1. Introduction
Lower respiratory tract infections (LRTIs) remain a major global cause of death and include etiologically diverse bacterial, viral, and mycobacterial disease [1]. Diabetes adds a clinically important host dimension: hyperglycemia perturbs innate immune-cell metabolism and airway defense, while pulmonary infection can rapidly couple local inflammation to systemic organ injury [2,3,4]. Recent research by investigators in the present team has linked iron handling in immunocytes to pneumonia and sepsis through genetic analysis, examined sepsis risk in hospital-acquired pneumonia, and characterized difficult-to-treat Klebsiella pneumoniae infection [5,6,7].
These observations motivate target-resolved analysis of infection susceptibility.
Type 2 diabetes affects a large and growing population [8]. In contemporary cohorts, poor glycemic control and obesity track with severe influenza and pneumonia, and experimental high-glucose conditions blunt macrophage NOD1–MAPK responses to Legionella [3,4]. Metabolic and inflammatory signaling also intersect through adipokines and insulin sensitivity [9].
These pathways make glucose-lowering target biology a plausible determinant of pathogen-specific responses. Smoking is a strong independent determinant of LRTI and sepsis and may alter the phenotype distribution in which genetic target effects are observed [10].
Clinical studies report different infection outcomes across glucose-lowering treatments. Recent active-comparator cohorts and randomized-trial syntheses have examined pneumonia and sepsis during SGLT2 inhibitor and GLP-1 receptor agonist use, while metformin cohorts provide separate evidence on pneumonia incidence and survival [11,12,13,14,15,16,17]. Treatment allocation reflects diabetes duration, cardiometabolic disease, renal function, and smoking history.
Genetic perturbations in target loci offer a complementary view of pathway biology, provided the exposure allele, outcome phenotype, and pharmacological mechanism are kept distinct.
Drug-target Mendelian randomization (MR) links cis-region variants to a molecular or downstream glycemic exposure and estimates associations with disease outcomes [18,19,20]. Investigators in this team have also applied MR to a respiratory symptom and lung cancer [21]. The approach is especially informative when a functional variant and a near-gene proxy can be compared within the same outcome GWAS.
PPARG illustrates the central mechanistic question: the Pro12Ala allele has reduced transcriptional activity, whereas thiazolidinediones activate PPARγ [22,23]. INSR variation likewise indexes receptor-mediated signaling, and KCNJ11/ABCC8 variation indexes an ATP-sensitive potassium-channel pathway. Their genetically proxied effects are therefore reported by target and variant, with treatment studies used for triangulation.
We evaluated genetically proxied INSR, PPARG, KCNJ11/ABCC8, GLP1R, SLC5A2, and putative metformin targets across bacterial, viral, tuberculosis, COVID-19, lower respiratory, and sepsis phenotypes. We combined the original target inventory with accession-matched re-estimation, functional-variant comparisons, and explicit accounting for case numbers. The scientific question is whether infection phenotypes differ across target pathways and whether those differences persist when outcome definitions and genetic proxies are aligned.
2. Materials and Methods
2.1. Study Design
We used two-sample drug-target MR to relate target-associated glycemic variants to respiratory infection and sepsis GWAS phenotypes (Figure 1). The analysis combined instrument characterization, a uniform T2D positive control, and target–infection estimation. Summary associations and study metadata were retrieved through the IEU OpenGWAS [24,25].
Each infection phenotype used one accession, retaining the same definition for primary and functional proxies. Supplementary Tables S1–S3 provide matched proxy results, outcome resources, and the T2D source. Reporting follows STROBE-MR [26].
2.2. Genetic Instruments for Glucose-Lowering Targets
We mapped glucose-lowering pathways to their encoding genes with DrugBank and ChEMBL and retained the primary and functional variants in Table 1 [19,27,28,29,30]. Source coefficients, effect alleles, tissue context, exposure units, and approximate F statistics are collected in Supplementary Table S4. INSR, PPARG, KCNJ11/ABCC8, and GLP1R glucose associations originated in a UK Biobank analysis of 309,895 White British participants [19].
The SLC5A2 six-variant panel came from a separate HbA1c analysis of 344,182 participants; its published six-SNP and two-SNP sets used r2 thresholds of 0.8 and 0.1, respectively [31]. Source-specific selection and LD adjustment were retained for each panel.
The primary variants were rs8103934 (INSR), rs1076733 (GLP1R), rs35240997 (PPARG), rs5219 and rs739688 (KCNJ11/ABCC8), and six SLC5A2-region variants. Functional comparisons used PPARG rs1801282 (Pro12Ala) and ABCC8 rs757110 (S1369A) as separate proxies [22,23,32,33]. The source effect alleles were T for rs5219 and rs739688 and C for rs757110.
Qin et al. also reported non-glycemic associations for selected variants [19]. These source details guided allele harmonization and the cross-trait sensitivity analyses.
PPARG Pro12Ala reduces receptor transcriptional activity, while thiazolidinediones activate PPARγ [22,23]. ABCC8 S1369A changes ATP-sensitive potassium-channel sulfonylurea sensitivity [32,33]. We compared each functional variant with its primary target panel within the same outcome GWAS.
This comparison tests consistency across biologically informative perturbations; the exposure throughout is the named genetic target pathway.
Metformin mechanisms include AMPK-dependent signaling, PEN2-linked lysosomal signaling, and mitochondrial glycerophosphate metabolism [34,35,36]. Metformin also increases the signaling metabolite Lac-Phe through mitochondrial effects in intestinal cells, linking it to appetite and energy balance [37]. The original 11-gene screen retained GPD1 rs12581493, PRKAG2 rs3793342, and three GPD2 variants (Supplementary Table S5).
GPD1 and PRKAG2 were selected with tissue-specific GTEx v8 expression evidence and calibrated with HbA1c associations in 344,182 participants [29,38,39]. GPD2 calibration used a separate HbA1c GWAS of 389,889 participants [40]. We retained separate gene and variant estimates to preserve these biological distinctions.
2.3. Outcome GWAS and Phenotype Definitions
Thirteen accession-matched phenotypes were retained: bacterial pneumonia (7987 cases/188,868 controls), pneumococcal pneumonia (648/188,868), lung abscess (170/218,125), confirmed respiratory tuberculosis (247/217,599), viral pneumonia (700/188,868), known-virus pneumonia (763/212,707), COVID-19 hospitalization (9986/1,877,672), severe COVID-19 (5101/1,383,241), influenza (4262/188,868), influenza-and-pneumonia composite (29,924/188,868), other acute lower respiratory infection (10,103/208,689), sepsis (11,643/474,841), and sepsis-related 28-day death (1896/484,588) (Supplementary Table S2) [41,42]. Values are cases/controls in the named GWAS, not treated-patient counts. The death phenotype uses population controls and represents a population-based death endpoint.
2.4. Diabetes Positive Control
The biological positive control used the European T2D dataset ebi-a-GCST90018926 from Sakaue et al., comprising 38,841 cases and 451,248 controls [43]. The exact retained variants were carried into this single accession with the same source alleles, glycemic direction, and estimator used for infection outcomes. A concordant nominal control was defined consistently as a negative T2D coefficient per lower glycemic exposure with two-sided p < 0.05.
We evaluated 12 target or variant groups, including individual GPD2 variants. This control describes glycemic relevance on the specified source scale.
2.5. Statistical Analysis
Approximate instrument strength was calculated as F = (βX/SEX)2. MR interpretation assumes that the variants predict the specified exposure, are independent of shared causes of exposure and infection, and influence infection through the nominated target pathway. Exposure and outcome coefficients were aligned to the published effect allele before estimating a Wald ratio for one SNP or generalized IVW for a multi-SNP panel.
For the latter, the outcome covariance matrix was Ω = diag(SEY) R diag(SEY), where R is the signed allele-dosage correlation matrix. The slope was (βXᵀΩ−1βY)/(βXᵀΩ−1βX). Its standard error used multiplicative random effects, with dispersion bounded below by one.
We reconstructed R from complete genotypes of 503 European participants in 1000 Genomes Phase 3, accessed through Ensembl [44]. The channel pair had r2 = 0.000427; the six SLC5A2 variants required explicit LD adjustment.
A glucose-decreasing contrast of 0.1 mmol/L was expressed as −0.1 times the native outcome-to-glucose slope, with its standard error multiplied by 0.1. HbA1c target estimates used one source-standard-deviation decrease; Dai et al. describe the SLC5A2 source SD as 0.62 percentage points [31]. GPD1/PRKAG2 and GPD2 retain their respective source HbA1c scales.
We calculated two-sided Wald p values and 95% intervals after calibration. Gene-specific metformin estimates preserve their separate molecular and tissue contexts.
Benjamini–Hochberg q values were calculated for the 52 matched PPARG/channel proxy comparisons and, separately, the broader 156-estimate target screen. Leave-one-out analyses refitted the LD-adjusted model after removing each variant and its covariance row and column. These analyses generated 26 channel and 77 SLC5A2 fits.
We also compared the six-SNP and published low-LD two-SNP SLC5A2 sets. Supplementary Table S6 records the diagnostic methods appropriate to each panel. Variant screening queried smoking initiation (ieu-b-4877; GSCAN) and ever smoking (ukb-b-20261), and the four principal/functional PPARG and channel variants were screened across OpenGWAS traits at p ≤ 5 × 10−8 [45].
Endpoint-choice sensitivity compared the calibrated single-accession results with the two PPARG bacterial-pneumonia reconstructions combining related endpoint releases. The present re-estimation additionally checked source alleles, applied the specified glycemic decrement, and incorporated the reconstructed reference LD matrices. Variant identity, outcome accession, source coefficients, and fitted estimates accompany the Supplement.
The analysis used R 4.5.3, TwoSampleMR 0.7.9, MRPRESSO 1.0, mr.raps 0.4.3, and ieugwasr 1.1.0. The allele-calibrated and LD-adjusted calculations used Python 3.12.14 and NumPy 2.5.3. The analysis specifies its 1000 Genomes reference, covariance model, source-specific selection rules, and retained variants.
Associations are reported as nominal unless a q value is given.
3. Results
3.1. Target Instruments Represent Distinct Glycemic Pathways
The source inventory contained one primary variant each for GLP1R, INSR, and PPARG; two for KCNJ11/ABCC8; six for SLC5A2; and five retained variants across GPD1, GPD2, and PRKAG2. The PPARG and KCNJ11/ABCC8 functional variants were analyzed separately. Instrument-level alleles and F statistics are provided in Supplementary Table S4 (Figure 2).
The PPARG primary and Pro12Ala F statistics were approximately 14.76 and 11.11; the two KCNJ11/ABCC8 primary F statistics were 54.39 and 12.25. These denominators help explain the broad intervals when an outcome has few cases.
The metformin screen retained five variants across GPD1, GPD2, and PRKAG2 (Supplementary Table S5). Their source HbA1c GWAS and tissue-selection contexts define distinct gene-level perturbations. GPD2 expression-selection labels were preserved, and its three retained variants were displayed separately on their source HbA1c scale.
The analysis therefore provides a target-resolved comparison of the molecular pathways associated with glucose regulation.
3.2. Target Perturbations Map Phenotype-Specific Infection Associations
Figure 3 displays calibrated target trajectories across the same 13 outcome definitions. Each panel retains its target-specific exposure unit, so direction and phenotype pattern can be read alongside uncertainty. Twelve target or variant groups produced 156 estimates.
Their distribution shows that respiratory infection associations differ across receptor, channel, transporter, and metformin-related pathways. The functional-proxy comparisons below identify the most repeated pattern.
3.3. A Common Diabetes GWAS Defines the Positive-Control Comparison
The common-accession T2D results are shown in Supplementary Figure S1 and Supplementary Table S7. Ten of 12 target or variant groups met the uniformly applied negative-direction and nominal-P criterion. Per 0.1 mmol/L lower glucose, PPARG primary and Pro12Ala ORs were 0.368 and 0.265, respectively.
The LD-adjusted SLC5A2 coefficient was −0.635 (SE 0.289; p = 0.028) per source HbA1c SD decrease. GPD1 and PRKAG2 were supported using their actual retained single variants. The two-SNP KCNJ11/ABCC8 panel had heterogeneous SNP directions and a positive pooled coefficient (0.388; SE 0.307; p = 0.206), while ABCC8 S1369A had a negative control coefficient.
GPD2 rs115087802 had a negative but imprecise coefficient (p = 0.192). These contrasts distinguish the genetic perturbations within each pathway.
3.4. Target Perturbations and Infection Phenotypes
Supplementary Table S1 reports all 52 accession-matched PPARG and channel proxy estimates with their exposure unit, confidence interval, nominal p value, and q values. The complete 156-estimate target screen is supplied with the harmonized association inputs.
3.4.1. PPARG-Mediated Perturbation
Per 0.1 mmol/L lower glucose, the PPARG primary proxy was associated with the influenza-and-pneumonia composite (finn-b-J10_INFLUPNEU; 29,924 cases): OR 1.53 (95% CI 1.16–2.02; p = 0.0028). Pro12Ala showed a closely aligned estimate, OR 1.53 (1.13–2.06; p = 0.0056). For other acute lower respiratory infection (finn-b-J10_LOWERINF; 10,103 cases), the corresponding estimates were OR 1.78 (1.14–2.77; p = 0.0109) and OR 1.87 (1.16–3.01; p = 0.0105).
Together, these two diagnostic groups show a repeated nominal PPARG-associated respiratory pattern.
Both PPARG variants also pointed toward higher odds of confirmed respiratory tuberculosis, based on 247 cases: primary OR 49.95 (95% CI 3.25–768.33; p = 0.0050) and Pro12Ala OR 111.61 (5.79–2152.96; p = 0.0018). The small case count and glycemic denominator make this a directionally informative but imprecise estimate. Experimental studies connect PPARγ activity with macrophage responses during tuberculosis and viral lung injury [46,47].
For the single bacterial-pneumonia accession finn-b-J10_PNEUMOBACT, PPARG primary and functional estimates were OR 1.44 (95% CI 0.87–2.39; p = 0.159) and OR 1.41 (0.82–2.44; p = 0.214). Their similar point directions extend the respiratory pattern, while the intervals include unity. The comparison with related-endpoint combinations shows how phenotype definition changes precision.
The matched PPARG primary estimates were OR 0.83 (95% CI 0.54–1.28; p = 0.394) for sepsis occurrence and OR 1.18 (0.41–3.38; p = 0.752) for sepsis-related 28-day death. Both are population-control contrasts with intervals spanning unity. Their different case definitions separate susceptibility from the population probability of sepsis-related death.
3.4.2. KCNJ11/ABCC8 Channel Perturbation
The LD-adjusted two-SNP KCNJ11/ABCC8 bacterial-pneumonia estimate was OR 0.83 (95% CI 0.64–1.08; p = 0.1570). The functional ABCC8 S1369A estimate pointed in the other direction, OR 1.27 (95% CI 0.95–1.69; p = 0.1042). Source-allele alignment therefore identifies different genetic channel perturbations within the same infection phenotype.
For severe COVID-19 (ebi-a-GCST011075), the primary channel estimate was OR 1.22 (95% CI 0.79–1.88; p = 0.3640), while the functional estimate was OR 0.63 (95% CI 0.35–1.14; p = 0.1251). These population-control GWAS combine susceptibility and progression components. The genotype-specific directions are interpreted together with their broad intervals and differing T2D control patterns.
Figure 4 compares primary and functional proxy directions and displays LD-adjusted sensitivity structure. PPARG proxies agreed in direction for 10/13 matched phenotypes; channel proxies agreed for 3/13. The six nominal PPARG associations involved confirmed tuberculosis, influenza and pneumonia, and other acute lower respiratory infection, with q values of 0.072–0.095 across 52 tests.
None of the 156 broader-screen estimates had q < 0.05. Supplementary Figures S2–S14 display all four proxy estimates within each matched phenotype.
Supplementary Figures S15–S20 show calibrated proxy contrasts, relations between case count and precision, and sensitivity to endpoint-release choice. These comparisons retain the PPARG direction for the influenza-and-pneumonia composite and distinguish agreement within PPARG from the channel-variant contrast.
3.4.3. INSR-Mediated Insulin Signaling
The INSR rs8103934 bacterial-pneumonia estimate was OR 0.64 (95% CI 0.37–1.09; p = 0.0986) per 0.1 mmol/L lower glucose. The point direction motivates investigation of receptor-mediated host responses, with the interval showing the precision available. This exposure concerns INSR signaling and is biologically distinct from exogenous insulin administration.
3.4.4. Metformin-Related Molecular Targets
GPD1, GPD2, and PRKAG2 were evaluated separately using their published glycemic coefficients and tissue-selection contexts. The matched INSR, GPD1, and PRKAG2 trajectories appear in Supplementary Figures S21–S23. Experimental models and clinical cohorts provide complementary evidence on metformin-linked macrophage, lung inflammatory, pneumonia, and septic-shock outcomes [12,48,49,50,51].
The present gene-level estimates define specific molecular comparisons within that broader biology.
3.4.5. Other Target Perturbations
The GLP1R and LD-adjusted SLC5A2 trajectories are presented in Supplementary Figures S24 and S25. SLC5A2 had a nominal association with known-virus pneumonia (coefficient −7.11; SE 2.37; p = 0.0027 per source HbA1c SD decrease), with a wide interval on the odds scale and q = 0.145 across the broader screen. Its common-GWAS diabetes control supported the intended glycemic direction.
Contemporary SGLT2 inhibitor cohorts and trial syntheses provide an external clinical comparison [13,14,15,17].
Supplementary Figure S26 shows variant-level T2D control associations using the same source alleles as the infection models. The aligned PPARG proxies both pointed toward lower T2D odds, while the two primary channel variants had different directions. This variant-level view explains the control pattern without averaging biologically different perturbations.
The LD-adjusted sensitivity analyses generated 103 leave-one-out fits, including 77 SLC5A2 fits across 13 phenotypes. Supplementary Figures S27–S39 show association fits, individual-SNP estimates, funnel distributions, whitened residual diagnostics, covariance-aware leave-one-out estimates, and information concentration for each SLC5A2 endpoint. Twelve outcomes retained six variants; COVID-19 hospitalization retained five.
The case count, source unit, and actual instrument count are reported for each comparison.
The genome-wide-significant cross-trait screen returned 475 association records for the four named PPARG and channel variants, spanning glycemic, blood-cell, lipid, anthropometric, and other phenotypes (Supplementary Figure S40). Smoking initiation and ever-smoking queries resolved 35 of 36 planned variant–trait pairs; six had nominal p < 0.05 and none had BH q < 0.05 (Supplementary Figure S41). These screens characterize shared genetic associations and identify specific pleiotropy questions.
The three GPD2 variants are displayed individually across all 13 infection accessions in Supplementary Figure S42. Supplementary Figure S43 compares the six-SNP and published two-SNP SLC5A2 sets, each fitted with its own LD submatrix. Together, these analyses examine instrument membership while retaining the source target and glycemic contrast.
4. Discussion
The clearest infection pattern in our matched analysis involved PPARG perturbation. The near-gene and functional variants showed aligned nominal associations with the influenza-and-pneumonia composite and other acute lower respiratory infection, and both pointed in the same direction for confirmed respiratory tuberculosis. Comparing the two proxies within the same outcome GWAS connects this repeated pattern to a defined target pathway.
Separate analyses of channel, receptor, and metformin-related targets reveal how infection associations vary across the biological pathways involved in glucose regulation.
Hyperglycemia affects pulmonary immune metabolism, pathogen handling, and the transition from local inflammation to systemic infection [2,3,4,52]. Autophagy and iron-dependent lipid damage further link antimicrobial defense to lung and systemic tissue injury [53]. The present team’s earlier genetic analysis connected immunocyte iron metabolism to pneumonia and sepsis, while its hospital-acquired-pneumonia work identified clinical predictors of sepsis [5,6].
These complementary lines suggest that infection susceptibility in diabetes reflects both host metabolic state and pathogen context. Our target-specific comparison adds genetic evidence to this framework.
PPARG biology connects receptor activity with cell type and pathogen context. Pioglitazone increased proinflammatory responses in bronchial epithelial cells during experimental Pseudomonas pneumonia, whereas macrophage PPARγ activity has been associated with control of influenza-induced lung inflammation [46,54]. Tuberculosis studies also implicate PPARγ in macrophage–pathogen interactions [47].
Pro12Ala reduces receptor transcriptional activity, while thiazolidinediones activate PPARγ. The aligned near-gene and functional-variant estimates motivate experiments that distinguish these perturbation directions across airway epithelial cells and macrophages.
The PPARG findings also identify a useful phenotype pattern. The influenza-and-pneumonia composite and other acute lower respiratory infection capture different parts of the respiratory disease spectrum, yet the two proxies point in the same direction in both. Confirmed tuberculosis adds a mycobacterial phenotype to this pattern.
This comparison focuses follow-up on the relationship between PPARG signaling and pulmonary host defense, with pathogen-specific phenotypes helping to identify the relevant cellular responses.
Sepsis occurrence and sepsis-related 28-day death describe two population-based phenotypes. Their different case definitions help explain why a target can show different point directions across these GWAS. Mortality conditional on a sepsis diagnosis is a separate case-only contrast.
Current sepsis immunology emphasizes changing mixtures of neutrophil, macrophage, and pyroptotic responses, and proteomic studies identify further target diversity [52,55,56]. Together, these observations motivate evaluation of target signaling across infection stages using outcome definitions that match each clinical question.
The channel findings illustrate genetic heterogeneity within a therapeutic pathway. The primary KCNJ11/ABCC8 panel and functional ABCC8 variant agreed in three of 13 phenotype directions and differed in their diabetes controls. Their opposing point directions for bacterial pneumonia and severe COVID-19 identify a variant-specific question about channel signaling.
INSR adds a separate receptor-mediated comparison. These results link respiratory phenotypes to defined genetic perturbations and complement clinical treatment studies.
Metformin’s GPD1, GPD2, and PRKAG2 axes involve cytosolic glycerol metabolism, mitochondrial redox biology, and AMPK signaling [34,35,36]. Metformin also increases the signaling metabolite Lac-Phe through mitochondrial effects in intestinal cells, linking the drug to appetite and energy-balance pathways [37]. Keeping the three genetic targets separate preserves their molecular and tissue contexts and supports direct examination of each infection trajectory.
Experimental models have identified macrophage antimicrobial and lung inflammatory responses to metformin, and clinical cohorts have reported pneumonia and septic-shock outcomes among metformin users [12,16,48,49,50,51]. Future studies can test each molecular axis using tissue-specific expression and infection-specific colocalization.
Smoking-related pathways merit consideration in target-region MR. The variant screen recovered six nominal smoking-trait associations among 35 resolved records; the minimum BH q value was 0.086. These results identify candidate behavioral or pleiotropic links for further study.
The broader cross-trait screen also connected target variants with blood-cell, lipid, and anthropometric traits. Smoking-stratified GWAS and multivariable analyses can test whether those pathways contribute to target–infection associations [10,20,57].
Interpretation and Future Work
This study brings together a curated target inventory, functional-variant comparisons, a T2D control recheck, and infection phenotypes matched to specific GWAS accessions. The comparison identifies a repeated PPARG-associated pattern across related lower respiratory phenotypes and preserves the biological identity of channel, receptor, and metformin-related pathways. Naming the exact genetic proxy and outcome accession makes the target-specific findings directly comparable with future molecular and clinical studies.
The statistical interpretation rests on exposure calibration and outcome precision. Applying the specified 0.1 mmol/L glucose decrement yields modest PPARG odds ratios for the two larger respiratory phenotypes, while tuberculosis remains imprecise because of its 247 cases and small genetic glucose coefficients. Winner’s curse and exposure-association error can affect Wald ratios despite approximate F statistics above 10.
The highlighted associations are nominal; q values summarize the multiple-testing context. Single-SNP comparisons and the smoking/cross-trait screens assess different aspects of proxy validity, and the smoking screen estimates variant associations and preserves the distinction from a smoking-adjusted infection coefficient. Potential overlap between UK Biobank-derived exposures and some outcome resources could correlate association errors; participant overlap could not be quantified from summary metadata.
With weaker instruments, this correlation can draw estimates toward confounded observational associations. The functional comparison uses the same source populations and is biological corroboration within the source data.
The PPARG findings prioritize tissue-resolved genetic and functional studies of pulmonary host defense. Independent-ancestry GWAS, infection-specific colocalization, and allele-specific assays can test whether receptor transcription and downstream immune metabolism account for the repeated respiratory pattern. Separate GPD1, GPD2, and PRKAG2 analyses can identify which components of metformin-related biology contribute to host responses.
Clinical cohorts with smoking, glycemic history, and medication timing will provide a complementary test of how these pathways operate during treatment.
5. Conclusions
Genetically proxied glucose-lowering targets show distinct respiratory infection association patterns. The clearest repeated nominal pattern, which did not meet the FDR threshold, involved PPARG rs35240997 and Pro12Ala for influenza and pneumonia and other acute lower respiratory infection, with additional imprecise directionality in confirmed tuberculosis. Source-allele calibration distinguishes the channel variants, and LD-adjusted analyses describe SLC5A2 sensitivity across phenotypes.
Separate metformin-related gene estimates preserve their molecular biology. These results focus further investigation on PPARG-related pulmonary host defense and provide defined genetic comparisons for clinical and functional studies.