Muscle atrophy is a significant concern for astronauts during prolonged spaceflight. While exercise countermeasures are employed, they are time-consuming and not fully effective in offsetting the effects of microgravity. One potential mechanism underlying muscle atrophy is the dysregulation of cytoplasmic calcium (Ca2+) levels due to impaired function of the Sarco Endoplasmic Reticulum Calcium ATPase (SERCA) pump, responsible for Ca2+ reuptake during muscle relaxation. Mammalian muscles are classified into slow-twitch (e.g., soleus, SOL) and fast-twitch (e.g., tibialis anterior, TA) types, both affected by spaceflight. Previous research indicates that Ca2+ uptake is impaired in SOL but enhanced in TA muscles during spaceflight, highlighting the differential effects on SERCA function. However, the underlying molecular mechanisms remain unclear. Machine learning (ML) offers a powerful approach for identifying patterns in complex biological data and discovering biomarkers, particularly in high-dimensional multi-omics datasets. Unlike traditional statistical methods, ML methods are less susceptible to distribution-specific effects, making them well-suited for space biology research where datasets are often small. This study uses an ML-based approach to identify molecular drivers of spaceflight effects on muscle physiology using multi-omics data (transcriptomic, proteomic, and epigenomic) and Ca2+ reuptake data from mice flown on the International Space Station. The QLattice symbolic regression algorithm, chosen for its explainability and ability to handle high-dimensional, low-sample-size data, was employed to model the relationships between multi-omics features and Ca2+ reuptake, aiming to identify potential biomarkers for muscle atrophy.

The literature review section highlights existing knowledge about muscle atrophy in microgravity and the role of calcium dysregulation. Studies show muscle atrophy in both slow-twitch (soleus) and fast-twitch (tibialis anterior) muscles during spaceflight. The dysregulation of cytoplasmic calcium levels, specifically due to abnormalities in SERCA pump function, is implicated as a contributing factor. Previous research has demonstrated differing responses in Ca2+ uptake in these muscle types under microgravity conditions, with impaired uptake in slow-twitch muscles and enhanced uptake in fast-twitch muscles. The molecular mechanisms underlying these differences remain unclear. Existing literature supports the use of machine learning (ML) for multi-omics data analysis, particularly its robustness in handling high-dimensional and potentially noisy data, a common issue with space biology experiments. The authors' selection of QLattice, a symbolic regression method, is justified by its focus on generating explainable models, which is crucial for generating biological insights rather than simply achieving high predictive accuracy.

The study utilized multi-omics datasets (transcriptomics, proteomics, and DNA methylation) and calcium reuptake data from the NASA Open Science Data Repository (OSDR), specifically from the Rodent Research missions (RR-1 and RR-9). The data included samples from both soleus (SOL) and tibialis anterior (TA) muscles of mice flown in space (FLT) and ground controls (GC). Data preprocessing involved filtering out lowly expressed genes and proteins, applying variance-stabilizing transformations (VST) using DESeq2 and DEP packages to account for library size differences and reduce heteroskedasticity. Missing values were imputed using K-nearest neighbor imputation, and batch effects were removed using Limma package for proteomics data. For bisulfite sequencing data, the Nextflow nf-core methylseq pipeline was used for processing. The area under the curve (AUC) of calcium reuptake measurements served as the target variable for regression analysis. QLattice, a symbolic regression algorithm, was employed to build models predicting either calcium reuptake levels (regression) or the experimental condition (classification: FLT vs GC). Model selection and feature importance were determined using cross-validation and by observing which features most frequently appeared in high-performing QLattice-generated models. The top features identified were used to infer potential biological interactions and mechanisms through gene set enrichment analysis. Statistical significance was assessed using the Mann-Whitney-Wilcoxon test.

The QLattice regression analysis revealed Acyp1 and Rps7 proteins as the top predictors of Ca2+ uptake in the TA muscle. Acyp1, an inhibitor of Ca2+ transporters in fast-twitch muscles, showed a negative correlation with Ca2+ reuptake, indicating that lower Acyp1 levels were associated with improved Ca2+ reuptake. Rps7, downregulated by nitrosative stress, showed a positive correlation with Ca2+ reuptake. Gene set enrichment analysis for the TA muscle indicated enrichment of biological processes related to apoptosis, endocytosis, and protein localization. In the SOL muscle regression analysis, the top features were all RNA-seq features, and Gene set enrichment analysis revealed enrichment of biological signaling involved in cellular differentiation, synapse organization, and neuron migration. The QLattice classification analysis aimed to identify features differentiating FLT from GC samples. In the TA muscle classification analysis, the top features included RNA-seq, proteomics, and methylation features, with gene set enrichment revealing pathways related to muscle biology, mitochondrial regulation, and actin/myosin structure. Actin, a key component of myofibrils, and Trak2, involved in myosin binding and muscle regeneration, were among the top features. In the SOL muscle classification analysis, RNA-seq features predominated, with gene set enrichment analysis showing enrichment of pre- and post-synaptic membrane organization. In summary, the study identified distinct sets of biomarkers in the TA and SOL muscles that were predictive of changes in calcium reuptake and overall spaceflight response. Notably, Acyp1 and Rps7 in TA and several RNA-seq genes in SOL were highlighted. Significant differences in weight were found between the SOL muscles in spaceflight and ground control samples, suggesting that the TA muscle is more resilient to space conditions than the SOL muscle.

The study's findings provide novel insights into the molecular mechanisms underlying muscle atrophy during spaceflight. The identification of Acyp1 and Rps7 as key biomarkers in TA muscle suggests potential therapeutic targets for mitigating muscle loss. The observed downregulation of Acyp1 in both TA and SOL muscles, even though QLattice was only trained on TA data, supports its potential role in SERCA function dysregulation. The use of QLattice, which provides explainable models, is a significant methodological contribution, allowing for clearer interpretation of complex biological interactions. The different sets of biomarkers identified for TA and SOL muscles highlight the distinct responses of different muscle types to spaceflight. The study's findings also contribute to a better understanding of the impact of spaceflight on various biological pathways beyond Ca2+ reuptake, such as mitochondrial regulation and synaptic organization. The overall results suggest that the TA muscle displays greater resilience to spaceflight compared to the SOL muscle.

This study successfully utilized explainable machine learning to identify multi-omics signatures associated with muscle response to spaceflight in mice. Acyp1 and Rps7 proteins emerged as crucial biomarkers for TA muscle resilience, while distinct RNA-seq based signatures were identified in SOL muscles. The findings highlight potential therapeutic targets for preventing muscle atrophy during space travel. Future research should focus on experimental validation of these biomarkers and investigation of the identified biological pathways.

The study's limitations include the use of data from different missions (RR-1 and RR-9) for comparing omics and calcium reuptake data due to data availability. This might have introduced some variability. Another limitation is the relatively small sample size, although the study effectively mitigated the risk of overfitting by employing QLattice. Further experimental validation studies are necessary to confirm the findings and fully elucidate the biological mechanisms.