The study protocol was described in detail elsewhere (16). Briefly, twenty volunteers underwent examination and material collection after the 90 min run (experimental Day 1) and at the baseline resting conditions (experimental Day 2). Time between experimental days was approximately 4 weeks. All volunteers were young, healthy (no history of chronic diseases or regular pharmacotherapy), physically active (≥1 h of intense, endurance exercise, ≥3x/week) nonsmokers. The participants were asked to refrain from an exhaustive physical activity, caffeine and alcohol forty-eight hours prior to both experimental days and examinations were performed after an overnight fast.

Day 1 (run) consisted of a 90 min monitored (Polar RS300X, Finland) run at the intensity of 75-80% maximal heart rate (HRmax), which was preceded by cognitive testing (computerized cognitive tests CogState & Memtrax). Blood was collected before, immediately after and 60 min after the 90 min lasting outdoor run. Cerebrospinal fluid (CSF) was collected 60min after the run by atraumatic lumbar puncture. During the run, each participant was provided with a drinking water (~3 dcl) to minimize effects of hemoconcentration on protein levels.

Day 2 (baseline/rest) included collection of fasting blood, anthropometric and metabolic (indirect calorimetry) measurements and an atraumatic lumbar puncture to collect CSF.

The assessment of cardiorespiratory fitness (VO₂max by cycle spiroergometry) was performed on a separate day, one week after the Day 1.

The study design and the analysis strategy of obtained CSF & blood is depicted in detail in the flow chart ( Supplementary Figures 1A, B ).

The study protocol was approved by the Ethics Committee of the University Hospital Bratislava and conforms to the ethical guidelines of the Declaration of Helsinki 2013. Volunteers received detailed information on the study protocol and signed written informed consent prior entering the study.

Body composition was measured with (quadrupedal bioelectrical impedance; BF511, Omron, Japan), resting energy expenditure (REE) and metabolic substrate preference (RQ; respiratory quotient) by indirect calorimetry (Ergostik, Geratherm Respiratory, Germany), blood pressure and heart rate by (Dimarson DM-14508, Taiwan) and cardiorespiratory fitness by cycle spiroergometry (PowerCube-Ergo, Ganshorn Medizin Electron GmbH, Germany) and cognitive functions by computerized cognitive tests (CogState and Memtrax). Cognitive testing was performed both before and after 90 min run. Body Mass Index (BMI) was calculated (body weight [kg]/height [m]²) and waist circumference measured at the mid-point between the lower border of the rib cage and the iliac crest. Habitual physical activity profile was evaluated by Baecke questionnaire (20).

Cerebrospinal fluid (5–8 ml) was taken by an atraumatic lumbar puncture technique in the lateral decubitus position using (Sprotte^®Pencil-Point needle, 22 G). To avoid blood contamination, the first 3 to 5 droplets of clear CSF have not been collected. The samples were immediately centrifuged at 400xg for 10 min at 4°C to remove the contaminating cells, aliquoted, placed on dry ice and stored at −80°C within 1h of collection. Paired CSF samples from both experimental days were obtained from 18 individuals, samples from additional 2 volunteers were taken only on Day 1 (after run). Post-puncture headache was reported in three out of thirty-eight individual punctures (7.9%), with a spontaneous resolution within 1–5 days.

Biochemical analyses were performed in a certified biochemical laboratory, and included analyses of serum and CSF glucose (ADVIA Chemistry glucose-hexokinase kit), albumin (N antiserum to human albumin) and insulin (ADVIA Centaur Insulin, all from Siemens Healthcare Diagnostics, UK). HOMA-IR was calculated according to formula fasting insulin (mU/l) × fasting glucose (mmol/l)/22.5.

Cerebrospinal fluid, plasma and/or serum (20; 2; 1 μL) were mixed with SDS and DTT containing loading dye, incubated (95°C, 5 min), separated on the 10%-SDS-PAGE and electroblotted to PVDF membrane (Millipore, USA). After blocking (intercept TBS, Licor, USA), membranes were incubated overnight with primary antibodies: (i) GDF11 N-terminus specific (rabbit polyclonal antibody, synthetic peptide GDF11 aa32-61, 1:1000, Abcam, ab71347, UK), (sample set A, n=14, Supplementary Figure 1B ; paired serum samples n=14; paired CSF samples n=12, unpaired CSF samples available only after run n=2), and (ii) GDF11/8 myostatin rabbit monoclonal Ab, synthetic peptide GDF11/8 aa350-407 (C-terminus), 1:1000, Abcam, ab124721, UK) (sample set A, n=14, Supplementary Figure 1B ; paired serum samples n=14; paired CSF samples n=12, unpaired CSF samples available only after run n=2). To corroborate the findings, GDF11 C-terminus specific monoclonal antibody was used (recombinant human GDF11 aa299-407, 1:1000, R&D systems, MAB19581, USA) (sample set B, n=13; overlap between A & B sample sets: n=7, Supplementary Figure 1B ). GDF11-C-terminus antibody was used in 5x concentrated CSF samples. Results obtained by GDF11 (N-terminus specific) antibody were used for the association analysis (sample set A). Secondary antibody IRDye 800CW (Li-Cor, USA, 1:10000) was used to visualize protein content (LI-COR, USA). GDF11/8 with molecular weight between 45-50 kDa was recorded with LiCor infrared imaging system and analyzed with Image Studio Lite (Ver 5.2, Li-Cor, USA). Since GDF8 (myostatin) and GDF11 share ~90% sequence identity in the mature, C-terminal signaling domains and ~52% identity in their N-terminal prodomains (21), C and N terminus specific antibodies and antibody recognizing both GDF11 and GDF8 were used.

Levels of adiponectin in serum and CSF were measured by ELISA [Biovendor, Czech Republic], as described in (16).

Endurance exercise effect was assessed by either Student’s paired t-test (two-time points) or multiple comparison testing with ANOVA and Tukey post-hoc test (three time-points) using JMP [version 4.0.4 academic; SAS Institute, USA]. A priori power analysis utilizing our pre-existing data on exercise-induced changes in albumin CSF/serum ratio (marker of blood brain barrier permeability) was performed, with α error probability set to 0.05 and power (1-β error probability) to 0.95. Total sample size of 11 would be needed to determine relevant exercise-induced difference in this parameter. Pearson linear correlation analysis model was used. Values are presented as mean ± SEM and the statistical significance was set at p<0.05. Changes induced by 90 min run were calculated as a fold change (ratio between post-run and pre-run levels).

The characteristics of the study population is presented in Table 1 . Intensive 90 min run induced a delayed decrease in glycaemia 60 min post-exercise, as compared to baseline and immediately post exercise (baseline (resting)) 4.9 ± 0.5 mmol/l, 0 min post exercise 5.2 ± 0.8 mmol/l, 60 min post exercise 4.4 ± 0.4 mmol/l, p<0.001), and an immediate increase of insulinemia and HOMA-IR after run, with a delayed decrease 60 min post-exercise, as compared to baseline and immediately post exercise (insulinemia: baseline (resting)) 5.8 ± 2.7 mIU/l, 0 min post exercise 8.4 ± 5.1 mIU/l, 60 min post exercise 3.5 ± 2.0 mIU/l, p=0.001; HOMA-IR: baseline (resting)) 1.32 ± 0.69, 0 min post exercise 2.00 ± 1.22, 60 min post exercise 0.72 ± 0.46, p=0.001). There was no running-induced change in glycemia and insulinemia in CSF when compared to baseline (resting) state (p>0.1 for all).

Serum lactate increased immediately after the run and near normalization was observed one hour post-run as compared to baseline (baseline (resting)) 1.1 ± 0.6 mmol/l, 0 min post exercise 3.3 ± 2.3 mmol/l, 60 min post exercise 1.6 ± 1.0 mmol/l; p=0.001). Similar pattern was observed for changes in serum albumin (baseline, resting) 45.1 ± 3.1 g/l, immediately post-exercise 49.1 ± 2.9 g/l, 60 min post-exercise 45.8 ± 2.7 g/l; p=0.001) and serum thrombocyte count (baseline (resting) 219 ± 43 10⁹/l, 0 min post exercise 267 ± 31 10⁹/l, 60 min post-exercise 209 ± 58 10⁹/l, p=0.036), which might reflect hemoconcentration present immediately after run. Interestingly, 90 min run increased lactate in CSF of all volunteers (baseline (resting)) 1.2 ± 0.1 mmol/l, 60 min post exercise 1.9 ± 0.4 mmol/l, p<0.001). Total serum creatinkinase increased immediately after run and compared to baseline remained elevated for another hour (baseline (resting)) 3.2 ± 2.4 umol/l, 0 min post exercise 4.5 ± 3.1 umol/l, 60 min post exercise 4.3 ± 2.9 umol/l, p=0.005).

An acute bout of exercise led to the reduction of GDF11 levels in CSF, which was confirmed in paired CSF samples with two GDF11-specific antibodies for C and N terminus, respectively (sample set A, N-terminus antibody: -20.6%, p=0.046, Figure 1A ; sample set B, C-terminus antibody: -11.2%, p=0.008, Figure 1B ). Trend towards a decrease in GDF was detected also by antibody against C-terminus which could not distinguish between GDF11 and GDF8 (-10.2%, p=0.065, Figure 1C ). Thus, the running-induced decrease of GDF11 in CSF was confirmed by 3 distinct GDF11 antibodies. The whole blots are depicted in Supplementary Figures 2 and 4 . There was no difference in running-induced change of GDF11 levels in CSF of 7 individuals, determined by two different GDF11-specific antibodies ( Supplementary Figure 3 ).

Concentration of GDF11 in CSF was lower than that in plasma (-29.7%, p=0.022) or serum (-21.9%, p=0.091). Plasma and serum GDF11 levels did not differ (p=0.261), and not surprisingly there was a positive correlation between GDF11 in plasma and serum (n=55, R=0.376, p=0.003), but not between serum or plasma and CSF levels of GDF11 (data not shown).

Neither serum nor plasma GDF11 levels changed immediately after or 60 min after the run ( Figures 1D–F ). However, there was a decrease in combined protein content for GDF11 and GDF8 in serum one hour after endurance exercise when compared to baseline (-13.4%, p=0.021). Notably, this effect appeared to be gender-specific, as the decrease of serum GDF11/GDF8 in serum one hour after run was significant in women (n=8, -18.5%, p=0.019), but not in men (n=6, -4.9%, p=0.553). This might indicate that the dynamic changes of GDF levels in blood induced by intensive endurance exercise are driven rather by GDF8 than by GDF11. The whole blots depicting dynamics of GDF11 in serum are presented as Supplementary Figure 4 .

The running-induced changes of GDF11 protein in CSF tended to negatively correlate with changes in its plasma (but not serum) levels (n=11, r=-0.57, p=0.057).

Blood-brain barrier permeability, as indirectly assessed by the ratio of albumin in CSF and serum, was decreased after the 90 min run by 27.1% (p<0.001; Figure 2A ). Running-induced changes in CSF levels of GDF11/GDF8 and albumin as well as CSF/serum albumin ratio were positively correlated (n=12, r=0.61/0.66, p=0.036/0.020). Change in GDF11 in CSF tended to positively correlate with a running-induced change in CSF albumin (r=0.54, p=0.069), but not with a change in CSF/serum albumin ratio. CSF/serum albumin ratio tended to positively correlate with GDF11 CSF/serum ratio (n=24, r= 0.39, p=0.060, Figure 2B ) and positively correlated with GDF11/GDF8 CSF/serum ratio (n=24, r=0.42, p=0.040, Figure 2C ), indirectly indicating a role of blood-brain barrier permeability in transporting GDF11 and/or GDF8 into the brain.

Albumin CSF/serum ratio correlated negatively with GDF11 levels in serum (n=26, r=-0.40, p=0.042) as well as in plasma (n=26, r=-0.44, p=0.026), but positively with GDF11/GDF8 in CSF (n=26, r=0.50, p=0.009).

Compared to individuals with higher physical fitness (n=7, M/F 3/4, mean VO₂max=47.2 mlO₂/kg_BW/min, cut off median of VO₂max=42.2 mlO₂/kg_BW/min), individuals with lower physical fitness (n=5, M/F 3/2, mean VO₂max=35.9 mlO₂/kg_BW/min) displayed greater exercise-induced decrease in CSF levels of GDF11 (-34.7% vs. -0.4%; p=0.025). However, there was no fitness-related difference in GDF11 protein content in CSF neither at baseline (resting) state nor after run. The link between physical fitness and exercise-induced regulation of GDF11 in CSF was supported also by a positive correlation between running-induced change in GDF11 levels in CSF and VO₂max (n=12, r=0.59, p=0.042) ( Figure 3 ). CSF levels of GDF11/GDF8 were not associated with physical fitness (data not shown).

Individuals with higher physical fitness decreased, whereas less fit volunteers rather increased GDF11 serum levels measured 60 min after 90 min run compared to baseline (high/low physical fitness; -18.3% vs. 22.5%; p=0.049). This observation was further supported by a trend towards negative correlation between exercise-induced change in serum GDF11 and self-reported habitual sport activity (n=13, r=-0.50, p=0.083). Moreover, baseline GDF11 serum levels positively correlated with physical fitness (n=13, r=0.63, p=0.020) ( Figure 3 ). There was no relationship between physical fitness and GDF11 in plasma (p=0.304).

The pattern of associations between GDF11 levels, clinical & biochemical parameters and their running-induced changes is depicted in the heat map ( Figure 3 ).

Briefly, there were several associations between GDF11 levels in CSF and cognitive functions assessed by sensitive computerized tests. GDF11 levels in CSF were positively associated with psychomotor performance and attention, visual learning, as well as learning and working memory (p<0.05). Visual memory was also associated with CSF levels of GDF11/GDF8 (r=0.40, p=0.049). On the other hand, running-induced change in serum GDF11/GDF8 levels correlated positively with reaction time in two computerized tests (Memtrax/CogState; r=0.52/0.65, p=0.055/0.012), and negatively with learning and working memory (r=-0.65, p=0.011).

Running-induced change in serum GDF11 was positively associated with body fat (p=0.022) and negatively with muscle mass (p=0.011). Furthermore, it correlated positively with serum insulin immediately after the run (p<0.01) as well as with running-induced change in serum insulin (p=0.024). Running-induced change in GDF11 in CSF correlated negatively with change in serum insulin 60 min after run (p=0.015). Lower serum lactate before run (baseline) was associated with higher GDF11 levels in serum after an acute exercise bout (p<0.05), as well as with a greater running-induced decrease in GDF11 (p=0.01) and GDF11/GDF8 in CSF (p=0.05). Similarly, lower baseline serum creatinkinase was linked with a greater running-induced change in serum GDF11 (p=0.045). Adiponectin, adipokine with a neuroprotective potential which we showed to be regulated by acute intensive exercise (16), was positively associated with levels of GDF11 in CSF, but only at baseline (before run) (p=0.009). Furthermore, there was a positive correlation between running-induced changes in CSF levels of GDF11 and adiponectin (p=0.028). In serum, on the other hand, the positive associations between adiponectin and GDF11 levels or their running-induced changes were present only immediately after the run (p<0.05). Baseline CSF/serum albumin ratio, an indicator of the blood-brain barrier permeability, correlated negatively with GDF11 serum levels before (p=0.042) as well as after the run (p<0.04), with a trend for GDF levels in CSF at baseline (p=0.075).