The metabolome as a diagnostic for maximal aerobic capacity during exercise in type 1 diabetes

Aims/hypothesis Our aim was to characterise the in-depth metabolic response to aerobic exercise and the impact of residual pancreatic beta cell function in type 1 diabetes. We also aimed to use the metabolome to distinguish individuals with type 1 diabetes with reduced maximal aerobic capacity in exercise defined by \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\dot{V}{\text{O}}_{\text{2peak}}$$\end{document}V˙O2peak. Methods Thirty participants with type 1 diabetes (≥3 years duration) and 30 control participants were recruited. Groups did not differ in age or sex. After quantification of peak stimulated C-peptide, participants were categorised into those with undetectable (<3 pmol/l), low (3–200 pmol/l) or high (>200 pmol/l) residual beta cell function. Maximal aerobic capacity was assessed by \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\dot{V}{\text{O}}_{\text{2peak}}$$\end{document}V˙O2peak test and did not differ between control and type 1 diabetes groups. All participants completed 45 min of incline treadmill walking (60% \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\dot{V}{\text{O}}_{\text{2peak}}$$\end{document}V˙O2peak) with venous blood taken prior to exercise, immediately post exercise and after 60 min recovery. Serum was analysed using targeted metabolomics. Metabolomic data were analysed by multivariate statistics to define the metabolic phenotype of exercise in type 1 diabetes. Receiver operating characteristic (ROC) curves were used to identify circulating metabolomic markers of maximal aerobic capacity (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\dot{V}{\text{O}}_{\text{2peak}}$$\end{document}V˙O2peak) during exercise in health and type 1 diabetes. Results Maximal aerobic capacity (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\dot{V}{\text{O}}_{\text{2peak}}$$\end{document}V˙O2peak) inversely correlated with HbA1c in the type 1 diabetes group (r2=0.17, p=0.024). Higher resting serum tricarboxylic acid cycle metabolites malic acid (fold change 1.4, p=0.001) and lactate (fold change 1.22, p=1.23×10−5) differentiated people with type 1 diabetes. Higher serum acylcarnitines (AC) (AC C14:1, F value=12.25, p=0.001345; AC C12, F value=11.055, p=0.0018) were unique to the metabolic response to exercise in people with type 1 diabetes. C-peptide status differentially affected metabolic responses in serum ACs during exercise (AC C18:1, leverage 0.066; squared prediction error 3.07). The malic acid/pyruvate ratio in rested serum was diagnostic for maximal aerobic capacity (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\dot{V}{\text{O}}_{\text{2peak}}$$\end{document}V˙O2peak) in people with type 1 diabetes (ROC curve AUC 0.867 [95% CI 0.716, 0.956]). Conclusions/interpretation The serum metabolome distinguishes high and low maximal aerobic capacity and has diagnostic potential for facilitating personalised medicine approaches to manage aerobic exercise and fitness in type 1 diabetes. Graphical Abstract Supplementary Information The online version contains peer-reviewed but unedited supplementary material available at 10.1007/s00125-024-06153-0.


Visit 3 Exercise Intervention
Individuals arrived at the CRF exercise laboratory at 8:30 A.M. following an overnight fast.
Participants maintained their normal basal insulin regimen and had abstained from exercise for 48 hr.A carbohydrate snack (belVita; Mondelez International) of 853.54 kJ (204 kcal; 31 g carbohydrate) was consumed, and participants were rested for 20 min.
Target capillary blood glucose was > 7 mmol/L during exercise, with participants given 10 g carbohydrate if capillary blood glucose decreased below this limit.
Each participant was cannulated and resting (baseline) blood samples (10 mL) were drawn.HbA1c was measured in baseline blood.Participants walked on an incline for 45 minutes at 60% V ̇O2Peak.Participants' treadmill velocity and gradient were calculated using Ȯ2, velocity, and gradient data from the preliminary V ̇O2Peak test.Breath-by-breath respiratory parameters (Metalyzer 3B-R3, Cortex) were recorded, with gradient adjusted at 10 and 30 minutes if V ̇O2 was >10% different than target V ̇O2.Upon exercise completion, blood samples were drawn from the cannula.Participants rested for 60 minutes before final blood samples were drawn.At each time point whole EDTA venous blood glucose was analysed using a YSI 2300 STAT PLUS Analyzer (YSI Inc, Xylem Analytics, USA).Blood was processed for serum and it was stored at -80°C.

Targeted Metabolomics LC−MS Analysis of Acylcarnitines, Non-esterified fatty acids, Bile Acids, Tryptophan Metabolism and TCA Cycle Metabolites -Chromatography
The binary solvent system used for the analysis of acylcarnitines, non-esterified fatty acids, bile acids and tryptophan metabolites was solvent A comprising LC−MS-grade water, 0.2 mM ammonium formate, and 0.01% formic acid and solvent B comprising analytical-grade acetonitrile/isopropanol 1:1, 0.2 mM ammonium formate, and 0.01% formic acid.For the analysis of acylcarnitines, non-esterified fatty acids and bile acids the mobile phase was set at a flow rate of 1.3 mL/min.For the tryptophan metabolite analysis the mobile phase was set at a flow rate of 0.45 mL/min.For acylcarnitine analysis, the column mobile phase was held at 2% solvent B for 0.1 min, followed by an increase from 2 to 98% solvent B over 1.2 min.The mobile phase was then held at 98% solvent B for 0.9 min.The mobile phase was then returned to 2% solvent B and held for 0.1 min to re-equilibrate the column.For the analyses of acylcarnitines 0.5 μL of sample was injected.
For non-esterified fatty acid analysis, the column mobile phase was increased from 50 to 98% solvent B over 0.7 min.The mobile phase was then held at 98% solvent B for 0.5 min.The mobile phase was then returned to 50% solvent B and held for 0.1 min to re-equilibrate the column.For the analyses of free fatty acids 2 μL of sample was injected.
For bile acid analysis, the column mobile phase was held at 20% solvent B for 0.1 min, followed by an increase from 20 to 55% solvent B over 0.7 min.The mobile phase was increased to 98% solvent B and held for 0.9 min.The mobile phase was then returned to 20% solvent B held for 0.1 min to re-equilibrate the column.For the analyses of bile acids 10 μL of sample was injected.The binary solvent system used for the analysis of TCA cycle metabolites was solvent A comprising LC−MS-grade water and 0.1% formic acid and solvent B comprising LC-MS-grade acetonitrile and 0.1% formic acid.The mobile phase was set at a flow rate of 0.4 mL/min.The column mobile phase was increased from 2% solvent B to 15% solvent B over 3 min.The mobile phase was then returned to 2% solvent B and held for 0.1 min to re-equilibrate the column.Three μL of sample was injected.

Multivariate -PLS-DA Permutation Test
In the Permutation Test the class labels for the dataset are randomly reassigned and a new classifier is built using the original data, its performance is then evaluated.This process is repeated with 2000 permutations to estimate the distribution of the group separation distance [3].By comparing the performance of the original model and the performance of the 2000 randomly assigned classifiers, it can be determined whether the original model is significantly different from the 2000 random permutations.The further to the right of the distribution formed by randomly permuted data in the permutation test statistic, the more significant the discrimination.The P-value is calculated from the proportion of times that class separation based on the randomly labelled sample is as good as the original data.

ANOVA Simultaneous Component Analysis (ASCA)
ASCA is a direct generalisation of ANOVA for univariate data to the multivariate setting, suitable for the analysis of datasets that incorporate both a group comparison and longitudinal structure.The results of this analysis can be visualised with a plot of leverage against the squared prediction error (SPE).Leverage measures the influence of each observation for a principal component.Score plots will identify observations with high leverages, ie.observations that tend to contribute mostly to separation in the PCA.SPE measures the expected squared distance between the predicted value and the true value (i.e. the quality of the predictor).

ESM
Group H = high (≥200 pmol/L) residual C-peptide (n = 10), group L = low (3-200 pmol/L) residual C-peptide (n = 9), group U = undetected (<3 pmol/L) C-peptide (n = 11)Table 13.ANOVA Simultaneous Component Analysis (ASCA) of the interaction between exercise and C-peptide status.Metabolites exhibiting a differential effect to exercise in people with Type 1 Diabetes dependent on C-peptide status are shown.Leverage measures the influence of each observation for a principal component.Score plots will identify observations with high leverages, ie.observations that tend to contribute mostly to separation in the PCA.SPE = squared prediction error.SPE measures the expected squared distance between the predicted value and the true value (i.e. it measures the quality of the predictor).The model was significant as determined by a permutation test statistic of p < 0.05.NEFA = Non-esterified Fatty Acid.AC = Acylcarnitine.High (≥200 pmol/L) residual C-peptide (n = 10), low (3-200 pmol/L) residual C-peptide (n = 9), undetected (<3 pmol/L) C-peptide (n = 11).
value.NEFA = Non-esterified fatty acid; AC = Acylcarnitine; MOVA = methyl-oxo-valeric acid; BAIBA = β-aminoisobutyric acid.N = 30.ESM Figures ESM Figure 1.Permutation test statistic for PLS-DA plot of metabolomic data from serum samples at baseline.The distribution formed by randomly permuted data in the permutation test statistic for the PLS-DA model of metabolomic data from serum samples at baseline against the observed statistic for the original model and the number of the 2000 random permutations that outperformed the original data (p = 0.0005) (1/2000).ESM Figure 2. Permutation test statistic for PLS-DA plot of metabolomic data from baseline, exercise and post exercise 1hr recovery of non-diabetes controls.The distribution formed by randomly permuted data in the permutation test statistic for the PLS-DA model of metabolomic data from serum samples from baseline, exercise and post exercise 1hr recovery of non-diabetes controls against the observed statistic for the original model and the number of the 2000 random permutations that outperformed the original data (p < 0.0005) (0/2000).ESM Figure 3. Permutation test statistic for PLS-DA plot of serum metabolomic data from baseline, exercise and post exercise 1hr recovery of people with type 1 diabetes .The distribution formed by randomly permuted data in the permutation test statistic for the PLS-DA model of metabolomic data from serum samples from baseline, exercise and post exercise 1hr recovery of people with type 1 diabetes against the observed statistic for the original model and the number of the 2000 random permutations that outperformed the original data (p < 0.0005) (0/2000).ESM Figure 4. Common metabolic response to aerobic exercise in non-diabetes controls (Control) and people with type 1 diabetes.Metabolite species from baseline (B, red), exercise (E, green) and post exercise 1hr recovery (blue, R) samples with common responses to exercise.Identified with multivariate empirical bayes analysis of variance (MEBA) for time series, which is designed to compare temporal profiles across different biological conditions.Corresponding univariate Two-way within subjects ANOVA data is given in ESM Table 9.Data is mean centred.Box and whisker plots show 25th and 75th percentile and the median.Upper whisker = Q3 + 1.5XIQR (Interquantile Range), lower whisker is Q1 -1.5XIQR.Data is mean centred.FFA = free fatty acid/ non-esterified fatty acid, AC = acylcarnitine, 5HIAA = 5-Hydroxyindoleacetic acid.Control n = 30, T1D = 30.**** p < 0.0001, *** p < 0.001, ** p < 0.01, * p < 0.05.from baseline with data grouped according to C-peptide status and compared with control.The distribution formed by randomly permuted data in the permutation test statistic for the PLS-DA model of metabolomic data from serum samples at baseline showing separation of control from people with type 1 diabetes and high plasma C-peptide (≥200 pmol/L), low plasma C-peptide ((3-200 pmol/L) and undetectable plasma C-peptide (<3 pmol/L).(p = 0.001) (2/2000).ESM Figure 6.Permutation test statistic for PLS-DA plot of serum metabolomic data from baseline of C-peptide high, low and undetected groups.The distribution formed by randomly permuted data in the permutation test statistic for the PLS-DA model of metabolomic data from serum samples at baseline showing separation of high plasma C-peptide (≥200 pmol/L), low plasma C-peptide ((3-200 pmol/L) and undetectable plasma C-peptide (<3 pmol/L) groups.(p = 0.0245) (49/2000).ESM Figure 7. C-peptide status effects the metabolic response to exercise in type 1 diabetes a) An ANOVA Simultaneous Component Analysis (ASCA) plot of the serum metabolomic data for baseline, exercise and 1 hr post exercise recovery in the non-diabetes control and type 1 diabetes volunteers.ASCA identifies major patterns with regard to the cpeptide status and exercise response and their interaction.Metabolites highlighted by the green circle distinguish the C-peptide status groups.Leverage measures the influence of each observation for a principal component.Score plots will identify observations with high leverages, ie.observations that tend to contribute mostly to separation in the PCA.SPE = squared prediction error.SPE measures the expected squared distance between the predicted value and the true value (ie. it measures the quality of the predictor).The model was significant as determined by a permutation test statistic of p < 0.05.b) Box and whisker plots of the concentration of the 4 metabolites identified by the ASCA model to exhibit a differential effect to exercise in people with T1D dependent on C-peptide status (Control, type 1 diabetes and high plasma C-peptide (≥200 pmol/L), type 1 diabetes and low plasma C-peptide ((3-200 pmol/L) and type 1 diabetes and undetectable plasma C-peptide (<3 pmol/L)).For corresponding Leverage and SPE see ESM Table 11.Box and whisker plots show 25th and 75th percentile and the median.Upper whisker = Q3 + 1.5XIQR (Interquantile Range), lower a b whisker is Q1 -1.5XIQR.Data are mean centred.B = baseline serum (red), E = aerobic exercise serum (green), R = 1 hour post exercise recovery serum (blue).FFA = free fatty acid / nonesterified fatty acid, AC = acylcarnitine, BAIBA = beta-aminoisobutyric acid.Non-diabetes controls (n = 30), high (≥200 pmol/L) residual C-peptide (n = 10), low (3-200 pmol/L) residual C-peptide (n = 9), undetected (<3 pmol/L) C-peptide (n = 11).** p < 0.01, * p < 0.05.

Table 1 .
Volunteer pharmacological treatments by group. ESM

Table 2 .
Multiple Reaction Monitoring Parameters for acylcarnitines species.
ESM Table3.Multiple Reaction Monitoring Parameters for Non-esterified fatty acid species.Non-esterified fatty acids are designated by acyl chain length in carbons and degree of unsaturated double bonds.Internal standard (IS).m/zmass-to-chargeratio.ESM Table4.Multiple Reaction Monitoring Parameters for bile acid species.Internal standard (IS).m/z mass-to-charge ratio.

Table 6 .
Multiple Reaction Monitoring Parameters for TCA cycle intermediates.ESM Table 11.ANOVA of metabolite species differentiating non-diabetes controls and m/z mass-to-charge ratio.ESM Table9.Two-way repeated measures within subjects ANOVA showing serum metabolites commonly altered by exercise in controls and people with type 1 diabetes (T1D).NEFA = Non-esterified Fatty Acid.AC = Acylcarnitine, 5HIAA = 5-hydroxyindoleacetic acid.Adj.P-values represent Bonferroni-correction for multiple hypothesis testing.Control n = 30, T1D = 30.
ESM Table12.ANOVA of metabolite species differentiating people with Type 1 Diabetes and undetected, low and high residual plasma C-peptide at baseline.p -value generated by One-Way ANOVA.Significant between group differences calculated using a Tukey's Posthoc test.

Table 15 .
Baseline serum metabolites significantly correlated with VO2Peak in the control population.Correlation analysis by Pearson r showing correlation, t-statistic and p- ESM Table 14.Baseline serum metabolites significantly correlated with VO2Peak in the total study population.Correlation analysis by Pearson r showing correlation, t-statistic and P-value.NEFA = Non-esterified Fatty Acid; AC = Acylcarnitine; MOVA = methyl-oxo-valeric acid; BAIBA = β-aminoisobutyric acid.N = 60.