Autophagy-Mitophagy Pathway-Linked Genetic Variants Associate with Systemic Inflammation and Interact with Dietary Factors in Asian and European Cohorts

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

• The topic is interesting and potentially important, specially with the use of two large cohorts. However, in its current form, the manuscript has major methodological, internal consistency, and reporting problems that weaken confidence in the findings. I would suggest major revision, and some of the errors are serious enough that the authors need to re-check the whole analysis and tables before resubmission.
• The main weakness is the definition of systemic inflammation. The outcome is defined differently in the two cohorts, with different WBC and hsCRP thresholds in KoGES and UKBB, and these thresholds were chosen based on their association with metabolic syndrome in each cohort. The paper then goes on to report a strong association between SI and MetS. This makes the cross-cohort comparison less pure than the manuscript suggests. It also means the outcome may represent different biological states in the two cohorts.
• The GRS construction and the “replication” strategy are not described clearly and are internally inconsistent. In the Methods, the authors state that the relaxed threshold in KoGES yielded four independent SNPs, and that UKBB SNPs were extracted using identical genomic coordinates for consistency. However, the abstract and results report 6 KoGES SNPs, and Table 3 lists a different set of UKBB variants rather than the same coordinates. This makes it difficult to understand what exactly was replicated: the same SNPs, the same genes, or only the same pathway. This needs to be rewritten very carefully.
• The GWAS reporting is not sufficiently complete for a genetics paper. The Methods describe standard SNP QC and model covariates, but there is no clear description of ancestry principal components, genotype imputation, genomic inflation, or relatedness handling. Even if these were done, they are not reported clearly enough for the reader to assess robustness against population stratification and technical bias.
• The interaction analysis needs stronger justification. Lifestyle variables were dichotomized, and the GRS was categorized differently in the two cohorts: tertiles in KoGES versus negative/zero/positive groups in UKBB. This reduces comparability across cohorts and may introduce arbitrary threshold effects. The analysis would be stronger if the authors presented continuous-scale sensitivity analyses and a more homogenous strategy between cohorts.
• The conclusions are too strong for the presented data. The authors themselves acknowledge that the study is observational, that SI was defined using clinical biomarkers rather than direct molecular measures, and that medication confounding was not fully addressed. Despite that, the conclusion speaks of a “central role,” “first population-level genetic evidence,” and even suggests precision prevention and treatment implications. That language needs adjustment of tone to be less confirmatory.
• There are several numerical inconsistencies that are too serious. The abstract reports the UKBB sample as n = 340,410 whereas the Methods state the final UKBB analytical sample was 343,892. There should be no difference between the abstract and main methods unless an additional exclusion step is included and explained.
• In KoGES, the final analytical sample is given as 30,599, but the SI categories shown in the Methods are 18,082 low-SI and 10,020 high-SI, which sum to 28,102, leaving 2,497 participants unexplained. This must be clarified in the flowchart and text.
• The Methods say the relaxed significance threshold in KoGES yielded 4 independent SNPs, but the Results and Table 3 present 6 SNPs for KoGES. This affects the interpretation of the GRS construction.
• Table 1 appears to contain mismatched count-percentage pairs. For example, in KoGES high-SI, MetS is reported as 2083 (49.7%), but 2083 out of 10,020 is not 49.7%. The allergy entries also do not match their listed percentages. This raises concern about the reliability of the rest of the numeric reporting.
• Table 4 has another serious issue. In KoGES, the group sizes are listed as 7,969, 15,435, and 13,187, which sum to 36,591, exceeding the whole KoGES cohort of 30,599. Also, the table labels these groups as PRS rather than GRS, which gives the impression that the table is not fully corrected.
• There is also inconsistency in how continuous data are described. The Methods state that continuous variables are presented as adjusted means ± standard deviations, while Table 1 says they are adjusted means and standard errors. These are not interchangeable and the manuscript must use one correctly and consistently.
• The abstract contains a gene-name error: AVPS33A is written instead of VPS33A, whereas Table 3 and the Results use VPS33A.
• The supplementary material contains a carry-over error: Figure S2 says the red dotted line indicates the cutoff for hypertension risk, even though the supplementary figure is supposed to concern systemic inflammation. ijms-4207899-supplementary
• The manuscript needs much proofreading. There are repeated language and formatting problems such as “Correspondence Authors,” “In conclusions,” inconsistent capitalization such as “Definition of Si,” and even “Ethnic Committee” instead of “Ethics Committee.”
• The figures and captions also need attention. Some main results are described only qualitatively in text and shown visually, without a clean companion table listing exact ORs, CIs, and adjusted P-values for the main GRS analyses.
• The figures (specially those embedded in the text) are hazy and difficult to re=produce. Please consider providing clear figures and increase their resolution.

Comments on the Quality of English Language

The manuscript is generally understandable, but the English language requires further improvement to express the research more clearly and professionally. There are multiple issues with grammar, sentence structure, word choice, capitalization, and proofreading throughout the text. In several places, the wording is imprecise, and some expressions appear inconsistent with standard scientific writing. The manuscript would benefit from careful language editing by a native English speaker or a professional editing service. 

Author Response

Reply to comments

We appreciate the associate editor and reviewers for their helpful and insightful comments and suggestions for our paper in improving the quality of our manuscript. We made a sincere effort to address each comment and incorporate the comments into the revisions. We have point-by-point replies to each comment. We believe that this revision substantially improved the quality of the manuscript.

 

Editor

1. 
   According to the journal's requirements, the article structure order is: Introduction, Results, Discussion, Materials and Methods, Conclusions

: We revised the order of the manuscript according to Introduction, Results, Discussion, Materials and Methods, Conclusions.
2. Please add the funding, Institutional Review Board Statement, Informed Consent Statement, and Data Availability Statement sections later in the backmatter section.

: We added the funding, Institutional Review Board Statement, Informed Consent Statement, and Data Availability Statement sections after conclusion section.

3. 
   We noticed that your reference [17] has a correction record. Could you please delete it or replace it with another reference?

: it was changed into other reference.

 

Reviewer 1

- The topic is interesting and potentially important, specially with the use of two large cohorts. However, in its current form, the manuscript has major methodological, internal consistency, and reporting problems that weaken confidence in the findings. I would suggest major revision, and some of the errors are serious enough that the authors need to re-check the whole analysis and tables before resubmission.

: Response - We appreciate your careful review and we revised the manuscript without errors.

 

- The main weakness is the definition of systemic inflammation. The outcome is defined differently in the two cohorts, with different WBC and hsCRP thresholds in KoGES and UKBB, and these thresholds were chosen based on their association with metabolic syndrome in each cohort. The paper then goes on to report a strong association between SI and MetS. This makes the cross-cohort comparison less pure than the manuscript suggests. It also means the outcome may represent different biological states in the two cohorts.

: response - We have addressed this concern through three major revisions:

 

1. Justification of Population-Specific Thresholds

We have clarified that different SI thresholds between cohorts reflect well-established ethnic differences in baseline inflammatory marker levels, not arbitrary choices based on MetS associations. Specifically, baseline median CRP differs substantially between populations (Korean ~0.06 mg/L vs European ~1.8 mg/L), necessitating population-specific cut points to capture comparable inflammatory states. This is now explained in Methods Section 2.3 (page 4, lines 150-164).

2. Comprehensive Sensitivity Analyses to Address Circularity

We conducted extensive sensitivity analyses demonstrating that genetic associations with SI are robust across multiple alternative definitions independent of MetS:

• Alternative binary definitions: WBC-only (no CRP), CRP-only (no WBC), and quartile-based SI (top 25% of either marker, defined independently of MetS status)
• Continuous outcomes: GRS associations with continuous WBC and CRP levels (eliminating any threshold effects)
• Independent SI-MetS validation: Using quartile-based SI definition (top 25%, completely independent of MetS), the SI→MetS association (OR=1.63) was nearly identical to the primary analysis (OR=1.65), demonstrating that the SI-MetS relationship is not an artifact of circular definitions

Results are presented in new Supplementary Tables S1A-C and described in Results Section 3.7 (page 17, lines 542-565). In all sensitivity analyses, GRS associations with SI remained statistically significant with comparable effect sizes, confirming robustness across operational definitions.

3. Cross-Cohort Validation with Continuous Markers

To address whether SI represents different biological states across cohorts, we analyzed continuous inflammatory markers in both populations:

• KoGES: GRS→WBC β=2.23 units (P<0.001), GRS→CRP 1.54-fold increase (P<0.001)
• UKBB: GRS→WBC β=2.57 units (P<0.001), GRS→CRP 1.20-fold increase (P<0.001)

The consistent direction and magnitude of genetic effects on continuous markers across populations (Supplementary Table S1B) demonstrates that the same genetic architecture influences inflammatory markers regardless of population-specific thresholds, supporting biological comparability.

We think our findings demonstrate that genetic associations are genuine and robust rather than artifacts of threshold selection or circular definition with MetS.

 

- The GRS construction and the “replication” strategy are not described clearly and are internally inconsistent. In the Methods, the authors state that the relaxed threshold in KoGES yielded four independent SNPs, and that UKBB SNPs were extracted using identical genomic coordinates for consistency. However, the abstract and results report 6 KoGES SNPs, and Table 3 lists a different set of UKBB variants rather than the same coordinates. This makes it difficult to understand what exactly was replicated: the same SNPs, the same genes, or only the same pathway. This needs to be rewritten very carefully.

: Response - We sincerely apologize for the confusion caused by inconsistent terminology and a typographical error. We have substantially revised Methods Section 2.8 to clearly describe our replication strategy and correct the errors.

1. Correction of SNP Number Error

The statement "four independent SNPs" in the original Methods was a typographical error. The correct number is six independent SNPs in KoGES, as accurately reported in the Abstract, Results, and Table 3. This has been corrected throughout the manuscript (Methods Section 2.8, page 5-6, lines 229-266).

2. Clarification of Replication Strategy

We have clarified that our study employed gene-level (pathway-level) replication, not SNP-level replication:

KoGES Discovery (n=30,599):

• 6 independent SNPs at P<5×10⁻⁵ after LD pruning (r²<0.01)
• Genes: INPP5D, ATG16L1, ATG7, AP3S1, OPTN, VPS33A
• These 6 SNPs were used to construct the KoGES GRS

UKBB Replication (n=343,892):

• Targeted analysis within the same 19 CLSV-2 gene regions (±20kb)
• 10 SNPs reached genome-wide significance (P<5×10⁻⁸)
• Genes: INPP5D, ATG7, RAB7A, ATG12, SQSTM1, VPS33A, VPS18, MAP1LC3B, BECN1
• These 10 SNPs were used to construct the UKBB GRS

Gene-Level Replication: Three genes identified in KoGES (INPP5D, ATG7, VPS33A) harbored genome-wide significant SNPs in UKBB, confirming cross-population replication at the gene level.

Why SNPs differ between cohorts:

• Different lead SNPs within the same genes reflect population-specific linkage disequilibrium (LD) structure
• Allele frequency differences between East Asian and European populations
• Larger UKBB sample (11-fold) enabled detection of additional variants

This pathway-level replication approach is standard for cross-population studies where LD structure differs substantially between ancestries.

 

- The GWAS reporting is not sufficiently complete for a genetics paper. The Methods describe standard SNP QC and model covariates, but there is no clear description of ancestry principal components, genotype imputation, genomic inflation, or relatedness handling. Even if these were done, they are not reported clearly enough for the reader to assess robustness against population stratification and technical bias.

: Response - We have added it in Methods Section 2.6 (page 5, lines 201-216) comprehensively addressing these issues.

1. Population Stratification Control

Ancestry restriction:

• KoGES: Korean ancestry only (homogeneous East Asian population)
• UKBB: White British ancestry verified using genetic principal component-based classification (field 22006, which uses the first 4 PCs to exclude ancestry outliers)

Genomic inflation:

• KoGES: λ=1.032, well below the 1.05 threshold indicating negligible stratification
• QQ plot shows excellent alignment with expected distribution (new Supplementary Figure S2, page 5)
• UKBB: Targeted gene region analysis (not genome-wide), λ not applicable

Principal components approach:

• Genetic PCs were used for ancestry-based sample selection in UKBB (field 22006)
• PCs were not included as GWAS covariates because: (1) homogeneous populations, (2) λ=1.032 indicates minimal stratification, (3) geographic covariates (residential area, assessment center) capture fine-scale structure
• This approach is standard for single-ancestry cohorts with low genomic inflation

2. Genotype Imputation

• KoGES: Korea Biobank Array (~833,000 directly genotyped variants); genotypes imputed to 1000 Genomes Project Phase 3 East Asian reference panel using IMPUTE2.
• UKBB: UK BiLEVE Axiom Array or UK Biobank Axiom Array (~820,000 directly genotyped variants); genotypes imputed to Haplotype Reference Consortium (HRC) and UK10K reference panels using IMPUTE4, yielding ~96 million variants.

3. Relatedness

  Relatedness was not explicitly filtered in either cohort

  The low genomic inflation (λ=1.032 in KoGES) and homogeneous ancestry restriction minimize confounding from cryptic relatedness

  Standard logistic regression models were used (not mixed models)

 

4. SNP and Sample QC

Expanded Methods Section 2.6 (page 5 lines 194-200) now includes:

• SNP QC: call rate ≥98%, HWE P>0.05, MAF≥0.005
• Sample QC: call rate ≥95%, heterozygosity < 30±3 SD, sex concordance verified

5. Cross-Population Validation

Gene-level replication across Korean and European populations provides additional evidence against stratification artifacts, as population structure would be ancestry-specific and would not replicate across continental groups.

 

• The interaction analysis needs stronger justification. Lifestyle variables were dichotomized, and the GRS was categorized differently in the two cohorts: tertiles in KoGES versus negative/zero/positive groups in UKBB. This reduces comparability across cohorts and may introduce arbitrary threshold effects. The analysis would be stronger if the authors presented continuous-scale sensitivity analyses and a more homogenous strategy between cohorts.

: response – We have substantially revised the gene-lifestyle interaction analysis to address concerns about categorization, comparability, and analytical rigor.

1. Continuous GRS × Lifestyle Interactions as Primary Analysis

We have re-analyzed all gene-lifestyle interactions using continuous GRS and continuous (or binary when appropriate) lifestyle variables to maximize statistical power and eliminate arbitrary categorization thresholds. These continuous analyses are now presented as the primary interaction results in new Supplementary Table S2 and described in Methods Section 2.9 (page 7, lines 294-302) and Results Section 3.6 (page 17, lines 536-554).

Model specification:

logit(P(SI=1)) = β₀ + β₁×GRS_continuous + β₂×Lifestyle + β₃×(GRS_continuous × Lifestyle) + covariates

Where β₃ represents the interaction effect. Negative β₃ indicates lifestyle attenuates genetic risk; positive β₃ indicates lifestyle amplifies genetic risk.

Key findings from continuous analysis (UKBB, n=343,892):

Significant dietary interactions:

• Coffee consumption: β=-0.010, 95% CI: -0.018 to -0.003, P=0.006
• Total fruit intake: β=-0.004, 95% CI: -0.008 to -0.0002, P=0.039
• Vegetable + fruit combined: β=-0.002, 95% CI: -0.004 to -0.0001, P=0.040
• Meat consumption: β=+0.023, 95% CI: 0.013-0.034, P<0.001

No significant interactions:

• Total vegetables alone: β=-0.002, P=0.224
• Alcohol consumption: β=-0.001, P=0.305
• Physical activity: β=+0.00002, P=0.126

These findings demonstrate selective dietary interactions (plant-based foods protective, meat harmful) while other lifestyle factors (alcohol, exercise) show independent effects without modifying genetic risk.

2. Harmonized Categorical Analysis for Cross-Cohort Comparison

To enable direct cross-cohort comparison and provide clinically interpretable effect sizes, we conducted sensitivity analyses using harmonized GRS tertiles (low/medium/high) in both cohorts, replacing the original negative/zero/positive categorization in UKBB. Results are presented in revised Table 4B (page 15) and Supplementary Table S2.

Harmonized approach:

• Both cohorts now use tertiles of continuous GRS distribution
• Lifestyle factors dichotomized using consistent criteria (e.g., high vs. low vegetable/fruit intake based on median or guidelines)
• Enables direct comparison of interaction patterns across populations

3. Robustness Across Analytical Approaches

The consistency of findings across continuous and categorical GRS operationalizations demonstrates robustness:

Example (Vegetable/Fruit Intake in UKBB):

• Continuous analysis: β=-0.002, P=0.040 (dose-response relationship)
• Categorical analysis (tertiles): SI prevalence gradient attenuated from 12 percentage points (low intake) to 3 percentage points (high intake), P-interaction <0.01
• Both approaches show the same biological pattern: favorable diet attenuates genetic risk

4. KoGES Interaction Analysis and Statistical Power

In KoGES (n=30,599), continuous dietary interactions did not reach statistical significance:

• Flavonoid intake: β=-0.0006, P=0.241
• Fat intake: β=+0.0001, P=0.967

However, categorical analyses showed borderline significance (flavonoid P=0.037, fat P=0.053), directionally consistent with UKBB findings but underpowered. This reflects the substantially larger sample size required for interaction detection—typically 4-fold larger than for main effects. With an 11-fold difference in sample size (KoGES 30,599 vs UKBB 343,892), KoGES lacks adequate power for interaction detection.

 

- The conclusions are too strong for the presented data. The authors themselves acknowledge that the study is observational, that SI was defined using clinical biomarkers rather than direct molecular measures, and that medication confounding was not fully addressed. Despite that, the conclusion speaks of a “central role,” “first population-level genetic evidence,” and even suggests precision prevention and treatment implications. That language needs adjustment of tone to be less confirmatory.

: Response – We have substantially revised the Conclusion section (page 20-21, lines 654-700) to appropriately use the language and acknowledge limitations.

We removed overstated language, explicit acknowledgment of limitations, use conditional language for clinical implications, added future research in the Conclusion, limitation, and future study sections.

 

- There are several numerical inconsistencies that are too serious. The abstract reports the UKBB sample as n = 340,410 whereas the Methods state the final UKBB analytical sample was 343,892. There should be no difference between the abstract and main methods unless an additional exclusion step is included and explained.

: Response - We made a mistake in the abstract. The participants’ exclusion was shown in Figure S1 in both cohorts.

 

- In KoGES, the final analytical sample is given as 30,599, but the SI categories shown in the Methods are 18,082 low-SI and 10,020 high-SI, which sum to 28,102, leaving 2,497 participants unexplained. This must be clarified in the flowchart and text.

: Response - The participants’ exclusion was shown in Figure S1A in KoGES and the final participants are 28,102.

 

- The Methods say the relaxed significance threshold in KoGES yielded 4 independent SNPs, but the Results and Table 3 present 6 SNPs for KoGES. This affects the interpretation of the GRS construction.

: Response - The 6 SNPs are used in KoGES and it was shown in Table 3 and PRS was changed into GRS.

 

- Table 1 appears to contain mismatched count-percentage pairs. For example, in KoGES high-SI, MetS is reported as 2083 (49.7%), but 2083 out of 10,020 is not 49.7%. The allergy entries also do not match their listed percentages. This raises concern about the reliability of the rest of the numeric reporting.

: Response - We apologize for the mistakes. It is corrected in Table 1.

 

- Table 4 has another serious issue. In KoGES, the group sizes are listed as 7,969, 15,435, and 13,187, which sum to 36,591, exceeding the whole KoGES cohort of 30,599. Also, the table labels these groups as PRS rather than GRS, which gives the impression that the table is not fully corrected.

: Response - We apologize for the mistakes. It is corrected in Table 4.

 

- There is also inconsistency in how continuous data are described. The Methods state that continuous variables are presented as adjusted means ± standard deviations, while Table 1 says they are adjusted means and standard errors. These are not interchangeable and the manuscript must use one correctly and consistently.

: Response - We used standard error instead of standard deviation. It was revised throughout the manuscript.

 

- The abstract contains a gene-name error: AVPS33A is written instead of VPS33A, whereas Table 3 and the Results use VPS33A.

: VPS33A is correct and abstract was changed.

 

- The supplementary material contains a carry-over error: Figure S2 says the red dotted line indicates the cutoff for hypertension risk, even though the supplementary figure is supposed to concern systemic inflammation. ijms-4207899-supplementary

: It was changed into systemic inflammation.

 

- The manuscript needs much proofreading. There are repeated language and formatting problems such as “Correspondence Authors,” “In conclusions,” inconsistent capitalization such as “Definition of Si,” and even “Ethnic Committee” instead of “Ethics Committee.”

: We changed it into Ethics Committee.

 

- The figures and captions also need attention. Some main results are described only qualitatively in text and shown visually, without a clean companion table listing exact ORs, CIs, and adjusted P-values for the main GRS analyses.

: We provided ORs, CIs, and adjusted P-values for the main GRS analyses in tables and also added them in the text in pages 17-18, lines 522-534..

 

- The figures (specially those embedded in the text) are hazy and difficult to re=produce. Please consider providing clear figures and increase their resolution.

: We increased resolution of the figures (330 dpi).

 

Comments on the Quality of English Language

The manuscript is generally understandable, but the English language requires further improvement to express the research more clearly and professionally. There are multiple issues with grammar, sentence structure, word choice, capitalization, and proofreading throughout the text. In several places, the wording is imprecise, and some expressions appear inconsistent with standard scientific writing. The manuscript would benefit from careful language editing by a native English speaker or a professional editing service.

: We had English language proof from the English editing company (Nurisco), which is the university assigned. We attached the certificate.

Reviewer 2 Report

Comments and Suggestions for Authors

This manuscript addresses an interesting and potentially important topic by linking autophagy/mitophagy-related genetic variants with systemic low-grade inflammation and lifestyle interactions in two large cohorts. The overall concept is novel. However, several major issues should be addressed.

1. SI was defined using different WBC and hsCRP cutoffs in KoGES and UK Biobank. This reduces direct comparability between cohorts and weakens the replication claim. The authors should provide sensitivity analyses using harmonized or alternative SI definitions (for example, WBC-only or hsCRP-only models) to test the robustness of the findings.
2. The manuscript contains an apparent inconsistency regarding the number of SNPs retained for the KoGES-based GRS. The SNP selection process, LD pruning, and final SNP set included in the GRS should be described more clearly, ideally with a supplementary table listing the selected variants and their effect estimates. The rationale for using different GRS categorization approaches across the two cohorts should also be explained.
3. It is unclear whether genetic principal components were included in the association models. This is essential for GWAS-based analyses. The authors should explicitly state how population structure was controlled and report basic GWAS quality-control information.
4. The manuscript emphasizes interaction effects, but the interaction terms themselves are not presented clearly enough. The authors should report interaction effect estimates with 95% confidence intervals and clarify the regression models used. This would make the interaction findings more convincing.
5. Given the observational design, the use of proxy inflammatory markers, and the relaxed threshold applied in part of the SNP selection process, some conclusions appear overstated. The discussion should be revised to present the findings as associations rather than strong evidence of a central or causal role.

Author Response

Reply to comments

We appreciate the associate editor and reviewers for their helpful and insightful comments and suggestions for our paper in improving the quality of our manuscript. We made a sincere effort to address each comment and incorporate the comments into the revisions. We have point-by-point replies to each comment. We believe that this revision substantially improved the quality of the manuscript.

Comments and Suggestions for Authors

This manuscript addresses an interesting and potentially important topic by linking autophagy/mitophagy-related genetic variants with systemic low-grade inflammation and lifestyle interactions in two large cohorts. The overall concept is novel. However, several major issues should be addressed.

1) SI was defined using different WBC and hsCRP cutoffs in KoGES and UK Biobank. This reduces direct comparability between cohorts and weakens the replication claim. The authors should provide sensitivity analyses using harmonized or alternative SI definitions (for example, WBC-only or hsCRP-only models) to test the robustness of the findings.

: Response – We agree that different thresholds could reduce direct comparability and have addressed this through the requested sensitivity analyses.

1. Rationale for Population-Specific Thresholds

Different SI thresholds reflect well-established ethnic differences in baseline inflammatory markers rather than arbitrary choices. Baseline median CRP differs ~30-fold between populations (Korean ~0.06 mg/L vs European ~1.8 mg/L), making identical absolute cutpoints biologically inappropriate. Our thresholds (KoGES: WBC >6.2 or CRP >1.0 mg/L; UKBB: WBC >7.0 or CRP >3.0 mg/L) were selected to capture comparable percentiles (~97th) within each population's distribution. This is now explained in Methods Section 2.3 (page 4, lines 150-164).

2. Sensitivity Analyses with Alternative SI Definitions

As requested, we conducted comprehensive sensitivity analyses testing robustness across multiple harmonized and alternative definitions:

1. a) Component-specific definitions (WBC-only, CRP-only):

• WBC-only SI (>6.2 in KoGES): GRS OR=1.19, P<0.001
• CRP-only SI (>1.0 mg/L in KoGES): GRS OR=1.31, P=0.003
• Results confirm genetic associations are not driven by a single biomarker

1. b) Quartile-based definition (harmonized approach):

• Top quartile of either WBC or CRP (25% prevalence in both cohorts)
• KoGES: GRS OR=1.10, P<0.001
• Provides harmonized prevalence across cohorts without absolute threshold dependency

1. c) Continuous outcomes (eliminates all threshold effects):

• KoGES: GRS→WBC β=+2.23 units (P<0.001); GRS→CRP 1.54-fold (P<0.001)
• UKBB: GRS→WBC β=+2.57 units (P<0.001); GRS→CRP 1.20-fold (P<0.001)
• Consistent genetic effects on continuous markers across populations demonstrate biological comparability

All sensitivity analyses showed statistically significant genetic associations with comparable effect sizes, confirming that findings are robust to SI operationalization. Results are presented in new Supplementary Tables S1A-C  and Results Section 3. 3.7 (page 17, lines 542-565).

3. Cross-Cohort Replication Remains Valid

The consistency of genetic effects across: (1) different operational definitions within each cohort, (2) continuous inflammatory markers across cohorts, and (3) gene-level replication (INPP5D, ATG7, VPS33A) in independent populations supports genuine biological replication rather than definition-dependent artifacts.

We think that our findings are robust across multiple harmonized and alternative SI definitions.

 

2) The manuscript contains an apparent inconsistency regarding the number of SNPs retained for the KoGES-based GRS. The SNP selection process, LD pruning, and final SNP set included in the GRS should be described more clearly, ideally with a supplementary table listing the selected variants and their effect estimates. The rationale for using different GRS categorization approaches across the two cohorts should also be explained.

: Response - 1. Correction of SNP Number Inconsistency

The Methods statement "four independent SNPs" was a typographical error. The correct number is six independent SNPs retained after LD pruning in KoGES, as accurately shown in Table 3 and reported in the Abstract and Results. This has been corrected in the revised Methods Section 2.8 (page 5-6, lines 230-266).

2. Detailed Description of SNP Selection and LD Pruning

We have substantially revised Methods Section 2.8 (page 5-6, lines 230-266) to provide step-by-step clarity:

KoGES Discovery:

1. Extracted all SNPs within CLSV-2 genes ±20kb flanking regions
2. Applied significance threshold P<5×10⁻⁵ (justified for pathway-focused analysis)
3. Performed LD pruning using PLINK (r²<0.2 within 500kb windows)
4. Result: 6 independent SNPs across 6 genes (INPP5D, ATG16L1, ATG7, AP3S1, OPTN, VPS33A)
5. Constructed weighted GRS: GRS = Σ(βᵢ × SNPᵢ), where βᵢ = cohort-specific log-odds ratios

UKBB Replication:

1. Extracted all SNPs within the same CLSV-2 gene regions (±20kb)
2. Identified SNPs reaching genome-wide significance (P<5×10⁻⁸)
3. Result: 10 genome-wide significant SNPs across 9 genes
4. Constructed weighted GRS using cohort-specific effect sizes
5. Table 3A included KoGES Discovery SNPs (n=6):

• rsID, chromosomal position, gene, effect allele/non-effect allele, MAF, OR (95% CI), P-value, genomic location

Table 3B - UKBB Replication SNPs (n=10):

• Same details as above, plus column indicating gene-level replication status (3 genes: INPP5D, ATG7, VPS33A)

4. Rationale for Different GRS Categorization

We have added explicit justification for different GRS categorization approaches (Methods Section 2.8, page 5-6, lines 230-266):

KoGES: Tertiles (Low/Medium/High)

• GRS distribution approximately normal
• Tertile categorization provides balanced group sizes (n~10,200 each)
• Standard approach for categorical risk stratification

UKBB: Negative/Zero/Positive

• GRS distribution centered near zero due to different SNP composition and larger sample size
• Natural categorization based on sign of weighted score
• Provides interpretable risk groups: below average (negative), average (zero), above average (positive)

Why approaches differ:

• Different SNP sets (6 vs 10 SNPs) with different effect sizes produce different GRS distributions
• Each categorization optimizes interpretability within its cohort's specific distribution
• Critical point: Primary interaction analyses used continuous GRS (Methods Section 2.3), making categorical differences irrelevant to main findings
• Categorical analyses serve illustrative purposes for clinical interpretation

5. Clarification of Cohort-Specific GRS Construction

The revised Methods now explicitly states that:

• Each cohort used cohort-specific SNP sets (not identical SNPs)
• Each cohort used cohort-specific effect sizes (to avoid overfitting)
• Both GRS capture variation in the same biological pathway (autophagy-mitophagy)
• This approach enables pathway-level replication while accounting for population-specific genetic architecture

 

3) It is unclear whether genetic principal components were included in the association models. This is essential for GWAS-based analyses. The authors should explicitly state how population structure was controlled and report basic GWAS quality-control information.

: Response - We have added a new Methods Section 2.6 (page 5, lines 199-232) explicitly addressing population structure control.

Principal components approach:

• Genetic PCs were used for ancestry-based sample selection (UKBB field 22006 uses the first 4 PCs to exclude ancestry outliers)
• PCs were not included as GWAS covariates in either cohort
• Justification: (1) Both cohorts restricted to genetically homogeneous populations (Korean-only, White British-only); (2) Genomic inflation λ=1.032 in KoGES, well below the 1.05 threshold, indicating negligible stratification; (3) Geographic covariates (residential area, assessment center) capture fine-scale structure
• This approach is standard practice for single-ancestry cohorts with low genomic inflation

GWAS quality control: We have expanded Methods Section 2.6 (page 5, lines 199-232) with complete QC reporting:

• Genotype imputation: 1000G Phase 3 EAS (KoGES); HRC+UK10K (UKBB)
• SNP QC: Call rate ≥98%, HWE P>1×10⁻⁶, MAF≥0.005, INFO≥0.8
• Sample QC: Call rate ≥95%, heterozygosity within ±3 SD, sex concordance
• Relatedness: Not explicitly filtered; low λ indicates minimal confounding
• Genomic inflation: λ=1.032 (Supplementary Figure S1)

Cross-population validation provides additional evidence against stratification artifacts, as population structure would not replicate across Korean and European ancestries.

 

4) The manuscript emphasizes interaction effects, but the interaction terms themselves are not presented clearly enough. The authors should report interaction effect estimates with 95% confidence intervals and clarify the regression models used. This would make the interaction findings more convincing.

: Response - We have substantially revised the interaction analysis presentation to clearly report interaction terms with complete statistical details. In Fig 3, since the values are SI prevalence, it cannot have 95% CI.

1. Complete Interaction Statistics Now Reported

We have created new Supplementary Table S2 presenting complete interaction analysis results with all requested statistical information:

For each lifestyle factor, the table reports:

• Main effect of GRS: β₁ (95% CI), P-value
• Main effect of lifestyle: β₂ (95% CI), P-value
• Interaction effect: β₃ (95% CI), P-interaction

2. Explicit Regression Model Specification

• We have added complete model specification to Methods Section 2.9 (page 6-7, lines 284-303):

3. Clear Presentation in Results

• We have revised Results Section 3.6 (page 14-15) to explicitly report interaction statistics in text:

4. Interaction Term Calculation Example

To further clarify interpretation, we have added a worked example in the Supplementary Methods:

Coffee × GRS Interaction

For an individual with:

• GRS = 2.0 (continuous score)
• Coffee consumption = 3 cups/day

 

 

5) Given the observational design, the use of proxy inflammatory markers, and the relaxed threshold applied in part of the SNP selection process, some conclusions appear overstated. The discussion should be revised to present the findings as associations rather than strong evidence of a central or causal role.

: Response - We have revised the Discussion to present findings as associations rather than causal relationships, with appropriate hedging throughout. We revised to use systematic language throughout discussion. In the limitation we explicit acknowledgment of observational nature and  acknowledged proxy measures and relaxed threshold. Gene-Lifestyle Interaction was revised in the discussion section. Future research was framed as Necessity.

Comments and Suggestions for Authors

This manuscript addresses an interesting and potentially important topic by linking autophagy/mitophagy-related genetic variants with systemic low-grade inflammation and lifestyle interactions in two large cohorts. The overall concept is novel. However, several major issues should be addressed.

1) SI was defined using different WBC and hsCRP cutoffs in KoGES and UK Biobank. This reduces direct comparability between cohorts and weakens the replication claim. The authors should provide sensitivity analyses using harmonized or alternative SI definitions (for example, WBC-only or hsCRP-only models) to test the robustness of the findings.

: Response – We agree that different thresholds could reduce direct comparability and have addressed this through the requested sensitivity analyses.

1. Rationale for Population-Specific Thresholds

Different SI thresholds reflect well-established ethnic differences in baseline inflammatory markers rather than arbitrary choices. Baseline median CRP differs ~30-fold between populations (Korean ~0.06 mg/L vs European ~1.8 mg/L), making identical absolute cutpoints biologically inappropriate. Our thresholds (KoGES: WBC >6.2 or CRP >1.0 mg/L; UKBB: WBC >7.0 or CRP >3.0 mg/L) were selected to capture comparable percentiles (~97th) within each population's distribution. This is now explained in Methods Section 2.3 (page 4, lines 150-164).

2. Sensitivity Analyses with Alternative SI Definitions

As requested, we conducted comprehensive sensitivity analyses testing robustness across multiple harmonized and alternative definitions:

1. a) Component-specific definitions (WBC-only, CRP-only):

• WBC-only SI (>6.2 in KoGES): GRS OR=1.19, P<0.001
• CRP-only SI (>1.0 mg/L in KoGES): GRS OR=1.31, P=0.003
• Results confirm genetic associations are not driven by a single biomarker

1. b) Quartile-based definition (harmonized approach):

• Top quartile of either WBC or CRP (25% prevalence in both cohorts)
• KoGES: GRS OR=1.10, P<0.001
• Provides harmonized prevalence across cohorts without absolute threshold dependency

1. c) Continuous outcomes (eliminates all threshold effects):

• KoGES: GRS→WBC β=+2.23 units (P<0.001); GRS→CRP 1.54-fold (P<0.001)
• UKBB: GRS→WBC β=+2.57 units (P<0.001); GRS→CRP 1.20-fold (P<0.001)
• Consistent genetic effects on continuous markers across populations demonstrate biological comparability

All sensitivity analyses showed statistically significant genetic associations with comparable effect sizes, confirming that findings are robust to SI operationalization. Results are presented in new Supplementary Tables S1A-C  and Results Section 3. 3.7 (page 17, lines 542-565).

3. Cross-Cohort Replication Remains Valid

The consistency of genetic effects across: (1) different operational definitions within each cohort, (2) continuous inflammatory markers across cohorts, and (3) gene-level replication (INPP5D, ATG7, VPS33A) in independent populations supports genuine biological replication rather than definition-dependent artifacts.

We think that our findings are robust across multiple harmonized and alternative SI definitions.

 

2) The manuscript contains an apparent inconsistency regarding the number of SNPs retained for the KoGES-based GRS. The SNP selection process, LD pruning, and final SNP set included in the GRS should be described more clearly, ideally with a supplementary table listing the selected variants and their effect estimates. The rationale for using different GRS categorization approaches across the two cohorts should also be explained.

: Response - 1. Correction of SNP Number Inconsistency

The Methods statement "four independent SNPs" was a typographical error. The correct number is six independent SNPs retained after LD pruning in KoGES, as accurately shown in Table 3 and reported in the Abstract and Results. This has been corrected in the revised Methods Section 2.8 (page 5-6, lines 230-266).

2. Detailed Description of SNP Selection and LD Pruning

We have substantially revised Methods Section 2.8 (page 5-6, lines 230-266) to provide step-by-step clarity:

KoGES Discovery:

1. Extracted all SNPs within CLSV-2 genes ±20kb flanking regions
2. Applied significance threshold P<5×10⁻⁵ (justified for pathway-focused analysis)
3. Performed LD pruning using PLINK (r²<0.2 within 500kb windows)
4. Result: 6 independent SNPs across 6 genes (INPP5D, ATG16L1, ATG7, AP3S1, OPTN, VPS33A)
5. Constructed weighted GRS: GRS = Σ(βᵢ × SNPᵢ), where βᵢ = cohort-specific log-odds ratios

UKBB Replication:

1. Extracted all SNPs within the same CLSV-2 gene regions (±20kb)
2. Identified SNPs reaching genome-wide significance (P<5×10⁻⁸)
3. Result: 10 genome-wide significant SNPs across 9 genes
4. Constructed weighted GRS using cohort-specific effect sizes
5. Table 3A included KoGES Discovery SNPs (n=6):

• rsID, chromosomal position, gene, effect allele/non-effect allele, MAF, OR (95% CI), P-value, genomic location

Table 3B - UKBB Replication SNPs (n=10):

• Same details as above, plus column indicating gene-level replication status (3 genes: INPP5D, ATG7, VPS33A)

4. Rationale for Different GRS Categorization

We have added explicit justification for different GRS categorization approaches (Methods Section 2.8, page 5-6, lines 230-266):

KoGES: Tertiles (Low/Medium/High)

• GRS distribution approximately normal
• Tertile categorization provides balanced group sizes (n~10,200 each)
• Standard approach for categorical risk stratification

UKBB: Negative/Zero/Positive

• GRS distribution centered near zero due to different SNP composition and larger sample size
• Natural categorization based on sign of weighted score
• Provides interpretable risk groups: below average (negative), average (zero), above average (positive)

Why approaches differ:

• Different SNP sets (6 vs 10 SNPs) with different effect sizes produce different GRS distributions
• Each categorization optimizes interpretability within its cohort's specific distribution
• Critical point: Primary interaction analyses used continuous GRS (Methods Section 2.3), making categorical differences irrelevant to main findings
• Categorical analyses serve illustrative purposes for clinical interpretation

5. Clarification of Cohort-Specific GRS Construction

The revised Methods now explicitly states that:

• Each cohort used cohort-specific SNP sets (not identical SNPs)
• Each cohort used cohort-specific effect sizes (to avoid overfitting)
• Both GRS capture variation in the same biological pathway (autophagy-mitophagy)
• This approach enables pathway-level replication while accounting for population-specific genetic architecture

 

3) It is unclear whether genetic principal components were included in the association models. This is essential for GWAS-based analyses. The authors should explicitly state how population structure was controlled and report basic GWAS quality-control information.

: Response - We have added a new Methods Section 2.6 (page 5, lines 199-232) explicitly addressing population structure control.

Principal components approach:

• Genetic PCs were used for ancestry-based sample selection (UKBB field 22006 uses the first 4 PCs to exclude ancestry outliers)
• PCs were not included as GWAS covariates in either cohort
• Justification: (1) Both cohorts restricted to genetically homogeneous populations (Korean-only, White British-only); (2) Genomic inflation λ=1.032 in KoGES, well below the 1.05 threshold, indicating negligible stratification; (3) Geographic covariates (residential area, assessment center) capture fine-scale structure
• This approach is standard practice for single-ancestry cohorts with low genomic inflation

GWAS quality control: We have expanded Methods Section 2.6 (page 5, lines 199-232) with complete QC reporting:

• Genotype imputation: 1000G Phase 3 EAS (KoGES); HRC+UK10K (UKBB)
• SNP QC: Call rate ≥98%, HWE P>1×10⁻⁶, MAF≥0.005, INFO≥0.8
• Sample QC: Call rate ≥95%, heterozygosity within ±3 SD, sex concordance
• Relatedness: Not explicitly filtered; low λ indicates minimal confounding
• Genomic inflation: λ=1.032 (Supplementary Figure S1)

Cross-population validation provides additional evidence against stratification artifacts, as population structure would not replicate across Korean and European ancestries.

 

4) The manuscript emphasizes interaction effects, but the interaction terms themselves are not presented clearly enough. The authors should report interaction effect estimates with 95% confidence intervals and clarify the regression models used. This would make the interaction findings more convincing.

: Response - We have substantially revised the interaction analysis presentation to clearly report interaction terms with complete statistical details. In Fig 3, since the values are SI prevalence, it cannot have 95% CI.

1. Complete Interaction Statistics Now Reported

We have created new Supplementary Table S2 presenting complete interaction analysis results with all requested statistical information:

For each lifestyle factor, the table reports:

• Main effect of GRS: β₁ (95% CI), P-value
• Main effect of lifestyle: β₂ (95% CI), P-value
• Interaction effect: β₃ (95% CI), P-interaction

2. Explicit Regression Model Specification

• We have added complete model specification to Methods Section 2.9 (page 6-7, lines 284-303):

3. Clear Presentation in Results

• We have revised Results Section 3.6 (page 14-15) to explicitly report interaction statistics in text:

4. Interaction Term Calculation Example

To further clarify interpretation, we have added a worked example in the Supplementary Methods:

Coffee × GRS Interaction

For an individual with:

• GRS = 2.0 (continuous score)
• Coffee consumption = 3 cups/day

 

 

5) Given the observational design, the use of proxy inflammatory markers, and the relaxed threshold applied in part of the SNP selection process, some conclusions appear overstated. The discussion should be revised to present the findings as associations rather than strong evidence of a central or causal role.

: Response - We have revised the Discussion to present findings as associations rather than causal relationships, with appropriate hedging throughout. We revised to use systematic language throughout discussion. In the limitation we explicit acknowledgment of observational nature and  acknowledged proxy measures and relaxed threshold. Gene-Lifestyle Interaction was revised in the discussion section. Future research was framed as Necessity.

 

Round 2

Reviewer 2 Report

Comments and Suggestions for Authors

Thanks for addressing all my comments.