Exercise Metabolomics Reveals Intensity-Dependent Metabolic Responses Associated with Cardiorespiratory Fitness in Adults with Type 1 Diabetes

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This manuscript examines exercise-related metabolomic responses at different exercise intensities in adults with type 1 diabetes. The topic is interesting, and the serial blood sampling during a graded maximal exercise test is a strength. However, several important methodological and interpretative issues should be addressed before the manuscript can be considered for publication.

Major comments

1. Very small sample size:  The metabolomic analysis included only 20 participants, while the authors examined 164 metabolites using several adjusted and interaction models. No specific sample-size or power calculation was performed for this secondary analysis. 

With such a small sample, the models may be overfitted and the estimated associations may be unstable. This is particularly important for the interaction analyses involving VO₂peak, HbA1c and resting glucose. The authors should clearly present this study as exploratory and reduce the strength of their conclusions.

2. Absence of a healthy control group

The study includes only adults with type 1 diabetes. Therefore, it is not possible to determine whether the observed metabolic responses are specific to type 1 diabetes or represent normal physiological responses to maximal exercise.

This is important because the manuscript links the findings to exercise intolerance and impaired cardiorespiratory fitness in type 1 diabetes. Without a healthy comparison group, such disease-specific conclusions are not fully supported. The authors should either moderate these claims or clearly explain this limitation.

3. Multiple testing and false-positive findings

A large number of metabolites, models, workload comparisons and interaction terms were examined. Although the authors applied the Benjamini–Hochberg correction, the correction was performed separately for different model terms. This approach may not adequately control the overall number of statistical tests.

In addition, the correlation analyses were not corrected for multiple testing. These correlations should be clearly described as exploratory, and strong biological conclusions should not be based on unadjusted p-values.

4. Interpretation of circulating metabolites as pathway activation

The manuscript frequently describes increased plasma metabolites as evidence of greater “activation” of glycolysis or the TCA cycle. However, plasma metabolite concentrations do not directly measure metabolic flux or tissue-specific pathway activity. The authors also acknowledge this limitation in the discussion. 

The wording should therefore be revised throughout the manuscript. It would be more accurate to state that the findings show changes in circulating metabolites related to glycolysis and the TCA cycle, rather than direct activation of these pathways.

5. Timing of blood sampling at the anaerobic threshold

Blood samples were collected every three minutes, and the sample closest to the individually identified anaerobic threshold was selected for analysis. 

This approach may introduce timing error because the selected sample could have been collected before or after the actual threshold. Rapidly changing metabolites, especially lactate, may be strongly affected by this issue. The authors should provide the average time difference between the blood sample and the identified anaerobic threshold and discuss its possible influence on the findings.

6. Data preprocessing requires stronger justification

Missing metabolomic values were imputed using a random forest method, followed by quantile normalization. 

The manuscript should report the percentage of missing values for each metabolite and explain why the values were missing. The authors should also justify the use of quantile normalization in an exercise study, since exercise may cause real global changes in metabolite distributions. A sensitivity analysis without imputation or with an alternative normalization method would improve confidence in the results.

7. Pathway enrichment analysis

The pathway analysis is based on assigning metabolites to nine broad and mutually exclusive metabolic categories. This approach may oversimplify biological pathways because one metabolite can participate in several metabolic processes.

The authors should explain why this classification method was selected instead of a standard metabolomic pathway enrichment method. The pathway findings should also be described cautiously.

8. Lack of independent validation

The study is a secondary exploratory analysis, and the findings were not confirmed in an independent cohort. Therefore, the reported associations between VO₂peak, lactic acid and malic acid should be considered hypothesis-generating.

The clinical relevance of metabolomic profiling is currently overstated. The conclusion should emphasize the need for confirmation in larger and independent populations.

 

 9. The authors may add a short paragraph about the effects of different exercise modalities on circulating metabolic markers in people with diabetes. This information would fit well in the Introduction, after the paragraph describing previous exercise metabolomics studies. It may help show that metabolic responses can depend on the type and intensity of exercise, although the available evidence is still limited.

The following reference may be useful:

Ravanbakhsh S, Pournemati P, Soori R. The Effect of Aerobic Exercise and Yoga on Serum Fetuin-A Levels in Obese Women with Type 2 Diabetes. International Journal of Sport Studies for Health. 2026;9(2):1–9. https://doi.org/10.61838/kman.intjssh.4809

 

10. ⁠�The authors may discuss the role of exercise intensity in changing circulating hormones and metabolic biomarkers. This point could be added in the Discussion, where the intensity-dependent metabolic responses are interpreted. It would support the idea that high-intensity exercise may produce different metabolic responses compared with lower-intensity exercise. However, the differences in population and study design should also be mentioned.

The following reference may be considered:

Almasoodi A, Amirsasan R, Vakili J, Khani M, Alwawi H. Effect of Eight Weeks of High-Intensity Interval Training on Serum Asprosin Levels and Body Composition in Overweight and Obese Men. International Journal of Sport Studies for Health. 2026;9(3):1–9.

Author Response

This manuscript examines exercise-related metabolomic responses at different exercise intensities in adults with type 1 diabetes. The topic is interesting, and the serial blood sampling during a graded maximal exercise test is a strength. However, several important methodological and interpretative issues should be addressed before the manuscript can be considered for publication.

Reviewer 1.

Major comments

1. Very small sample size: The metabolomic analysis included only 20 participants, while the authors examined 164 metabolites using several adjusted and interaction models. No specific sample-size or power calculation was performed for this secondary analysis. 

With such a small sample, the models may be overfitted and the estimated associations may be unstable. This is particularly important for the interaction analyses involving VO₂peak, HbA1c and resting glucose. The authors should clearly present this study as exploratory and reduce the strength of their conclusions.

Author response:

Manuscript location: Abstract; Sections 2.1, 2.2, 2.5, 5.2, 5.3, and 6.

We thank the reviewer for raising this important concern. We agree that the small number of participants limits the precision and stability of the estimated associations, particularly for the interaction analyses.

The study was an exploratory, pooled secondary analysis of screening data obtained from two randomized crossover trials. Both parent trials used the same cardiopulmonary exercise-testing and blood-sampling protocol, at the screening visit. For this analysis, we included only measurements collected during this common screening procedure; that is, we combined data collected at rest, anaerobic threshold, peak, and recovery. We did not include data from the subsequent randomized treatment periods.

To clarify the modelling approach in relation to the reviewers' concern about potential overfitting, we did not include the 164 metabolites simultaneously as predictors in a single model. Instead, we fitted a separate linear mixed-effects model for each metabolite, with metabolite abundance as the outcome. Each participant contributed repeated measurements across four exercise stages, and a participant-specific random effect accounted for the correlation among measurements from the same individual.

This design allowed each participant to serve as their own control by comparing their metabolite abundance at each exercise stage with their own resting value. The random intercept also accounted for differences in the overall metabolite abundance between participants.

Although this modelling structure avoids fitting all 164 metabolites simultaneously in a single model, it doesn’t eliminate the potential for overfitting. The dataset still contained only 20 independent participants, and the adjusted and interaction models contained several parameters relative to this sample size. We therefore acknowledge that the estimated associations, particularly those involving VO2peak, HbA1c and resting glucose, may be unstable. We therefore describe these analyses as exploratory and hypothesis-generating and interpret their findings cautiously.

We could not perform a separate a priori sample-size calculation because the available sample size was fixed by the number of eligible participants from the two parent trials who completed the common screening protocol and had available metabolomic data. We have now stated this explicitly in the main Methods section. We have also expanded the limitations section and softened the Abstract, Discussion, and Conclusion to avoid implying that these findings are definitive or immediately clinically applicable. We now present the interaction results as preliminary signals requiring confirmation in larger independent cohorts.

To limit the false positive findings across the 164 metabolites we apply the Benjamini–Hochberg false-discovery-rate correction across metabolites within each model term. We acknowledge that this addresses multiplicity but does not eliminate the uncertainty or potential instability associated with the small sample size.

We have also acknowledged the small sample was as an admitted primary weakness in our limitations section ‘The principal limitation of this exploratory secondary analysis is the small sample size’.’

2. Absence of a healthy control group

The study includes only adults with type 1 diabetes. Therefore, it is not possible to determine whether the observed metabolic responses are specific to type 1 diabetes or represent normal physiological responses to maximal exercise.

This is important because the manuscript links the findings to exercise intolerance and impaired cardiorespiratory fitness in type 1 diabetes. Without a healthy comparison group, such disease-specific conclusions are not fully supported. The authors should either moderate these claims or clearly explain this limitation.

Author response:

Manuscript location: Section 5.2 (Limitations).

We thank the reviewer for highlighting this important information. We agree that because the study included only adults with type 1 diabetes and did not include a healthy comparison group, we cannot determine whether the observed metabolic responses are specific to type 1 diabetes or represent physiological responses to maximal exercise that would also occur in individuals without diabetes.

However, the purpose of this study was to characterize the exercise-induced metabolic changes and examine their associations with cardiorespiratory fitness within a cohort of adults with type 1 diabetes. The study was not designed to establish disease-specific metabolic abnormalities or determine the mechanisms responsible for the lower cardiorespiratory fitness commonly reported in type 1 diabetes.

We have acknowledged this in our limitations section ‘The principal limitations of this exploratory secondary analysis is the small sample size and absence of a healthy control group, the latter precluding determination of whether the observed exercise-induced metabolic responses are specific to type 1 diabetes or reflect physiological responses to exercise more generally.’

3. Multiple testing and false-positive findings

A large number of metabolites, models, workload comparisons and interaction terms were examined. Although the authors applied the Benjamini–Hochberg correction, the correction was performed separately for different model terms. This approach may not adequately control the overall number of statistical tests.

In addition, the correlation analyses were not corrected for multiple testing. These correlations should be clearly described as exploratory, and strong biological conclusions should not be based on unadjusted p-values.

Author response:

Manuscript location: Section 2.5; Section 3.5; Section 5.2.

We thank the reviewer for the comment. We agree that the multiplicity framework and its scope require clearer explanation. For each model-specific coefficient, the scientific question was whether that workload, covariate or interaction effect was associated with any of the 164 metabolites. We therefore defined each model-coefficient combination as a testing family and applied the Benjamini–Hochberg false discovery rate correction across metabolites within that family. This approach preserves the interpretation of q-values for a specific biological contrast. A single correction across all coefficients from all six models would combine distinct hypotheses, including correlated and partially redundant model terms, and was not the prespecified multiplicity framework.

We now explicitly acknowledge that this term-wise adjustment controls the false discovery rate within each model–coefficient family and not across the complete collection of models and terms. Accordingly, results from secondary models are described as exploratory, and interaction findings are described as hypothesis-generating.

The correlation analyses were secondary, post-selection analyses involving metabolites selected based on BH-FDR-significant workload effects. In the correlation heatmaps, asterisks denote nominal Pearson correlation p-values. We have clarified in the Methods, Results, figure legends and Discussion that these correlations are exploratory and are not used as confirmatory biological evidence.

4. Interpretation of circulating metabolites as pathway activation

The manuscript frequently describes increased plasma metabolites as evidence of greater “activation” of glycolysis or the TCA cycle. However, plasma metabolite concentrations do not directly measure metabolic flux or tissue-specific pathway activity. The authors also acknowledge this limitation in the discussion. 

The wording should therefore be revised throughout the manuscript. It would be more accurate to state that the findings show changes in circulating metabolites related to glycolysis and the TCA cycle, rather than direct activation of these pathways.

Author Response:

Manuscript location: Highlights; Sections 3.6 and 4.

Thank you for raising this. We have been careful to emphasize this by changing the wording throughout, including the article highlights where it was most cited:

What are the new findings?

• Exercise intensity is a key determinant of the circulating metabolome, with dynamic changes in metabolites related to glycolysis, the tricarboxylic acid (TCA) cycle and purine metabolism occurring predominantly at higher physiological workloads.

• Higher cardiorespiratory fitness was associated with greater exercise-induced changes in circulating glycolysis-related and TCA-cycle metabolites.

5. Timing of blood sampling at the anaerobic threshold

Blood samples were collected every three minutes, and the sample closest to the individually identified anaerobic threshold was selected for analysis. 

This approach may introduce timing error because the selected sample could have been collected before or after the actual threshold. Rapidly changing metabolites, especially lactate, may be strongly affected by this issue. The authors should provide the average time difference between the blood sample and the identified anaerobic threshold and discuss its possible influence on the findings.

Author response:

Manuscript location: Section 2.3 and Section 5.2.

We thank the reviewer for raising this important methodological consideration. To further examine this, we quantified the temporal difference between the modeled AT using the exercise threshold app and the corresponding blood sample in a paired analysis.

In grouped mean analyses, there was no significant difference in timing between the modeled AT and corresponding blood sample (11.47 ± 2.42 mins vs. modeled AT: 11.21 ±  2.80 mins, respectively, p=0.181). At the individual level, the mean absolute difference was 43.2 ± 27.3 s (median 45.0 s [IQR 19.8–62.4]), with 73.7% of samples collected within 60 s and all samples within 90 s of the modeled AT.

Thus, the actual temporal discrepancy was generally smaller than the theoretical maximum suggested by the reviewer and the absolute difference between the two was non-significant- Even so, we agree that blood sampling did not coincide precisely with the modeled AT which may have attenuated transient metabolic changes occurring around the threshold. We have therefore acknowledged this as a potential contributor to the relatively limited metabolic signal observed at AT.

We have added this information into the relevant methodology section for full transparency

‘Among participants with both timing measurements available (n = 19), the modeled AT occurred at 11.47 ± 2.42 min, and the corresponding blood sample was collected at 11.21 ± 2.80 min. There was no evidence of a systematic difference between the modeled AT and blood-sampling times(p = 0.181). At the individual level, the mean absolute timing difference was 43.2 ± 27.3 s (median, 45.0 s; IQR, 19.8–62.4 s). Fourteen of the 19 samples (73.7%) were collected within 60 s of the modeled AT, and all samples were collected within 90 s of the modeled AT.

6. Data preprocessing requires stronger justification

Missing metabolomic values were imputed using a random forest method, followed by quantile normalization. The manuscript should report the percentage of missing values for each metabolite and explain why the values were missing. The authors should also justify the use of quantile normalization in an exercise study, since exercise may cause real global changes in metabolite distributions. A sensitivity analysis without imputation or with an alternative normalization method would improve confidence in the results.

Author response:

Manuscript location: Sections 2.4 and 2.5; Appendices A13–A15; Section 5.2.

We thank the reviewer for highlighting the need to describe the preprocessing more precisely. We re-audited the data and code used to generate the final modelling matrix. The dataset comprised 80 biological samples from 20 participants at four exercise workloads and 164 metabolites, corresponding to 13,120 expected metabolite measurements. Thirty-four values were missing overall (0.259%), and all missing values involved cysteine (34/80 measurements; 42.5%). Cysteine missingness was missing from 6 of 20 samples at rest, 10of 20 at AT, 8 of 20 at peak and 10 of 20 during recovery. The other 163 metabolites were complete.

The missing cysteine measurements were already absent from the targeted analytical export. The available metadata did not allow us to determine whether individual missing values resulted from concentrations below the detection or quantification threshold, unsuccessful peak integration, or insufficient confidence in metabolite identification. We have therefore avoided assigning a specific cause that cannot be verified.

Re-examination of the final modelling matrix confirmed that the missing cysteine measurements had not been imputed. They remained missing, and the cysteine-specific linear mixed-effects models were fitted using the available observations. Therefore, the primary analysis already represents an available-case analysis without imputation, and a separate no-imputation sensitivity analysis was not required. The previous statement that random-forest imputation had been applied was incorrect and has been removed from the manuscript.

The metabolomic data were processed using the established laboratory preprocessing pipeline. Following normalization to the relevant internal standards and quality-control filtering, the retained metabolite matrix was quantile normalized across samples to harmonize the empirical abundance distributions. Quantile normalization was not used as a direct batch-effect correction. Nevertheless, because this distribution-based procedure makes the overall abundance distributions more similar across samples, we acknowledge that it could potentially attenuate genuine global changes in metabolite abundance induced by exercise. We have clarified the purpose of quantile normalization in the Methods and acknowledged this potential limitation in Section 5.2.

An alternative-normalization sensitivity analysis was not performed because quantile normalization had been applied as part of the established analytical preprocessing pipeline before statistical modelling.

7. Pathway enrichment analysis

The pathway analysis is based on assigning metabolites to nine broad and mutually exclusive metabolic categories. This approach may oversimplify biological pathways because one metabolite can participate in several metabolic processes.

The authors should explain why this classification method was selected instead of a standard metabolomic pathway enrichment method. The pathway findings should also be described cautiously.

Author response:

Manuscript location: Sections 2.5 and 3.6; We thank the reviewer for this important comment. We selected the nine-category classification to provide a simple, interpretable overview of the measured metabolites. Because the dataset included targeted metabolites and tentatively annotated untargeted features, many metabolites could not be mapped reliably to standardized KEGG pathway identifiers. We therefore used a predefined, manually reviewed classification based on established biochemical functions.

Each metabolite was assigned to one of nine broad categories. The assignments generated by the classification code were manually checked against established biochemical information and KEGG annotations where available. We acknowledge that this simplified classification does not capture the fact that an individual metabolite can participate in multiple biological pathways.

Over-representation was assessed using a one-sided hypergeometric test, with BH-FDR correction applied separately within each workload contrast. The analysis identified over-representation of the TCA-cycle category at Peak and Recovery and the nucleotide-metabolism category at Peak. However, these findings should be interpreted as over-representation within our broad, study-specific classification and not as evidence of pathway activation or altered metabolic flux.

We have clarified the rationale and methodology in the revised manuscript and have moderated the interpretation of these results. The term “pathway enrichment” has been replaced, where appropriate, with “metabolic-category over-representation.”

8. Lack of independent validation

The study is a secondary exploratory analysis, and the findings were not confirmed in an independent cohort. Therefore, the reported associations between VO₂peak, lactic acid and malic acid should be considered hypothesis-generating.

The clinical relevance of metabolomic profiling is currently overstated. The conclusion should emphasize the need for confirmation in larger and independent populations.

Author response:

Manuscript location: Abstract; Sections 5.2, 5.3, and 6; clinical-impact highlight.

We agree with this as a point and so have added a sentence stating the exploratory nature of this work throughout the manuscript as well as in the future research section: ‘The results of this study should be considered as hypothesis generating, and their potential clinical relevance requires confirmation in larger, independent populations across different exercise stimuli.’

9. The authors may add a short paragraph about the effects of different exercise modalities on circulating metabolic markers in people with diabetes. This information would fit well in the Introduction, after the paragraph describing previous exercise metabolomics studies. It may help show that metabolic responses can depend on the type and intensity of exercise, although the available evidence is still limited.

The following reference may be useful:

Ravanbakhsh S, Pournemati P, Soori R. The Effect of Aerobic Exercise and Yoga on Serum Fetuin-A Levels in Obese Women with Type 2 Diabetes. International Journal of Sport Studies for Health. 2026;9(2):1–9. https://doi.org/10.61838/kman.intjssh.4809

Author response:

Manuscript action: No additional citation or paragraph was added because the suggested chronic type 2 diabetes intervention is outside the scope of this acute type 1 diabetes study.

We would like to thank the reviewer for this suggestion and agree that the metabolic responses to exercise may vary according to exercise modality and intensity. We carefully considered the above reference but felt that it falls outside of the scope of the present study. The suggested reference was to examine the effects of chronic aerobic exercise and yoga on serum fetuin-A, a glycoprotein and hepatokine rather than a metabolite, concentrations in women with type 2 diabetes. In contrast, the present study. In contrast, the present study investigated acute changes in the circulating metabolome across increasing exercise intensities during a fixed exercise modality (cycle ergometry) in adults with type 1 diabetes. Given these substantial differences in interventional exposure, population and measured outcomes, we have elected not to include the suggested reference or expand the introduction in this direction. We have referenced, where possible, the limited evidence on the metabolomic responses to acute physical exercise in people with type 1 diabetes in both he introduction and discussion sections.

10. ⁠The authors may discuss the role of exercise intensity in changing circulating hormones and metabolic biomarkers. This point could be added in the Discussion, where the intensity-dependent metabolic responses are interpreted. It would support the idea that high-intensity exercise may produce different metabolic responses compared with lower-intensity exercise. However, the differences in population and study design should also be mentioned.

The following reference may be considered:

Almasoodi A, Amirsasan R, Vakili J, Khani M, Alwawi H. Effect of Eight Weeks of High-Intensity Interval Training on Serum Asprosin Levels and Body Composition in Overweight and Obese Men. International Journal of Sport Studies for Health. 2026;9(3):1–9.

Author response:

Manuscript location: Section 5.3 (Future research).

Again, we wish to thank the reviewer for this suggestion and agree with their statements of the importance of exercise characteristics in influencing metabolic responses. Indeed, the intensity-dependent nature of the metabolomic response to acute exercise is a central finding of the present study and is discussed in detail throughout the article. We have also referred to previous type 1 diabetes specific literature on this topic, including research by both Brugnara et al. and Bally et al., which directly investigated the influence of exercise intensity and/or modality on circulating metabolic responses. We carefully considered the suggested reference; however, we felt that it does not directly align with the purpose of the present investigation. Specifically, the referenced study examined changes in serum asprosin following eight weeks of high-intensity interval training in overweight and obese men, whereas our study investigated acute changes in the circulating metabolome across progressively increasing exercise intensities during a single graded exercise test in adults with type 1 diabetes. Thus, the studies differ substantially in population, exercise exposure (chronic training versus an acute exercise bout), and measured outcomes (a specific circulating hormone versus broad metabolomic profiling). With this in mind, we felt that the existing studies by Brugnara et al. and Bally et al. provide more directly relevant evidence for the specific context of our investigation and have therefore not included the suggested reference. Even so, we acknowledge that more work ought to be done in this area, and so have added a sentence to support this in the future research section using stimuli to encompass different exercise modalities and intensities: ’The results of this study should be considered as hypothesis generating, and their potential clinical relevance requires confirmation in larger, independent populations across different exercise stimuli’

 

Author Response File: Author Response.docx

Reviewer 2 Report

Comments and Suggestions for Authors

McCarthy et al investigated here the metabolic responses in plasma and cardiorespiratory fitness indexes of type-1 diabetic adults upon maximal/submaximal exercises.

The reviewed proposal demonstrates strong scientific merit, with a clear research question and a well‑structured methodology (GC-MS metabolomic). The authors employ appropriate analytical techniques and integrate quantitative and qualitative approaches, enhancing the robustness of their framework. However, concerns arise regarding the limited sample size (the authors are aware of this fact), which may restrict the generalizability of findings. The absence of a healthy control group further complicates interpretation, as a comparison to glycemic-controlled individuals upon exhaustive exercise is almost intuitive here. Additionally, the gender distribution among participants is insufficiently addressed, raising questions about potential bias in outcomes (e.g. iron metabolism in men versus women, especially under strenuous exercises). While the study is commendable in design, metodology, and execution, addressing these limitations in future work (as the authors indeed mentioned here) would significantly strengthen its credibility and impact within the scientific community. 

Nevertheless, there are still MINOR questions that have to be answered before the MS acceptance for publication in Metabolites/MDPI:

(i) How similar the exhaustive cycle ergometer test was to the classic Wingate test? Please insert comments or justify the opted exercise protocol;

(ii) The first 03 paragraphs in Materials & Methods were mistakenly copied from Authors Guide. Please delete them.

(iii) Table 2: please format the Font size within lines/columns for a whole-number presentation of values in just one line of each table cell, e.g. p-value <0.0001, etc. 

(iv) Discussion: please improve the metabolic discussion/comparison of the observed data in type-1 diabetic individuals with the available results in healthy adults, despite the differences of exercise or sampling protocols. Maybe an additional figure (metabolic map) would help readers to deepen the discussion here;

(v) Future Research: please include a brief comment about redox metabolism, especially in terms of oxidative stress and antioxidant defenses. Your results here strongly aligns with the hypothesis that exhaustive exercise activate purine catabolism (due to severe changes in ATP-AMP ratios) that culminate in a late-onset accumulation of URIC ACID in plasma, which is a powerful iron-chelating agent that protects blood, cells, and tissues against further oxidative stress promoted by labile iron ions during the post-exercise recovery period. 

Wiecek M, Maciejczyk M, Szymura J, Kantorowicz M, Szygula Z. Impact of single anaerobic exercise on delayed activation of endothelial xanthine oxidase in men and women. Redox Rep. 2017 Nov;22(6):367-376. doi: 10.1080/13510002.2016.1238991. Epub 2016 Oct 7. PMID: 27715604; PMCID: PMC6837425.

Souza-Junior T, Lorenço-Lima L, Ganini D, Vardaris C, Polotow T, Barros M. Delayed uric Acid accumulation in plasma provides additional anti-oxidant protection against iron-triggered oxidative stress after a wingate test. Biol Sport. 2014 Dec;31(4):271-6. doi: 10.5604/20831862.1120934. Epub 2014 Sep 12. PMID: 25435669; PMCID: PMC4203843.

 

Author Response

Reviewer 2 - Round 1

McCarthy et al investigated here the metabolic responses in plasma and cardiorespiratory fitness indexes of type-1 diabetic adults upon maximal/submaximal exercises.

The reviewed proposal demonstrates strong scientific merit, with a clear research question and a well‑structured methodology (GC-MS metabolomic). The authors employ appropriate analytical techniques and integrate quantitative and qualitative approaches, enhancing the robustness of their framework. However, concerns arise regarding the limited sample size (the authors are aware of this fact), which may restrict the generalizability of findings. The absence of a healthy control group further complicates interpretation, as a comparison to glycemic-controlled individuals upon exhaustive exercise is almost intuitive here. Additionally, the gender distribution among participants is insufficiently addressed, raising questions about potential bias in outcomes (e.g. iron metabolism in men versus women, especially under strenuous exercises). While the study is commendable in design, metodology, and execution, addressing these limitations in future work (as the authors indeed mentioned here) would significantly strengthen its credibility and impact within the scientific community. 

Nevertheless, there are still MINOR questions that have to be answered before the MS acceptance for publication in Metabolites/MDPI:

• How similar the exhaustive cycle ergometer test was to the classic Wingate test? Please insert comments or justify the opted exercise protocol;

Author response:

Manuscript location: Section 2.3. The existing graded CPET description now makes the distinction from a Wingate test explicit in this response.

We would like to clarify that the exercise protocol used in the present study was not a Wingate test and differs substantially in both design and physiological purpose. The Wingate test is a short-duration, supramaximal all-out exercise test primarily used to assess anaerobic power and capacity. In contrast, participants in the present study completed a graded incremental cardiopulmonary exercise test on a cycle ergometer, with progressively increasing workload until volitional exhaustion and continuous breath-by-breath gas-exchange measurement. A full methodological breakdown of this is in section 2.3 of the article method section. This protocol was selected to determine V̇O_₂peak and to characterize metabolic responses across progressively increasing physiological exercise intensities, including the anaerobic threshold and peak exercise. A supplementary figure has now been added to clearly illustrate the methodological procedures of the exercise test.

• The first 03 paragraphs in Materials & Methods were mistakenly copied from Authors Guide. Please delete them.

Author response:

Manuscript location: Opening of Section 2; the template instructions were deleted.

 

Thank you very much for picking up on this. We have since deleted the section within the revised manuscript.

• Table 2: please format the Font size within lines/columns for a whole-number presentation of values in just one line of each table cell, e.g. p-value <0.0001, etc. 

 

Author response:

Manuscript location: Section 3.4, Table 2. The column widths, font size, numerical alignment, and headers were reformatted so numerical values remain on one line.

Thank you for catching this. All formatting has been rechecked for consistency in sizing.

• Discussion: please improve the metabolic discussion/comparison of the observed data in type-1 diabetic individuals with the available results in healthy adults, despite the differences of exercise or sampling protocols. Maybe an additional figure (metabolic map) would help readers to deepen the discussion here;

Author response:

Manuscript location: Section 4 (Discussion). No additional metabolic-map figure was added.

We have expanded the Discussion to provide a clearer comparison between the exercise-induced metabolic responses observed in the present type 1 diabetes cohort and those previously reported in the existing literature, including changes in circulating glycolysis-related metabolites, TCA-cycle intermediates and purine metabolites. We have also strengthened the comparison of the CRF-associated metabolic responses with those reported in healthy adults, while acknowledging differences in study methodologies. Given these additions, we felt that an additional metabolic map was not necessary and that the expanded discussion provides a more appropriate interpretation of the observed circulating metabolite responses.

‘The exercise-induced metabolic responses observed in the present study broadly align with those reported in the general population, encompassing changes in circulating glycolysis-related metabolites, TCA-cycle intermediates and purine metabolites [29, 30]. In keeping with previous studies in healthy individuals [31–34], adults with type 1 diabetes with higher CRF demonstrated greater exercise-induced changes in selected metabolites involved in central energy metabolism including L-(+)-lactic acid and malic acid. ‘

 

• Future Research: please include a brief comment about redox metabolism, especially in terms of oxidative stress and antioxidant defenses. Your results here strongly aligns with the hypothesis that exhaustive exercise activate purine catabolism (due to severe changes in ATP-AMP ratios) that culminate in a late-onset accumulation of URIC ACID in plasma, which is a powerful iron-chelating agent that protects blood, cells, and tissues against further oxidative stress promoted by labile iron ions during the post-exercise recovery period. 

Wiecek M, Maciejczyk M, Szymura J, Kantorowicz M, Szygula Z. Impact of single anaerobic exercise on delayed activation of endothelial xanthine oxidase in men and women. Redox Rep. 2017 Nov;22(6):367-376. doi: 10.1080/13510002.2016.1238991. Epub 2016 Oct 7. PMID: 27715604; PMCID: PMC6837425.

Souza-Junior T, Lorenço-Lima L, Ganini D, Vardaris C, Polotow T, Barros M. Delayed uric Acid accumulation in plasma provides additional anti-oxidant protection against iron-triggered oxidative stress after a wingate test. Biol Sport. 2014 Dec;31(4):271-6. doi: 10.5604/20831862.1120934. Epub 2014 Sep 12. PMID: 25435669; PMCID: PMC4203843.

Author response:

Manuscript location: Section 5.3 (Future research).

Thank you for raising this as a matter for contemplation and for providing the suggested references. We see that the observed changes in purine-related metabolites during peak exercise and recovery could serve as an interesting potential link between exercise-induced purine metabolism and redox homeostasis. However, the present study did not directly assess ATP/AMP turnover, xanthine oxidase activity, oxidative stress, labile iron, or antioxidant capacity. We therefore believe that a direct mechanistic interpretation of our findings in this context would extend beyond the measurements performed. Nevertheless, we agree that this represents an important avenue for further investigation and have added a brief comment to the Future Research section highlighting the potential relationship between exercise-induced purine metabolism and redox homeostasis during and after exercise.

‘Integrating serial exercise metabolomics with direct assessments of skeletal muscle mitochondrial function, in vivo metabolic flux, purine turnover, oxidative stress and antioxidant defences, alongside longitudinal changes in CRF following training would further elucidate the biological mechanisms underlying exercise capacity and recovery.’’

 

 

Author Response File: Author Response.docx

Reviewer 3 Report

Comments and Suggestions for Authors

1. Introduction

C1. The background is accurate but relatively thin for the journal's expectations.

The Introduction (three short paragraphs) correctly identifies the gap the study addresses — prior T1D exercise-metabolomics work has relied on fixed-intensity protocols or rest/pre–post sampling (refs [15–18]), precluding characterization of how the metabolome evolves across intensity. However, none of these four directly relevant precedent studies are named or briefly contrasted until the Discussion section. Naming at least the two or three most relevant prior T1D exercise metabolomics studies in the Introduction (even in one sentence each) and stating an explicit a priori hypothesis (e.g., "we hypothesized that metabolic perturbation would scale with exercise intensity and that its magnitude would be positively associated with V̇O2peak") would strengthen the framing and allow the reader to evaluate the novelty claim earlier.

C2. Please double-check the reference pairing for the sentence on chronic hyperglycemia and diabetes duration.

Lines 63–66: "this impairment has been primarily linked to chronic hyperglycemia and long-standing disease duration [7–9]" cites references 7–9. Reference 9 (McCarthy et al., blood glucose responses during CPET) appears, from its title, to concern acute glycemic responses during exercise testing rather than chronic hyperglycemia or disease duration as determinants of CRF impairment. Please confirm that this citation is correctly placed or cite the specific finding being referenced.

2. Materials and Methods

C3 (major). Unedited MDPI template instructions remain in Section 2, and the requested study design schematic is missing.

The opening paragraph of Section 2 ("The main experimental stages should be explained... Please provide a scheme summarizing the experimental design, including the time needed to complete every stage") is boilerplate reviewer guidance from the MDPI protocol template and was evidently not deleted before submission. Beyond needing removal, the substance of the instruction was also not followed: the manuscript has no schematic figure showing the timeline of the CPET (3-min rest → 3-min warm-up → 1-min incremental stages → 3-min active + 3-min passive recovery) with the rest/AT/peak/recovery blood-draw points and gas exchange sampling overlaid. Given that the entire analytical structure of this study depends on this four-workload framework, a simple timeline figure would materially improve accessibility for readers unfamiliar with CPET protocols and should be added as Figure 1 (or an early supplementary figure).

C4. Circulating insulin exposure and residual β-cell function were not reported, despite being later identified as explanatory factors.

In insulin-treated T1D, circulating insulin concentration is a first-order determinant of lipolysis, hepatic glucose output, and substrate selection during exercise, and the Discussion explicitly cites Taylor et al. [18] to argue that residual β-cell function shapes metabolomic responses to exercise. However, the present study neither measured plasma insulin/C-peptide around the test nor reported whether/how insulin dosing was adjusted for exercise (only that participants "maintained their usual insulin regimen"). Please report what is available (e.g., total daily insulin dose, time since last bolus, any pump suspension/reduction protocol used during CPET) and add this as an explicit limitation if direct insulin measurements were not collected.

C5. The pre-exercise meal composition was not standardized or recorded; please state this as a limitation.

Appendix A3 notes only that a meal was consumed ≥2 h before testing and that "the composition of the pre-exercise meal was not standardized." Because macronutrient composition (particularly carbohydrate/fat content) measurably shifts baseline and exercise-responsive lipid- and glucose-related metabolites, this should be explicitly added to Section 4.2 (Limitations) alongside the other acknowledged sources of residual confounding, rather than only appearing in the appendix.

C6. The anaerobic-threshold sample was not drawn at the modeled AT, which may partly explain the sparse AT findings.

AT was identified from continuous breath-by-breath gas-exchange data via the Exercise Threshold App, but venous blood was only drawn every 3 min, and the "AT" sample used for analysis was simply "the exercise sample collected closest in time" to the modeled threshold (Appendix A6). This is a reasonable pragmatic compromise, but the resulting timing offset (potentially up to ~1.5 min) could dilute the genuine AT-related signal and may contribute to the finding that only 2/164 metabolites changed significantly at AT versus 16 and 21 at peak/recovery. Please add a sentence acknowledging this as a possible contributor to the limited AT signal (in addition to or instead of interpreting it purely as evidence of a true physiological threshold effect).

C7. Justify the use of maximum likelihood rather than restricted maximum likelihood (REML) for variance component estimation with n = 20.

Appendix A17 states that models were "fitted using maximum likelihood rather than restricted maximum likelihood." ML is the correct choice if the intention is likelihood-based model comparison, but ML variance component estimates are known to be more downward-biased than REML in small samples. Since no formal likelihood-ratio model comparisons appear to be reported for the fixed effects presented (Satterthwaite/Wald tests are used instead), please clarify the rationale for ML over REML, or consider whether REML would be more appropriate for the reported inferential statistics.

C8. Consider the statistical interpretability of single- or double-member metabolic categories in the overrepresentation analysis.

As shown in Supplementary Table S6, the "Pyruvate metabolism" category contains only one metabolite in the 164-metabolite background universe (K = 1) and "Nucleotide metabolism" contains only two (K = 2). A hypergeometric enrichment test on a category of size 1 has essentially no meaningful null distribution and will mechanically yield either a trivial or an extreme result, depending on whether that single metabolite is significant. Please either note this limitation explicitly or consider merging very small categories (e.g., folding pyruvate metabolism into glycolysis/lactate, as done biochemically) before running the over-representation analysis.

C9. The processed metabolomics data will be deposited in a public repository.

The Data Availability Statement appropriately restricts participant-level clinical data, given the Danish data protection requirements. However, the retained template text in Section 2 also asks authors to "specify where [large] data are deposited and provide the relevant accession numbers," and many metabolomics journals/communities (following the Metabolomics Standards Initiative) expect de-identified processed data (e.g., the 164-metabolite feature matrix) to be deposited in a repository such as MetaboLights or the Metabolomics Workbench, independent of the participant-level clinical datasets. Please clarify whether this is planned, or explain why it is not applicable.

3. Results

C10. Please clarify the selection rule used for Table 2.

Table 2 is titled "Selected significant metabolite associations" but includes only 11 of the 16 significant peak exercise and a subset of the 21 significant recovery associations reported in the text (the remainder — e.g., 2-hydroxybutyric acid, dihydrouracil, glutamic acid, glycerol, palmitic acid, stearic acid at peak; alpha-tocopherol, alanine, aminomalonic acid, cholesterol, dodecanoic acid, ketovaline, L-(+)-erythrulose, L-(−)-sorbose at recovery — appear only in Figure 3/Supplementary Figure S8 and Supplementary Table S2A). This is a legitimate editorial choice for a main-text table, but a one-line footnote stating the selection criterion (e.g., "restricted to core glycolytic/TCA-cycle/purine metabolites; the complete list of 39 significant associations is provided in Supplementary Table S2A") would prevent readers from perceiving this as selective reporting.

C11. The correlation analysis in Figure 2 / Section 3.5 is hypothesis-generating due to selection-then-test.

Metabolites entering the exploratory Pearson correlation analysis were pre-selected on the basis of having a BH-FDR-significant Model 1 effect for that same workload contrast (Appendix A22), and correlation p-values were not corrected for multiple testing (explicitly and appropriately stated by the authors). Because the same significant change metabolites were then tested against several correlated exercise/fitness variables (V̇O2peak, peak power in W/kg, time to AT, time to peak), there is a meaningful risk of overinterpreting the number and pattern of asterisks in Figure 2. The authors already call this analysis "exploratory," which is good practice; I would simply suggest stating explicitly, once, that findings from Section 3.5 should be considered hypothesis-generating rather than confirmatory, to pre-empt over-reading by less statistically minded readers.

4. Discussion and Conclusions

C12. V̇O2peak and peak power were not independent in this design; the fitness-response interpretation should be qualified accordingly.

The Discussion attributes the larger lactate/malate excursions in fitter participants to a "greater capacity to mount the metabolic response required to sustain maximal exercise." Because fitter participants also reach a higher absolute peak workload and exercise for longer (as shown by the correlations with peak power and exercise duration in Section 3.5, which parallel those with V̇O2peak), part of the larger metabolite excursion likely reflects the higher absolute mechanical/metabolic demand attained at "peak" rather than a fitness-specific difference in metabolic responsiveness per se at a matched relative or absolute workload. The manuscript partially acknowledges this ("fitter individuals can sustain higher external workloads"), but a more explicit statement that V̇O2peak and peak power cannot be fully disentangled in this design (both reflect the same underlying maximal-effort test) would improve precision.

C13. The translational potential paragraph cites the UK Biobank study.

The penultimate Discussion paragraph cites a UK Biobank study (ref [34]) of metabolomic/proteomic CRF signatures predicting mortality and disease risk as support for the translational potential of the present findings. That study was conducted in a large, largely non-diabetic general population using a different metabolomic platform and metabolite panel; therefore, its relevance to a 20-person, T1D-specific, GC-MS-based exercise metabolome study is illustrative at best. Consider softening the linkage (e.g., "this raises the possibility that..." rather than implying direct support) to avoid overstating generalizability.

C14. The comparatively homogeneous" cohort description should be reconciled with the wide ranges reported in Table 1.

Section 4.2 (Limitations) describes the cohort as "comparatively homogeneous with respect to demographic and diabetes-related characteristics," yet Table 1 shows age 26–72 years, diabetes duration 12–57 years, and V̇O2peak 12.1–45.0 mL·kg⁻¹·min⁻¹ — a nearly four-fold range in fitness and a 45-year range in disease duration. This heterogeneity is actually a strength for the V̇O2peak-interaction analysis (Model 5) and should not be understated; suggest rephrasing to something like "participants were homogeneous in that all had long-standing, pump-treated, well-characterized T1D, but spanned a wide range of age, diabetes duration, and fitness" so the claim is not self-contradicted by the reader's own inspection of Table 1.

5. Figures, Tables and Presentation

C15. Duplicate section numbering.

"4. Discussion" is followed later by "4. Strengths, limitations, and perspectives for future research" and then "5. Conclusion" — the second "4." should be "5." and "Conclusion" should become "6." Please renumber the references before typesetting.

C16. Figures and tables are otherwise clear, high resolution, and internally consistent.

I cross-checked Figures 1–3 and Supplementary Figures S1–S8 against the underlying values in Supplementary Tables S1B, S2A, and S6 (e.g., TCA cycle enrichment q = 0.0027 at peak / 0.0085 at recovery; nucleotide metabolism q = 0.0404 at peak; 39 significant workload associations; four significant V̇O2peak × workload interactions) and found no discrepancies. The forest plot (Figure S2), volcano plots (Figures S3–S5), and heatmap (Figure S1) are well-designed and legible. Minor suggestion: Figure 3 and Supplementary Figures S7–S8 would benefit from a shared, fixed y-axis scale across the three workload panels (AT, peak, recovery) so that readers can visually compare the magnitude of change across intensities at a glance rather than needing to consult three separately scaled figures.

6. Language and Style

C17 (minor). Inconsistent American/British spelling conventions.

The manuscript predominantly uses American English ("characterized," "randomized," "analyzed," "hemoglobin," "-ize/-ization" endings) but intermittently switches to British forms ("normalisation," p. 4; "glycaemia"/"glycaemic," Discussion; "Whilst," Introduction). This does not impede comprehension, but should be standardized to a single variety during copy-editing.

Summary of Required Actions

1. The remaining MDPI template instructions were removed from Section 2, and a study-design/timeline schematic figure (C3) was added.
2. The absence of insulin/C-peptide/β-cell function data and pre-exercise meal standardization should be reported or explicitly discussed as limitations (C4, C5).
3. Add a methodological caveat regarding the AT blood-sampling timing resolution relative to the modeled threshold (C6).
4. Justify ML vs. REML model fitting, given the small sample size (C7).
5. Address the very small metabolic category sizes used in the over-representation analysis (C8).
6. Clarify data deposition plans for the processed metabolomics feature matrix (C9).
7. Add a footnote clarifying the selection rule for Table 2 (C10).
8. Please correct the duplicate "4." section numbering (C15).
9. Standardize English spelling conventions throughout (C17).

• The remaining comments (C1, C2, C11–C14, C16) are suggested but are non-essential improvements.

Author Response

Reviewer 3 Round 1

1.Introduction

C1.The background is accurate but relatively thin for the journal's expectations.

The Introduction (three short paragraphs) correctly identifies the gap the study addresses — prior T1D exercise-metabolomics work has relied on fixed-intensity protocols or rest/pre–post sampling (refs [15–18]), precluding characterization of how the metabolome evolves across intensity. However, none of these four directly relevant precedent studies are named or briefly contrasted until the Discussion section. Naming at least the two or three most relevant prior T1D exercise metabolomics studies in the Introduction (even in one sentence each) and stating an explicit a priori hypothesis (e.g., "we hypothesized that metabolic perturbation would scale with exercise intensity and that its magnitude would be positively associated with V̇O2peak") would strengthen the framing and allow the reader to evaluate the novelty claim earlier.

Author response:

Manuscript location: Section 1 (Introduction).

We thank the reviewer for this suggestion and believe it has much improved the rationale. As such, we have reworked the introduction to expand upon the existing research base in people with type 1 diabetes. Specifically, we have highlighted the evidence showing differences in the exercise metabolome in those with versus without type 1 diabetes as well as according to variations in exercise modality, glycemic concentrations and residual beta cell function in those with type 1 diabetes. The new text is as follows.

‘Recent developments in mass spectrometry-based metabolomics enable comprehensive in vivo profiling of these dynamic bioenergetic responses, providing physiological insight beyond that offered by conventional clinical markers [12–14]. Though limited in number, previous studies applying metabolomics around acute physical exercise in those with type 1 diabetes have demonstrated distinct metabolomic responses compared with healthy controls [15], as well as differences according to exercise modality [16], glycemic conditions [17] and residual beta-cell function [18].’

We feel this has highlighted the gap in literature and the resultant need for the present study. Thank you for your guidance. Given the exploratory secondary nature of the analysis, however, we have retained the relationship between exercise-induced metabolic responses and V̇O_₂peak as an exploratory objective rather than retrospectively describing this analysis as an a priori hypothesis.

C2. Please double-check the reference pairing for the sentence on chronic hyperglycemia and diabetes duration.

Lines 63–66: "this impairment has been primarily linked to chronic hyperglycemia and long-standing disease duration [7–9]" cites references 7–9. Reference 9 (McCarthy et al., blood glucose responses during CPET) appears, from its title, to concern acute glycemic responses during exercise testing rather than chronic hyperglycemia or disease duration as determinants of CRF impairment. Please confirm that this citation is correctly placed or cite the specific finding being referenced.

Author response:

Manuscript location: Section 1 (Introduction), first paragraph.

 

This is a very fair point. The original reference was conducted by the research group, and the supportive statement is stated clearly in the discussion of that article ‘ Our results demonstrated that HbA_1c was a significant, independent, and contributory factor combined with gender, age, and body mass in explaining 53.4% of the variance in peak power in our cohort.’ However, we recognize that the included article is perhaps not the optimal choice to support the current study rationale and so have instead replaced it with references that definitively support claims as a point of primary focus with more accuracy in our wording. The rewritten section is as follows:

Cardiorespiratory fitness (CRF) is recognized as one of the strongest predictors of all-cause and cardiovascular mortality [1–4]. Compared with healthy individuals, adults with type 1 diabetes exhibit significantly lower CRF [5–7]. Although multifactorial, this impairment has been associated with chronic hyperglycemia and the presence of diabetes related complications [7–10]. Whilst informative, these long-term clinical markers provide limited insight into the metabolic processes underlying CRF.

 

Again, thank you for this suggestion as we believe it has improved the precision of our writing.

2.Materials and Methods

C3 (major). Unedited MDPI template instructions remain in Section 2, and the requested study design schematic is missing.

The opening paragraph of Section 2 ("The main experimental stages should be explained... Please provide a scheme summarizing the experimental design, including the time needed to complete every stage") is boilerplate reviewer guidance from the MDPI protocol template and was evidently not deleted before submission. Beyond needing removal, the substance of the instruction was also not followed: the manuscript has no schematic figure showing the timeline of the CPET (3-min rest → 3-min warm-up → 1-min incremental stages → 3-min active + 3-min passive recovery) with the rest/AT/peak/recovery blood-draw points and gas exchange sampling overlaid. Given that the entire analytical structure of this study depends on this four-workload framework, a simple timeline figure would materially improve accessibility for readers unfamiliar with CPET protocols and should be added as Figure 1 (or an early supplementary figure).

Author response:

Manuscript location: Opening of Section 2 and Section 2.3 as Supplementary Figure S1

Thank you very much for picking up on this. We have since deleted the protocol template section within the revised manuscript. Furthermore, we have produced a supplementary figure S1 to illustrate the methodological protocol for a CPET in alignment with the reviewer’s suggestion.

C4. Circulating insulin exposure and residual β-cell function were not reported, despite being later identified as explanatory factors.

In insulin-treated T1D, circulating insulin concentration is a first-order determinant of lipolysis, hepatic glucose output, and substrate selection during exercise, and the Discussion explicitly cites Taylor et al. [18] to argue that residual β-cell function shapes metabolomic responses to exercise. However, the present study neither measured plasma insulin/C-peptide around the test nor reported whether/how insulin dosing was adjusted for exercise (only that participants "maintained their usual insulin regimen"). Please report what is available (e.g., total daily insulin dose, time since last bolus, any pump suspension/reduction protocol used during CPET) and add this as an explicit limitation if direct insulin measurements were not collected.

Author response:

Manuscript location: Section 5.2 (Limitations).

 

We thank the reviewer for raising this totally fair response which we agree needs addressing in the limitations section. Participants were only instructed to avoid consuming a meal with their accompanying insulin bolus dose for ≥2 hours prior to their scheduled appointment. The exercise test usually started 45 to 60 minutes after arriving to the laboratory, depending on the initial pre-exercise screening procedures and medical consolations. Hence, by the time they begin cycling, it is highly probable that the peak action of meal-time insulin had subsided. Unfortunately, detailed information regarding individual pump suspension and/or insulin-reduction strategies was not available for this secondary analysis. Furthermore, we did not measure circulating insulin concentrations before and/or during the test. C-peptide concentrations were available in 13 of the 21 participants (62%); however, concentrations were very low in the majority of these participants, with several values (7 out of the 13) below the assay detection limit. Given the small number of available measurements and limited variability in C-peptide concentrations, meaningful subgroup or association analyses according to residual endogenous insulin secretion were not considered feasible. We have therefore acknowledged the inability to comprehensively assess the influence of exogenous and endogenous insulin exposure as a limitation of the study.

‘Despite implementing a highly standardized pre-testing protocol that controlled  the timing of dietary intake and insulin administration as well as exposure to recent hypoglycemia and physical activity, residual confounding by these factors could not be fully excluded. Furthermore, circulating insulin concentrations were not measured, and C-peptide concentrations were available in only 13 (62%) of participants with very low concentrations, limiting our ability to assess the influence of exogenous and residual endogenous insulin exposure on the observed exercise-induced metabolomic responses.’

C5. The pre-exercise meal composition was not standardized or recorded; please state this as a limitation.

Appendix A3 notes only that a meal was consumed ≥2 h before testing and that "the composition of the pre-exercise meal was not standardized." Because macronutrient composition (particularly carbohydrate/fat content) measurably shifts baseline and exercise-responsive lipid- and glucose-related metabolites, this should be explicitly added to Section 4.2 (Limitations) alongside the other acknowledged sources of residual confounding, rather than only appearing in the appendix.

Author response:

Manuscript location: Section 5.2 (Limitations), with supporting detail retained in Appendix A3.

 

Yes, we completely agree with this point. Thus, we have highlighted this in our limitation section, drawing upon the fact that only the timing of dietary intake was standardized. ‘‘Despite implementing a highly standardized pre-testing protocol that controlled  the timing of dietary intake and insulin administration as well as exposure to recent hypoglycemia and physical activity, residual confounding by these factors could not be fully excluded.’

C6. The anaerobic-threshold sample was not drawn at the modeled AT, which may partly explain the sparse AT findings.

AT was identified from continuous breath-by-breath gas-exchange data via the Exercise Threshold App, but venous blood was only drawn every 3 min, and the "AT" sample used for analysis was simply "the exercise sample collected closest in time" to the modeled threshold (Appendix A6). This is a reasonable pragmatic compromise, but the resulting timing offset (potentially up to ~1.5 min) could dilute the genuine AT-related signal and may contribute to the finding that only 2/164 metabolites changed significantly at AT versus 16 and 21 at peak/recovery. Please add a sentence acknowledging this as a possible contributor to the limited AT signal (in addition to or instead of interpreting it purely as evidence of a true physiological threshold effect).

Author response:

Manuscript location: Section 2.3

We thank the reviewer for raising this important methodological consideration. To further examine this, we quantified the temporal difference between the modeled AT using the exercise threshold app and the corresponding blood sample in a paired analysis.

In grouped mean analyses, there was no significant difference in timing between the modeled AT and corresponding blood sample (11.47 ± 2.42 mins vs. modeled AT: 11.21 ±  2.80 mins, respectively, p=0.181). At the individual level, the mean absolute difference was 43.2 ± 27.3 s (median 45.0 s [IQR 19.8–62.4]), with 73.7% of samples collected within 60 s and all samples within 90 s of the modeled AT.

Thus, the actual temporal discrepancy was generally smaller than the theoretical maximum suggested by the reviewer and the absolute difference between the two was non-significant- Even so, we agree that blood sampling did not coincide precisely with the modeled AT which may have attenuated transient metabolic changes occurring around the threshold. We have therefore acknowledged this as a potential contributor to the relatively limited metabolic signal observed at AT.

We have added this information into the relevant methodology section for full transparency

‘Among participants with both timing measurements available (n = 19), the modeled AT occurred at 11.47 ± 2.42 min and the corresponding blood sample was collected at 11.21 ± 2.80 min, with no significant systematic difference between the two times (p = 0.181). However, the mean absolute difference between the modeled AT and the selected blood sample was 43.2 ± 27.3 s (median 45.0 s, IQR 19.8–62.4). Fourteen of the 19 samples (73.7%) were collected within 60 s of the modeled AT, and all samples were collected within 90 s).’’

C7. Justify the use of maximum likelihood rather than restricted maximum likelihood (REML) for variance component estimation with n = 20.

Appendix A17 states that models were "fitted using maximum likelihood rather than restricted maximum likelihood." ML is the correct choice if the intention is likelihood-based model comparison, but ML variance component estimates are known to be more downward-biased than REML in small samples. Since no formal likelihood-ratio model comparisons appear to be reported for the fixed effects presented (Satterthwaite/Wald tests are used instead), please clarify the rationale for ML over REML, or consider whether REML would be more appropriate for the reported inferential statistics.

Author response:

Manuscript location: Appendix A17 (Linear mixed-effects modeling framework).

We thank the reviewer for raising this point. The use of maximum likelihood was intentional. Models 0–5 contained different fixed-effect structures but used the same random-intercept structure. Full maximum likelihood therefore provided a consistent likelihood-based estimation framework across the complete model series. In contrast, the REML criterion depends on the fixed-effect design matrix and is not directly comparable between models containing different fixed effects.

We acknowledge that the manuscript did not report formal likelihood-ratio comparisons between Models 0–5. Nevertheless, the models formed a predefined analytical series, and ML was retained consistently rather than changing the estimation procedure between model specifications.

Importantly, the inferential targets were the fixed-effect coefficients representing workload, clinical-variable and interaction associations. The participant-specific random intercept was included only to account for the correlation among the four repeated measurements from each participant. Random-effect variance estimates were not interpreted, compared or presented as study findings. Fixed-effect inference was based on Wald tests with Satterthwaite-adjusted degrees of freedom.

We recognize that REML may reduce small-sample bias in variance-component estimation. However, REML is not mandatory for fixed-effect inference, and the potential advantage of REML primarily concerns estimation of the random-effect variance, which was not an inferential target in this study. Given the balanced repeated-measures design, the simple random-intercept structure and the consistent use of ML throughout the original analysis pipeline, we retained the ML-based results. We have clarified this rationale and acknowledge the small sample size as a limitation, particularly for the interaction analyses.

 

Linear mixed-effects models were fitted using maximum likelihood to maintain a consistent estimation framework across models with different fixed-effect structures. Participant was included as a random intercept to account for repeated measurements. Random-effect variance was treated as a nuisance parameter and was not interpreted as a study outcome. Fixed-effect p-values were obtained using Satterthwaite-adjusted degrees of freedom.

 

C8. Consider the statistical interpretability of single- or double-member metabolic categories in the overrepresentation analysis.

As shown in Supplementary Table S6, the "Pyruvate metabolism" category contains only one metabolite in the 164-metabolite background universe (K = 1) and "Nucleotide metabolism" contains only two (K = 2). A hypergeometric enrichment test on a category of size 1 has essentially no meaningful null distribution and will mechanically yield either a trivial or an extreme result, depending on whether that single metabolite is significant. Please either note this limitation explicitly or consider merging very small categories (e.g., folding pyruvate metabolism into glycolysis/lactate, as done biochemically) before running the over-representation analysis.

Author response:

Manuscript location: Section 3.6

We thank the reviewer for raising this point. We agree that small categories require careful interpretation, but we respectfully note that the hypergeometric test remains mathematically valid for categories of any finite size because it calculates an exact probability conditional on the measured background universe and the number of significant metabolites. Thus, a category containing one or two metabolites does have a defined null distribution, although its p-value is necessarily highly discrete and strongly influenced by individual metabolites.

We elected not to merge pyruvate metabolism with glycolysis/lactate. The categories were defined before the over-representation calculation and were kept mutually exclusive to prevent double counting. Moreover, pyruvate represents a distinct metabolic junction connecting glycolysis, lactate metabolism, the TCA cycle and amino-acid metabolism. Combining these categories would produce a broader and more heterogeneous category and would change the biological question being tested.

Importantly, the single pyruvate-metabolism metabolite did not produce a significant result after BH-FDR correction at Peak, Recovery or AT. Therefore, no pathway-level conclusion is based on this one-metabolite category.

The nucleotide-metabolism result at Peak was based on both metabolites assigned to this category; hypoxanthine and dihydrouracil being significant. Within the measured 164-metabolite background, this corresponded to a fold enrichment of 10.25, an exact hypergeometric p-value of 0.009 and a BH-FDR-adjusted q-value of 0.040. We therefore retained this result as evidence of over-representation within the measured metabolite panel. Nevertheless, we clarify that it represents coordinated findings for two measured nucleotide-related metabolites and not comprehensive activation of the entire nucleotide-metabolism pathway.

We have added the category sizes to the interpretation and clarified that findings from small categories should be interpreted within the defined measured-metabolite universe.

Results

At peak exercise, we observed significant changes in both hypoxanthine and dihydrouracil, the two measured metabolites classified under nucleotide metabolism. Consequently, this category was significantly over-represented within the measured metabolite panel (fold enrichment = 10.25, p = 0.009, q = 0.040). However, because the category contained only these two metabolites, the finding should not be interpreted as evidence of changes across the entire nucleotide-metabolism pathway.

 

C9.The processed metabolomics data will be deposited in a public repository.

The Data Availability Statement appropriately restricts participant-level clinical data, given the Danish data protection requirements. However, the retained template text in Section 2 also asks authors to "specify where [large] data are deposited and provide the relevant accession numbers," and many metabolomics journals/communities (following the Metabolomics Standards Initiative) expect de-identified processed data (e.g., the 164-metabolite feature matrix) to be deposited in a repository such as MetaboLights or the Metabolomics Workbench, independent of the participant-level clinical datasets. Please clarify whether this is planned, or explain why it is not applicable.

Author response:

Manuscript location: Data Availability Statement.

We thank the reviewer for raising this important point. We agree that public deposition of processed metabolomics data is generally desirable. However, the 164-metabolite matrix contains repeated participant-level measurements collected at four exercise stages. Although direct identifiers can be removed, the metabolomics data remain linked through pseudonymized participant identifiers and cannot be regarded as fully anonymous independently of the associated study data.

The data were collected and processed under the data-governance framework of Steno Diabetes Center Copenhagen and the relevant study approvals. Unrestricted public deposition of the participant-level metabolomics matrix is not covered by the applicable institutional data-governance arrangements and may create a risk of participant re-identification when combined with study characteristics.

Consequently, the processed participant-level metabolomics data will not be deposited in an unrestricted public repository. However, the data may be made available to qualified researchers upon reasonable request. Access will require evaluation and approval by the responsible data controller at Steno Diabetes Center Copenhagen, compliance with the applicable Danish and European data-protection requirements, and establishment of an appropriate data-sharing or data-processing agreement between the institutions.

We have revised the Data Availability Statement accordingly and removed the remaining template wording requesting a public repository accession number.

 

The pseudonymized participant-level metabolomics and clinical data underlying this study are not publicly available because unrestricted public disclosure is not covered by the applicable participant-consent, institutional data-governance and data-protection arrangements. Data may be made available to qualified researchers upon reasonable request to the corresponding author, subject to approval by the responsible data controller at Steno Diabetes Center Copenhagen, fulfilment of applicable ethical and legal requirements, and establishment of an appropriate institutional data-sharing or data-processing agreement. Aggregate results supporting the conclusions of this study are provided in the article and its Supplementary Materials.

 

 

3. Results

C10. Please clarify the selection rule used for Table 2.

Table 2 is titled "Selected significant metabolite associations" but includes only 11 of the 16 significant peak exercise and a subset of the 21 significant recovery associations reported in the text (the remainder — e.g., 2-hydroxybutyric acid, dihydrouracil, glutamic acid, glycerol, palmitic acid, stearic acid at peak; alpha-tocopherol, alanine, aminomalonic acid, cholesterol, dodecanoic acid, ketovaline, L-(+)-erythrulose, L-(−)-sorbose at recovery — appear only in Figure 3/Supplementary Figure S8 and Supplementary Table S2A). This is a legitimate editorial choice for a main-text table, but a one-line footnote stating the selection criterion (e.g., "restricted to core glycolytic/TCA-cycle/purine metabolites; the complete list of 39 significant associations is provided in Supplementary Table S2A") would prevent readers from perceiving this as selective reporting.

Author response:

Manuscript location: Section 3.4

We thank the reviewer for identifying the need to clarify the selection criterion. Table 2 was intentionally designed as a concise main-text summary rather than a complete presentation of all significant Model 1 associations. It includes 11 metabolites selected because they represent the principal exercise-responsive metabolic patterns discussed in the Results and Discussion, including TCA-cycle intermediates, lactate and pyruvate metabolism, hypoxanthine, taurine and representative lipid and organic-acid responses. The table also includes all four BH-FDR-significant workload × V̇O₂peak interactions identified in Model 5.

This editorial selection focused the main-text table on the metabolites central to the biological interpretation. No additional statistical threshold based on effect size, p-value or q-value was used to select these metabolites. The complete list of all 39 BH-FDR-significant Model 1 workload–metabolite associations is provided in Supplementary Table S2A and is also represented in the corresponding figures.

 

C11.The correlation analysis in Figure 2 / Section 3.5 is hypothesis-generating due to selection-then-test.

Metabolites entering the exploratory Pearson correlation analysis were pre-selected on the basis of having a BH-FDR-significant Model 1 effect for that same workload contrast (Appendix A22), and correlation p-values were not corrected for multiple testing (explicitly and appropriately stated by the authors). Because the same significant change metabolites were then tested against several correlated exercise/fitness variables (V̇O2peak, peak power in W/kg, time to AT, time to peak), there is a meaningful risk of overinterpreting the number and pattern of asterisks in Figure 2. The authors already call this analysis "exploratory," which is good practice; I would simply suggest stating explicitly, once, that findings from Section 3.5 should be considered hypothesis-generating rather than confirmatory, to pre-empt over-reading by less statistically minded readers.

Author response:

Manuscript location: Section 3.5

We thank the reviewer for this constructive suggestion. We agree that the correlations presented in Figure 2 and Section 3.5 are hypothesis-generating rather than confirmatory. Metabolites were first selected because they demonstrated BH-FDR-significant workload effects in Model 1 and were subsequently examined for correlations with several related exercise and fitness variables. Therefore, the correlation analysis represents a secondary selection-then-test analysis and is not statistically independent of the initial metabolite selection.

The asterisks in Figure 2 represent nominal Pearson correlation p-values and should not be interpreted as evidence from a confirmatory multiplicity-controlled analysis. We have added an explicit statement to Section 3.5 and the Figure 2 legend clarifying that these findings are exploratory and require confirmation in an independent cohort. The correlation findings are not used to determine the statistical significance of the primary workload effects.

 

4.Discussion and Conclusions

C12. V̇O2peak and peak power were not independent in this design; the fitness-response interpretation should be qualified accordingly.

The Discussion attributes the larger lactate/malate excursions in fitter participants to a "greater capacity to mount the metabolic response required to sustain maximal exercise." Because fitter participants also reach a higher absolute peak workload and exercise for longer (as shown by the correlations with peak power and exercise duration in Section 3.5, which parallel those with V̇O2peak), part of the larger metabolite excursion likely reflects the higher absolute mechanical/metabolic demand attained at "peak" rather than a fitness-specific difference in metabolic responsiveness per se at a matched relative or absolute workload. The manuscript partially acknowledges this ("fitter individuals can sustain higher external workloads"), but a more explicit statement that V̇O2peak and peak power cannot be fully disentangled in this design (both reflect the same underlying maximal-effort test) would improve precision.

Author response:

Manuscript location: Section 4 (Discussion), paragraph comparing CRF, peak power, and exercise demand.

 

We completely agree that V̇O₂peak and peak power cannot be fully disentangled in the present study design. We have therefore revised the Discussion to explicitly address this: “Due to their interconnectedness, the effects of internal versus external ‘work’ cannot be fully disentangled in this design.”

 

C13.The translational potential paragraph cites the UK Biobank study.

The penultimate Discussion paragraph cites a UK Biobank study (ref [34]) of metabolomic/proteomic CRF signatures predicting mortality and disease risk as support for the translational potential of the present findings. That study was conducted in a large, largely non-diabetic general population using a different metabolomic platform and metabolite panel; therefore, its relevance to a 20-person, T1D-specific, GC-MS-based exercise metabolome study is illustrative at best. Consider softening the linkage (e.g., "this raises the possibility that..." rather than implying direct support) to avoid overstating generalizability.

Author response:

Manuscript location: Section 4 (Discussion), UK Biobank paragraph.

We thank the reviewer for this important clarification and agree that the UK Biobank study provides illustrative evidence of the broader prognostic relevance of CRF-associated metabolomic signatures rather than direct support for the translational potential of our findings. We have therefore softened the wording of the Discussion to avoid implying direct generalizability of these findings to our T1D-specific cohort.

‘Although the clinical utility of implementing metabolomics in routine care for individuals with type 1 diabetes remains to be established, growing evidence suggests that CRF-associated metabolomic signatures may have broader prognostic relevance. Indeed, a recent UK Biobank study demonstrated that metabolomic signatures associated with CRF were independently associated with substantially lower risks of all-cause mortality, cardiovascular disease, type 2 diabetes and colorectal cancer [35].’

 

Thank you for raising this as a matter for caution.

 

C14.The comparatively homogeneous" cohort description should be reconciled with the wide ranges reported in Table 1.

Section 4.2 (Limitations) describes the cohort as "comparatively homogeneous with respect to demographic and diabetes-related characteristics," yet Table 1 shows age 26–72 years, diabetes duration 12–57 years, and V̇O2peak 12.1–45.0 mL·kg⁻¹·min⁻¹ — a nearly four-fold range in fitness and a 45-year range in disease duration. This heterogeneity is actually a strength for the V̇O2peak-interaction analysis (Model 5) and should not be understated; suggest rephrasing to something like "participants were homogeneous in that all had long-standing, pump-treated, well-characterized T1D, but spanned a wide range of age, diabetes duration, and fitness" so the claim is not self-contradicted by the reader's own inspection of Table 1.

Author response:

Manuscript location: Section 5.1 (Strengths); the contradictory homogeneous-cohort sentence was removed from Section 5.2.

We thank the reviewer for highlighting this inconsistency and agree that describing the cohort as comparatively homogeneous did not appropriately reflect the substantial interindividual variability in age, diabetes duration and CRF. We have therefore removed this characterization from the Limitations and, as suggested, now recognize this variability as a strength of the study. Specifically, we have added the following to the Strengths section:‘Additionally, participants spanned a broad range of age, diabetes duration and CRF, providing substantial interindividual variability in the characteristics relevant to the present analyses.’

5.Figures, Tables and Presentation

C15. Duplicate section numbering.

"4. Discussion" is followed later by "4. Strengths, limitations, and perspectives for future research" and then "5. Conclusion" — the second "4." should be "5." and "Conclusion" should become "6." Please renumber the references before typesetting.

Author response:

Manuscript location: Sections 5 and 6; numbering was corrected.

Thank you for picking up on this. We have since renumbered correctly.

C16. Figures and tables are otherwise clear, high resolution, and internally consistent.

I cross-checked Figures 1–3 and Supplementary Figures S1–S8 against the underlying values in Supplementary Tables S1B, S2A, and S6 (e.g., TCA cycle enrichment q = 0.0027 at peak / 0.0085 at recovery; nucleotide metabolism q = 0.0404 at peak;39 significant workload associations; four significant V̇O2peak × workload interactions) and found no discrepancies. The forest plot (Figure S2), volcano plots (Figures S3–S5), and heatmap (Figure S1) are well-designed and legible. Minor suggestion: Figure 3 and Supplementary Figures S7–S8 would benefit from a shared, fixed y-axis scale across the three workload panels (AT, peak, recovery) so that readers can visually compare the magnitude of change across intensities at a glance rather than needing to consult three separately scaled figures.

Author response:

We thank the reviewer for this helpful suggestion. We considered using a shared y-axis scale; however, the magnitude of the AT response was substantially smaller than the Peak and Recovery responses. A common scale considerably reduced the visibility of the AT results. We therefore retained separate y-axis scales for legibility and clarified the scale differences in the figure legends. Exact values are provided in Supplementary Tables S2A and S6.

 

6.Language and Style

C17 (minor). Inconsistent American/British spelling conventions.

The manuscript predominantly uses American English ("characterized," "randomized," "analyzed," "hemoglobin," "-ize/-ization" endings) but intermittently switches to British forms ("normalisation," p. 4; "glycaemia"/"glycaemic," Discussion; "Whilst," Introduction). This does not impede comprehension, but should be standardized to a single variety during copy-editing.

Author response:

Manuscript location: Throughout the manuscript; American English was applied consistently.

Thank you for being diligent in checking consistency. We have gone through the article for incongruencies and hope any remaining issues will be resolved in copy-editing.

Summary of Required Actions

1. The remaining MDPI template instructions were removed from Section 2, and a study-design/timeline schematic figure (C3) was added.
2. The absence of insulin/C-peptide/β-cell function data and pre-exercise meal standardization should be reported or explicitly discussed as limitations (C4, C5).
3. Add a methodological caveat regarding the AT blood-sampling timing resolution relative to the modeled threshold (C6).
4. Justify ML vs. REML model fitting, given the small sample size (C7).
5. Address the very small metabolic category sizes used in the over-representation analysis (C8).
6. Clarify data deposition plans for the processed metabolomics feature matrix (C9).
7. Add a footnote clarifying the selection rule for Table 2 (C10).
8. Please correct the duplicate "4." section numbering (C15).
9. Standardize English spelling conventions throughout (C17).

• The remaining comments (C1, C2, C11–C14, C16) are suggested but are non-essential improvements.

Author response: Thank you for summarizing these, we have since actioned each accordingly and hope the revised version is suitably adapted. 

 

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

Accepted