Serum ω-6/ω-3 polyunsaturated fatty acids ratio and diabetic retinopathy: A propensity score matching based case-control…
] to 463 million in 2019, and the number is expected to rise to more than 690 million by 2045 [
]. If poorly treated, DM can lead to multiple organ damages, which will inevitably reduce quality of life and increase early mortality. As a frequent complication of DM, diabetic retinopathy (DR) is a chronic DM-induced eye disease and a leading cause of global visual impairment and blindness burdens [
]. Pang et al. [
] have reported that nearly all type 1 and the majority (over 60%) of type 2 diabetic (T2D) patients eventually suffer from some degree of DR after 20 years of diabetes. Besides, due to its silent clinical characteristics, almost all DR patients are diagnosed at a moderate or advanced stage of disease [
]. This not only greatly decreases its clinical efficacy, but also leads to a huge waste of limited healthcare budgets. So, we need a simple and effective approach to distinguish T2D patients with DR from those without DR in time, which will be fruitful for its early diagnosis, timely treatment, and clinical administration.
]. And available evidence suggests that lipids metabolism may play a central role in the initiation and progression of DR [
]. However, routine lipid profiles such as low-density lipoprotein cholesterol (LDL-C), high-density lipoprotein cholesterol (HDL-C), total cholesterol (TC), and triglyceride (TG) are inadequate to serve as sensitive and specific biomarkers for the timely detection of DR because of the complex relationships between lipids and DR [
]. Available lines of evidence suggest that polyunsaturated fatty acids (PUFAs) are closely linked to the risk of DR. Nevertheless, overall results from both intervention and observational studies remains inconclusive. Although it is generally conceptualised that ω-3 PUFAs are beneficial [
], no association between the supplement of ω-3 PUFAs and DR has been detected in several studies [
]. While ω-6 PUFAs are reported to be positively associated with the risk of DR due to their inductions of adhesion molecule expression and leucocyte adhesion in human retinal endothelial cells [
]. However, two clinical studies conducted in Europe reveal that ω-6 PUFAs were inversely associated with DR occurrence [
]. In a recently published paper, ω-6 and ω-3 PUFAs are considered to be competitors combining the same desaturase and elongase in the process of metabolism [
], and different ω-6/ω-3 PUFAs ratio (PUFAR) may result in different severity of inflammation and angiogenesis [
]. All these findings strongly indicate that PUFAR may be linked to the pathogenesis of DR. To test this hypothesis and avoid potential underfitting or overfitting as well as collinearity, the association of PUFAR instead of individual PUFAs with DR should be carefully investigated. However, to the best of our knowledge, studies on the relationship between PUFAR and DR are rare. Limited evidence only comes from observational studies and levels of PUFAs are mainly evaluated by self-reported dietary intake, which inevitably leads to unsatisfactory precision of PUFAs measurements and significantly reduces the credibility of relevant conclusions.
To overcome these limitations, this study aims to comprehensively quantify the association of serum PUFAR with DR, as well as to evaluate its role when using PUFAR as a sensitive and specific biomarker in the detection of DR. An ideal cut-off value of PUFAR is also proposed.
2.1 Study design and participants
]. In brief, we ascertained 195 type 2 diabetic (T2D) patients (83 with DR and 112 without DR) from two affiliated hospitals of Wenzhou Medical University and Anhui Medical University, respectively, from August 2017 to June 2018. The DR group included 60 patients with non-proliferative diabetic retinopathy (NPDR: 9 mild, 31 moderate, and 20 severe) and 9 patients with proliferative diabetic retinopathy (PDR). DR status was independently assessed by two experienced ophthalmologists strictly following the guidelines for clinical diagnosis and image screening [
- Han X.
- Yang K.
- Gross R.W.
]. Details of grading DR were presented in Supplementary Methods. To adjust for the potential impacts due to possible confounders and improve comparability of the results to some extent, 69 pairs of T2D patients with DR and those without DR were matched by age, gender, body mass index (BMI) and glycosylated haemoglobin A1c (HbA1c) at a ratio of 1:1 using propensity score matching (PSM) approach. To fully describe the changes of ω-3 and ω-6 PUFAs in participants with different health states, we additionally matched another 69 healthy volunteers from the routine physical examination cohort from the Second Affiliated Hospital of Wenzhou Medical University with T2D patients by age, gender and BMI via PSM algorithm at a ratio of 1:1 (Supplementary Fig. 1).
2.2 Demographic and clinical covariates
Demographic variables containing age, gender, diabetes duration, smoking, drinking, history of diabetes treatment, hyperlipidaemia and others were obtained via a standardised questionnaire. To improve and guarantee the quality of information acquired, a 10% sub-sample of participants were re-interviewed within 3 weeks of the first interview by a full-time investigator strictly following the protocol of quality control and standardised operation procedure (SOP). Clinical features including fasting plasma glucose (FPG), HbA1c, routine lipid profiles such as HDL-C, LDL-C, TC, TG and others were detected by the automatic biochemical analyser (Roche, Cobas c311).
2.3 Anthropometric measurements
All participants accepted physical examinations by trained investigators according to the SOP. Sitting systolic blood pressure (SBP) and diastolic blood pressure (DBP) were measured three times by standard mercury sphygmomanometers after at least 5 min of rest during the enrolment, and average SBP and DBP values were used for further analyses. In addition, weight and height were also measured to calculate BMI using the following formula: BMI (kg/m2) = weight (kg) / squared height (m2).
2.4 Polyunsaturated fatty acids data
]. In brief, serum samples were tested in a random order using an ultra-performance liquid chromatography-electrospray ionisation-tandem mass spectrometry system (UPLC-MS/MS) at the central laboratory of Metware, Inc, a professional and experienced metabolomics institution in China. Technicians were also blind to the characteristics of subjects during the measurement of PUFAs. Individual PUFAs levels were expressed as percentages of the total fatty acids identified (%) and each PUFAs was classified into ω-6 or ω-3 family according to the position of the first double bond from the methyl end of the acyl chain. Total ω-3 and ω-6 PUFAs were determined by the sum of all associated PUFAs in each category, respectively. Finally, the ω-6/ω-3 PUFAs ratio (PUFAR) was calculated using total ω-6 PUFAs divided by total ω-3 PUFAs for clinical interests in balancing the two major PUFAs subclasses.
2.5 Sample size estimation
2.6 Statistical analysis
]. Then, population characteristics of participants were compared as follows: (1) Normally distributed variables were described as mean ± standard deviation, and analysis of variance (ANOVA) or student t-test was applied to assess the differences amongst or between groups. While obviously skewed data would be expressed as median (1st quartile, 3rd quartile) and Kruskal-Wallis H test or Mann–Whitney U tests would be performed for the comparisons. (2) All categorical variables were presented as frequency (percentage) and chi-square or Fisher’s exact tests were conducted to compare the differences of proportions amongst or between groups.
To investigate the independent association of PUFAR with the odds of DR, multivariable logistic regression models adjusting for confounders screened by a directed acyclic graph (DAG) (Supplementary Fig. 2) were used in the following two ways: with PUFAR as continuous variables [scaled to interquartile range (IQR)] and as categorical variables (tertiles), in which the linear trend tests were also carried out. Furthermore, a potential nonlinear association between PUFAR and the presence of DR was additionally evaluated using the restricted cubic spline (RCS) regression model, with 3 knots at the 5th, 50th, and 95th percentiles. To test the robustness and consistency of the results in different subgroups, we repeated all analyses stratified by age (≤ 55, > 55 years), gender (male, female), BMI (≤ 23.9, > 23.9 kg/m2), FPG (≤ 7, > 7 mmol/L), hypertension (no, yes), and lipid lowering therapy (no, yes), respectively. In addition, possible modifications induced by above stratified factors on the association between PUFAR and DR were also carefully evaluated by including their interaction terms with PUFAR in the logistic regression models. Depending on the potential causal relationships of PUFAR and DR as well as hypertension (Supplementary Fig. 2), we assumed that hypertension might affect the links between PUFAR and DR to some extent. To test our hypothesis, mediation analysis was further performed. Odds ratio (OR) and 95% confidence interval (CI) of the total effect (TE), natural direct effect (NDE), and natural indirect effect (NIE) were estimated by the ‘mediation’ package in R software.
To develop a simple and effective predictive model for evaluating the value of PUFAR in DR detection, 69 pairs of cases and controls was randomly split into independent training and testing sets at a ratio of 7:3 by scikit-learn package of Python version 3.8.8 (Copyright © 2001–2021 Python Software Foundation). Goodness-of-fit of models were assessed by calibration curves and Hosmer-Lemeshow tests. While discrimination ability of the predictive model was evaluated by the area under the receiver operating curve (AUC). Then, depending on the cut-off value determined by RCS, the sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) were also calculated in both the training and testing sets.
Furthermore, sensitivity analyses were also conducted to examine the robustness and consistency of our results. First, we repeated the analysis using data without any missing values in covariates to assess the influence of missing data on our findings. Second, single imputation approach, median for continuous variables and mode for categorical ones, was also performed to assess the influence of different imputation methods. Third, covariates associated with metabolic syndrome such as dyslipidemia and hypertension were further controlled to avoid their possible confounding influences on our results. Forth, dietary information such as different types of cooking oils, consumption of oils and salt were also adjusted to evaluate whether the association of PUFAR with DR remained robust when controlling for dietary variations.
Data management and statistical analyses were carried out using Stata/MP 15.1 for windows (© 1985–2017 Stata Corp LLC, College Station, Texas 77,845, USA), Python Version 3.8.8 (© 2001–2021 Python Software Foundation), and R Version 4.0.4 (R Foundation for Statistical Computing, Vienna, Austria). All tests were two-sided and the significance level was set as p ≤ 0·05.
2.7 Ethic statement and approval
The protocol of the present study had been approved by the Eye hospital of Wenzhou Medical University ethics committee (Number: KYK  46). All participants were informed about the study, participation was voluntary, and all signed written informed consent.
2.8 Role of the funding source
The funders of this study had no role in study design, data collection, data analysis, data interpretation, or writing of the report. The corresponding authors had full access to all data in the current study and had final responsibility for the decision to submit for publication.
To the best of our knowledge, this was the first study to comprehensively quantify the association of PUFAs and PUFAR with DR depending on a multi-centre, PSM designed study, and the levels of PUFAs were determined by UPLC-MS/MS system instead of participants’ self-report. Based on the principle of data mining approach, the potential role of PUFAR in DR detection was also carefully examined. Our findings strongly suggest that exposure to higher PUFAR is an important protective factor of DR, and PUFAR with an appropriate cut-off value (PUFAR ≤ 35) can serve as an ideal biomarker for DR identification though its alteration was not observed amongst different stages of DR. In addition, our findings advance an understanding of the possible differential links between serum PUFAR and different disease statuses of diabetes.
]. However, no relevant study has further explored the possible changes of PUFAs in patients from DM to DR. This study found that the levels of four ω-6 PUFAs (EDA, GLA, DGLA, and AA) and three ω-3 PUFAs (ALA, EPA, and DHA) would be apparently down-regulated from healthy controls to DM and DR while two plant-derived ω-3 PUFAs (SDA and ETA) showed contrary trends. And the relationships of PUFAs with DR was complex, simply linear trends could not account for the variations differ from healthy controls to DM to DR. Although comparison across studies of PUFAs and disease is complicated by differences in study design and exact exposure and outcomes examined, consistent with our findings, results based on previous epidemiological studies have suggested that ALA, an essential precursor of long chain ω-3 PUFAs, has protective effects on improving glucose homoeostasis [
]. EPA and DHA, known as metabolites of ALA, are potent biological regulators with therapeutic and preventive effects on human health. And SDA, an intermediate metabolite of ALA, helps to enhance the tissue levels of EPA and is a potential sustainable source to generate EPA and DHA. However, it is evident that EPA and DHA are more potent than SDA in reducing the risk of chronic diseases [
]. So, negative effects of SDA on DR detected in our study should be interpreted with caution, as it might be due to the inhibition of conversion from SDA to EPA and DHA in DR patients, rather than SDA itself being deleterious. As the potential effect of ETA on DR was rarely explored before, insufficient evidence could be obtained to compare with our findings. Depending on the comparable changes with SDA amongst the three groups in this study, it is possible to hypothesize that ETA has a similar biological effect as SDA on the initiation and development of DR.
]. Available evidence from a large European cohort demonstrated the inverse association between EDA and diabetes [
]. Besides, Shen et al. [
] reported that a decrease in AA was closely correlated with a decrease in anti-oxidants and an increase in pro-inflammatory molecules in diabetic patients. Additionally, a parabola trend had also been observed between LA and the progression of the disease. Two clinical trials confirmed that LA-rich diet could inhibit the development of microangiopathy or the deterioration of DR [
]. While a cross-sectional study observed that higher LA was positively associated with the risk of DM [
]. This variability between findings can variously be attributed to the complex interplay of metabolic factors and their different disease statuses. It is believed that PUFAs of ω-6 and ω-3 families compete for the same set of enzymes (Δ6 and Δ5 desaturases and elongases) and metabolic pathway [
]. Besides, the conversion of LA (a chief member of ω-6 PUFAs), and ALA (an essential member of ω-3 PUFAs) into their long-chain PUFAs depend on the ratio of ingested ω-6/ω-3 PUFAs [
]. Therefore, it is not strange that PUFAR will be more important in comparison of single PUFAs.
] while quite contrary to the negative relationship between PUFAR and DR illustrated in our study. It has been generally accepted that high PUFAR promotes the pathogenesis of many chronic diseases [
]. However, type 2 diabetes mellitus is a disturbance of metabolic homoeostasis with highly variable aetiology and progression [
]. The changes in PUFAR caused by metabolic disorders may not be a simple process that continues to increase with disease progression. In addition, the levels of individual PUFAs and PUFAR are maintained in a dynamic equilibrium. The ratio is influenced by individual PUFAs, and the conversion of essential PUFAs, in turn, depends on the ratio [
]. These may account for the results that a parabolic trend of PUFAR was observed with the progression of diabetes, and although the level of PUFAR between the healthy controls and the DR was similar, the composition of individual PUFAs was completely different. On the other hand, the level of PUFAR showed in our study was relatively higher than the others. The main reason was that LA (ω-6 PUFAs) was found to be significantly higher while DHA (ω-3 PUFAs) was significantly lower in comparison of the results showed in other studies. As ω-6 and ω-3 PUFAs are essential fatty acids that must be consumed in the diet and the conversion of LA and ALA to their higher metabolites is limited by the rate-limiting enzyme (△6-desaturase), we speculate that the dietary intake of ω-3 PUFAs, especially DHA, is relatively low in the Chinese population, whereas the dietary intake of ω-6 PUFAs, especially LA, is generally high. Therefore, an optimal threshold of 35 for PUFAR may be more applicable to populations with similar dietary habits as Chinese.
], SBP was consistently shown to be associated in the majority of population-based studies, while the association of diastolic blood pressure (DBP) was less consistent [
]. PUFAs were previously found to be significantly associated with SBP [
]. But the possible intermediate effects for SBP between PUFAs and DR have not been recognised yet. Therefore, we conducted mediation analyses of SBP in the relationship between PUFAs and DR in this study and found that LA alone, total ω-6 PUFAs, and PUFAR met the criteria for further mediation analyses. The weak mediation effects observed might be attributed to the actual situation that hypertensive patients included in our study had only a slight increase in blood pressure, or there may be other significant and larger effects to mediate the association between PUFAR and DR. Therefore, further studies are expected to verify the results.
The main strengths of this study may be summarised as follows. Firstly, the patients diagnosed with DR in our study were from endocrinology department and had no evident ocular symptoms, which guaranteed the target of early DR identification. Secondly, the cases and controls were matched by a PSM approach, which might largely increase the comparability of participants in many potential confounding factors. Thirdly, the level of serum PUFAs were determined by UPLC-MS/MS system instead of participants’ self-reports, which greatly facilitated accurate examination of serum PUFAs exposure. Fourthly, study population was enroled from two study centres covered over 150 million people in Zhejiang and Anhui provinces, China, which might decrease potential selection bias to some extent. Fifthly, adjusted confounding variables in the study were screened via a directed acyclic graph (DAG), which strongly indicated that the management for confounders was rigorous and considerate because of the combination of DAG and PSM. Finally, potential mediating effects of hypertension on the association between DR and PUFAR are also firstly investigated. All measures mentioned above will obviously improve the robustness and credibility of our findings.
Our study also inevitably has several limitations. The case-control design prevents us from clarifying the causal relationship and mechanisms between DR occurrence and PUFAR. Although our findings remain to be confirmed by large scale longitudinal studies, the results clearly revealed that serum PUFAs might be linked to the risk of DR since different PUFAs levels were observed in participants from healthy controls to DM and DR. Besides, our sample size was relatively small, which may affect the results to some extent. However, a small sample size does not necessarily mean that the sample size is insufficient. Our results clearly revealed that the current sample size was sufficient enough to meet the requirements of statistics (power >0.8) and to guarantee the reliability of our findings. In addition, the cut-off value of PUFAR at 35 to identify patients at high risk of DR was determined based on a case-control study and needs to be confirmed in additional prospective cohort studies. Moreover, confounding variables attained from the self-reported questionnaire may be at risk of some misclassification which may not be balanced across the groups. And no specific nutrition information was obtained via questionnaire in this study. This may partly prevent us from completely removing the potential influences due to dietary intake on the final conclusion.
In conclusion, this was the first propensity score matching based study to thoroughly qualify the relationship between DR and serum PUFAR as well as ω-6 and ω-3 families of PUFAs. We detected a negative monotonic relationship between PUFAR exposure and the presence of DR. Furthermore, PUFAR could be used as a specific and sensitive biomarker to distinguish type 2 diabetic patients with early DR from those without DR though its alteration was not observed amongst different stages of DR, and PUFAR = 35 may be an optimal cut-off value. These findings may provide new insights into the effective administration of DR prevention and control.
Shuzhen Zhao and Dongzhen Jin conducted the research, analysed the data, and wrote the manuscript. Shengyao Wang, Huihui Li, Yujie Chang, Yange Ma and Yixi Xu analysed the data. Ruogu Huang, Mengyuan Lai, Zhezheng Xia, Mingzhu Che and Jingjing Zuo contributed to the epidemiological investigation, sample handling, data management, and analysis. Chengnan Guo, and Fang Peng repeated the data analysis independently. Guangyun Mao, Chao Zheng, and Depeng Jiang designed the study, thoroughly reviewed and edited the manuscript. All authors contributed to critical revision of the manuscript, and approved the final version. The corresponding authors had full access to all the data in the study and had final responsibility for the decision to submit for publication.