Introduction
The fatty acid (FA) profile of blood and tissues integrates the complex interplay between dietary intake of FAs and endogenous FA metabolism, and has been shown to be associated with risk of type 2 diabetes in a number of studies [
1‐
6]. As a potential mechanism, there is a long-held view that the fluidity of the cell membrane, which is importantly determined by its FA composition, affects cellular functions [
7]. Despite this notion of an important role of membrane fluidity for diabetes development, there is scarce data to support the relevance of this hypothesis. In cell culture studies, it has been shown that membrane fluidity affects glucose transport across membranes as well as the properties of the insulin receptor [
8‐
10]. However, we are not aware of human studies investigating the relevance of alterations of the fluidity of cell membranes for the incidence of type 2 diabetes.
Membrane fluidity is strongly determined by the van der Waals forces between FAs in the phospholipid bilayers. These van der Waals forces are dependent on the chemical structure of the FA molecules. The longer and straighter the FA chain, the higher the van der Waals forces and the lower the FA fluidity, as reflected by a high FA melting point. Consequently, long-chain saturated FAs (SFAs) with a straight FA chain are characterised by a relatively high melting point, whereas polyunsaturated fatty acids (PUFAs) with a high number of double bonds, leading to a complex three-dimensional structure, generally have lower melting points.
Only recently, the lipophilic index has been proposed as a measure of overall FA fluidity, which can be easily derived from the FA composition of biological tissues. The lipophilic index is characterised as the mean of the melting points of FAs in biological tissues weighted by their specific concentrations [
11,
12]. Recent studies have suggested that a higher lipophilic index in plasma and adipose tissue, reflecting decreased FA fluidity, is associated with a higher risk of CHD [
11,
12]. However, no significant association was observed for an erythrocyte lipophilic index [
11]. We are not aware of studies that have related the lipophilic index to the incidence of type 2 diabetes.
In the current study, we aimed to investigate prospectively the lipophilic index as a measure of the FA fluidity of erythrocyte membranes in relation to the incidence of type 2 diabetes.
Results
Median subcohort proportions of the individual erythrocyte membrane FAs as well as Spearman correlation coefficients of the individual FAs with the lipophilic index are presented in Table
1. The highest proportions in erythrocyte membranes were observed for 16:0 (22.3%), 18:0 (13.8%), 18:1
n-9 (12.7%), 20:4
n-6 (13.3%) and 18:2
n-6 (10.8%). As expected, SFAs generally showed positive correlations with the lipophilic index (exception: 18:0), whereas PUFAs were inversely correlated with the index (exception: 18:3
n-6). Correlation coefficients for MUFAs and
trans-FAs were usually close to zero.
Table 1
Melting points of FAs and median proportions of erythrocyte FAs, and correlation coefficients of erythrocyte FAs with lipophilic index for the subcohort of the EPIC-Potsdam study (n = 1,406)
SFAs |
14:0 | 53.9 | 0.38 (0.30, 0.47) | 0.34**** |
15:0 | 52.3 | 0.21 (0.17, 0.26) | 0.27**** |
16:0 | 63.1 | 22.3 (21.2, 23.4) | 0.49**** |
17:0 | 61.3 | 0.32 (0.29, 0.36) | 0.16**** |
18:0 | 69.6 | 13.8 (12.8, 14.5) | −0.09* |
20:0 | 76.8 | 0.39 (0.34, 0.44) | 0.50**** |
21:0 | 74.3 | 0.04 (0.03, 0.05) | 0.22**** |
22:0 | 81.5 | 1.53 (1.32, 1.78) | 0.62**** |
23:0 | 79.1 | 0.27 (0.23, 0.32) | 0.36**** |
24:0 | 87.8 | 4.1 (3.5, 4.7) | 0.70**** |
MUFAs |
16:1 n-7 | 0 | 0.44 (0.35, 0.55) | 0.16**** |
18:1 n-7 | 15.0 | 1.0 (0.9, 1.1) | −0.08* |
18:1 n-9 | 16.0 | 12.7 (12.0, 13.4) | 0.08* |
20:1 n-9 | 23.3 | 0.28 (0.25, 0.32) | −0.13**** |
22:1 n-9 | 34.7 | 0.29 (0.21, 0.45) | 0.08* |
24:1 n-9 | 42.8 | 4.0 (3.5, 4.6) | 0.59**** |
PUFAs |
18:3 n-3 | −11.2 | 0.15 (0.13, 0.18) | −0.14**** |
20:5 n-3 | −54.1 | 0.77 (0.61, 0.94) | −0.40**** |
22:5 n-3 | −54.1 | 2.4 (2.1, 2.6) | −0.55**** |
22:6 n-3 | −44.2 | 4.8 (4.1, 5.4) | −0.62**** |
18:2 n-6 | −5.0 | 10.8 (9.9, 11.7) | −0.03 |
18:3 n-6 | −11.2 | 0.05 (0.04, 0.07) | 0.15**** |
20:4 n-6 | −49.5 | 13.3 (12.3, 14.2) | −0.71**** |
trans-FAs |
16:1 n-7t | 31.0 | 0.17 (0.14, 0.21) | −0.08* |
18:1 n-9t + 18:1 n-7t | 44.8 | 0.51 (0.44, 0.59) | 0.12**** |
Table
2 shows baseline characteristics of the subcohort members by quartiles of the lipophilic index. Participants with a high lipophilic index were more likely to be men and had a slightly higher BMI compared with participants with a low index, although this association did not reach statistical significance. Leisure-time physical activity, smoking status and educational attainment were not significantly associated with the lipophilic index. With regard to dietary factors, a high lipophilic index was associated with obtaining a slightly higher proportion of energy from fat at the expense of protein or carbohydrates.
Table 2
Baseline characteristics by quartiles of the lipophilic index for the subcohort of the EPIC-Potsdam study (n = 1,406)
n (subcohort) | 351 | 352 | 351 | 352 | |
General characteristics |
Age (years) | 47.9 | 47.5 | 48.6 | 48.9 | 0.16 |
Men (%) | 31.3 | 35.8 | 39.3 | 42.3 | 0.02 |
BMI (kg/m2) | 25.1 | 25.2 | 25.5 | 25.6 | 0.13 |
WCa
|
Men | 91.0 | 92.8 | 91.8 | 93.6 | 0.31 |
Women | 78.0 | 78.0 | 78.5 | 78.0 | 0.43 |
Sport activities and biking (h/week) | 2.0 | 1.5 | 2.0 | 2.0 | 0.99 |
Never smokers (%) | 49.9 | 45.7 | 45.3 | 46.3 | 0.45 |
College/university (%) | 35.3 | 39.2 | 39.3 | 40.9 | 0.39 |
Dietary intake |
Fat (% energy) | 38.5 | 39.2 | 39.1 | 40.3 | 0.001 |
Carbohydrates (% energy) | 42.2 | 42.6 | 42.7 | 41.2 | 0.05 |
Protein (% energy) | 13.8 | 13.9 | 13.6 | 13.7 | 0.03 |
PUFA/SFA ratio | 0.42 | 0.43 | 0.42 | 0.43 | 0.74 |
Alcohol (g/day) | 9.2 | 8.6 | 8.9 | 9.2 | 0.82 |
Coffee (g/day) | 302 | 302 | 450 | 450 | 0.16 |
HRs for the association of quartiles of the lipophilic index with diabetes incidence are presented in Table
3. In the age- and sex-controlled model (model 1), we observed a higher diabetes risk when comparing top with bottom quartile of the lipophilic index (HR 1.76 [95% CI 1.25, 2.47],
p for trend <0.001). This association slightly strengthened after further adjustment for lifestyle factors, education and dietary factors (HR 1.85 [95% CI 1.29, 2.65],
p for trend <0.001). The HR was moderately attenuated after further adjustment for BMI and waist circumference (model 3), but still statistically significant (HR 1.59 [95% CI 1.08, 2.34],
p = 0.005). Additional adjustment for diabetes-related FAs did not weaken this association (HR 1.90 [95% CI 1.25, 2.90],
p<0.001). In our sensitivity analyses, the exclusion of participants with treated hypertension at baseline, baseline HbA
1c ≥6.5% (47.5 mmol/mol) or incident cases diagnosed with diabetes within the first 2 years of follow-up did not materially change the results. The inclusion of participants with a history of cardiovascular disease or cancer at baseline did moderately attenuate the HRs (HR comparing top with bottom quartile 1.34 [95% CI 0.96, 1.87]), however, there was still a significant positive trend across the quartiles (
p = 0.02). The test for interaction of the lipophilic index with sex was not significant (
p = 0.46).
Table 3
Risk of type 2 diabetes by quartiles of the lipophilic index in a case-cohort study embedded in the EPIC-Potsdam cohort (n = 1,740)
Median index | 21.5 | 23.2 | 24.7 | 26.8 | |
n (cases/ subcohort) | 68/351 | 70/352 | 88/351 | 136/352 | |
Model 1a
| 1.00 (ref.) | 0.97 (0.67, 1.40) | 1.28 (0.89, 1.84) | 1.76 (1.25, 2.47) | <0.001 |
Model 2b
| 1.00 (ref.) | 0.97 (0.65, 1.43) | 1.36 (0.93, 1.99) | 1.85 (1.29, 2.65) | <0.001 |
Model 3c
| 1.00 (ref.) | 0.94 (0.61, 1.43) | 1.18 (0.78, 1.78) | 1.59 (1.08, 2.34) | 0.005 |
Model 3 + diabetes-related FAsd
| 1.00 (ref.) | 0.96 (0.62, 1.49) | 1.36 (0.88, 2.08) | 1.90 (1.25, 2.90) | <0.001 |
In further analyses, we investigated whether the association of the lipophilic index with diabetes risk could be explained by individual FAs, for which a relation to diabetes risk has already been shown. However, we did not observe a substantial attenuation of the HR in our model 3 after additional adjustment for proportions of 15:0 (HR comparing top with bottom quartile of lipophilic index 1.77 [95% CI 1.18, 2.65]), 17:0 (HR 1.85 [95% CI 1.24, 2.76]), 18:0 (HR 1.59 [95% CI 1.08, 2.34]), 16:1 n-7 (HR 1.44 [95% CI 0.97, 2.13]), 18:2 n-6 (HR 1.58 [95% CI 1.07, 2.33]), 18:3 n-6 (HR 1.51 [95% CI 1.02, 2.23]), 20:3 n-6 (HR 1.81 [95% CI 1.22, 2.67]), or total SFAs (HR 3.36 [95% CI 1.68, 6.75]), total trans-FAs (HR 1.55 [95% CI 1.05, 2.30]), total MUFAs (HR 1.31 [95% CI 0.79, 2.18]) and total PUFAs (HR 2.27 [95% CI 1.12, 4.60]).
ESM Figures
1 and
2 depict mean baseline biomarker levels according to quartiles of the lipophilic index for men and women of the subcohort. In general, we did not observe strong associations of the lipophilic index with the biomarkers. After multivariable adjustment, including body size (model 3), there was a positive association of the lipophilic index with fetuin-A in men (
p for trend across quartiles <0.001) and a slight positive association with HDL-cholesterol in women (
p = 0.002). Baseline levels of triacylglycerol, adiponectin, CRP and GGT showed no significant trend across quartiles in both men and women. Excluding participants with treated hypertension at baseline or with baseline HbA
1c ≥6.5% (47.5 mmol/mol) did not substantially change these findings. Furthermore, the association between the lipophilic index and triacylglycerol remained materially unchanged after the exclusion of participants in the unfasted state at blood collection (results not shown).
ESM Figures
3 and
4 show plasma random glucose and HbA
1c values by quartiles of the lipophilic index for men and women of the subcohort. In women, we detected a positive association of random glucose with the lipophilic index (
p for trend = 0.01), whereas a slight inverse association was observed for HbA
1c (
p for trend = 0.01). In men, the positive association of the lipophilic index with random glucose was only borderline significant (
p for trend = 0.05) and the association with HbA
1c did not reach statistical significance (
p for trend = 0.08).
Discussion
In this prospective study of middle-aged men and women, a high lipophilic index, indicating lesser fluidity of erythrocyte membranes, was associated with a higher risk of developing type 2 diabetes after multivariable adjustments, including body size.
The lipophilic index has been proposed recently as a measure of the FA fluidity of biological samples [
11,
12]. It is determined as the mean of the melting points of the individual FAs weighted by their concentration, and hence can easily be applied in large-scale epidemiological studies subject to the availability of FA profile data in biological samples. In our study, we were able to investigate the lipophilic index calculated from the FA profile of actual membranes, namely erythrocyte membranes. Although the FA composition of erythrocyte membranes is not identical compared with other cells important for glucose metabolism such as hepatocytes or muscle cells, membranes of different cell types share a common feature, which lies in the exchange of phospholipids with the plasma phospholipid pool. Nevertheless, it should be noted that membrane fluidity is not only determined by its FA composition, but also by other factors such as the cholesterol content of the membrane, the degree of phospholipid methylation and calcium binding [
21‐
23].
To our knowledge, this is the first study that investigated the lipophilic index as a measure of FA fluidity in relation to type 2 diabetes risk. We observed a higher diabetes risk for participants with a high lipophilic index reflecting lesser fluidity of the erythrocyte membranes. This finding is in line with the long-held notion that membrane fluidity is an important mediator that links intake and metabolism of FAs with diabetes risk. In cell culture studies, it was shown that membrane fluidity affects the glucose transport across membranes as well as the properties of the insulin receptor [
8‐
10], which is in agreement with our findings.
Two earlier epidemiological studies have investigated the lipophilic index in relation to myocardial infarction and CHD [
11,
12]. In a matched case–control study from Costa Rica, the lipophilic index for adipose tissue was significantly positively associated with myocardial infarction in multivariable adjusted models [
12]. Similarly, the lipophilic index of plasma phospholipids was significantly associated with a higher risk of CHD in the Health Professionals Follow-up Study. The erythrocyte lipophilic index, however, showed no significant association with CHD risk in this study, although there was a tendency towards a positive relation. The authors speculated that higher measurement error for the FA measurements in erythrocytes compared with plasma phospholipids in their study may have led to an attenuation of the statistical association [
11].
In addition to the investigation of the biological meaning of the lipophilic index, it is of interest whether the index is an independent predictor of diabetes risk. After multivariable adjustments, including body size measures, the effect size of the association between lipophilic index and diabetes risk was at least as strong in our study as those for a number of individual FAs, which are considered as established diabetes risk markers including 15:0, 17:0, 18:0, 18:2
n-6 and 20:3
n-6 (see our earlier publication [
3] for details on results). Only 16:1
n-7 and 18:3
n-6 showed a somewhat stronger association according to our earlier quintile analysis [
3]. Of note, adjustment for these FAs individually and simultaneously did not lead to a substantial attenuation of the HR for the association between the lipophilic index and diabetes risk, indicating that the lipophilic index may provide predictive value beyond individual FAs with regard to diabetes risk. Similarly, adjustment for total SFAs, total
trans-FAs and total PUFAs did not lead to an attenuation of the HR between the lipophilic index and diabetes risk, and adjustment for total MUFAs slightly attenuated the HR likely reflecting that some MUFAs showed associations with risk of diabetes (see our earlier publication [
3]). As expected, we have observed positive correlations of individual SFAs and negative correlations of PUFAs with the lipophilic index. Among those individual FAs with the strongest correlation with the lipophilic index, the long-chain SFA 24:0 showed a strong positive association with diabetes risk, whereas the PUFAs 20:5
n–3, 22:5
n–3 and 20:4
n–6 tended to be inversely associated with risk in our study, although not significantly [
3]. Interestingly, adjustment for these individual FAs also did not attenuate the effect estimates for the association of the lipophilic index with diabetes risk in our study (results not shown), which further corroborates the hypothesis that the lipophilic index is an independent predictor of diabetes incidence.
We observed only very slight, if any, associations of our lipophilic index with baseline levels of metabolic biomarkers. Similar observations were made in the Health Professionals Follow-up Study for the erythrocyte lipophilic index [
11]. As FA proportions and metabolic biomarkers have been assessed in a cross-sectional manner (using baseline blood samples), it was not possible to establish the correct time sequence (i.e. FA proportions precede levels of metabolic markers) in this kind of analysis. Still, clear positive cross-sectional associations of the plasma and adipose tissue lipophilic index with triacylglycerol have been detected in earlier studies, whereas less consistent findings were obtained for inflammatory markers [
11,
12]. Surprisingly, we observed higher HDL-cholesterol concentrations for women with a high lipophilic index, which is counterintuitive. This finding is in contrast to earlier studies that detected inverse associations of the plasma and dietary lipophilic index with HDL-cholesterol [
11,
12]. However, no significant relations were seen for the erythrocyte and adipose tissue lipophilic index in these studies [
11,
12]. Given these mixed findings, further studies should be performed to evaluate this relationship, preferably with a prospective design.
Changed insulin sensitivity is a biological mechanism that may link membrane fluidity to diabetes risk. Unfortunately, we could not perform analyses with specific indicators of insulin sensitivity, such as the HOMA index, owing to the small number of fasted participants in our study. Instead, we have performed analyses with other measures of glucose metabolism, namely random glucose and HbA1c, which are, however, not specific indicators of insulin sensitivity. The results for the cross-sectional association of the lipophilic index with random glucose and HbA1c were mixed. However, when accounting for the correct time sequence in our prospective analyses, there was a relatively strong, significant, positive relationship between the lipophilic index and diabetes incidence. This relationship turned out to be robust in various sensitivity analyses, suggesting that changed membrane fluidity is an important factor that precedes the development of type 2 diabetes.
Our analysis has several limitations that should be discussed. The lipophilic index is not a direct measurement of cell membrane fluidity. Although the FA composition is a strong determinant of membrane fluidity, other factors also play a role [
21‐
23]. Further studies using direct methods to determine membrane fluidity are warranted to confirm our findings. Furthermore, melting points of some FAs, especially those with very low proportions in biological tissues (see Methods), were not available in the LipidBank database and hence could not be considered in the calculation of the lipophilic index. However, our results remained substantially the same in a sensitivity analysis for which we excluded specific FAs (proportion of less than 0.5% of total FAs) from the lipophilic index (data not shown). We considered only clinically apparent type 2 diabetes and did not screen our study population for diabetes at baseline, thus it is possible that prevalent but undiagnosed cases remained in our analyses. Still, excluding participants with a baseline HbA
1c higher than 6.5% (47.5 mmol/mol) or incident cases diagnosed with diabetes within the first 2 years of follow-up did not appreciably change our results, suggesting that reverse causation should not have had major effects on our findings. Further, we did not screen participants for incident diabetes during follow-up. However, all self-reports of diabetes during follow-up were verified through the treating physician in our study. Given the resulting high specificity and positive predictive value of the disease classification, the remaining misclassification (unidentified cases) should not have biased the estimated risk [
24]. Strengths of our study include the wide profile of FAs (
n = 25 individual FAs) considered in the calculation of the lipophilic index as well as the use of actual membranes to reflect FA fluidity. The prospective design and high rate of follow-up made reverse causation and bias through loss to follow-up less likely. Comprehensive data on diet, lifestyle and other risk factors allowed us to account for potential confounders in detail.
In conclusion, our data suggest that a high lipophilic index reflecting a lower fluidity of erythrocyte membranes is associated with a higher risk of type 2 diabetes. Our findings corroborate the hypothesis that membrane fluidity may be an important mediator that links intake and metabolism of FAs to diabetes risk. Hence, interventions aiming at an improvement of cell membrane fluidity may have the potential to lower diabetes risk. Our findings also suggest that the lipophilic index is of value as an independent predictor of the incidence of type 2 diabetes.