondrial disorders [36, 37]. PGC‐1α in particular was known to promote oxidative phosphorylation and counteract ROS accumulation [38].

Similarly, the miRNAs identified here, including miR‐34a, miR‐20b, and miR‐130b, have been implicated in metabolic disorders. miR‐34a was consistently elevated in NAFLD, alcoholic liver disease, and hepatocellular carcinoma, where it promoted apoptosis and lipid dysregulation [39, 40]. miR‐20b had been associated with hypoxia signaling and metabolic reprogramming in cardiovascular and cancer models [41]. miR‐130b was frequently dysregulated in obesity and type 2 diabetes, where it directly targeted PGC‐1α and other genes regulating lipid oxidation, highlighting its role in mitochondrial dysfunction [42, 43]. The rapid normalization of these miRNAs during the recovery period suggested that they might serve as sensitive molecular regulators governing hepatocellular adaptation, consistent with prior evidence in metabolic disease models.

Taken together, the phenotypic, omics, and targeted Western blot data support a coherent model in which recovery is evident at the tissue level but remains incomplete at the molecular level. First, the energy‐related lipid/mitochondrial regulatory aspect is reflected by hepatic atrophy and histological injury, AMPK suppression with dynamic PPAR‐γ/PGC‐1α changes, and mitochondrial complex I/TCA cycle‐related omics alterations. Second, the fatty‐acid utilization aspect is supported by lipid‐related miRNA changes, persistent ACSM5/CRAT‐associated protein alterations, and acylcarnitine‐related metabolic changes, indicating that lipid handling remains actively remodeled during recovery. Finally, the structural repair and anabolic aspect is supported by partial histological recovery together with TGF‐β/IGF1 dysregulation and NASA GeneLab‐supported extracellular matrix suppression, suggesting that morphological improvement may precede complete normalization of repair‐related signaling. These observations provide a bridge from targeted validation to metabolomic signatures.

A key finding from our integrated proteomics and metabolomic analysis was the significant enrichment of pathways typically associated with central nervous system (CNS) function—such as GABAergic synapse and retrograde endocannabinoid signaling—within the liver. This strongly suggested that spaceflight stress disrupted the bidirectional liver‐brain communication [44, 45]. The accumulation of long‐chain acylcarnitines, coupled with the depletion of TCA cycle intermediates (e.g., succinate, malate, and α‐ketoglutarate) [46], reflected severe impairment of mitochondrial β‐oxidation, while perturbations in arachidonic acid and glutathione metabolism indicated sustained inflammatory and oxidative stress. During recovery, phospholipid species such as phosphatidylcholine (PC 40:7) and phosphatidylethanolamine (PE 18:1/20:4) were replenished, suggesting restored membrane integrity and reactivated lipid catabolic flux [27, 47]. Notably, liver‐derived metabolites—including GABA and glutamate [48], may in turn influence central neural function, potentially contributing to the neurocognitive alterations frequently reported during spaceflight.

Overall, our findings suggest that simulated spaceflight does not induce a simple injury‐and‐return‐to‐baseline process, but rather a transition from hepatic metabolic injury toward partial recovery. Mitochondrial energy stress, fatty‐acid utilization remodeling, and insufficient structural/anabolic repair appear to represent the major components of this response, supported across histology, qPCR, proteomics, metabolomics, NASA GeneLab comparison, and targeted Western blot validation. This integrated interpretation helps explain why phenotypic recovery after simulated spaceflight is accompanied by unresolved molecular alterations and provides a more focused basis for future development during deep space exploration.

5. Conclusions and Perspectives

This study established the hepatic response to simulated spaceflight as a biphasic process—initial metabolic injury followed by active self‐repair. The injury phase was characterized by impaired mitochondrial energy metabolism, disrupted TCA cycle activity, lipid‐metabolic dysregulation, and oxidative/inflammatory stress, consistent with a NAFLD‐like phenotype. During recovery, hepatic mass and histological lesions were markedly improved, accompanied by partial restoration of mitochondrial complex I‐related energy metabolism, PPAR‐associated lipid regulation, and fatty acid β‐oxidation. Machine learning‐based proteomics prioritized mitochondrial complex I‐ and lipid‐metabolic proteins that distinguished injury from recovery states. External validation from different platforms and different datasets using NASA GeneLab data confirmed extracellular matrix suppression in actual spaceflight. Together with targeted Western blot validation, these findings summarize the hepatic response across (1) energy‐metabolic regulation, (2) fatty‐acid utilization, and (3) structural repair/anabolic signaling, indicating that recovery was active but incomplete. These results establish a molecular framework for developing targeted methods to safeguard health during extended missions.

6. Limitations

Despite the breadth of our investigation, several limitations warrant consideration. First, although our ground‐based rat model incorporated critical factors of the space environment—including proton irradiation and simulated microgravity—it may provide an insufficient consideration of the highly complex, chronic, and mixed‐field nature of the actual space radiation environment. While our one‐time acute proton irradiation (1 Gy at 0.5 Gy/min) served as a necessary “stress test” to identify synchronized molecular responses and metabolic vulnerability hubs, it may not fully capture the nuanced dose‐rate effects or the cumulative impact of low‐dose galactic cosmic radiation exposure. Second, although the moderate sample size allowed us to detect major molecular alterations, it may have limited the ability to capture phenotypic changes, weaker omics signals, and inter‐individual variability. Additionally, targeted Western blot validation was performed only for representative proteins along the three main response axes, and therefore does not constitute complete pathway‐level functional validation. Nevertheless, our findings collectively provided a multidimensional picture of hepatic adaptation to spaceflight, establishing a foundational mechanistic framework for developing countermeasures to preserve astronaut hepatic health during deep space exploration.

Author Contributions

Conceptualization: Bo Li; Methodology: Bo Li, Han Zhang and Boyang Li; Formal analysis: Han Zhang, Boyang Li and Feng Yao; Investigation: Feng Yao and Yufan Zou; Data curation: Han Zhang and Boyang Li; Writing – original draft preparation: Bo Li and Han Zhang; Writing – review and editing: Han Zhang, Boyang Li, Weijun Su and Bo Li; Visualization: Han Zhang and Boyang Li; Supervision: Bo Li; Project administration: Bo Li; Funding acquisition: Bo Li and Weijun Su. All authors have read and agreed to the published version of the manuscript.

Funding

This research was funded by the National Natural Science Foundation of China (grant number 82202072 and grant number 32471197).

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Figure S1: (A) Topological overlap matrix (TOM) plot showing the 21 injury‐related co‐expression modules. (B) TOM plot showing 20 co‐expression modules during recovery.

Figure S2: (A–D) Enrichment of differentially expressed genes from mice livers of space (NASA GeneLab GLDS‐137). Top enriched terms are shown. (A) GO cellular component, (B) GO biological process, (C) GO molecular function, and (D) KEGG enriched pathways.

Figure S3: The UpSet plot illustrates the inter‐cross‐relationships among proteins in KEGG pathways enriched by proteins that were first upregulated and then downregulated, as well as those that were first downregulated and then upregulated.

Figure S4: Western blot validation of selected mRNA/proteomics‐derived proteins. (A) ACSM5, (B) CRAT, and (C) IGF1. β‐actin served as the loading control. The blots were used as targeted validation of protein‐level changes across CON, SCE, and REC groups.

Table S1: Primer sequences for the microRNA qPCR reaction.

Table S2: Fold change and p‐value for 18 key proteins in SCE–CON and REC–SCE comparison.

Data Availability Statement

The NASA GeneLab dataset used in this study is publicly available (accession: GLDS‐137, https://osdr.nasa.gov/bio/repo/data/studies/OSD‐137). All other data generated or analyzed in this study are available from the corresponding author upon reasonable request.