Introduction

Physical activity is a cornerstone in the prevention and treatment of several chronic diseases like obesity, non-alcoholic fatty liver disease (NAFLD), atherosclerotic vascular disease, and type 2 diabetes mellitus ¹. Both acute and long-term exercise may enhance insulin sensitivity and thereby improve glucose tolerance ². Both resistance and endurance exercises enhance insulin sensitivity, although the most pronounced effect is observed when combining these training modalities ³.

Metabolic adaptations to exercise encompass intricate inter-organ communication facilitated by molecules referred to as exerkines ^4–7. These exerkines are secreted from various tissues and include a variety of signal molecules released in response to acute and/or long-term exercise with endocrine, paracrine and/or autocrine functions ^4,5. Although there has been considerable emphasis on exerkines originating from skeletal muscle (SkM) ^8,9, it is also known that exerkines can originate from organs such as white ^6,10,11 and brown adipose tissue ¹² or the liver ¹³. The established bona fide exerkine, interleukin-6 (IL6), is released during muscle contractions, contributing to improved overall glucose homeostasis ^4,14. In addition, a range of other exerkines are recognized, including IL7 ¹⁵, 12,13-diHOME ¹², myonectin ¹⁶, myostatin ^17,18, METRNL ¹⁹, CSF1 ⁸, decorin ²⁰, SFRP4 ¹⁰, fetuin-A ^13,21, and ANGPTL4 ^22,23, among many others ⁵.

Extensive screening aimed at discovering novel exercise responsive blood proteins have faced considerable challenges, primarily due to the technical challenges in quantifying the blood proteome on a large scale. However, recent advances in multi-plex technology, such as the proximity extension assay (PEA), have made it possible to quantify more than 3000 proteins in blood samples more reliably than traditional untargeted mass spectrometry (https://olink.com/application/pea). Some recent studies have used other proteomic platforms, such as aptamer-based techniques (https://somalogic.com/somascan-platform/), to show that acute and long-term aerobic exercise affected several hundred serum proteins^24–28, but the downstream causal effects of such changes on clinical phenotypes are not known. Furthermore, no studies have used the PEA technology to identify exerkines potentially underlying the mechanisms through which long-term physical activity, including strength exercise, enhances glucose homeostasis.

We performed the ‘physical activity, myokines and glucose metabolism’ (MyoGlu) study ²⁹, which was a controlled clinical trial aiming to identify novel secreted factors (‘exerkines’) that may serve as links between physical activity and glucose metabolism. We conducted a comprehensive serum screening of 3072 proteins in normal weight and overweight men both before and after combined endurance and strength exercise. Rigorous phenotyping was carried out, including hyperinsulinemic euglycemic clamping, assessments of maximum oxygen uptake, maximum muscle strength, and ankle-to-neck MRI/MRS scans.

Exerkines identified with potential effects on glucose homeostasis in the MyoGlu study were subsequently subject to analysis across several external data sets. Using data from 47,747 participants in the UK Biobank ³⁰, we assessed correlations between candidate proteins and estimates of physical activity and glucometabolic outcomes. These associations were then tested for causality using Mendelian randomization (MR). Exerkines of interest were also assessed in a knockout mouse model and in a exercise mouse model to further assess potential links with glucose homeostasis.

Results

Cohort characteristics

We studied 26 male participants, including 13 with normal weight, and another 13 with overweight, as described previously ²⁹. They were subjected to 12-weeks of high-intensity resistance and endurance exercise (Figure 1). The overweight participants had lower glucose tolerance and insulin sensitivity compared to the normal weight participants (Supplementary Table 1). After the 12-week intervention, body fat mass decreased and lean body mass increased, together with significant improvements in insulin sensitivity (∼40%), maximum oxygen uptake and muscle strength in both groups (Supplementary Table 1).

Figure 1.

Serum proteome responses to prolonged exercise

Recognizing that circulating proteins could mediate exercise-induced metabolic improvements, we next investigated alterations in the serum proteome in response to the 12-week intervention using PEA technology. Of the 3072 proteins quantified, we detected increased serum concentrations of 126 proteins, and decreased serum concentrations of 157 proteins following the 12-week intervention, at a false discovery rate (FDR) below 5% (Figure 2 A-C; Supplementary Table 2-4). Among these, 20 proteins increased exclusively in normal weight men, whereas 19 proteins increased exclusively in overweight men (Figure 2 D). Four proteins were uniquely reduced in normal weight men, and 66 proteins were uniquely reduced in overweight men (Figure 2 E).

Figure 2.

Several of the exercise-responsive proteins had potential roles in muscle adaptation and metabolism. For example, platelet-derived growth factor subunit B (PDGFB) and IL7, are both myokines with potential effects on muscle differentiation ^15,31. Further, fibroblast growth factor-binding protein 3 (FGFBP3) may influence running capacity ³² and muscle strength ³³. NADH-cytochrome b5 reductase 2 (CYB5R2) can preserve SkM mitochondria function in aging mice ³⁴. FGFBP3 and switch-associated protein 70 (SWAP70) may protect against weight gain ³⁵ and cardiac hypertrophy ³⁶, respectively. Finally, dual specificity protein phosphatase 13 isoform A (DUSP13A) is highly specific to SkM ³⁷, making it a potential novel muscle-specific marker for long-term exercise. Detailed results for 2885 proteins in response to prolonged exercise are shown in Supplementary Table 2.

A proteomic liver fat signature in overweight men

In response to the 12-week exercise intervention, a larger number of serum proteins responded in overweight men than in normal weight men (Figure 2 B-C). In particular, 66 proteins decreased in serum after 12 weeks in overweight men (Figure 2 E). Gene ontology analyses revealed known pathways only for the proteins that decreased in overweight men (Figure 3 A-B), and one of the most enriched pathways is related to metabolism of organic acids (Figure 3 C). A key driver analysis of the 66 proteins identified SLC22A1, a hepatocyte transporter related to liver fat content (Figure 3 D). Furthermore, the 66 proteins also displayed a 24% overlap with a known human serum proteomic signature of non-alcoholic fatty liver disease (NAFLD; Figure 3 E) ³⁸, but no common proteins with signatures of specific liver cells (The Human Liver Cell Atlas: ³⁹). Baseline serum protein concentrations in the identified signature of 66 proteins were higher among men with overweight compared to those with normal weight, but were reduced or normalized in overweight men following prolonged exercise (Figure 3 F). Using principal component analysis of the 66 proteins, the first component correlated positively to liver fat content at baseline (Figure 3 G), but not after prolonged exercise (Figure 3 H). Similarly, the first component also correlated positively with several liver-related markers at baseline (Figure 3 L) and negatively to insulin sensitivity at baseline (Figure 3 I), but not after prolonged exercise (Figure 3 J). The first component mediated 37% of the association between baseline insulin sensitivity and liver fat content (Figure 3 K).

Figure 3.

Secretory proteins

Among the 96 up-regulated and 110 down-regulated serum proteins responding to the 12-week exercise intervention (Figure 2 D-E), 37 are curated secretory proteins, and another 46 proteins are predicted as highly likely secretory proteins (Figure 4 A-C). We assessed the corresponding mRNA responses in SkM and subcutaneous white adipose tissue (ScWAT) following the 12-week intervention (Figure 4 A-C). In total, 19.7 % of the serum secretory proteins displayed a directionally consistent significant change mRNA levels in SkM, whereas 12.1 % of the serum secretory proteins exhibited a corresponding mRNA response in ScWAT (Figure 4 C). COL1A1 was the most responsive SkM mRNA that also had a corresponding increase in serum COL1A1 after prolonged exercise (Figure 4 D-E). CCL3 was the most responsive ScWAT mRNA that also had a corresponding decrease in serum after prolonged exercise (Figure 4 F-G). To prioritize proteins for follow-up analyses, we focused on SMOC1 and CD300LG, which had similar exercise responses in blood, SkM and ScWAT (Figure 4 H). SMOC1 is a known hepatokine with effects on insulin sensitivity in mice ⁴⁰, but probably with no causal link to insulin sensitivity in humans ^40,41. Thus, we focused on CD300LG in subsequent analyses.

Figure 4.

Cd300 lg

CD300LG displayed increased concentration in serum (+20%, p<0.001) together with increased levels in both SkM (+60%, p<0.001) and scWAT (+13%, p=0.01) mRNA following the 12-week exercise intervention (Figure 5 A-C). Changes in serum CD300LG correlated positively with changes in insulin sensitivity after the intervention (rho=0.59, p=0.002; Figure 5 D). In addition, serum CD300LG concentration was lower in overweight than normal weight men (−51%, p=0.014) and positively correlated with insulin sensitivity before as well as after the 12-week intervention (pre-trained: r=0.50, p=0.001, and post-trained: r=0.43, p=0.028).

Figure 5.

To investigate the potential effect of serum CD300LG on SkM and ScWAT, we performed an overrepresentation analysis on the top 500 mRNAs that were positively correlated (p<0.05) with serum CD300LG levels in each tissue (Figure 5 E-H). Pathway analyses revealed that the change in serum CD300LG concentrations correlated with changes in expression of genes involved in oxidative phosphorylation, G2M check point and hypoxia both in ScWAT and SkM (Figure 5 E-F). In ScWAT, serum CD300LG levels also showed the strongest enrichment with angiogenesis pathways (Figure 5 E). In ScWAT, the change in ScWAT CD300LG mRNA levels correlated positively with the change in ScWAT mRNA of genes related to angiogenesis/vasculature development (Figure 5 G). Similar correlations between CD300LG mRNA and angiogenesis genes were observed in SkM as well (Figure 5 H). For example, 60% of the mRNAs in the angiogenesis pathway correlated with CD300LG (Figure 5 H). However, serum CD300LG levels were also correlated positively with pathways related to fatty acid metabolism in both ScWAT (Figure 5 E) and SkM (Figure 5 H).

To explore tissue specific expression of CD300LG, we utilized data from a publicly available human tissue panel ⁴². CD300LG is highly expressed in adipose tissue compared to other tissues (Figure 6 A) supporting our observation that ScWAT expression was substantially higher than in SkM (Figure 5 B-C). To further investigate which cells in ScWAT that express CD300LG, we utilized data from a single cell mRNA sequencing atlas of human ScWAT (https://singlecell.broadinstitute.org/single_cell) generated by Emont et al. ⁴³. CD300LG mRNA in ScWAT was primarily expressed in venular endothelial cells, but not adipocytes or other cell types present in ScWAT (Figure 6 B-E).

Figure 6.

We next explored whether CD300LG mediates tissue-tissue cross-talk using data from the GD-CAT (Genetically-Derived Correlations Across Tissues) database ⁴⁴, which is a tool for analyzing human gene expression correlations in and across multiple tissues. In men, ScWAT CD300LG correlated strongly with ScWAT, SkM and aortic gene expression (Supplementary Figure 2). Consistent with our observations in the MyoGlu exercise intervention study, the top network of gene expression in ScWAT related to ScWAT CD300LG mRNA was angiogenesis (Supplementary Figure 2 A). Like ScWAT, SkM CD300LG also correlated strongly with ScWAT, SkM and aortic gene expression (Supplementary Figure 2 B). The proteasome complex was the top network of gene expression related to SkM CD300LG mRNA Supplementary Figure 2 B). In contrast, running the same analyses in women did not reveal associations between CD300LG and angiogenesis (Supplementary Figure 3 A-B).

We then evaluated serum CD300LG levels in up to 47,747 samples in the UK biobank (see Methods). Descriptive statistics of the UK biobank cohort are presented in Supplementary Table 5. Serum CD300LG levels were positively associated with several measures of physical activity (all metabolic equivalent tasks, results from the international physical activity questionnaire and meeting the recommended amount of weekly physical activity or not; Table 1). Interestingly, serum CD300LG levels were most strongly related to vigorous activity (Table 1). Furthermore, the associations between serum CD300LG and physical activity were significantly stronger in men than in women (Table 1). Serum CD300LG levels were also positively associated with fat mass and fat free mass, and negatively associated with glucometabolic traits including serum glucose levels, Hb1Ac and the risk of having type 2 diabetes (Table 1). These associations were independent of body mass index (BMI).

Table 1.

GWAS of serum CD300LG levels

GWAS analyses of CD300LG levels detected 43 independent genome-wide significant genetic associations across the genome (Supplementary Figure 4). The genomic inflation factor (λ=1.0966) and LD score intercept (1.039) were consistent with our GWAS being well controlled for population stratification and other possible biases (Supplementary Figure 5). The most significant SNPs lay along chromosome 17 with these SNPs mapping to the genomic region encoding the CD300LG gene (Supplementary Figure 4). Follow-up analyses revealed three significant, independent cis-pQTLs associated with the protein CD300LG (Supplementary Table 6) and a number of trans-pQTLs (Supplementary Table 7).

Mendelian randomization (MR) analysis

The independent genome-wide significant SNPs from the CD300LG GWAS were used for two-sample MR (see Methods for details), where 39 SNPs were also available in the outcome GWAS ⁴⁵. We first performed Inverse Variance Weighted (IVW) MR analysis to test the causal relationship between CD300LG and fasting glucose, 2-hour post oral glucose tolerance test (OGTT) glucose levels and HbA1c using only cis-SNPs (Supplementary Table 8) and all SNPs (Supplementary Table 9). The cis IVW MR analysis showed some evidence for a negative causal effect of CD300LG on fasting insulin (p=0.01), but due to only three SNPs in these analyses, we could not perform additional sensitivity analyses (except for tests of heterogeneity in estimates of the causal effect across SNPs) and could not determine whether the absence of strong evidence for a causal effect on the glycaemic parameters was genuine or whether our analyses just lacked power. Although some of the analyses involving all the genome-wide significant SNPs indicated a potential causal link between increased serum CD300LG concentration and these outcomes (Supplementary Table 9) the analysis showed significant heterogeneity. We did not detect strong evidence of directional pleiotropy (significant MR Egger intercept, Supplementary Table 9). The heterogeneity in the analysis is possibly due to the fact that many of the SNPs found in the GWAS of CD300LG are associated with related phenotypes which could exert pleiotropic effects on diabetes related outcomes, and so the results should be interpreted with care (Supplementary Table 10). Due to the heterogeneity in our results we therefore performed MR PRESSO to account for outliers. The MR PRESSO analysis showed a significant negative effect of CD300LG on all outcomes (Table 2).

Table 2.

Mouse models

To functionally validate association of CD300LG with metabolic homeostasis, we leveraged phenotypic data for exercising mice and for Cd300lg deficient (Cd300lg^-/-) mice that both were publicly available through the MoTrPAC ⁴⁶ study and the international mouse phenotyping consortium (PhenoMouse) ⁴⁷.

There is a 51% homology between human CD300LG and mouse Cd300lg ⁴⁸, and also in mice, Cd300lg is predominantly expressed in adipose tissue endothelial cells ⁴³.

In MoTrPAC ⁴⁶, Cd300lg levels in scWAT from n=12-15 male and female mice were increased after exercise for 8 weeks (∼30% in both female (p=0.03) and male (p=0.01) mice) (Supplementary Figure 5 A-C). Based on data from n=3050 mice from PhenoMouse male, but not female, mutants for the Cd300lg^tm1a(KOMP)^Wtsi allele displayed impaired glucose tolerance (Supplementary Figure 5 D), but no change in fasting glucose and insulin (Supplementary Figure 5 E-F). Mutant male, but not female, mice also displayed increased lean mass (Supplementary Figure 5 G) and less fat mass (Supplementary Figure 5 H). Detailed PhenoMouse results are presented in Supplementary Table 11.

Discussion

In the present study, we characterized the effects of strength and endurance exercise on the serum proteome of sedentary normal weight and overweight men. We identified significant changes in 283 serum proteins related to many signaling pathways after the 12-week intervention. Some of these proteins were related to the mitochondria, muscle differentiation and exercise capacity. Among known secretory proteins, 19.7% and 12.1% displayed corresponding mRNA changes in SkM and ScWAT, respectively. Although some proteins may be myokines, others may be adipokines or other types of exerkines. A multi-tissue responding protein was CD300LG, which also correlated positively to insulin sensitivity. CD300LG was particularly interesting because we could replicate the finding in an external cohort, find evidence of a causal link to glucose homeostasis, and perform functionally validation in mice models. Furthermore, the association between CD300LG, physical activity and glycemic traits might display sex dimorphic relationships.

One of the protein signatures observed in response to exercise was based on strong associations with markers of liver fat content in overweight men. This was related to SLC22A1, which regulates the hepatic glucose-fatty acid cycle affecting gluconeogenesis and lipid metabolism ⁴⁹, and may influence liver fat accumulation ⁵⁰. This signature also shared many common proteins with a known serum NAFLD proteomic signature ³⁸. However, we did not detect overlaps between proteins in this signature and specific gene expression patterns of liver cells (e.g. hepatocytes, immune cells etc.) ³⁹. This observation suggests that the proteins detected do not relate to liver protein synthesis per se, but may accumulate in serum due to being released in the blood stream as a result of impaired liver protein catabolism or cell damage as a consequence of overweight/obesity. Notably, this protein signature in overweight men normalized after 12 weeks of exercise and resembled the signature observed in normal weight men. These data suggest prolonged exercise leads to improvements of liver function in overweight men.

Several proteins responding to prolonged exercise had a known signal sequence. These secretory proteins are of particular interest because they could mediate inter-tissue adaptations to exercise. For example, COL1A1 was substantially increased in serum and its corresponding mRNA level was increased in SkM. However, COL1A1 is a collagen peptide that is related to muscle damage, turn-over and extracellular matrix remodeling in response to exercise ⁵¹ and may mostly reflect muscle restructuring and not represent signaling effects to distant tissues. The large overlap between serum proteins and SkM mRNA most likely suggests a similar phenomenon, where tissue restructuring following exercise is reflected in blood. However, there are probably also several myokines with distant signaling effects among the identified proteins. CCL3 was reduced in serum in parallel with a reduction in its mRNA level in ScWAT. CCL3 is a monocyte chemoattractant protein that may be related to immune cell infiltration in adipose tissue ⁵². Hence, this may reflect a positive effect of prolonged exercise on adipose tissue inflammation, which is in line with our previous results showing normalization of adipose tissue inflammation following prolonged exercise ¹⁰.

A particularly interesting protein was CD300LG, which responded to prolonged exercise in serum, and, judged by its mRNA levels, in SkM and ScWAT. Serum CD300LG levels were lower in overweight compared to normal weight men. Furthermore, the exercise-induced response in CD300LG correlated positively to improvements in insulin sensitivity, and there was also a significant correlation between serum CD300LG and insulin sensitivity both before and after the intervention. We therefore analyzed CD300LG in an external data set, the UK Biobank, and again we observed positive associations between especially vigorous exercise and serum CD300LG. Moreover, serum CD300LG levels were negatively associated with glucose levels and type 2 diabetes in the UK Biobank, and these associations might be causal based on MR analysis. These findings were functionally corroborated by the alterations in glucose tolerance and parameters related to insulin sensitivity observed in Cd300lg^-/- mice. Thus, CD300LG may represent an exerkine with a causal link to glucose homeostasis. However, whether CD300LG can mediate tissue-tissue crosstalk is unknown. CD300LG is a cell surface protein with a transmembrane domain, but is also a predicted secretory protein ⁵³. Whether the protein is released from the cell surface in a regulated manner to mediate cross-tissue signaling needs further investigation. Furthermore, the exact link between CD300LG and glucose metabolism is not clear, but possibly related to the fact that CD300LG is expressed in endothelial cells ⁵⁴, linked to blood pressure ⁵⁵, lymphocyte binding ⁵⁶, blood triacylglycerol levels ^57,58 and molecular traffic across the capillary endothelium ⁵⁹. Both MyoGlu and GD-CAT data also suggested that CD300LG may be related to angiogenesis in ScWAT ⁶⁰ and SkM ^60,61, at least in men. Hence, we speculate that the link between CD300LG and glucose metabolism is related to improved tissue capillarization/vascular function following prolonged exercise. Furthermore, since vigorous exercise leads to angiogenesis in ScWAT and SkM ⁶⁰, serum CD300LG may be a maker of exercise intensity.

Limitations and strengths

Although MyoGlu included only 26 sedentary men, they were extensively phenotyped with euglycemic hyperinsulinemic clamp, fitness tests, whole body imaging (MRI/MRS) and mRNA sequencing of ScWAT and SkM. We also supplied our study with data from 47747 persons in the UK biobank to enhance the validity and generalization of the results. Furthermore, to assess sex differences we stratified analyses for men and women in the UK Biobank, in external data bases (GD-CAT) and analyzed data from both male and female mice. Since correlations with the clamp data only imply a role for a protein with regard to glucose homeostasis, so we also tested associations with related glucometabolic traits in the UK Biobank and tested these associations for causality using MR. We also included data from exercised mice, and mutant mice to further strengthen the results. Our serum proteome study assessed 3072 proteins, and therefore we do not cover the complete human proteome. However, the Olink platform is based on dual recognition of correctly matched DNA-labeled antibodies, and DNA sequence-specific protein-to-DNA conversion to generate a signal. This is a highly scalable method with an exceptional specificity (https://olink.com/application/pea/). Previous exercise-proteomic studies has looked at ∼600 proteins in overweight men after endurance exercise ²⁵, and three papers were published from the HERITAGE study analyzing ∼5000 proteins in response to endurance exercise ^26–28. However, our study is the first and largest exercise study using PEA in both overweight and normal weight men, and also including strength exercise.

Conclusion

Our study provided a detailed analysis of serum proteins responding to three months of strength and endurance exercise in both normal weight and overweight men. Our results identified a novel NAFLD-related serum protein signature in overweight men that was normalized after prolonged exercise. We also identified hundreds of tissue-specific and multi-tissue serum markers of e.g., mitochondrial function, muscle differentiation, exercise capacity and insulin sensitivity. Our results were enriched for secretory proteins (exerkines), such as CD300LG, which may be a marker of exercise intensity especially in men, and may also have causal roles in improved glucose homeostasis after physical activity.

Methods

The MyoGlu study was conducted as a controlled clinical trial (clinicaltrials.gov: NCT01803568) and was carried out in adherence to the principles of the Declaration of Helsinki. The study received ethical approval from the National Regional Committee for Medical and Health Research Ethics North in Tromsø, Norway, with the reference number 2011/882. All participants provided written informed consent before undergoing any procedures related to the study. The UK biobank has ethical approval from the North West Multi-Centre Research Ethics Committee (MREC), which covers the UK, and all participants provided written informed consent. This particular project from the UK biobank received ethical approval from the Institutional Human Research Ethics committee, University of Queensland (approval number 2019002705).

Participants

The MyoGlu study enrolled men aged 40 to 65 years who were healthy but sedentary (having engaged in fewer than one exercise session per week in the previous year) ^29,62. These participants were divided into two groups based on their BMI and glucose tolerance: overweight (with an average BMI of 29.5 ± 2.3 kg/m²) and normal weight controls (with an average BMI of 23.6 ± 2.0 kg/m²). The overweight men had reduced glucose tolerance and/or insulin sensitivity (Supplementary Table 1). Both groups, consisting of 13 individuals each, underwent a 12-week regimen of combined strength and endurance training.

Exercise protocols

This 12-week training intervention included two weekly sessions of 60 minutes each for endurance cycling and two sessions of 60 minutes each for whole-body strength training. Prior to and after the 12-week exercise intervention, a 45-minute bicycle test at 70% of their maximum oxygen uptake (VO₂max) was conducted. Muscle (m. vastus lateralis) and subcutaneous white adipose tissue biopsies were taken 48 hours after the last exercise session, both before and after the 12-week intervention period ^29,62.

Clinical data

The euglycaemic hyperinsulinemic clamp was performed after an overnight fast ^29,62. A fixed dose of insulin 40 mU/m²·min^-1 was infused, and glucose (200 mg/mL) was infused to maintain euglycaemia (5.0 mmol/L) for 150 min. Insulin sensitivity is reported as glucose infusion rate (GIR) during the last 30 min relative to body weight. Whole blood glucose concentration was measured using a glucose oxidase method (YSI 2300, Yellow Springs, OH) and plasma glucose concentration was calculated as whole blood glucose x 1.119. Magnetic resonance imaging/spectroscopy (MRI/MRS) methods were used to quantify fat and lean mass. The ankle-to-neck MRI protocol included a 3D DIXON acquisition providing water and lipid quantification, data were then analysed using the nordicICE software package (NordicNeuroLab, Bergen, Norway), and the jMRUI workflow. VO₂max tests were performed after standardized warm-up at a workload similar to the final load of an incremental test in which the relationship between workload (watt) and oxygen uptake was established. Participants cycled for one min followed by a 15 watt increased workload every 30 s until exhaustion. Test success was based on O₂ consumption increased <0.5 mL·kg⁻¹·min⁻¹ over a 30 watt increase in workload, respiratory exchange ratio values >1.10, and blood lactate >7.0 mmol/L. We obtained scWAT, SkM biopsies and blood samples as described previously ²⁹. Biopsies were obtained from the periumbilical subcutaneous tissue and from m. vastus lateralis. After sterilization, a lidocaine based local anaesthetic was injected in the skin and sub cutis prior to both SkM and scWAT biopsies. Biopsies were dissected on a cold aluminium plate to remove blood etc. before freezing. For standard serum parameters, measurement were either conducted using standard in-house methods or outsourced to a commercial laboratory (Fürst Laboratories, Oslo, Norway).

The Olink proteomics explorer 3072 platform

We utilized antibody-based technology (Olink Proteomics AB, Uppsala, Sweden) to conduct profiling of serum samples through the Olink Explore 3072 panel. This PEA technique involves using pairs of DNA oligonucleotide-labeled antibodies to bind to the proteins of interest. When two matching antibodies attach to a target protein, the linked oligonucleotides hybridize and are extended by DNA polymerase, forming a unique DNA “barcode”. This barcode is then read using next-generation sequencing. The specificity and sensitivity of the PEA technology are notably high because only accurately matched DNA pairs generated detectable and measurable signals. To refine the dataset, proteins that were not detected or were duplicated were removed, resulting in an analysis of 2886 proteins. Only a single assay was conducted, eliminating inter-assay variability. Data are presented as normalized protein expression (NPX) units, which are logarithmically scaled using a log₂ transformation.

Proteomics validations

Duplicate measurements of IL6 and leptin in plasma were conducted using ELISA kits (Leptin; Camarillo, CA and IL6; R&D Systems, Minneapolis, MN) following the manufacturer’s instructions. The correlations between PEA or enzyme linked immunosorbent (ELISA) assays were r=0.94 (p=1.4×10^-11), and r=0.92 (p=2.2×10^-11) for IL6 and leptin, respectively (Supplementary Figure 1).

mRNA sequencing

Biopsies were frozen in liquid nitrogen, crushed to powder by a pestle in a liquid nitrogen-cooled mortar, transferred into 1 mL QIAzol Lysis Reagent (Qiagen, Hilden, Germany), and homogenized using TissueRuptor (Qiagen) at full speed for 15 sec, twice ^29,62. Total RNA was isolated from the homogenate using miRNeasy Mini Kit (Qiagen). RNA integrity and concentration were determined using Agilent RNA 6000 Nano Chips on a Bioanalyzer 2100 (Agilent Technologies Inc., Santa Clara, CA). RNA was converted to cDNA using High-Capacity cDNA Reverse Transcription Kit (Applied Biosystems, Foster, CA). The cDNA reaction mixture was diluted in water and cDNA equivalent of 25 ng RNA used for each sample. All muscle and scWAT samples were deep-sequenced using the Illumina HiSeq 2000 system with multiplex at the Norwegian Sequencing Centre, University of Oslo. Illumina HiSeq RTA (real-time analysis) v1.17.21.3 was used. Reads passing Illumina’s recommended parameters were demultiplexed using CASAVA v1.8.2. For prealignment quality checks, we used the software FastQC v0.10.1. The mean library size was ∼44 millions unstranded 51 bp single-ended reads for muscle tissue and ∼52 millions for scWAT with no differences between groups or time points. No batch effects were present. cDNA sequenced reads alignment was done using Tophat v2.0.8, Samtools v0.1.18, and Bowtie v2.1.0 with default settings against the UCSC hg19 annotated transcriptome and genome dated 14^th of May 2013. Post-alignment quality controls were performed using the Integrative Genome Viewer v2.3 and BED tools v2.19.1. Reads were counted using the intersection strict mode in HTSeq v0.6.1.

Statistics and bioinformatics

Olink data were analyzed using the ‘AnalyzeOlink’ R package for pre-processing, testing using mixed linear regression and annotation. Pathway and gene ontology overrepresentation analyses were performed using MSigDB data sets (Hallmark pathways and biological processes). mRNA sequencing data were normalized as reads per kilobase per million mapped read (RPKM) and analyzed using mixed linear regression from the ‘lme4’ R package. Normality was determined by quantile-quantile plots. P-values were corrected using the Benjamini-Hochberg approach set at a false discovery rate (FDR) of 5%. For univariate correlations, Pearsons’ or Spearmans’ method were applied as appropriate. Principal component analysis was performed using the ‘prcomp’ R package. Key driver analysis was performed using the ‘Mergeomics’ R package. Mediation analysis was performed using the ‘Mediation’ R package with 1000 bootstraps and the set.seed function to ensure reproducibility.

UK Biobank

The UK biobank is a large prospective population-based cohort containing ∼500,000 individuals (∼273,000 women), with a variety of phenotypic and genome-wide genetic data available ⁶³. The UK biobank has ethical approval from the North West Multi-Centre Research Ethics Committee (MREC), which covers the UK, and all participants provided written informed consent.

We utilized imputed genetic data from the October 2019 (version 3) release of the UK biobank for our analyses (Application ID: 53641). In addition to the quality control metrics performed centrally by the UK biobank⁶⁴, we defined a subset of unrelated “white European” individuals. We excluded those with putative sex chromosome aneuploidy, high heterozygosity or missing rate, or a mismatch between submitted and inferred sex as identified by the UK biobank (total N = 1932). We excluded individuals who we did not identify as ancestrally European using K-means clustering applied to the first four genetic principal components generated from the 1000 Genomes Project ⁶⁵. We also excluded individuals who had withdrawn their consent to participate in the study as of February 2021.

The Olink proteomics explorer 1536 platform

All analysis were done using the UK Biobank Olink data containing a total of 58699 samples and 54309 individuals, after excluding individuals as mentioned above we had 47747 samples with measured serum CD300LG levels. Data was generated according to Olink’s standard procedures.

Observational analyses

For the physical activity measurements we investigated if the degree of physical activity was associated with serum levels of protein (serum levels of protein regressed on physical activity), alternatively for the metabolic measurements we investigated if the protein expression affected the metabolic measurements (trait regressed on serum levels of protein), for both we used a linear regression model. We performed analyses stratified by sex and adjusting for age, protein batch, UK Biobank assessment centre, the time the sample was stored and BMI. All analyses were performed in R version 3.4.3.

Genome-wide association analysis

A GWAS of serum CD300LG levels (log₂ transformed) measured in the UK Biobank was performed using BOLT-LMM ⁶⁶ on individuals of European descent who had proteomic data available (N=45788). We included sex, year of birth, protein and genotyping batch, time from sample collection to processing time (in weeks) and five ancestry informative principal components as covariates in the analysis.

Post GWAS quality control included the removal of SNPs with MAF ≤0.05 and info score ≤0.4 (n_SNPs=6,945,819). The previously generated LD reference panel for clumping consisted of a random sample of 47674 unrelated British UK biobank individuals identified using GCTA ⁶⁷ with identity by state (IBS)<0.025 and identity by descent (IBD) sharing of <0.1. LD score regression analysis ⁶⁸ was used to investigate whether genomic inflation was likely due to polygenicity or population stratification/cryptic relatedness.

Prior to gene annotation palindromic SNPs were excluded (n_SNPs=6,882,889 remaining). Variants were classified as either cis- or trans-pQTLs based on SNP proximity to the protein-encoding gene (CD300LG) of interest. Variant annotation was performed using ANNOVAR ⁶⁹, labelling genes +/-500kb from variants. A pQTL was considered a cis-pQTL if the gene annotation in the 1Mb window matched the protein name, all remaining variants were considered trans-pQTLs.

To extract independent genome-wide significant pQTLs (p<5 × 10⁻⁸) clumping was performed using the PLINK v1.90b3.31 software package ⁷⁰; variants with r²>0.001 with the index SNP were removed using a 1 Mb window. Variants which lied within the human major histocompatibility complex (MHC) region were removed, excluding pQTLs on chromosome 6 from 26Mb to 34Mb.

Mendelian Randomization (MR)

To obtain valid instrumental variables (SNPs) for our analysis we assessed them against the three core assumptions for MR analysis: 1) That the SNPs were robustly associated with the exposure of interest. For that we obtained summary result statistics on genome-wide significant SNPs from our own GWAS. 2) That the SNPs were not associated with any known or unknown confounders. This is not an assumption that can be fully tested, however we used PhenoScanner ^71,72 to assess whether any SNPs were associated with known confounders (described below). 3) That the SNPs were not associated with the outcomes through any other path than through the exposure. To test this assumption, we searched PhenoScanner ^71,72 (detailed below) to see if our exposures of interest were associated with other potentially pleiotropic phenotypes.

MR statistical analysis

We used the TwoSampleMR package ⁷³ (https://github.com/MRCIEU/TwoSampleMR) in R version 4.2.2 (https://cran.r-project.org/). The outcome studies were obtained from https://www.magicinvestigators.org/ ⁴⁵ and were external to the UK biobank. Specificially, we used the outcomes “fasting glucose adjusted for BMI” (mmol/L, n=200622), “2-hour post OGTT glucose adjusted for BMI” (mmol/L, n=63396), “fasting insulin adjusted for BMI” (pmol/L, n=151013) and “HbA1c” (%, n=146806) ⁴⁵.

We performed a two-sample inverse variance weighted (IVW) analysis to assess the causal effect of CD300LG on metabolic factors (Supplementary Table 8 and 9). To explore potential violations of the core assumptions when using the full set of SNPs, we performed a heterogeneity test using Cochran’s Q, and a test for directional pleiotropy was conducted by assessing the degree to which the MR Egger intercept differed from zero ⁷⁴. We also performed additional sensitivity analyses using MR Egger regression ⁷⁴, weighted median ⁷⁵, simple and weighted mode estimation methods ⁷⁶. Effect estimates from the different sensitivity analysis were compared as a way of assessing the robustness of the results. To assess potential heterogeneity in the MR estimates we further performed MR-PRESSO ^45,77 to detect (MR-PRESSO global test) and correct for horizontal pleiotropy via outlier removal (MR-PRESSO outlier test).

Investigation of potentially pleiotropic SNPs

SNPs robustly associated with the exposure investigated in the MR analyses (serum CD300LG levels) were checked for other possible associations (PhenoScanner v2 ^71,72, http://www.phenoscanner.medschl.cam.ac.uk/) which may contribute to a pleiotropic effect on the metabolic outcomes. Supplementary Table 10 lists the SNPs used in our analysis that many influence related phenotypes. Phenotypes from PhenoScanner were listed if they were associated with the SNPs or nearby variants in high LD (r²=0.8) at p-value level <1 x 10^-5 and could have potential pleiotropic effects in the analysis.

Data availability

mRNA sequencing data from MyoGlu can be found at https://exchmdpmg.medsch.ucla.edu/app/ as well as in GEO:GSE227419. Secretory proteins are available in the MetazSecKB data base at http://proteomics.ysu.edu/secretomes/animal/. The human serum proteomic NAFLD signature is available in the study of Govaere et al ³⁸. Expression profiles in human liver cells are available in the Human Liver Cell Atlas ³⁹. Data obtained from the UK biobank (Olink explore 1536 and measures of physical activity ⁷⁸ can be found at https://biobank.ndph.ox.ac.uk/ukb/. Glucometabolic outcomes used in MR analyses are available at: https://www.magicinvestigators.org/ ⁴⁵. Data from the GD-CAT database is available from: https://pipeline.biochem.uci.edu/gtex/demo2/. Mice exercise data are available at https://motrpac-data.org/ and knock-out data at https://www.mousephenotype.org/. CD300LG expression values from a human tissue panel were obtained from Uhlén et al. ⁴². The single nuclei mRNA sequencing data from human adipose tissue was plotted in Seurat v. 4 by downloading processed data from the Single Cell Portal ⁴³. The data can also be explored at: https://singlecell.broadinstitute.org/single_cell/study/SCP1376/a-single-cell-atlas-of-human-and-mouse-white-adipose-tissue). UK Biobank (https://www.ukbiobank.ac.uk/) data are available to researchers upon application to the individual cohorts via their websites. All other data used are publicly available and referenced according in the main text. For additional details and data inquiries, please contact Sindre Lee-Ødegård.

Data Availability

mRNA sequencing data from MyoGlu can be found at https://exchmdpmg.medsch.ucla.edu/app/ as well as in GEO:GSE227419. Secretory proteins are available in the MetazSecKB data base at http://proteomics.ysu.edu/secretomes/animal/. The human serum proteomic NAFLD signature is available in the study of Govaere et al 38. Expression profiles in human liver cells are available in the Human Liver Cell Atlas 39. Data obtained from the UK biobank (Olink explore 1536 and measures of physical activity 78 can be found at https://biobank.ndph.ox.ac.uk/ukb/. Glucometabolic outcomes used in MR analyses are available at: https://www.magicinvestigators.org/ 45. Data from the GD-CAT database is available from: https://pipeline.biochem.uci.edu/gtex/demo2/. Mice exercise data are available at https://motrpac-data.org/ and knock-out data at https://www.mousephenotype.org/. CD300LG expression values from a human tissue panel were obtained from Uhlen et al. 42. The single nuclei mRNA sequencing data from human adipose tissue was plotted in Seurat v. 4 by downloading processed data from the Single Cell Portal 43. The data can also be explored at: https://singlecell.broadinstitute.org/single_cell/study/SCP1376/a-single-cell-atlas-of-human-and-mouse-white-adipose-tissue). UK Biobank (https://www.ukbiobank.ac.uk/) data are available to researchers upon application to the individual cohorts via their websites. All other data used are publicly available and referenced according in the main text. For additional details and data inquiries, please contact SL.

Acknowledgements

South-Eastern Norway Regional Health Authority, Simon Fougners fund, Diabetesforbundet, Johan Selmer Kvanes’ legat til forskning og bekjempelse av sukkersyke. This research has been conducted using the UK Biobank resource (Reference 53641). DME is funded by an Australian National Health and Medical Research Council Investigator Grant (APP2017942). GHM is the recipient of an Australian Research Council Discovery Early Career Award (Project number: DE220101226) funded by the Australian Government and supported by the Research Council of Norway (Project grant: 325640 & Mobility grant: 287198).

Author contributions

Conceptualization: SLØ, CAD, KIB. Methodology: SLØ, GHM, ED, DME. Data Collection: SLØ, FN. Data analysis: SLØ, TO, MH, GHM, ED. Visualization: SLØ, GHM and MH. Writing original draft: SLØ. Writing, review and editing: SLØ, MH, TO, GHM, ED, DME, JKV, HLG, FN, CAD, KIB. Supervision: CAD, KIB. Funding acquisition: SLØ, CAD, KIB. Project administration: KIB.

Competing interests

The authors declare no conflict of interest.