Study design and study population

The Kumasi Diabetes and Hypertension (KDH) study was conducted as a non-matched hospital-based case-control study in urban Ghana between August 2007 and June 2008 [9]. In brief, patients from the diabetes center and the hypertension clinic were recruited, and preliminary controls came from friends, neighbors and parishioners, outpatients and hospital staff.

The study protocol conformed to the principles embodied in the Declaration of Helsinki and was reviewed and approved by the Ethics Committee of the School of Medical Sciences, University of Science and Technology, Kumasi. All participants gave informed written consent before they participated in the study. On the examination day, after 10-h over-night fast, venous blood samples were collected. Following breakfast, a personal interview was conducted on dietary intake, demographics, socio-economic status, medical history and lifestyle; and anthropometric measurements were taken. T2D was defined as fasting plasma glucose >7.0 mmol/L or as documented intake of anti-diabetic medication. Accordingly, there were 688 T2D cases and 778 controls without T2D. The latter were used for the present analysis, because diabetes status and anti-diabetic drugs can impact on FA metabolism [10]. Further 125 participants were excluded, because of missing data on dietary intake, demographics, socio-economics or anthropometrics. The final sample size was 653 (Fig. 1).

Serum fatty acids and biochemical analyses

The proportions of the most commonly measured FAs in epidemiological studies [11] were analyzed in serum phospholipids. Fasting serum phospholipids reflect the dietary intake of FAs during the past weeks and are not influenced by recent FA intake [6]. The details of FA analysis are presented in Additional file 1. In brief, FAs were extracted from serum samples with tert-butyl methyl ether/methanol, followed by a solid phase separation, hydrolysis and methylation with trimethyl sulfonium hydroxide. The FA methyl esters were separated by their retention time in the gas chromatograph with a 100 m capillary column (HP-88) and detected by flame ionization. The 28 FAs were identified by standard substances and quantified as area percentage of each FA relative to the total area of all detected FAs.

Fasting glucose was measured in full venous blood in mmol/L by on-site photometry (Glucose 201⁺ Analyzer, HemoCue, Germany); inter-assay coefficients of variation ranged between 1.7% and 6.1%. Serum triglycerides and high-density lipoprotein (HDL)-cholesterol were measured by colorimetric assays (ABX Pentra400, Horiba Medical, Germany). The inter-assay coefficients of variation were 4.5% and 1.8%, respectively. Low-density lipoprotein (LDL)-cholesterol was calculated according to the Friedewald equation [12].

Dietary assessment

In face-to-face interviews, trained study personnel applied a Ghana-specific food frequency questionnaire (FFQ) to capture the usual food intake of all participants over the last 12 months and to ensure, that the influence of daily and seasonal variation was minimized. The FFQ comprised 51 food items. According to their culinary use and similarities in their nutrient profile, these items were collapsed in the following 10 categories: starchy roots and tubers; cereals and cereal products; animal products; legumes, nuts and seeds; fruits; vegetables; fats and oils; salt and spices; sweets; and beverages (Additional file 2). The weekly intake frequencies were captured by six categories: never, seldom (<time per week), 1–2 times per week, 3–4 times per week, 5–6 times per week, and daily.

Assessments of covariates

In personal interviews, we obtained demographic and socio-economic data: age, sex, education (none, primary, secondary, tertiary, other), literacy (not able, able with difficulties, able), occupation (subsistence farmer, commercial farmer, casual laborer, artisan, trader, business men, public servant, unemployed, other), presence of 11 household assets, and number of people living in the household. A socio-economic status (SES) sum score was constructed comprising the three major domains education, occupation and income. Details are explained in Additional file 3. Medical history included own and family history of diabetes and the use of lipid-lowering drugs. Smoking status and self-reported physical activity (work-related, transportation-related, leisure-time physical activity) were documented. Daily energy expenditure (kcal/day) was calculated as the sum of metabolic equivalents corresponding to activity intensity as metabolic equivalents (MET-hours) × body weight (kg) × duration (min).

Anthropometric data were obtained by trained personnel (all devices SECA, Germany). Weight (kg) was measured with a person scale, height (cm) with a stadiometer and waist circumference (cm) and hip circumference (cm) with a measuring tape. Body Mass Index (BMI) was calculated as weight/(height)² (kg/m²) and waist-to-hip ratio (WHR) as waist circumference/hip circumference.

The common proxy markers education, occupation and income were used to construct a SES sum score ranging from 0 to 10 points. First, a new variable was constructed by combining the information on education and literacy. This new variable with four characteristics covered information about having formal education and being able to write and read; points from 0 to 3 were given. Occupation, originally a variable with nine characteristics, was condensed to a new variable with five characteristics, given the points 0 to 4. Due to differences in household structures and inflation rates of the local currency, income was assessed using a list of 11 household assets. An income score ranging from 0 to 12 points was constructed based on these assets and the number of people living in the household. The income score was divided into quartiles, given the points 0 to 3. To create the overall SES sum score, the points of education, occupation and the income score were summed up to a score ranging from 0 to 10 points.

Statistical analysis

General characteristics of the study population and the proportions of single serum phospholipid FAs are presented for normally distributed metric variables as mean ± standard deviation, for non-normally distributed variables as median with interquartile range, and for categorical variables as percentage. Summarized proportions for the following major FA groups were calculated: SFAs, mono-unsaturated FAs (MUFAs), n-3 PUFAs, n-6 PUFAs, and TFAs. For comparisons between groups, Mann-Whitney-U test was applied for continuous variables and χ²-test was used for categorical variables.

Dietary patterns were constructed applying PCA for participants who had no T2D but FA measurements (n = 653), to evaluate the internal validity of previously identified dietary patterns. Details of the PCA analysis in the KDH study have previously been reported [4]. In brief, the 51 food items were collapsed into 33 food groups (Additional file 2) and were subjected to PCA using the PROC FACTOR procedure in SAS with an orthogonal rotation. The following criteria were applied to extract the optimal number of factors: eigenvalue >1, scree plot, and plausibility of the components. Standardized food intake weighted by factor loadings was summed to be able to rank the participants according to their adherence to each dietary pattern.

The distribution of general characteristics and the FA profile were examined across tertiles of the dietary pattern scores using χ²-test and trend test. For the associations of dietary patterns with FAs, Box-Cox-transformed FAs were calculated as median with interquartile range across pattern score tertiles. For those FAs that showed a significant trend across tertiles, linear regression models were fitted and adjusted for age, sex, family history of diabetes, SES sum score, energy intake, energy expenditure and WHR. For the associations of FAs with diabetes-related biomarkers, adjusted means and 95% confidence intervals (CI) of serum triglycerides, HDL-cholesterol and LDL-cholesterol were calculated across FA tertiles using the same set of adjustment variables. As a sensitivity analysis, multivariate linear regression models were calculated to analyze the associations of serum phospholipid FAs with fat-containing foods, characteristic of the respective dietary patterns.

Study population

Fatty acids profile

As presented in Fig. 2, SFAs contributed the highest proportion to total serum FA concentration with 52%, followed by PUFAs (36%), MUFAs (12%) and TFAs (<1%). In the group of SFAs, palmitic acid (16:0) and stearic acid (18:0) showed the highest proportions. In the group of MUFAs, oleic acid (18:1n-9) yielded the highest proportion, followed by vaccenic acid (18:1n-7). Eicosapentaenoic acid (20:5n-3) (EPA) and docosahexaenoic acid (22:6n-3) (DHA) made up almost 8%, and therefore highly contributed to n-3 PUFAs. The highest proportion in the group of n-6 PUFAs was observed for linoleic acid (18:2n-6) (LA), followed by arachidonic (20:4n-6) and dihomo-γ-linolenic acid (20:3n-6). Within the group of TFAs, the highest contribution was seen for linolelaidic acid (18:2n-6 t).

Exploratory dietary patterns

Dietary patterns and fatty acids profile

For the “purchase” pattern, there was no significant association with SFAs, except for lower proportions of heptadecanoic acid in higher tertiles, but this was not statistically significant after multiple adjustment for potential confounders (β per 1 score-SD increase: -0.06; SE: 0.05; p = 0.16). For MUFAs, lower proportions were observed for palmitoleic (16:1n-7), oleic and vaccenic acid across the “purchase” pattern tertiles. The n-3 PUFAs EPA and DHA, were significantly lower; this was still discernible after multiple adjustment (EPA [β: -0.24; SE: 0.05; p < 0.0001]; DHA [β: -0.25; SE: 0.04; p < 0.0001]). Within n-6 PUFAs, significantly higher proportions were observed for LA, which remained after adjustments (β: 0.24; SE: 0.04; p < 0.0001). While dihomo-γ-linolenic and arachidonic acid were lower across tertiles of the “purchase” pattern, no further n-6 PUFAs were significantly associated. In a sensitivity analysis, the positive association of the “purchase” pattern with LA was confirmed for the characteristic fat-containing foods margarine (β: 0.81; SE: 0.22; p < 0.0003), vegetable oil (β: 0.60; SE: 0.23; p < 0.01) and groundnut (β: 1.21; SE: 0.21; p < 0.0001).

Across tertiles of the “traditional” pattern, significant lower proportions of arachidic acid were discernible and remained after adjustment (β: -0.10; SE: 0.03; p = 0.001). EPA, docosapentaenoic acid (DPA), and DHA proportions were nominally higher across the tertiles of the “traditional” pattern. In the group of n-6 PUFAs, significantly higher proportions were observed for eicosadienoic acid, whereas proportions tended to be higher for arachidonic acid and lower for LA (Table 4). Also, in a sensitivity analysis, we verified the inverse association of the “traditional pattern” with LA for the characteristic fat-containing food items cassava (β:-0.95; SE: 0.22; p < 0.0001) and plantain (β: -1.09; SE: 0.22; p < 0.0001). After adjustment for respective confounders, no significant association remained for n-3 and n-6 PUFAs with the “traditional” pattern. Proportions of TFAs neither differed across tertiles of the “purchase” nor of the “traditional” pattern (Table 4).

Fatty acids profile and blood lipids

Fatty acids profile

Lately, the recent epidemiologic literature covering the importance of circulating blood fatty acids has been reviewed [13]. Taken together, 13 observational studies with at least 100 healthy adults were published. The number of FAs under investigation differed greatly and focused largely on n-3 and n-6 PUFAs. None of them were from sub-Saharan Africa, but two studies described FAs among African-American individuals [14, 15]. More recently, Forouhi and colleagues added substantial findings from the European Prospective Investigation into Cancer and Nutrition (EPIC)-InterAct cohort and nine longitudinal studies among Caucasians for the relationships between PUFAs and T2D risk [16]. In comparison to Caucasians, the proportions were higher for SFAs, similar for MUFAs, higher for n-3 PUFAs, and lower for LA in the present study [13–16]. Still, comparability is constrained because FAs composition was measured in differential compartments, including adipose tissue [6], erythrocyte membranes [17] or serum free FAs [18].

Compared to African-American individuals, n-6 PUFAs appeared to be lower in the present study population [14, 15]. The differences between Caucasians and black US citizens are attributed to genetically determined activities of converting enzymes [15]. For n-3 PUFAs, the proportions were similar to those reported among blacks in a multi-ethnic cohort [14].

As to the few data from sub-Saharan Africa, proportions of serum phospholipid SFAs and MUFAs in the present analysis were similar to those in a small cross-sectional study among middle-aged Nigerians [19]. Compared to our study population, n-3 PUFAs (EPA 0.4% and DHA 3.1%) were remarkably lower in this indigenous population adhering to a dairy-based diet. TFA proportions were similar to those in our study [19]. Similar results for phospholipid FAs were seen in a study conducted among adolescent girls in Mozambique who rely on a maize- and rice-based diet [20]. Among Kenyan Massai [21], proportions of SFAs and n-3 PUFAs in lipids of red blood cells were lower than in the present study; MUFAs and n-6 PUFAs were higher.

Dietary patterns, fatty acids profile and blood lipids

The “traditional” pattern, which was related to higher risk of T2D, was inversely associated with arachidic acid. In a European-wide study, arachidic acid was positively correlated with the intake of nuts and seeds, margarine, dairy products and poultry and inversely associated with T2D [22]. In fact, we and others observed an inverse association of arachidic acid with serum triglycerides [23]. Possibly, higher circulating concentrations of very long-chain FAs, including arachidic, behenic and lignoceric acid, compete with palmitic acid for the integration in respective ceramides and therefore might decelerate insulin resistance and β-cell dysfunction that are linked to ceramides with a high content of palmitic acid [24]. While the lack of association between arachidic acid and HDL-cholesterol in our study accords with recent findings [23], we could not confirm the positive association of arachidic acid with LDL-cholesterol [23]. The significant direct association of the “purchase” pattern and its fat-containing components with LA could possibly explain the inverse association of this pattern with T2D. We observed a more favorable lipid profile of lower triglycerides and higher HDL-cholesterol in the highest LA tertile. Indeed, in EPIC-InterAct, an inverse association of LA with T2D was discernible [16], possibly conveyed by cholesterol-lowering effects of LA-rich vegetable oils [25]. Yet, in our study population, LDL-cholesterol increased also across the tertiles of LA, arguing against a general cholesterol-lowering effect of LA. Moreover, controversy remains about the health impact of LA, which can also act as a precursor of pro-inflammatory metabolites and oxidized lipoproteins [26].

The “purchase” pattern was furthermore inversely associated with EPA and DHA proportions. Evidence for a risk reduction of T2D by dietary n-3 PUFAs are so far inconclusive [27]. Heterogeneous results may possibly stem from contaminated fish oil (supplements) [28]. In our study, EPA was positively associated with HDL-cholesterol and inversely with triglycerides, while results for LDL-cholesterol were not significant. DHA was positively associated with triglycerides and LDL-cholesterol, which is partly in line with observations from RCTs, suggesting a triglyceride-lowering effect [29, 30].

Only few studies investigated the role of FAs profiles for T2D among populations in sub-Saharan Africa. Thus, our findings contribute essentially to the knowledge in this field. Owing to the cross-sectional observational design of the present analysis, reverse causation might partly explain the deviation of our results from the findings in intervention studies. Nevertheless, the likelihood of reverse causation was minimized, because we have excluded participants with T2D and those who took lipid-lowering medication to account for diabetes-related changes of and interferences with lipid metabolism. We acknowledge that the metabolic profile of this specific study population might not be representative and hence, not generalizable to the Ghanaian population. Residual and unmeasured confounding may have distorted our findings, but regression models were adjusted for a large variety of relevant confounders. While the Ghana-specific FFQ has not been validated yet, it provides culture-specific dietary information.

FAs were expressed as percentages of total measured FAs from gas chromatography. Thus, we cannot interpret individual FAs independently of other FAs and imprecise estimates of FAs in very small proportions (e.g. TFAs) are likely [13]. Still, flame ionization detection enables to directly measure FA proportions and is a well-established measurement method. Other biological compartments with a slow turnover might better reflect the habitual fat intake, such as adipose tissue. However, the time frame of fat intake covered by the FFQ overlaps with the period reflected in fasting serum phospholipid fatty acids [6, 7]. Therefore, we consider the choice of the latter as a good compromise to investigate the associations between dietary intake and fatty acid composition.