Responsiveness to high intensity interval training depends on insulin sensitivity and protein content and composition of small extracellular vesicles

High intensity interval training (HIIT) improves cardiorespiratory tness (VO2max), but its impact on metabolism remains unclear. We hypothesized that 12-week HIIT improves insulin sensitivity in people with or without type 2 diabetes (T2D, NDM). However, despite identically improved VO2max, mainly insulin-resistant persons (T2D, IR NDM) responded with improved insulin sensitivity and circulating small extracellular vesicles (SEV), along with reduced myocellular protein kinase Cε activity (T2D) or inammation (IR NDM). These changes related to the SEV proteome, characterized by downregulated phospholipase C pathway (T2D) and upregulated antioxidant capacity (IR NDM). Thus, SEV cargo likely contributes to modulating exercise responsiveness in humans. combination with the respective horseradish peroxidase (HRP)-conjugated secondary anti-rabbit antibody, diluted 1:2500, or anti-mouse diluted 1:1000. The membranes were nally coated with Immobilon Western Chemiluminescent HRP Substrate (Millipore) and the proteins were detected using a Bio-Rad ChemiDocTM MP Imaging System in combination with the software ImageLab 6.0.1 (Bio-Rad 199 Laboratories) for densitometric analysis. Primary antibodies were purchased from Cell Signaling Technology: LC3 (4108); p38 MAPK (9212); p44/p42 MAPK (9102); NF-κB (8242); AMPKα (2532); phospho-AMPKα(Thr172) (2535); NQO1 (3187); phospho-IRS1(Ser307) (2381); phospho-IRS1(Ser1101) (2385); GAPDH (glyceraldehyde 3-phosphate dehydrogenase) (2118) as housekeeping protein for the soluble and cytosolic fractions. p62 (610833), PKCθ (610090) and PKCε (610086) were obtained from BD Biosciences; IRS1 from Millipore (06-248), NRF2 from Santa Cruz Biotechnology (sc-365949) and Na + /K + -ATPase, used as loading control for the membrane fraction, from Abcam (Ab76020). Data are expressed in arbitrary units and normalized to housekeeping protein.


Introduction
Regular exercise training not only reduces cardiovascular risk, but also helps to prevent and treat type 2 diabetes (T2D) 1,2 . However, up to 20% of all T2D participants in exercise interventions fail to respond to physical training with improved glucose metabolism 3 . Even rst-degree relatives of persons with T2D do not necessarily increase their insulin sensitivity despite increased muscle ATP synthase ux after 6month exercise training 4 . Among other factors, exercise volume and intensity predict exercise responsiveness 5 . High intensity interval training (HIIT) represents a time-saving highly e cient alternative to moderate training modalities and may exert superior bene cial effects 6 , such as improved insulin sensitivity and mitochondrial function in elderly people 7 . However, 9-days HIIT failed to improve insulin sensitivity in insulin-resistant offspring of T2D despite increased mitochondrial function 8 . It remains to be determined whether longer term HIIT ameliorates mitochondrial changes and insulin sensitivity in insulin-resistant people with or without T2D.
Metabolic effects of exercise result from inter-tissue communication via metabolites, hormones, myokines/exerkines, microRNA or medium-sized extracellular vesicles [9][10][11][12] . Recent studies suggest that one exercise bout also leads to the release of small  nm) extracellular vesicles (SEV) by skeletal muscle in healthy volunteers 13,14 . However, the role of SEV in insulin-resistant humans and their impact on the metabolic response to a supervised exercise program per se, independent of diet or body weight changes, is not known. SEV can shuttle their functional cargo to target tissue to activate cellular signaling 15 , but it is also unclear whether and if so, which SEV cargo relates to changes in insulin sensitivity and underlying cellular pathways 16 .
Thus, this study examined the effect of a supervised 12-week HIIT on metabolic features and SEV release in sedentary insulin-sensitive (IS NDM) and insulin-resistant (IR NDM) glucose tolerant as well as T2D individuals. We originally hypothesized that HIIT would improve insulin sensitivity independently of glucose tolerance. Surprisingly, the responders to HIIT, as de ned by improved peripheral insulin sensitivity (M-value), were mainly present among the insulin-resistant groups. This study therefore investigated next whether quantity and proteome of SEV help to differentiate between responders (T2D-R and IR-R from the T2D and IR NDM groups) and non-responders (IS-NR from the IS NDM group). SEV were isolated by size exclusion chromatography (SEC), allowing for subsequent downstream analysis 17 , and characterized by nanoparticle tracking analysis and mass spectrometry for protein identi cation and quanti cation. Finally, expression of selected SEV candidates was validated in skeletal muscle biopsies.
This study revealed that HIIT indeed differently affects SEV concentration and their proteome depending on baseline insulin sensitivity, indicating that SEV cargo can modulate the metabolic responsiveness to exercise training.
De ned by a cutoff-value of 4.9 mg*kg -1 *min -1 for the M-value 18 , which re ects insulin-stimulated skeletal muscle glucose uptake 16 , 11 healthy persons were insulin-resistant (IR NDM) and 12 insulin-sensitive (IS NDM). Cardiorespiratory tness (VO 2 max) was comparable between the three groups ( Fig. 1a). M-value was lower in T2D and IR NDM than IS NDM (p<0.001, Fig. 1b). Insulin-mediated suppression of endogenous glucose production (iEGP), re ecting hepatic insulin sensitivity was lower (p<0.001, Fig. 1c), whereas liver fat content was higher in T2D than in both NDM groups (p<0.001, Fig. 1d). As expected, T2D also had higher HbA1c and lower HDL-cholesterol than both NDM groups. Compared to IS NDM, T2D had higher body mass index (BMI), partly due to due to higher visceral fat mass (Table 1), whereas the higher BMI of IR NDM resulted from both higher subcutaneous and visceral fat mass.
HIIT uniformly improves whole body oxidative and skeletal muscle mitochondrial capacity After 12-week of HIIT, all participants uniformly increased their VO 2 max (Fig. 1a) despite unchanged body weight and whole-body fat content (Table 1). Maximal uncoupled respiration, as assessed ex vivo from permeabilized skeletal muscle bers tended to be lower in T2D than in IS NDM at baseline (p=0.06), but rose by 28-45% in all groups after HIIT (Fig. 1e). HIIT neither affected leak control ratio (LCR), re ecting uncoupling at constant electron transport capacity, nor mitochondrial e ciency, assessed from respiratory control ratio (RCR), or reactive oxygen species (ROS) emission in any group (Fig. 1f,g). Muscle citrate synthase activity (CSA), as a surrogate marker of mitochondrial content 19 , at least doubled after HIIT in all groups without any difference between groups (Fig. 1h).
HIIT differently affects myocellular pathways of insulin sensitivity in the insulin-resistant responders (T2D-R, IR-R) To further examine the heterogeneous responses of the peripheral insulin sensitivity (M-value) within the groups, we employed skeletal muscle biopsies from subgroups of insulin-resistant responders, (T2D-R, N=16; IR-R, N=9) and insulin-sensitive non-responders (IS-NR, N=7) (Fig. 2a). We primarily addressed key mechanisms known to underlie human myocellular insulin resistance, such as abnormal mitochondrial function and inhibitory lipid signaling via activation of the diacylglycerol/novel protein kinase C isoform (DAG/nPKC) pathway and/or inhibitory in ammatory pathways 16,20 .
Similar to the whole groups, HIIT increased mitochondrial function (Fig. 2b), but did not affect ROS in all subgroups, although ROS was lower in IR-R than in T2D-R (Fig. 2c).
Activation of the nPKCε and θ isoforms was assessed from translocation of the respective proteins to the myocellular membrane (Fig. S2a). PKCε activity negatively associated with M-value (log transformed; β=-0.55; p=0.002) at baseline, but not after HIIT. At baseline, PKCε activity was doubled in T2D-R compared to IR-R and IS-NR (Fig. 2d). After HIIT, T2D-R exhibited 50% and 40% decreases in PKCε and θ activities, respectively (Fig. 2d,e). Additional measurement of the nPKC activities in the whole groups of T2D, IR NDM and IS NDM, showed an identical pattern (Fig. S2b,c), indicating that the improvement in insulin resistance by HIIT is mediated by the nPKC pathway, at least in overt T2D.
In ammatory pathways, speci cally nuclear factor kappa-light-chain-enhancer of activated B cells (NF-κB), not only relate to insulin resistance and oxidative stress, but may be also downregulated by chronic exercising 21,22 . Thus, we also analyzed muscle NF-κB protein levels in all groups, which did not differ between responders and non-responders at baseline, but were reduced by 24% after HIIT exclusively in IR-R (Fig. 2f). This exercise effect was con rmed in the whole groups of T2D, IR NDM and IS NDM, although baseline NF-κB levels were 40% higher in T2D (Fig. S2d). However, circulating pro-and anti-in ammatory cytokines (interleukin-1β, -6, -15, -1ra, tumor necrosis factor (TNF)α) were not signi cantly different between groups and after HIIT (data not shown). Collectively, these data suggest that the metabolic response to HIIT is at least partly mediated by reduction of nPKC activity in responders with diabetes (T2D-R) and by lower activation of NF-κB-dependent pathways in responders without diabetes (IR-R).
HIIT induces an increase in circulating SEV in the insulin-resistant responders (T2D-R, IR-R) To clarify the role of SEV in exercise response, we measured the effect of HIIT on SEV release in representative subgroups of T2D-R (N=8), IR-R (N=8) and IS-NR (N=6). These subgroups showed overall similar changes as the whole cohort, i. e. increased VO 2 max in all, but increased peripheral and hepatic insulin sensitivity only in T2D-R and IR-R (Fig. S3a,b,c). Mitochondrial function increased in responders and was nominally higher compared to baseline in IS-NR (Fig. S3d), whereas ROS production remained unchanged in all three subgroups. Also, only T2D-R showed reduced PKCε activation (Fig. S3e), while only IR-R had lower NF-κB levels (Fig. S3f).
We measured the size and the number of circulating SEV in serum of responders and non-responders, collected at baseline and 72 h after the last bout of the 12-week HIIT. The purity of the SEV preparations was validated by nanoparticle tracking analysis (NTA) (Fig. 3a). The mean diameter of the isolated SEV was about 100 nm with a peak at about 80 nm and comparable between groups before and after HIIT (Fig. 3b). Interestingly, the estimated SEV concentration (number of circulating SEV per protein) was higher in IS-NR compared to T2D-R and numerically higher than in IR-R at baseline (Fig. 3). After HIIT, the SEV concentration rose only in the insulin-resistant groups, T2D-R and IR-R (Fig. 3d), with an increase of 52% (expressed as log2 fold change, FC) for both groups (Fig. S4a).
As HIIT increased the number of circulating SEV only in the insulin-resistant responders, we hypothesized that SEV and their cargo could contribute to the different cellular metabolic adaptations to this mode of chronic exercise training. We therefore characterized the proteome of the SEV isolated from T2D-R (N=5), IR-R (N=5) and IS-NR (N=5) individuals at baseline and after HIIT and identi ed a total of 1707 proteins including the 24 exosomal-enriched proteins (Table S1). Furthermore, 1589 of the identi ed proteins (98%) overlapped with the human exosome-associated proteins, previously identi ed in the Vesiclepedia database (13550 unique protein entries), whereas 39 were newly discovered as SEV-carried proteins (Fig.   4a, Table S2). Among the total SEV-associated proteins, we then selected the candidates with low variability (q value <0.05) and high degree of regulation (absolute log FC between groups or baseline vs HIIT >0.5) and we found that the proteomic pro le differed between groups at baseline (Supplementary information) and after HIIT.
HIIT affects the proteomic pro le of SEV Quantitative proteomic analysis revealed that HIIT regulates the expression of 262 SEV proteins (n=122 in T2D-R, n=130 in IR-R, n=89 in IS-NR), of which 102 were downregulated and 160 upregulated (Fig. 4b, Table S3). Among the regulated SEV proteins, we identi ed proteins typically associated with exosomes 23 , such as biogenesis markers (ALIX), signaling proteins (GTPase, Ras-related protein), proteins associated with membrane tra cking and fusion (Rab proteins, annexins), lipid rafts ( otilin), cytoskeleton components (moesin, tubulin) and cell adhesion molecules (integrins). Moreover, the cellular component (CC) enrichment analysis of regulated SEV proteins con rmed a signi cant enrichment of proteins associated with extracellular exosomes (Fig. 4c). Of note, 13 of the 262 SEV proteins, differentially expressed after HIIT, were shared between all groups (Fig. 4d, Table S3), including antithrombin III, kininogen I, histidine-rich glycoprotein and α1-antitrypsin, which relate to in ammatory and immune responses 24 . Moreover, 29 SEV proteins were up-or down-regulated after HIIT exclusively in responders, but not in IS-NR (Fig. 4d, Table S3). Among these proteins, brinogen α, b and g chains (FGA, FGB, FGG) were similarly upregulated in SEV isolated from T2D-R and IR-R after HIIT. These acute-phase proteins are not only associated with insulin resistance and acutely increased by insulin in T2D 25 , but have been also described as myokine candidates carried by EV and released from the exercising limb after recovery 14 , suggesting that exercise might activate insulin sensitizing pathways.
Proteomic pro ling of SEV suggests a new mode of myokine release for the metabolic adaptation to exercising Since 12-week HIIT triggered the release of SEV, we assumed that SEV might represent an alternative to the release of biologically active proteins by classical secretory pathway. Indeed, we found that only 34% (89/262) of the proteins differentially expressed after HIIT had a predicted secretory signal peptide (SP) and only 12% (32/262) were predicted to follow a non-classical secretion pathway, based on the bioinformatics tools SignalP and SecretomeP (Fig. 4e, Table S3). These ndings suggest that a large number of proteins can enter the circulation within SEV and that SEV can thereby contribute to inter-organ communication affecting cellular metabolism. In order to examine whether HIIT-induced increase in circulating SEV -at least partially -originated from skeletal muscle, we performed another quantitative proteomic analysis of SEV released from primary human skeletal muscle cells (hSkMC) using electrical pulse stimulation (EPS) to simulate the in vivo exercise intervention under in vitro conditions (Fig. 4f, Table S4). Interestingly, 44 SEV proteins regulated after HIIT were also present in SEV collected from the media of EPS-trained hSkMC, suggesting that these SEV proteins could represent novel myokines. Some of the SEV proteins upregulated, both after HIIT in vivo and after EPS in vitro, are indeed involved in signal transduction (GO:0007165; Ras-related protein Rap-2b and 1b), in oxidation-reduction processes (GO:0055114; aldehyde dehydrogenase family 16 member A1) and in cellular response to oxidative stress (GO:0034599; protein/nucleic acid deglycase DJ1,peroxiredoxin-2).
HIIT enriches SEV with proteins related to insulin sensitivity in T2D-R and in ammation and oxidative metabolism in IR-R The 262 SEV proteins differentially regulated by HIIT were subsequently subjected to gene ontology (GO) analysis of biological processes (BP) and molecular function (MF) and to Ingenuity Pathway Analysis (IPA), as described in the supplementary information. In order to assess whether SEV protein cargo is responsible for the different metabolic adaptations induced by exercise in responders and nonresponders, we performed a functional analysis of the exercise-regulated SEV proteins in each group (T2D-R, IR-R and IS-NR) and found different enriched GO-terms and pathways (Table S5).
In T2D-R, functional analysis revealed a signi cant enrichment of proteins linked to glycolytic process and the inhibition of glycolysis (z-score -2), PLC (z-score -2), mitogen-activated protein kinase (ERK/MAPK) (z-score -2.236) and protein kinase A (PKA) signaling (z-score -2.249), since the proteins associated to these pathways were downregulated after HIIT (glycolysis: ENO1 In IR-R, we found an overrepresentation of the non-canonical NF-κB and TNF-mediated pathways with upregulation of the 20S core proteasome complex (PSMA1, PSMA3, PSMA5, PSMA6, PSMA7, PSMB1, PSMB5, PSMB8) 26,27 as well as the process "cellular response to oxidative stress" with an upregulation of proteins belonging to the cellular antioxidant system (CAT, CCS, G6PD, NME2, PRDX1, PRDX2, SOD2) (Fig. 5b). In line, IPA predicted the overrepresentation of pentose phosphate pathway, with an upregulation of glucose-6-phosphate dehydrogenase (G6PD), the rate-controlling enzyme of this pathway 28 , and the activation of the upstream molecule NFE2L2 (z-score 2.607) and the NRF2-mediated oxidative stress response (z-score 2) since the downstream targets of NRF2 (CAT, PRDX1, PRKCA, RAB1B, SOD2, AKR7A2) were upregulated after HIIT. IPA revealed also IL15 as activated upstream regulator (zscore 2.224), probably stimulated by the antioxidant system 29 . Finally, in the IR groups we found an enrichment of proteins associated to "response to calcium ion", extending the concept of the involvement of Ca 2+ in EV release 30 (Fig. 5a,b).
In IS-NR, we found a "lipid metabolism and transport" signature ( Fig. 5c) and IPA revealed "autophagy of cells" as a cellular function associated only with SEV derived from IS-NR after 12-week HIIT (B-H p-value <0.05; z-score 1.432).
Taken together, these results indicate that HIIT (i) stimulates the insulin downstream pathway by inhibiting the MAPK, PLC and PKA signaling in the SEV from T2D-R and (ii) enhances the antioxidant system in the SEV released by IR-R, which eventually leads to improved peripheral insulin sensitivity, while (iii) it does not affect any pathway in SEV from IS-NR.
SEV cargoed proteins released after HIIT may directly affect cellular pathways in skeletal muscle Finally, for investigating whether the speci c SEV proteins also affect molecular pathways in skeletal muscle, we measured selected candidates related to insulin signaling for validation in muscle biopsies of T2D-R, IR-R and IS-NR individuals. Activating Thr172 phosphorylation of AMP activated protein kinase (AMPK) was higher in T2D-R than in IR-R at baseline and after HIIT (Fig. 6a). Inhibitory Ser307phosphorylation of insulin receptor substrate 1 (IRS1) was numerically lower in IR-R after HIIT (p=0.13), whereas Ser1101-phosphorylation of IRS1 was not different between the groups (Fig. 6b,c). In addition, NRF2 and its downstream target NADPH quinone dehydrogenase 1 (NQO1) levels were lower at baseline in IR-R compared to T2D-R and tended to rise only in IR-R (Fig. 6d,e). After HIIT, also expression of both p38 mitogen-activated protein kinase (MAPK) and p44/p42 MAPK, as in ammation mediators target of TNFα 31 , were lower in IR-R and in IS-NR (Fig. 6f,g). Expression levels of microtubule-associated proteins 1A/1B light chain 3B (LC3) and ubiquitin-binding protein p62 (p62) was increased after HIIT only in IS-NR (Fig. 6h,i).

Discussion
This study found that (i) the majority of insulin-resistant, but not insulin-sensitive humans respond to HIIT with improved insulin sensitivity, (ii) the myocellular mechanism underlying the improvement differs between nondiabetic and diabetic responders (NF-κB vs. nPKC), (iii) only the responders also increase their circulating SEV and that (iv) the SEV proteome composition is differently affected between nondiabetic and diabetic responders (upregulated NRF2 vs. downregulated PLC, PKA and ERK pathways), probably mediating changes in muscle metabolism.
First, this study showed that HIIT similarly improves whole body and muscle maximal oxygen uptake and muscle mitochondrial mass, at least when assessed from CSA, in metabolically healthy as well as in insulin-resistant groups (IR NDM, T2D). This is in line with the exercise-induced increases in PGC1-α, mitochondrial density, oxidative capacity and mitochondrial complex proteins reported for glucosetolerant and T2D individuals 32,33 . Exhaustive exercising may not only increase oxidative capacity, but also cause oxidative stress with bene cial effects on mitochondrial adaptation and insulin sensitivity 34 . In the present study, HIIT did not induce oxidative stress in any group, but led to improved peripheral and hepatic insulin sensitivity in the insulin-resistant groups. Previous HIIT studies reported heterogeneous results in individuals with metabolic diseases albeit mostly using less accurate methods than the hyperinsulinemic-euglycemic clamp to assess insulin sensitivity 35 . In contrast, a study implying a twostep clamp protocol showed that HIIT improved peripheral insulin sensitivity but not EGP in young and elderly people with similar baseline insulin sensitivity 7 . Of note, one 8-week HIIT study found improved HOMA-IR in insulin-resistant humans 36 and a 6-week intensive exercise intervention led to greater insulinstimulated muscle glucose transport/phosphorylation in insulin-resistant offspring of parents with T2D than in glucose tolerant humans 37 . These ndings already indicate an ability of insulin-resistant individuals to (over)compensate for insulin resistance by adequate exercise training.
Of note, it has been estimated that 20% of T2D are non-responders to exercise 3 and 31% do not improve their metabolic control and insulin sensitivity after hypocaloric dietary intervention 38 . Genetic predisposition could play an important role regarding response to lifestyle interventions, as reported in several studies 39,40 . For example, non-response among glucose-tolerant relatives of parents with T2D has been attributed to a polymorphism in the NDUFB6 gene, which encodes a subunit of mitochondrial complex I 41 , supporting the concept of a tight association between mitochondrial function and changes in insulin sensitivity in T2D patients 33 . Of note, the present study found that mainly insulin-resistant individuals responded to HIIT with improved insulin resistance independently of muscle mitochondrial function. In this context, another study reported that non-responders to exercise training also had higher baseline insulin sensitivity along with different DNA methylation status and RNA expression patterns 42 .
Interestingly, the present study showed that the improvement of insulin resistance upon HIIT associated with different pathways in T2D and IR NDM. More speci cally, in T2D -but not in IR NDM-we found a reduction of nPKCε and θ activities. This indicates that DAG/nPKC pathway, which is well known to mediate lipid-induced insulin resistance 16,43 , is downregulated by HIIT at least in overt diabetes. In line, HIIT also led to reductions in liver fat content paralleled with improved hepatic insulin sensitivity in both T2D and IR NDM and these changes were in contrast to divergent effects of HIIT in non-alcoholic fatty liver disease (NAFLD) 44 .
The T2D group also showed activation of the NF-κB pathway, which is another feature of long-standing obesity and T2D 31 . HIIT reduced NF-κB protein expression in IR NDM, but not in T2D, indicating the operation of different mechanisms of metabolic exercise response in individuals with or without diabetes. One might suggest that improvement of the more severe insulin resistance requires reduction of the lipotoxic pathways in skeletal muscle. Of note, the reduction in myocellular in ammatory pathways in IR NDM was not accompanied by any changes in circulating in ammatory markers, suggesting that lowgrade (subclinical) in ammation may not be the primary inter-organ crosstalk mechanism explaining HIIT-induced metabolic changes.
To elucidate other mechanisms of inter-organ crosstalk mediating the metabolic effects of HIIT 12 , we isolated circulating SEV and found that their number was already lower in insulin-resistant responders than in BMI-matched insulin-sensitive non-responders before exercise training. This is in contrast to a recent study reporting higher number of EV in T2D than lean healthy humans 45 , which may be explained by the differences in BMI between the groups, since circulating EV are signi cantly increased in obesity 46 . In addition, the previous study reported an increase of total (large and small vesicles) and large EV, but not speci cally of SEV due to a different method of vesicle isolation.
Previous studies have only reported on the acute release of EV upon a single exercise bout in healthy mice and humans [12][13][14] . This study shows that 12-week HIIT induces secretion of SEV in insulin-resistant humans responding with improved insulin resistance (IR-R and T2D-R). Also, several SEV proteins regulated after HIIT were found also in exosomes released from hSkMC after EPS in vitro, supporting the concept that skeletal muscle is likely the major contributor to exercise-induced SEV release. The large skeletal muscle mass, its central involvement in exercise responses and high secretory activity in terms of myokines may serve to support this contention 47 .
Analysis of SEV proteins differentially expressed after HIIT in T2D-R revealed inactivation of PLC, PKA and ERK/MAPK signaling, which can be translated into a functional IRS-PI3K/Akt-GLUT4 signaling pathway and improved insulin sensitivity 16,48 . Indeed, inactivation of PLC and PKA in SEV might serve as link to the reduced nPKC activity and the upregulated AMPKα activity observed in skeletal muscle of T2D-R. Although we did not verify pathways in other target tissues, it is conceivable that SEV mediate metabolic improvements also in other organs, such as liver, and contribute with its cargo to improved hepatic insulin sensitivity observed in T2D-R via inhibition of PKCε and activation of AMPKα, as previously described in PKCε knock-out mice 49 and primary hepatocytes treated with metformin 50 .
In IR-R, the SEV proteome after HIIT displayed an upregulation of the 20S core proteasome complex of the non-canonical NF-κB and TNF-mediated pathways, which reduce in ammation via degradation of NF-κB and transcriptional termination of target genes 26,27 , and activation of the NRF2-mediated oxidative stress response, which potentiate antioxidative response in target tissues. Of note, skeletal muscle of IR-R showed reduced NF-κB, p38 and p44/p42 MAPK proteins and upregulation of NRF2 and NQO1 proteins, which decrease the inhibitory Ser307-phosphorylation of IRS1 and protect against insulin resistance 16,51 .
SEV proteins might reduce activation of NF-κB also in liver and thereby reduce hepatic in ammation, insulin resistance and fat content 52 , as observed in IR-R after HIIT.
Finally, SEV proteome isolated from IS-NR did not reveal any signi cant inhibition or activation of pathways in line with the non-responder status, i. e. the absence of changes in insulin sensitivity. However, expression of the autophagy markers, LC3 and p62, was markedly increased after 12-week HIIT in their skeletal muscle, which may re ect enhanced endurance performance 53 .
Taken together, this study shows that (i) even highly e cient HIIT exercise training does not generally improve peripheral insulin sensitivity, but allows to identify responders, who are more likely insulinresistant humans at baseline, (ii) exercise response of insulin sensitivity relates to different mechanisms among individuals with or without T2D, i. e. reduction of myocellular lipotoxic signaling via DAG/nPKC or downregulation of NF-κB pathways and that (iii) different amount and proteome of circulating SEV likely contribute to the variation in metabolic exercise response. These ndings might help to better understand mechanisms of lifestyle intervention and to identify novel targets for the tailored prevention and treatment of insulin resistance and T2D in the concept of precision medicine.

Study participants and study design
This study included 20 male patients with type 2 diabetes (T2D) as well as 12 age-matched insulinsensitive (IS) and 11 insulin-resistant (IR) glucose tolerant male participants (NDM). All participants underwent a screening procedure with detailed physical examination, interview, blood sampling and spiroergometry. Exclusion criteria were performance of more than 60 minutes of endurance training per week, acute or chronic cardiovascular, renal or liver diseases, use of insulin-sensitizing medication or beta-blockers, alcohol intake of more than 30 g per day and smoking. Volunteers of the control group needed to have neither any family history of T2D nor dysglycemia during a standardized 75-g oral glucose tolerance test. All participants gave written informed consent prior to inclusion in the study, which was approved by the institutional review board of Heinrich-Heine University Düsseldorf (NCT02039934) and performed according to the World's Medical Association Declaration of Helsinki.
Exercise training protocol and diet control All participants took part in a progressive 12-week supervised cycle ergometer training protocol of a total duration of 35 minutes per session consisting of four 4-minute high-intensity intervals (HIIT), during which the participant trained at an intensity corresponding to 90% of their maximal heart rate as determined during a baseline spiroergometry. These intervals were interspersed by three 3-minute intervals, during which the participant trained at 70% of their maximal heart rate. Exercise training sessions were repeated three times weekly on non-consecutive days. Over the 12-week training period, workload was progressively adjusted to maintain the target heart rate. All participants were instructed to maintain their body weight (change up to 5% of initial body weight was tolerated) throughout the training period. In turn, body weight was checked every 4 weeks and participants presenting with a 3% change of their initial body weight received adequate dietary counselling by our expert clinical nutritionists.

Spiroergometry
Each participant performed an incremental exhaustive exercise test on an electronically braked cycle ergometer (Ergoline ergometrix 900, Bitz, Germany) at 60 revolutions/min 54 . Respiratory gas exchange measurements were determined by open-air spirometry (Masterscreen CPX; Jäger/Viasys, Hoechberg, Germany) 55 . During exercising, work rate was continually increased by 15 W/min increments, and the incremental part of the test lasted 8-12 minutes. Blood pressure, heart rate and a 12-lead electrocardiogram (ECG) were recorded every 2 minutes during the test. Capillary blood was drawn from the ear every 2 minutes and then every minute after the anaerobic threshold for the measurement of lactate.

Hyperinsulinemic-euglycemic clamp tests
Two-step hyperinsulinemic-euglycemic clamp tests were performed for assessment of peripheral and hepatic insulin sensitivity before intervention and 72 h after the last training session. Patients with T2D did not take their oral glucose lowering medication for three days before the clamp to avoid any interference from metabolic drug effects. The clamp test were performed as previously described in detail 56  Liver fat content was quanti ed by volume selective proton magnetic resonance spectroscopy ( 1 H-MRS) using a stimulation echo acquisition mode in a whole-body 3-Tesla magnetic resonance scanner, as described 57 . Whole-body, subcutaneous and visceral fat mass were quanti ed by application of a T1weighted axial fast spin echo 54 .

High resolution respirometry
Mitochondrial respiration was assessed in permeabilized muscle bers from skeletal muscle biopsies 43 . Leak respiration (LEAK) after addition of glutamate and malate, state 3 respiration at saturating ADP concentrations (OXPHOS), LEAK state induced by inhibition of ATP synthase by oligomycin (LEAKomy) and electron transport (ET) capacity to obtain maximum oxygen ux at optimum concentration of the uncoupler carbonyl cyanide-4-(tri uoromethoxy)phenylhydrazone (FCCP) were assessed in a 2-chamber Oxygraph-2k (OROBOROS Instruments, Innsbruck, Austria). Respiratory control ratio (RCR) was calculated as the ratio of OXPHOS/LEAKomy, whereas leak control ratio (LCR) was calculated from the ratio of LEAK/ET capacity.

Skeletal muscle biopsy
At baseline as well as after 12-week HIIT, a skeletal muscle biopsy was performed on the day of clamp procedure, directly before initiation of insulin infusion. The region above the muscle vastus lateralis was anesthetized using local anesthetics (Lidocain 2%) and 100-500 mg of skeletal muscle tissue was obtained using a Bergstrom needle 43  and M41]) were obtained to generate hSkMC (PromoCell, Heidelberg Germany) and cultured as described 60 . For an individual experiment, myoblasts were seeded in six-well culture dishes and cultured to 90% of con uence in α-modi ed Eagle's medium (αMEM)/Ham's F-12 medium (Gibco, Berlin Germany) containing supplement for skeletal muscle cell growth medium (PromoCell). The cells were then differentiated in αMEM supplemented with 2% horse serum for 5 days, followed by overnight starvation in αMEM without serum. Differentiated cells were subjected to electrical pulse stimulation (EPS) cells 60 and after 24 h of EPS the SEV were isolated from conditioned medium as reported 61 .

Small extracellular vesicles (SEV) isolation from serum
Serum samples were collected before the intervention and 72 h after the last training. Small extracellular vesicles were isolated from serum by size exclusion chromatography, using the IZON columns (qEV2/70 nm) coupled to successive ultracentrifugation 62 . Brie y, the serum samples (starting material 750 µl) were rst cleared at 1500 g for 10 min followed by centrifugation at 10000 g for 10 min to remove particulate matter. The serum supernatant was then loaded into an IZON column previously equilibrated with PBS. Once the serum entered the column, 50 ml of PBS were loaded on top of the column. The rst 13 ml of ow-through were discarded and the elution fractions 14-21 ml were collected. 300 µl of PBSeluted EV were used for Nanoparticles Tracking Analysis, while the rest was centrifuged at 100000 g for 70 min. The supernatant was discarded and the transparent pellet containing SEV was lysed for mass spectrometry sample processing. Mass spectrometry (MS) Extracellular vesicles pellets derived from human serum samples and skeletal muscle cells were lysed in denaturing SDS buffer (62.5 mM Tris-HCl pH 6.8, 10% glycerol, 2 mM EDTA, 2% SDS and 100 mM DTT) and loaded onto SDS-PAGE (10% polyacrylamide, 0.5 cm separation distance) as previously described 61 . Subsequent to protein quanti cation of the Coomassie blue stained protein bands against a BSA standard, bands were excised and subjected to in-gel protein digestion. Therefore, gel slices were alternately washed twice with 25 mM ammonium bicarbonate and 25 mM ammonium bicarbonate and 50% (v/v) acetonitrile (ACN). Protein reduction was performed in 65 mM DTT for 15 min, shaking at 350 rpm and 50°C. Proteins were then alkylated in 216 mM iodoacetamide for 15 min in the dark at room temperature. Followed by an additional washing step (25 mM ammonium bicarbonate, 25 mM ammonium bicarbonate, 50% ACN (v/v)), gel slices were shrinked in 100% (v/v) ACN. Digestion was performed with 100 ng trypsin (Promega) in 25 mM ammonium bicarbonate and 2% (v/v) ACN over night at 37°C. Resulting peptides were eluted rst with 1% (v/v) tri uoroacetic acid (TFA) followed by elution with 0.1% TFA/90% (v/v) ACN and lyophiliation. For MS analysis, lyophilised peptides were reconstituted in 1% TFA (v/v) including iRT (indexed retention time) peptides (Biognosys) and separated by liquid chromatography (Ultimate 3000, ThermoFisher Scienti c) using an EASYspray ion source equipped to an Orbitrap Fusion Lumos mass spectrometer (ThermoFisher Scienti c). Peptides were trapped and desalted on an Acclaim PepMap C18-LC-column (ID: 75 μm, 2 cm length; ThermoFisher Scienti c) and subsequently separated via EASY-Spray C18 column (ES803; ID: 50 cm x 75 μm inner diameter; ThermoFisher Scienti c) using a 100 min linear gradient from buffer A (0.1% formic acid) to 4-34% buffer B (80% ACN, 0.1% formic acid) at a ow rate of 300 nl/min followed by a 20 min linear gradient increasing buffer B to 50% and a 1 min linear gradient increasing buffer B to 90%. Column temperature was set to 40 C°. MS data for label free quanti cation were acquired in a DIA (data independent acquisition) mode. Full scan MS spectra were obtained at 120,000 resolution, m/z range of 400-1200, and an AGC (automatic gain control) target value of 5 5  Analysis of mass spectrometry data DDA (data dependent acquisition) and DIA data were imported into Spectronaut Pulsar (Version 12, Biognosys) to generate a sample speci c library using the default settings (modi cations: carbamidomethyl (C) ( xed); oxidation (M), acetyl (protein N-term) (variable); enzyme: Trypsin/P; max. missed cleavages: 2) 63 . Pulsar search was done against a human FASTA le (UniProtKB database, reviewed SwissProt, Homo sapiens TaxID 9096 canonical and isoforms, downloaded 07-2018), results were ltered by an FDR of 1% on precursor and protein group level (q value <0.01).
For quantitative analysis the DIA data were analyzed in Spectronaut (Version 12, Biognosys) using the generated library (obtained from the corresponding DDA and DIA runs) with default settings. Candidate list was created using an average log2 ratio ≤0.58 and q value ≤0.05.

Bioinformatics analysis
The proteins information was extracted from Swiss-Prot database (http://www.uniprot.org/). The prediction of a secretory signal peptide (SP) was done with the server SignalP 4.1 using as cutoff 0.450 64,65 and the proteins without a SP were further analyzed with SecretomeP 2.0 to identify nonclassically secreted proteins (score >0.6) 64 . The comparison between our study and the vesicle database Vesiclepedia was done using FunRich (v3.1.3; released on 2018) 65 . Gene ontology (GO) enrichment analysis of the SEV proteins regulated during exercise was performed for cellular component (GO-CC), molecular function (GO-MF) and biological process (GO-BP) versus annotations derived from Uniprot, with cutoff enriched p-value <0.05 (Benjamini-Hochberg, B-H, corrected FDR value).
In addition, the proteomic dataset, with UniProt identi ers, FC 12-week HIIT vs baseline and q value of each comparison, were uploaded into Ingenuity Pathway Analysis (IPA, Ingenuity™, Qiagen, Hilden, Germany) for core analysis and overlaid with the Ingenuity pathways knowledgebase, in order to categorize the SEV-cargoed proteins in biological functions and pathways. IPA predicted also the upstream regulators for the proteins in the dataset, which are responsible for the pathways and networks most signi cant to the dataset.

Statistical analysis
Data are presented as mean and standard error of mean (±SEM) or median (1st and 3rd quartile).
Variables with a skewed distribution (M-value, triglycerides, liver fat content, SEV number, protein levels) were log-transformed before analysis. ANCOVA-like linear regression analyses of baseline to 12-week HIIT changes allowing for different residual variances between the investigated groups and t-tests were performed. Logistic regression analysis was used to estimate the probability of responder-status in each group and to investigate associations between PKC activity and M-value.
Analysis in IPA were performed using the B-H method of multiple testing correction, based on the Fisher's exact test p-value with a threshold value of 0.05, and the z-score algorithm, which predicts the activation or inhibition of pathways and functions after HIIT according to the molecule expression changes in our dataset. We considered only functions and pathways with a z-score > 2 or <2.
The B-H corrected p-value (threshold <0.05) was used also to determine signi cantly enriched GO-terms.
Duplicate or highly similar GO-terms were removed and only those with the highest statistical power or with the highest number of genes in the background data set were selected. GraphPad Prism 8 was used to plot the graphs when comparing datasets (GraphPad Software, Inc).