Transcriptomic links to muscle mass loss and declines in cumulative muscle protein synthesis during short‐term disuse in healthy younger humans

Muscle disuse leads to a rapid decline in muscle mass, with reduced muscle protein synthesis (MPS) considered the primary physiological mechanism. Here, we employed a systems biology approach to uncover molecular networks and key molecular candidates that quantitatively link to the degree of muscle atrophy and/or extent of decline in MPS during short‐term disuse in humans. After consuming a bolus dose of deuterium oxide (D2O; 3 mL.kg−1), eight healthy males (22 ± 2 years) underwent 4 days of unilateral lower‐limb immobilization. Bilateral muscle biopsies were obtained post‐intervention for RNA sequencing and D2O‐derived measurement of MPS, with thigh lean mass quantified using dual‐energy X‐ray absorptiometry. Application of weighted gene co‐expression network analysis identified 15 distinct gene clusters (“modules”) with an expression profile regulated by disuse and/or quantitatively connected to disuse‐induced muscle mass or MPS changes. Module scans for candidate targets established an experimentally tractable set of candidate regulatory molecules (242 hub genes, 31 transcriptional regulators) associated with disuse‐induced maladaptation, many themselves potently tied to disuse‐induced reductions in muscle mass and/or MPS and, therefore, strong physiologically relevant candidates. Notably, we implicate a putative role for muscle protein breakdown‐related molecular networks in impairing MPS during short‐term disuse, and further establish DEPTOR (a potent mTOR inhibitor) as a critical mechanistic candidate of disuse driven MPS suppression in humans. Overall, these findings offer a strong benchmark for accelerating mechanistic understanding of short‐term muscle disuse atrophy that may help expedite development of therapeutic interventions.


| INTRODUCTION
Reduced physical activity occurring during injury, illness, spaceflight, or with certain lifestyle choices, results in muscle disuse. Even in healthy people, muscle disuse leads to muscle atrophy, 1,2 the health consequences of which include reduced strength and endurance, 3 impaired insulin sensitivity, 4 and reduced anabolic responsiveness to protein intake. 5 Temporally, muscle disuse atrophy is a rapidly occurring phenomenon, with the rate of muscle loss being most pronounced during the initial stages (ie, ≤7 days) of disuse. 6,7 This has significant clinical bearing, not only since the typical length of hospital stay or home-based inactivity following illness/injury is ≤1 week, [7][8][9][10][11] but also because the aggregation of such short periods of disuse throughout life accelerates age-related muscle loss (ie, "sarcopenia"). 6 With acute hospital admissions and the aging population expanding, 12,13 the contribution of short-term muscle disuse to muscle dysfunction is of increasing concern. Muscle disuse atrophy is a phenomenon intrinsic to the affected muscle(s) (thus termed "simple" atrophy), 14,15 yet there are often comorbidities present that may contribute further to atrophy, such as whole-body insulin resistance 16 or systemic inflammation 17 as seen in aging 18 or metabolic disease. 19 The most frequently used models for studying muscle disuse atrophy in humans are bed rest, and unilateral lower-limb immobilization (ULLI; as with musculoskeletal rehabilitation) via, for example, bracing/casting or limb suspension. 6 Although bed rest is arguably more clinically relevant, whether it is an appropriate mechanistic model to study disuse atrophy has been debated. 1,20,21 Indeed, systemic influences arising during bed rest (eg, systemic inflammation, hormonal alterations) make it difficult to isolate the intrinsic mechanisms driving disuse atrophy. 1,20 In contrast, this issue is moderated with ULLI, since disuse remains isolated to a single limb or muscle group. Such a model also permits use of the contralateral leg as a co-temporal control, enabling a more robust physiologic/molecular signal to be obtained within individuals. As a consequence, ULLI can be considered a highly appropriate experimental model to better understand the mechanisms driving short-term disuse atrophy per se. 1 Muscle mass is maintained through a dynamic balance between muscle protein synthesis (MPS) and muscle protein breakdown (MPB). 22 Disuse atrophy must, therefore, be underpinned by net imbalance between the two. MPS (both fed and fasted) has been shown to decline with short-term immobilization (5 days) in younger, healthy individuals, 5 while there remains a lack of direct evidence for any quantifiable change in MPB in humans. 14,15 On this basis, attenuated MPS is considered the predominant physiological mechanism of (non-diseased) short-term disuse atrophy 23 and hence a primary target for therapeutic intervention. 23,24 A critical limiting factor in this regard, however, is that the molecular regulation of MPS remains poorly understood in the context of short-term disuse. Indeed, of the few benchmark molecular investigations that exist to date (ie, those utilizing ULLI in younger, healthy cohorts), most remain as targeted analysis of "established" MPS markers (eg, Akt/mTOR), with either a lack of direct relation to, or inclusion of, measures of MPS/muscle mass across most previous research efforts: hindering the ability for one to draw more robust conclusions on molecular causality. 5,[25][26][27][28][29][30] In any case, the molecular complexity of muscle is overlooked by viewing molecular changes as isolated events, as is the case for targeted molecular profiling or transcriptome-wide gene-level differential expression efforts that have been undertaken to date. 5,[25][26][27][28][29][30] Further worthy of consideration is that, while muscle mass and MPS do indeed decline as a result of disuse, there also appears to exist notable inter-individual variability in muscle physiologic responses to unloading, 7,26,31 implying that molecular regulation of muscle disuse is too somewhat heterogeneous. Accounting for such molecular/physiological heterogeneity consequently helps identify molecules that transcriptional regulators) associated with disuse-induced maladaptation, many themselves potently tied to disuse-induced reductions in muscle mass and/or MPS and, therefore, strong physiologically relevant candidates. Notably, we implicate a putative role for muscle protein breakdown-related molecular networks in impairing MPS during short-term disuse, and further establish DEPTOR (a potent mTOR inhibitor) as a critical mechanistic candidate of disuse driven MPS suppression in humans. Overall, these findings offer a strong benchmark for accelerating mechanistic understanding of short-term muscle disuse atrophy that may help expedite development of therapeutic interventions.

K E Y W O R D S
atrophy, disuse, gene network analysis, muscle protein synthesis, skeletal muscle regulate human muscle phenotypic adaptation. [32][33][34] Studies employing more untargeted approaches that both take into account the molecular complexity of muscle and facilitate linkage between molecular, MPS, and muscle mass changes therefore remain necessary before a complete blueprint of the intrinsic mechanisms regulating short-term human disuse atrophy can be attained.
Recently, we confirmed the efficacy of deuterium oxide (D 2 O; "heavy water") techniques for quantification of MPS in free-living conditions over extended durations (ie, daysto-weeks). [35][36][37] This approach permits measurement of MPS over an entire period of disuse, providing valuable data to determine the mechanistic underpinnings of disuse atrophy. 26 More recently, we have demonstrated the power of network analysis as an advanced bioinformatic tool to uncover novel pathways and molecular drivers of human muscle (un)loading phenotypes. 34,38 As a systems biology concept, network analysis enables for the molecular complexity of a system to first be encapsulated by modeling interactions between its molecular components, with downstream network-driven analyses thereafter facilitating the discovery process of new, physiologically relevant molecules by providing an optimal basis to make direct connections between molecular responses and endpoint physiological changes. 39 Network analysis thus offers a judicious strategy to probe for novel molecular correlates of the short-term muscle disuse phenotype. Using data obtained as part of a new clinical study of healthy younger volunteers, we herein combined the abovementioned bioinformatic and metabolic techniques with robust measures of muscle mass to establish key molecules quantitatively linked to the degree of muscle atrophy and/ or extent of MPS suppression following 4 days of ULLI-in turn providing new insights into possible intrinsic mechanisms of short-term disuse atrophy in humans.

| Subject characteristics
Eight young, healthy male volunteers (mean ± SD: age, 22 ± 2 years; body mass index, 25 ± 3 kg.m −2 ) were involved in this study. Volunteers were initially screened by medical questionnaire, with exclusions for history of any neuromuscular disorder or muscle/bone wasting disease, acute or chronic metabolic, respiratory or cardiovascular disorder, or any other signs of ill health. All subjects were recreationally active upon entry into the study, but were not routinely undertaking heavy, structured exercise. Subjects did not use tobacco-containing products or consume excessive alcohol (>21 units per week). The experimental procedures outlined herein were approved by the Hamilton Integrated Research Ethics Board (HiREB #2192) as part of a wider clinical trial registered at trialregister.nl (NTR6099) and conformed to the Declaration of Helsinki. Written informed consent was obtained from all subjects prior to their participation.

| Overview of experimental procedures
This study involved a bilateral leg protocol, with one leg randomly assigned to be immobilized and the other consequently used as a co-temporal contralateral control. Upon inclusion, subjects were asked to visit the laboratory on two separate occasions, both of which followed an overnight fast. On the first visit, participants had the thigh lean mass of both legs measured using dual-energy X-ray absorptiometry (DXA) (Lunar iDXA; GE Medical Systems, Mississauga, ON), before ingesting a bolus dose of D 2 O (3 mL.kg −1 ) for the measurement of integrated MPS rates. A saliva sample was obtained prior to and 2 hours following D 2 O consumption for measurement of deuterium enrichment in body water. Thereafter, a knee brace (X-Act ROM; DonJoy, Dallas, TX, USA) was fitted to the immobilization leg and fixed at an angle of 60° of knee flexion. Participants wore the knee brace as described for a total duration of 4 full days, ambulating on crutches throughout the immobilization period and returning to the laboratory for their second visit exactly 4 days after their first visit. Upon arrival for visit 2, participants had their knee brace removed, the thigh lean mass of both legs again measured, and a further saliva sample was obtained. Finally, a skeletal muscle sample was collected from the m. vastus lateralis of each leg under local anesthesia using the Bergström needle technique, 40 with extracted tissue snap-frozen in liquid nitrogen and stored at −80℃ until analysis.

| Quantification of thigh lean mass and MPS
DXA-derived thigh fat-and bone-free (ie, lean) mass (referred to herein as "muscle mass" for simplicity) was determined from the lowest point of the ischium to the knee space. 41 Myofibrillar MPS was quantified using the same protocol as is described in full elsewhere. 41 Postintervention DXA data were unable to be obtained for one participant due to a technical issue, with all muscle mass-related analyses herein consequently based on data from seven of the eight participants. Physiologic data were analyzed via IBM SPSS (v26.0) using paired t test (MPS) or two-way repeated measures ANOVA with Fisher's LSD post-hoc analysis (muscle mass), with data presented as mean ± SD and statistical significance set at P < .05.

| RNA-sequencing data generation
Muscle tissue (>30 mg per sample) was homogenized in Trizol reagent (Invitrogen) following the manufacturer's directions. After the addition of chloroform and subsequent centrifugation step the aqueous phase was transferred to RNeasy columns (Qiagen) for mRNA isolation following manufacturer directions. RNA was quantified using the Denovix DS11 FX+spectrophotometer (Denovix, UK). RNA quality was examined using the RNA iQ assay and the Qubit 4 instrument (Thermofisher, UK). All samples returned quality scores ≥8.1. Strand-specific mRNA libraries were prepared using the TruSeq Library Kit (Illumina, UK) and RNA sequenced to generate paired end reads using the NovaSeq 6000 instrument (Ilumina, UK) by Edinburgh Genomics, UK. All samples passed initial quality control. The RNA-sequenced reads were aligned to the human genome (version GrCH38, from Ensembl release 96) and summary counts generated using the R programming language and the Rsubread library. 42

| Differential gene-level expression analysis
Differential gene expression analysis was undertaken using the R programming language and the edgeR library. 43 The count data were filtered using a counts-per-million bases threshold of 10 as recommended by the edgeR authors 44 and the filtered data normalized using the trimmed mean of M values algorithm as implemented in edgeR. To detect differential gene expression between the immobilized and non-immobilized legs we used a design matrix to account for the paired nature of the samples following guidance in the edgeR user's manual (Section 3.4-Additive models and blocking). Differential expression of genes was assessed with a likelihood ratio test using the glmLRT function of edgeR. The Benjamini-Hochberg procedure was used to control for false discovery rate (FDR). Genes with a Benjamini-Hochberg corrected P-value <.05 were defined as being differentially expressed.

| Co-expression network analysis: Network construction and module detection
A gene-wise network was constructed using the weighted gene co-expression network analysis (WGCNA) R library. 45 Before network construction, raw read counts of filtered genes were normalized using a variance-stabilizing transformation 46 in order to attain homoscedastic expression data. An adjacency matrix quantifying the connection strength between each pair of genes was then defined, in which entries equaled the 12th power of the Pearson's correlation coefficient between a given gene pair when the coefficient was positive, and zero otherwise. Such an adjacency structure is the basis for building a signed hybrid coexpression network-selected on the premise that signed variants appear more effective in identifying biologically coherent network modules than unsigned counterparts. 47 Moreover, the power of 12 was chosen following the scalefree topology criterion 48 and was the lowest exponent for which the corresponding scale-free topology fitting index metric (signed-R 2 ) achieved an appropriately high value (≥0.9). A relative interconnectedness measure for each gene pair (dissimilarity topological overlap) was then quantified from the adjacency matrix and subject to average linkage hierarchical clustering to generate a network tree, with network modules in turn determined using the cutreeDynamic algorithm. 49 Of note, a minimum module size of 50 genes was selected when defining modules, which were subsequently merged if their composite expression (module eigengene; first principal component of module gene expression) was very similar (correlation ≥ 0.75)-undertaken to obtain moderately large and distinct modules. 47,50 Modules were then assigned a unique numerical label "Mi" for identification. Genes which could not be distinctly clustered were assigned to a module labeled "M0" and excluded from all subsequent analyses (but included in background gene lists where applicable).

| Co-expression network analysis:
Defining disuse-associated modules Molecular networks associated with short-term muscle disuse were characterized in two ways. First, molecular networks responsive to disuse were established by applying differential analysis (immobilization vs control) to the module eigengene. Such analysis was performed using empirical Bayes-moderated paired t tests as implemented in the limma package for R, 51 with differentially regulated modules selected as those with a Benjamini-Hochberg corrected P-value <.05. Second, module eigengene patterns were correlated with disuse-induced muscle mass (% loss, pre-to-post intervention of the immobilized leg) and MPS (% decrease, control vs immobilized leg) changes to elucidate molecular networks of possible physiologic relevance in the context of short-term disuse muscle adaptation. Physiologically relevant modules were subsequently defined as any network module for which the difference (Δ) in eigengene expression between control and immobilized legs was significantly correlated (|r| > 0.7, P < .05) with a given physiologic variable.

| Ontological analysis of disuseassociated expression changes
The functional characteristics of differentially expressed genes and disuse-associated network modules were derived by testing the relevant gene lists for enrichment of Gene Ontology (GO) terms. Analyses was undertaken using the clusterProfiler package in R, 52 with the corresponding background gene list used in each instance being the genes used as input during differential expression testing/network construction. Each of the three separate GO categories (biological process, BP; cellular component, CC; and molecular function, MF) were considered during analyses, and enriched terms selected as those with a Benjamini-Hochberg corrected P-value <.05.

| Network-driven identification of key molecular drivers of disuse muscle adaptation
Primary candidate targets of disuse-induced human muscle adaptation were determined via two approaches. First, key molecular drivers within disuse-associated modules (module "hub" genes) were established as based on their scaled within-module connectivity (kIM), 53 such that genes within a given module with a scaled kIM value ≥ 0.7 were classified as module hubs. 38 Hub genes of modules associated with muscle mass and/or MPS changes were also further prioritized by those ranked in the upper quartile of their module's gene set on the basis of their individual "gene significance" (GS; |correlation coefficient|) to the given endpoint measure. Second, putative transcriptional regulators of disuse-associated modules were predicted by testing their corresponding gene lists for enriched transcription factor binding site(s) (TFBS) in the 5 kb upstream/downstream region encompassing transcription start sites. In particular, the oPOSSUM-3 webserver 54 was used to query JASPAR core vertebrae profiles with a minimum specificity of 8 bits meeting a conservation cut-off of 0.4 and similarity matrix score threshold of 85%, with the corresponding background gene list in each case constituting the genes inputted during network construction. Of note, large modules (>1000 genes) were represented by their top 75% most connected genes. Enriched TFBS were selected as those with a corresponding Z-score and Fisher score ≥ the mean + 1.5 SD of their respective distributions.

| Physiologic characteristics
Thigh muscle mass tended to change over time in a legdependent manner (leg x time interaction P = .076). At baseline, thigh muscle mass did not differ between legs (P = .424) and remained unchanged in the control leg throughout the study (P = .457). However, 4 days of immobilization lead to a 1.68 ± 1.71% decrease in thigh muscle mass of the immobilized leg (P = .036), with daily MPS rates concomitantly 16.16 ± 9.57% lower in the immobilized leg compared to the control leg (P = .004). Nevertheless, as expected, inter-individual variability in physiologic responses to immobilization was observed ( Figure 1), with declines in muscle mass ranging from −0.4% to 4% ( Figure 1A) and concomitant MPS deficits ranging from 3% to 30% ( Figure 1B).

| Gene-level expression changes following 4 days of leg immobilization
We first compared individual gene expression changes in immobilized muscle vs control muscle and identified a total of 595 genes differentially expressed between conditions (FDR < 5%), the majority of which were found to be downregulated in immobilized muscle (455 genes; Table S1). GO term enrichment analysis of differentially expressed gene sets revealed a clear ontological profile only for genes downregulated following immobilization (Figure 2A,B, Table S2). Downregulated genes were enriched for BP terms related to energy metabolism and muscle development/contraction (Figure 2A), CC terms mainly of contractile apparatus/sarcoplasmic origin ( Figure 2B), and MF terms related to actin binding and ion transporter activity ( Figure 2B). To provide additional functional insight of gene-level expression changes, we then determined the top-ranking up-and downregulated genes based on their log2 fold change (Figure 3)-noting several plausible candidates in the process. For example, among genes most potently upregulated by immobilization is that which encodes the well-established cytokine interleukin 18 (IL18). The top-ranked downregulated gene was identified to be the Leucine Rich Repeat Containing 52 gene (LRRC52). Several novel transcripts were also identified ( Figure 3).

| Network-level expression changes following leg immobilization
To gain a systems-level understanding of the muscle transcriptional response to short-term muscle disuse, we next implemented WGCNA. 48 In doing so, we identified a total of 49 distinct groups of co-regulated genes (ie, network "modules") in our dataset (Table S3), nine of which were found to have an expression profile differentially regulated by immobilization vs control (FDR < 5%; Figure 4, Table S4). Of these, seven could be subsequently labeled for function via enrichment analysis of GO terms (Figure 4, Table S5). In keeping with the overarching themes identified by genelevel differential analysis, network analysis also revealed gene signatures related to mitochondrial function (M49) and myogenesis (M35) as being downregulated in immobilized muscle. Nevertheless, network analysis also uncovered several molecular features of immobilization that were not established using standard gene-level differential analysis alone. In particular, immobilization was additionally found to upregulate muscle gene networks related to ribosome biogenesis (M2), proteasomal ubiquitin-(in)dependent protein catabolism (M28) and ribonucleoprotein complex organization/mRNA processing (M42).

| Molecular networks associated
with the short-term disuse phenotype WGCNA also facilitates discovery of the molecular signatures that underpin phenotypic change in a given state. 39 Capitalizing on the observed heterogeneity in muscle physiologic responses, we used WGCNA to identify network modules with expression change correlating with muscle mass/MPS declines. We identified a total of four network modules that were regulated in proportion to the extent of muscle lost after immobilization (M5, M15, M20, and M31) and a further three that were regulated in proportion to the magnitude of MPS suppression (M8, M10, and M33) ( Figure 4, Table S4). Of those modules regulated in proportion to muscle mass declines, one had expression positively scaling with the extent of muscle mass lost (M15) and was enriched for genes involved in mitochondria function. The remaining muscle mass-related modules all had expression negatively scaling with the magnitude of muscle mass lost and were enriched for genes involved in protein folding (M5), histone acetyltransferase activity (M20), and extracellular matrix (ECM) organization (M31). All three MPS-related network modules had expression positively scaling with the extent of MPS suppression following disuse. Of these, only one could be labeled for function via enrichment analysis of GO terms (M10-ubiquitinprotein transferase activity).

| Network-driven identification of key disuse-associated candidates
We established key molecular drivers of the short-term muscle disuse phenotype by searching for highly connected "hub" genes within disuse-associated modules. 53 In doing so, we identified 242 module hubs from 7158 genes across the 15 disuse-associated network modules (Table S6). Notably, the top-ranked hub gene of each module regulated in proportion to the extent of muscle mass ( Figure 5) or MPS declines ( Figure 6) was also found to be strongly associated with consequent muscle mass/ MPS decrements following immobilization ( Figures 5  and 6). These particular seven genes (AGO2, ADAMTS2, DEPTOR, FOXO3B, LRRC30, MTATP6P1, and TP53BP1), along with the other module hubs that tightly connect to muscle mass/MPS declines ( Figures 5 and 6), thus F I G U R E 3 Top ranking differentially expressed genes following muscle disuse. Top 10 significantly upregulated (red shading) and downregulated (blue shading) genes, as based on log2 fold change (log2 FC). Strength of shading depicts the magnitude of significant differential expression for a given gene, given by the negative log10 of its FDR P-value (with darker shading corresponding to a stronger FDR P-value) represent strong physiologically relevant mechanistic candidates in the context of short-term disuse atrophy.

| Putative transcriptional regulators of co-expression
Finally, we established putative regulatory factors driving co-expression of disuse-associated modules, by searching module genes for conserved TFBS. 54 A core set of 31 putative transcriptional regulators was subsequently identified, with each pertinent module displaying enrichment for at least one TFBS ( Figure 4). Interestingly, just over two-thirds (21 of the 31) of the identified transcription factors were predicted to regulate only a single of the disuse-associated network modules, perhaps implicating the upstream mechanisms that regulate critical molecular responses during short-term muscle disuse are largely unique. Moreover, within six of the seven muscle mass/ MPS-related modules, most of the hub genes strongly associated with muscle mass/MPS declines were predicted to contain a binding site for at least one of the predicted transcription factors for that module (Figures 5 and 6). In module M15 (mitochondria function), just one of the hub genes (DLD) was predicted to contain a binding site for this module's only predicted transcription factor (NKX2-5). Overall, these specific transcription factors may therefore play a central role in the mediating molecular mechanisms of short-term disuse atrophy.

| DISCUSSION
Periods of short-term muscle disuse occur regularly throughout life (eg, with minor injury or illness) and often result in muscle atrophy, 6,7 with the amalgamation of such short periods of muscle disuse consequently regarded as an important driver of sarcopenia. 6 Nevertheless, irrespective of age, the mechanisms that underlie short-term disuse atrophy remain poorly defined, hindering the development of optimal preventative strategies that might in turn help to mitigate sarcopenic progression. Here, for the first time, this study combined robust muscle morphological (muscle mass) and metabolic (MPS) assessment with data-driven network analysis to elucidate new mechanistic candidates of short-term disuse atrophy in humans, namely by establishing molecular networks and key regulatory molecules quantitatively linked to muscle mass and/or MPS changes following 4 days of immobilization. Muscle loss ensues at a rapid rate following the onset of disuse, occurring in a matter of days, 7,55 although there does exist notable inter-individual variability, 7,31 implying that regulation of short-term disuse atrophy is heterogeneous. Consistent with this, we identified a number of molecular networks that were not globally regulated by disuse per se, but rather, were "activated" directly in proportion to the extent of disuse muscle atrophy experienced. Networks that showed the greatest expression reduction in proportion to losses of muscle mass were enriched for genes involved in ECM organization (M31), protein folding (M5), and histone acetyltransferase activity (M20). In recent network analyses of a multi-study tissue biobank, Stokes et al similarly show that gene networks centered upon ECM remodeling form to become a top feature in muscle under different loading states, 34 with decreased ECM gene expression being further noted in both rodent and human models of short-term immobilization. 29,56 ECM remodeling and protein folding, in particular, are both highly mechanosensitive events, and consequently, play a central role in regulating muscle mass by maintaining structural integrity as well as via downstream mechanotransduction of other signaling cascades responsible for muscle growth/maintenance. [57][58][59][60] Collectively, such F I G U R E 5 Hub gene networks for modules regulated in proportion to the extent of muscle mass lost with disuse. Each network visualization comprises all corresponding hub genes of the given module, with the top ranked hub gene given by a larger node. Red borders illustrate hub genes within each module that are also highly linked to the extent of mass loss (ie, within the upper quartile of corresponding module's genes ranked by their gene significance to declines in mass), with node fill colors illustrating whether the corresponding hub gene is among the module genes that are enriched with binding site(s) for the given transcription factor (TF). Visualizations were generated using Cytoscape (v3.7.1) 92 findings indicate that the downregulation of processes involved in muscle structural dynamics is part of potent intrinsic mechanisms contributing to short-term disuse atrophy, presumably as a direct result of the musculature rapidly adapting to alterations in muscle loading.
Unloading also plays a fundamental role in regulating MPS, 61-63 the decrement in which is proposed to be the dominant mechanism underpinning short-term disuse atrophy. 23 Here, we obtained D 2 O-derived measurements of MPS to assess integrated MPS changes across the entire period of disuse, in contrast to traditional "acute" (ie, stable isotopically labeled amino acid infusion) tracer methods. 35,64 The D 2 O method remains novel in human disuse research. 26,65 Exploiting this approach in tandem with untargeted transcriptomic profiling thus provided a powerful basis through which to directly connect key molecular and MPS changes and, in turn, identify new mechanistic candidates of disuse-induced MPS suppression. In doing so, we identified a molecular network that was both activated by disuse and positively correlated with declines in MPS (M8)-that is, those with greater upregulation of this network also experienced a greater suppression of MPS. While this network demonstrated no particular ontological annotation, the top-ranked "hub" gene was DEPTOR, which is a potent inhibitor of mTOR: 66 a regulatory complex through which MPS is positively controlled in response to mechanical (un) loading. 67 Work in rodents has shown DEPTOR activity to be tightly linked to MPS, 68 with DEPTOR knockdown ameliorating disuse atrophy in vivo via enhanced MPS. 69 However, to our knowledge the current study is the first to provide evidence for DEPTOR as a key mechanistic target of disuse driven MPS suppression directly in humans. While presumably mTOR related, the precise action by which DEPTOR might inhibit MPS during short-term disuse warrants elucidation. Indeed, DEPTOR is a common inhibitory component of both mTOR complexes 1 and 2 (mTORC1/2), [70][71][72] and could thus theoretically inhibit MPS during disuse directly upstream of Akt (via mTORC2 inhibition), at the intersection of Akt and mTORC1 (via PRAS40 inhibition), and/ or directly at the level of mTORC1 (ie, independent of F I G U R E 6 Hub gene networks for modules regulated in proportion to the extent of MPS suppression with disuse. Each network visualization comprises all corresponding hub genes of the given module, with the top ranked hub gene given by a larger node. Red borders illustrate hub genes within each module that are also highly linked to the extent of MPS decline (ie, within the upper quartile of corresponding module's genes ranked by their gene significance to declines in MPS), with node fill colors illustrating whether the corresponding hub gene is among the module genes that are enriched with binding site(s) for the given transcription factor (TF). Visualizations were generated using Cytoscape (v3.7.1) 92 Akt altogether). [69][70][71][72] Nevertheless, augmentation of both 4E-BP1 and S6K1 signaling upon DEPTOR knockdown 69 suggests that DEPTOR may suppress MPS during disuse by impairing both translation initiation and elongation. Given that this network was not associated with muscle mass changes, it remains to be established how any such DEPTOR-induced MPS suppression might relate to corresponding atrophy. Nonetheless, our direct linking of transcriptomics to MPS suggests DEPTOR-dependent, mTOR-related suppression of MPS could be an important mechanism of atrophy during disuse in humans.
Two additional molecular networks were found to quantitatively connect to alterations in MPS, with both tending to display greater expression increases in subjects with a more pronounced decline in MPS (M10-ubiquitinprotein transferase activity, M33-undetermined function). Interestingly, more than half of the top-ranked hub genes contained within M10 putatively serve to function in proteolytic-related pathways (FOXO3B, ATG2B, WDR7, TTC37, RBM25, BIRC6, and UFL1), 73 while FBXO32 (ie, MAFbx/Atrogin-1) was among the top-ranked hub genes of M33. The untargeted identification of FBXO32 is particularly striking, given that it is widely considered a universal atrophy-related gene ("atrogene") involved in initiating proteolysis across a wide range of catabolic conditions 74 and is unsurprisingly found upregulated often in human models of simple disuse. 26,27,55,[75][76][77][78] Nevertheless, our data extend beyond previous studies by suggesting a central regulatory role for FBXO32 in suppressing MPS during short-term human disuse atrophy. More generally, our above-mentioned data seem to imply an associative link between increased proteolytic marker expressionand thus ostensibly MPB processes-and declines in MPS during short-term disuse. Whether or not MPB plays a direct role in human muscle disuse atrophy remains up for debate. Secondary evidence for increased MPB during human disuse can indeed be derived from other "static" markers of myofibrillar breakdown (eg, increased interstitial 3-methylhistidine and proteasomal/lysosomal molecule expression). 26,77,79 On the other hand, both model-derived and direct quantifications, although scant, have found no appreciable change in "bulk" MPB, 80,81 with mathematical estimates further showing that attenuated MPS alone is sufficient to explain muscle loss during disuse. 23,25 Thus, if observed increases in markers of MPB play a mechanistic role in human disuse atrophy, it seems likely to be beyond functionally regulating bulk myofibrillar proteolysis, which itself does not appear to be a necessity of human disuse atrophy. An alternative contention which may then both explain and be supported by our current data is that increases in static markers of MPB are actually acting to degrade specific components of the protein-synthetic machinery (and thus regulating the capacity for MPS), rather than regulating bulk MPB per se. 14 This indeed seems plausible especially in the case of FBXO32, which holds a strong potential to suppress MPS during disuse via degradation of EIF-3F: an essential protein in instigating translation initiation. 14 Although, such inferences should remain taken with caution given the lack of available data that currently exists on dynamic MPB changes during human disuse as opposed to changes in MPS. 14,15 It is also important to acknowledge those molecular networks that did not link to muscle mass/MPS changes but were nevertheless found to be regulated in direct response to the disuse intervention. Such networks could theoretically contribute to short-term disuse atrophy via mechanisms beyond those we have considered and/or be molecular events that precede atrophy occurring beyond 4 days of disuse. Of note are the functional expression patterns that remained consistent across both our differential module-level and gene-level analyses; namely the concordant downregulation of genes functionally related to mitochondrial function, muscle cell differentiation and ion transmembrane transporter activity. Tight coordination of the mitochondrial transcriptome corroborates with recent large-scale network analysis of differential muscle loading responses to resistance training/immobilization, in spite of a longer duration of disuse (14 days) having been employed. 34 In fact, downregulation of mitochondria-related genes appears a general feature of muscle disuse (ie, independent of duration or model), [28][29][30]82,83 and on the whole is likely indicative of a disuse-induced impairment in mitochondrial function. The downregulation of genes involved in ion transmembrane transporter activity was also notable, given that it remains a relatively novel finding in the context of human muscle disuse. Interestingly, the top-ranking hub gene of this specific molecular network was identified as ACHE, the primary function of which is to hydrolyze acetylcholine at neuromuscular junctions. ACHE deficiency is associated with impaired neuromuscular junction function in the form of reduced neuromuscular transmission. 84 The possible physiological relevance of this specific network is thus perhaps more likely to lie within the realms of disuse-induced muscle dysfunction (eg, impaired muscle contractibility).
Worthwhile appraising are the relative merits of our study design. By subjecting healthy younger individuals to disuse via ULLI, we were able to isolate the effects of disuse on muscle tissue per se independently of any comorbid factors. By integrating muscle mass and MPS measurements with network-driven analysis, we were subsequently able to generate new insight into potential pathways and key molecules associated with the short-term human muscle disuse phenotype; crucially, direct links between MPS deficits and both DEPTOR and molecular MPB biomarker upregulation. Nevertheless, we acknowledge that our design of comparing end-intervention biopsies between legs to assess immobilization effects could, in theory, be complicated by having a weight bearing "control" limb that is supporting more of a person's bodyweight. In which instance, we would be comparing biopsies from a limb bearing more weight to one bearing no weight and thus the non-immobilized leg biopsy would not be a true control nor our model consequently a true representation of immobilization per se. While we cannot discount such a possibility, if this were indeed the case then it might be reasonable to postulate improved conditioning in the non-immobilized leg. Regarding our physiologic variables, however, in line with others, 7,85,86 muscle mass was unchanged in the non-immobilized leg herein, with further evidence demonstrating that D 2 Oderived integrative MPS rates remain stable over 7 days in a non-immobilized leg while declining in the immobilized leg. 26 Equally, it may even be that the non-immobilized leg is bearing less weight across the immobilization phase of the study as people on crutches simply do less and are more sedentary as the work of ambulation on crutches is taxing compared to normal ambulation. Some support for this comes from our previous observation that both legs during a ULLI model showed evidence of reduced vascular response to shear stress, demonstrating at least some level of deconditioning can take place in both limbs. 87 Evidence also shows that muscles, including the m. vastus lateralis, can do less work (ie, reduced loading) during crutching and single-leg immobilization and thus may not necessary always be loaded to a greater extent. 88 Either way, however, it is likely that initial responses of the immobilized and non-immobilized leg muscles are manifest at the mRNA level. As a result, we cannot fully exclude the possibility of some transcriptomic perturbation having occurred in the non-immobilized leg (convergent, divergent, or even entirely distinct from the immobilized leg) if it was in fact either over-or under-loaded throughout, and which theoretically could have still arisen in spite of muscle mass (and plausibly MPS) of this leg remaining stable. Relatedly, given the number of genes analyzed, the potential for at least some within-individual molecular variability and/or cross-leg "transfer" response also cannot be ruled out entirely. Clearly, to resolve these issues we could have taken a baseline biopsy from each leg to derive and compare temporal mRNA changes in each leg. Without these biopsies we thus acquiesce that our results might not provide an entirely clear transcriptomic interpretation of the impact of immobilization per se and further work should add pre-intervention biopsies to account for the potential confounding of using a post-intervention biopsy from the non-immobilized leg as a control.
We also recognize a further need for caution when inferring true intervention effects of this work due to the relatively small number of individuals that were included herein. Although, the number of samples used was within the WGCNA developers recommendations, 89 and the network approaches employed shown to perform strongly even when applied to smaller (~20 samples)/paired design datasets. 90,91 Thus, while further work incorporating larger cohort sizes and deeper quantitative/mechanistic investigation is ultimately warranted as a next step, a major novel outcome of our current work is the networkdriven identification of an experimentally tractable set of physiologically relevant mechanistic targets: key candidates that quantitatively connect to the primary physiological/metabolic consequences of muscle disuse-that is, muscle atrophy and blunted MPS. Our findings therefore offer a strong benchmark to expedite mechanistic understanding of short-term human disuse atrophy and, in turn, may help to accelerate the development of effective countermeasure approaches.

ACKNOWLEDGMENTS
The authors thank all of the volunteers for giving up their time to partake in the human clinical experiment. Craig R. G. Willis is supported by the Biotechnology and Biological Sciences Research Council-funded South West Biosciences Doctoral Training Partnership (BB/ J014400/1; BB/M009122/1). Tanner Stokes is supported by the Alexander Graham Bell Canadian Graduate Scholarship from the Natural Sciences and Engineering Research Council. This work was partially supported by funding from the Biotechnology and Biological Sciences Research Council (BB/N015894/1). This research was also supported by the MRC-Versus Arthritis Centre for Musculoskeletal Ageing Research (MR/P021220/1; MR/ R502364/1) and National Institute for Health Research Nottingham Biomedical Research Centre. The views expressed are those of the author(s) and not necessarily those of the NHS, the NIHR or the Department of Health and Social Care.

DISCLOSURES
SMP reports that he is an inventor on a patent (WO/2018/157258) held by Exerkine Corporation and holds equity in Exerkine (all proceeds donated to charity). SMP is a member, and receives an honorarium, of the scientific advisory board of Enhanced Recovery (https://erspo rtsdr ink. com/; all proceeds donated to charity). SMP has received, in the last 5 years, honoraria and travel expenses from the US National Dairy Council, the National Cattlemen's Beef Association, and the US Dairy Export Council. The other authors declare that they have no conflicts of interest.