Between November 2019 and August 2021, 1,120 unrelated adult Lebanese individuals of Arab descent with type 2 diabetes mellitus (T2DM) were enlisted from outpatient clinics at LAU Medical Center-Rizk Hospital, Rafic Hariri University Hospital, and St. Marc Medical Center in the Greater Beirut area. Of the 1,120 participants screened, 90 were excluded for not meeting inclusion criteria, 64 for incomplete clinical data, and 18 for withdrawing consent (Figure 2). Of the 948 enrolled, 16 were excluded due to genotyping failure and 24 for incomplete neuropathy assessment, yielding a final analytical cohort of 908 T2DM patients: 382 with DPN and 526 without (DwPN) (Figure 2). Reporting followed STROBE recommendations, with a completed checklist provided in Supplementary Table S1.

T2DM was diagnosed based on clinical and laboratory criteria, with no reports of ketoacidosis by the patients. Treatments included diet, oral antidiabetic agents, and insulin; patients who initially required insulin received oral medications for at least 2 years. Blood pressure was checked twice in a seated position, with hypertension defined as BP >140/90 mm Hg on two occasions or the use of antihypertensive medication. Obesity was categorized as a body mass index (BMI) of 30 kg/m² or higher. The control group consisted of 695 healthy, euglycemic Lebanese individuals, matched by gender and geographic origin, with no personal or family history of diabetes. Given the practice of consanguinity in Middle Eastern populations, including Lebanon, we screened for close relatives using family IDs and recruitment records; none were identified, so clustering by family structure was unnecessary. Recorded demographic data included age, gender, ethnicity, BMI, age at diabetes onset, diabetes duration, family history of diabetes, and history of chronic complications and systemic illnesses.

DPN was assessed through clinical, electrophysiological, and patient-reported measures, including Quantitative Sensory Testing (QST) and Nerve Conduction Studies (NCS). Performed by clinicians blinded to genotype status, Nerve Conduction Studies (NCS), and the Michigan Neuropathy Screening Instrument (MNSI), were performed at all study sites. Inter-rater reliability was high (Cohen’s κ = 0.89 for NCS interpretation; κ = 0.85 for clinical examination). Neuropathy assessments and laboratory investigations were conducted during the same clinical visit, with genotyping performed on the same blood sample. All procedures occurred within a 7-day window, ensuring temporal alignment. Clinical evaluation included a structured symptom questionnaire assessing numbness, tingling, and neuropathic pain (burning, shooting, stabbing, or pins-and-needles), accompanied by examination of ankle reflexes, vibration perception (128-Hz tuning fork), and 10-g monofilament sensation. Symptoms were considered present when two or more 2 features were bilateral, distal-predominant, and persistent for more than 3 months. The MNSI questionnaire score was used to quantify symptom severity; scores >4 was deemed abnormal. Ten percent of NCS studies (n = 91) were independently re-reviewed by a second neurologist (κ = 0.89; 95% CI: 0.83–0.95).

QST assessed thermal (cold/warm) thresholds on the dorsum of the foot using the method of limits (1 °C/s ramp; baseline 32 °C), mechanical detection using von Frey filaments, and vibration perception using a biothesiometer. NCS evaluated motor and sensory nerve conduction parameters (amplitude, latency, conduction velocity), with electromyography performed when clinically indicated. The Neuropathy Disability Score (NDS) was computed from clinical and QST findings. Toronto Criterion 2 was fulfilled when two or more motor or sensory nerves demonstrated two or more abnormal electrophysiological parameters. Individuals with alternative causes of neuropathy (vitamin B12 deficiency, hypothyroidism, alcohol abuse, chemotherapy) were excluded.

A standardized multimodal algorithm, aligned with Toronto consensus guidelines, was used to ensure reproducible DPN classification. Diagnosis required abnormal findings in at least two independent modalities: clinical exam, MNSI, QST, or NCS. For NCS, the worst-affected limb was used; studies were considered abnormal if ≥2 parameters in one nerve or ≥1 in two nerves exceeded age-adjusted limits. QST values within ±5% of normative cutoffs were retested in triplicate and deemed abnormal if ≥2 repeated measures fell outside reference ranges. MNSI scores ≥4 were classified as abnormal. Single-modality findings were insufficient. This approach harmonized site-level assessments and reduced misclassification. Pain descriptors and thermal thresholds were available for a subset, allowing limited stratification by pain status and fiber type. No substantial APOE effect differences emerged across these subgroups, though small sample sizes warrant cautious interpretation.

Venous blood samples were drawn via venipuncture from both patients and controls after an overnight fast. Glucose levels were measured using the hexokinase method on Roche Cobas Integra 800 (Mannheim, Germany), while total hemoglobin and HbA1c levels were determined through colorimetric and immunoturbidimetric methods, respectively. Serum lipids, including total cholesterol, HDL, LDL, and triglycerides (TG), were enzymatically measured. Creatinine levels were assessed by the Jaffe reaction, and additional tests of liver and renal function, as well as serum electrolytes, were performed using Dade-Behring instruments.

Genomic DNA was extracted from the leukocyte-rich layer of EDTA-anticoagulated blood using the phenol-chloroform method. APOE genotyping was performed by PCR-RFLP using specific primers and the CfoI restriction enzyme. The digested PCR products were separated on a 5% NuSieve agarose gel. Genotyping quality control measures included running positive control DNAs representing the APOE genotypes (ϵ2/ϵ2, ϵ2/ϵ3, ϵ2/ϵ4, ϵ3/ϵ3, ϵ3/ϵ4, ϵ4/ϵ4) in every batch, and re-genotyping a subset of control and case samples by Sanger sequencing; concordance was ≥99%. Call rates exceeded 98% for all samples. Duplicate samples (n = 45, 2.8% of total) demonstrated 100% concordance. The distribution of genotypes in controls satisfied Hardy-Weinberg equilibrium (χ² = 2.14, p = .71). Laboratory personnel were blinded to case-control status during genotyping. Concordance exceeding 99.5% were maintained for inter-plate and intra-plate controls.

All analyses were performed using SPSS v29 (IBM, Armonk, NY). Continuous variables are presented as mean ± standard deviation (SD), while categorical variables are shown as counts and percentages. Missing data for covariates (<5%) and APOE and DPN (<2%) were minimal. Since the patterns aligned with missing-at-random, we used multiple imputation for variables with >1% missingness, analyzing exposure and outcome as complete cases. Sample size calculations were performed using GPower 3.1.9.7. Based on APOE allele frequencies previously reported in Lebanese populations and targeted detectable effect sizes of OR 1.5–2.0 at α = 0.05. pilot data indicating ϵ4 allele frequencies of 9% in controls and 15% in DPN cases (effect size w = 0.10), we determined that 382 DPN cases and 526 non-DPN controls would provide 85% power to detect this difference at α = 0.05 (two-tailed). Owing to the rarity of ϵ2/ϵ2 and ϵ4/ϵ4 (<1%), recessive models were underpowered. Interaction analyses (e.g., ϵ4 × TG) were exploratory and not powered a priori, and thus their results should be interpreted with caution. For logistic regression models with 10 covariates and an anticipated OR of 1.5–2.0, our sample size exceeds the recommended minimum of 10 events per variable, providing adequate power (>80%) for multivariable analyses.

Allele and genotype frequencies were calculated by direct gene counting, and the differences between groups were assessed using two-tailed Student’s t-tests (continuous variables) and Pearson’s χ² or Fisher’s exact tests, depending on expected cell counts or sample size (categorical variables), as appropriate. We prespecified a single primary contrast comparing ϵ4 carriers (ϵ3/ϵ4 + ϵ4/ϵ4) with the ϵ3/ϵ3 reference genotype to minimize type I error. Exploratory analyses involving ϵ2-containing genotypes and full multi-category models were FDR-corrected, and the Hardy-Weinberg equilibrium (HWE) was assessed separately in controls, DwPN, and DPN groups using exact mid-P correction. APOE-DPN analyses were primarily modeled additively, with ϵ2, ϵ3, and ϵ4 dosage assessed via logistic regression. Dominant models grouped ϵ2- (ϵ2/ϵ3, ϵ2/ϵ4) and ϵ4-containing (ϵ3/ϵ4, ϵ4/ϵ4, ϵ2/ϵ4) genotypes against ϵ3/ϵ3 (OR = 1.00). Recessive models (ϵ2/ϵ2, ϵ4/ϵ4) were used only for sensitivity analyses due to low frequencies.

Associations between APOE variants and DPN outcomes were examined through multivariate logistic regression, reporting odds ratios (ORs) and 95% confidence intervals (CIs). We compared minimally and lipid-adjusted models, recognizing lipids as mediators and confounders. The primary model included sex, HbA1c, lipids, smoking, hypertension, medications, age at T2DM onset, and duration to assess lipid effects, adjusted for HbA1c/glycemic control. TG were log-transformed due to skewed distributions; values >5.6 mmol/L were flagged, with three retained despite no secondary hypertriglyceridemia. Estimates reflect direct APOE effects, with all models adjusted for statin use. Sensitivity excluding statin users was consistent. Type I error was controlled by grouping analyses: APOE-DPN associations were corrected with Holm-Bonferroni, and lipid/exploratory analyses were corrected with Benjamini-Hochberg FDR. Adjusted p-values are reported, with primary inference based on Holm results. Statistical significance was defined as p < 0.05.

We examined effect modification by testing the interaction between ϵ4-carrier status (ϵ3/ϵ4 or ϵ4/ϵ4 vs ϵ3/ϵ3) and log-transformed TG, centered at the sample mean to reduce collinearity, using a logistic regression model adjusted for age, sex, T2DM duration, HbA1c, cholesterol measures, hypertension, smoking, and statin use. Stratum-specific adjusted OR (aORs) were reported for clinically relevant TG categories (<1.7 vs ≥1.7 mmol/L) and validated with TG quartiles. Marginal effects with 95% CIs across the TG distribution were estimated from the interaction model with covariates fixed at sample means or reference levels, employing robust standard errors and a two-sided α=0.05.

This study was approved by the Institutional Review Board (IRB) of St. Marc Medical Center (protocol number SMMC-2019-0103, dated October 17, 2019) and adhered to the Declaration of Helsinki. Genetic data were pseudonymized by replacing personal identifiers with unique study codes, with the key securely stored separately and accessible only to authorized personnel. Informed consent was obtained from all participants to ensure confidentiality and compliance with ethical standards. De-identified data underlying the findings, the full data dictionary, analysis code, and workflow documentation are available at the Mendeley Repository: https://doi.org/10.17632/b9mmtpyv4k.1.

Table 1 summarizes the clinical characteristics of T2DM patients with DPN, those without DPN (DwPN), and normoglycemic controls. While the Lebanese population exhibits low genetic substructure, we tested whether regional variation may bias APOE-DPN associations by stratifying participants by birthplace (North, Mount Lebanon/Beirut, Bekaa/South). While ancestry-informative principal components were unavailable, regional matching and prior population-genetic data suggest that meaningful stratification is unlikely. Between DPN and DwPN, SMDs indicated small imbalances for sex (SMD = 0.16) and HbA1c (SMD = –0.13), and moderate imbalances for age (SMD = 0.40) and TG (SMD = 0.23). Other traits showed minimal differences (|SMD|<0.10). A higher proportion of females (p = 0.016) and age (p < 0.001) were seen in the DPN group compared to the DwPN and control groups.

A higher mean BMI and a higher prevalence of obesity (BMI > 30 kg/m²) were observed in both DPN and DwPN groups compared with controls (p = 0.005), but not between T2DM patient subgroups (p = 0.118). Hypertension was significantly elevated in both DPN and DwPN groups compared to controls (p < 0.001). DPN and DwPN cases were matched for BMI, family history of T2DM, age of onset, fasting glucose, total cholesterol, HDL-cholesterol and LDL-cholesterol. Significant differences between the T2DM subgroups were noted in gender (p = 0.016), age at inclusion in the study (p < 0.001), waist-hip ratio (p = 0.049), hypertension (p < 0.001), HbA1c (p = 0.028), TG (p = 0.003), and urea (p < 0.001). While triglyceride levels were significantly higher in the DPN group than in both DwPN and controls (p = 0.003), total cholesterol (p = 0.479), LDL (p = 0.579), and HDL (p = 0.508) levels were not significantly different between DPN and DwPN groups compared to controls. In addition, urea levels were significantly elevated in the DPN group compared to DwPN and controls (p < 0.001). Creatinine levels were higher in the DPN (101.9 ± 63.0 μmol/L) and DwPN (97.3 ± 67.5 μmol/L) groups than in controls (61.9 ± 37.7 μmol/L). However, this difference was not statistically significant (p = 0.294).

The frequencies of the APOE alleles among Lebanese were comparable to Middle Eastern populations (Iran, Saudi, Arabia, Egypt and Turkey), as well as South Americans and Africans (Supplementary Table S2). Results presented in Table 2 demonstrated significantly higher APO-ε2 and APO-ε4 allele frequencies, but lower APO-ε3 allele frequency, among T2DM patients compared to non-diabetic control subjects (all at p < 0.001). This resulted from the significantly higher ε2/ε3, ε4/ε4, ε3/ε4, and ε2/ε4, and significantly lower ε3/ε3 genotype frequencies among T2DM cases (all at p < 0.001), which assigned positive and negative associations to these genotypes, respectively (Table 2). These differences persisted after applying the Bonferroni correction for multiple comparisons, with Holm-adjusted p-values reported for primary APOE–DPN comparisons and Benjamini-Hochberg FDR-adjusted p-values reported for secondary analyses.

The distribution of APOE alleles and genotypes was examined in T2DM patients with DPN and those with DwPN, who served as controls. HWE analyses showed that controls (p = 0.64) and the DwPN group (p = 0.21) were in equilibrium, and a mild HWE departure was noted for the DPN group (p = 0.048). Significantly higher APO-ε2 (p < 0.001) coupled with lower APO ε3 (p < 0.001) allele frequencies were seen in DPN compared to DwPN patient subgroups (Table 2). APO ε4 allele frequency was not significantly different between the two T2DM patient subgroups (p = 0.269). Significantly higher frequencies of ε2/ε3 (p = 0.001) and ε2/ε4 (p < 0.001) and significantly lower frequencies of ε3/ε3 (p < 0.001) APOE genotypes were seen among DPN cases, which conferred DPN-susceptible and -protective nature to these genotypes, respectively (Table 2). Compared to the ϵ3 allele, both ϵ2 and ϵ4 alleles conferred increased odds of DPN.

At the genotype level, ϵ2/ϵ3 (OR = 1.85 [1.27–2.69]), ϵ2/ϵ4 (OR = 1.87 [1.29–2.70]), and ϵ3/ϵ4 (OR = 1.62 [1.08–2.44]) were significantly associated with increased odds of DPN, while ϵ4/ϵ4 (OR = 1.01 [0.57–1.78]) showed no significant effect (Figure 3). In logistic regression, ϵ4 carriers (ϵ3/ϵ4+ϵ4/ϵ4) had higher odds of DPN versus ϵ3/ϵ3 in the minimally adjusted model (age, sex, duration, HbA1c) and the lipid-adjusted model (adding total cholesterol, HDL, LDL, log-TG, hypertension, smoking, statin use) (report aOR, 95% CI, p for both models). Pooled ϵ2-containing genotypes (ϵ2/ϵ3+ϵ2/ϵ4) also showed higher odds (Figure 3).

Genotype-specific lipid patterns are shown in Table 3. Compared to DwPN cases, DPN cases carrying the ϵ2/ϵ3 genotype had significantly higher total cholesterol (p = .018) and LDL-cholesterol (p = 0.045), and lower HDL-cholesterol (p = 0.042) (Table 3). Similarly, ϵ3/ϵ4 genotype was linked to significantly higher total cholesterol (p = 0.002) and TG (p < 0.001) (Table 3). DPN cases with ϵ3-carrying genotypes (ϵ3/ϵ3 + ϵ2/ϵ3 + ϵ3/ϵ4) had higher total cholesterol (p = 0.030) and TG (p < 0.001), but lower HDL-cholesterol (p = 0.017), while ϵ4-carrying genotypes (ϵ4/ϵ4 + ϵ3/ϵ4 + ϵ2/ϵ4) showed similar trends (p = 0.004 for cholesterol; p < 0.001 for TG and p = 0.032 for HDL-cholesterol). Additionally, higher LDL-cholesterol was associated with ϵ3-carrying genotypes in the Apo E3 group (p = 0.009) and the ϵ2/ϵ3 genotype (p = 0.045). A significant ϵ4 × log(TG) interaction was observed, with ϵ4 carriers showing higher DPN odds in the high TG group (≥1.7 mmol/L) but attenuated association in the low/normal group (<1.7 mmol/L); marginal effects indicated a monotonic rise in ϵ4-related risk across TG, reaching significance near the upper tertile. Sensitivity analyses confirmed robustness, with exclusion of lipid-lowering therapy users or lipid adjustment yielding similar or stronger effects, suggesting partial lipid mediation. Results were unchanged with stricter DPN definitions or exclusion of extreme TG values (>5.6 mmol/L), and sex-stratified models showed comparable effects in men and women.

Logistic regression was performed with two adjustment models. Model 1 adjusted for age of onset, lipid profile, fasting glucose, and HbA1c, while Model 2 additionally adjusted for nephropathy, retinopathy, gender, and hypertension (Figure 3). Apo ϵ2-containing genotype group (ϵ2/ϵ3 + ϵ2/ϵ4) were not significantly associated with altered odds of DPN according to Model 1 (p = 0.343) and Model 2 (p = 0.196). Although not statistically significant, the Apo ϵ3-containing genotype group (ϵ3/ϵ3+ϵ2/ϵ3+ϵ3/ϵ4) indicated a trend towards increased odds of DPN in Model 1 (p = 0.092) and Model 2 (p = 0.065). In contrast, the Apo ϵ4-containing genotype group (ϵ4/ϵ4+ϵ3/ϵ4+ϵ2/ϵ4) was significantly associated with a higher odds of DPN in both Model 1 (p = 0.013) and Model 2 (p = 0.028).

Figure 1 illustrates the correlations between APOE genotypes and clinical features of DPN. Significant positive correlations were observed between ϵ2/ϵ3 and both retinopathy (ρ = 0.60, p < 0.01) and diabetes duration (ρ = 0.54, p < 0.05). The ϵ4/ϵ4 genotype showed a strong positive correlation with BMI (ρ = 0.68, p <.05) but an inverse association with disease duration (ρ = –0.72, p < 0.05). Conversely, the ϵ3/ϵ3 and ϵ2/ϵ4 genotypes demonstrated weaker or nonsignificant correlations across most clinical parameters. These results highlight distinct genotype-specific patterns that influence DPN-related clinical characteristics.

Stratified analyses and formal interaction testing were performed to evaluate the contribution of key clinical factors to the APOE effects on DPN risk. Consistent association between ϵ4-containing genotypes and increased odds of DPN was noted across the diabetes low duration (<10 years: OR [95% CI] = 1.58 [1.12–2.23]) and prolonged duration (≥10 years: OR [95% CI] = 1.67 [1.21–2.31]) strata (p interaction = 0.76), HbA1c categories (<8%: OR = 1.61; ≥8%: OR = 1.64; p interaction = 0.89), and hypertension (present: OR = 1.59; absent: OR = 1.66; p interaction = 0.82). A significant interaction was observed with hypertriglyceridemia status (TG >2.0 mmol/L), in which ϵ4 effects were amplified among hypertriglyceridemic patients (OR = 2.34, 95% CI: 1.62–3.38) compared to normotriglyceridemic patients (OR [95% CI] = 1.21 [0.78–1.87], p interaction = 0.03), suggesting that lipid-mediated mechanisms may partially explain the association between ϵ4 variants and altered odds of DPN. Notably, these associations persisted after adjustment for HbA1c, suggesting that lipid pathways may operate independently of glycemic control to modulate genetic susceptibility to DPN.