Evening Sleep Patterns and Poor Mediterranean Diet Linked to Depression in Older Adults and Faster Aging

We conducted a cross-sectional analysis at baseline involving participants from the PREDIMED Plus-Valencia study (n = 465), a field center of the multi-center PREDIMED Plus study; an ongoing randomized primary prevention trial in Spain [60]. We collected data and samples during the baseline visit for each participant. We recruited participants consecutively from March 2014 to December 2016. No longitudinal follow-up was conducted in this study. The trial was registered at https://www.isrctn.com/ISRCTN89898870↗ (accessed on 1 December 2024). The eligible participants, recruited from several primary care health facilities in the Valencia field center, were community-dwelling adults (men, 55–75 years; women, 60–75 years) with body-mass index (BMI) in the overweight or obesity range and metabolic syndrome. The inclusion and exclusion criteria have been published elsewhere [60]. Briefly, the inclusion criteria included the above-mentioned age range, the presence of overweight or obesity (BMI for overweight: 27–29; for obesity: 30–39 kg/m²), and people who, at baseline, met at least three components of the metabolic syndrome and had no cardiovascular disease at enrollment. The exclusion criteria included the following: inability to give written informed consent or communicate with study staff; institutionalization; history of previous cardiovascular diseases; active malignant cancer; any severe co-morbid condition with less than 24-month life expectancy; impossibility to follow the recommended interventions; body weight loss > 5 kg during the 6 months prior to the screening visit; food allergy to any MedDiet component; immunodeficiency or HIV-positive status; liver cirrhosis or chronic renal failure; current use of weight loss medication; concurrent participation in another randomized clinical trial; and any other condition that may interfere with the completion of the study protocol.

We conducted the analysis on these participants, recruited in the Valencia field center, because the chronotype data and DNA methylation results [61] had been assessed only at the Valencia site. The demographic, clinical, lifestyle, and depression data were obtained from all participants recruited at the Valencia site (n = 465). From them, additional chronotype data (not included in the general multicenter study) were obtained (n = 426). DNA methylation analyses and the calculation of accelerated aging were performed for 414 subjects.

Participants provided written informed consent. The study protocols and procedures were approved according to the ethical standards of the Helsinki Declaration by the Human Research Ethics Committee of Valencia University, Valencia (approval codes H1373255532771, 15 July 2013; and H1509263926814, 6 November 2017).

At baseline, socio-demographic data, lifestyle (smoking, sleep duration, etc.), medication, and clinical characteristics were assessed by a questionnaire, as previously described [60]. The depression diagnosis was obtained through this questionnaire and was defined as a self-reported lifetime medical diagnosis of depression [45,46]. In addition, we gathered the age at which depression received the diagnosis. Type 2 diabetes was assessed as previously reported [60]. Anthropometric measurements and blood pressure were assessed by trained personnel following the PREDIMED Plus protocol [62,63]. Weight and height were assessed using calibrated scales and a wall-mounted stadiometer, respectively. The Body Mass Index (BMI) was determined by dividing weight in kilos by the square of height in meters. Waist circumference was measured at the midpoint between the lowest rib and the iliac crest, following normal exhalation, with an anthropometric tape. Blood pressure was determined using a validated semiautomatic oscillometer (Omron HEM-705CP, Netherlands) while the individual remained seated for 5 min. After an overnight fast, venous blood samples were collected by venipuncture. Fasting plasma total cholesterol, HDL-C, LDL-C, triglycerides, and glucose were determined as previously described [63]. A complete blood count (CBC) in peripheral blood samples (including the total leukocyte count and the types of white blood cells) was additionally determined as previously described [61].

The validated REGICOR short Physical Activity questionnaire [64] was employed to measure leisure-time physical activity. The questionnaire included questions regarding the type of activity, frequency (number of days), and duration. The total energy expenditure associated with leisure-time physical activity was calculated as the sum of the frequency, duration, and intensity of each activity, divided by 30 days, and was estimated in Metabolic Equivalent of Tasks (METs)/min/day as a continuous variable. A dichotomous variable for physical activity was also derived, which was based on the median values. We estimated adherence to the MedDiet using the validated 17-item screening questionnaire for the PREDIMED-Plus study [65]. This questionnaire consisted of 17 questions about compliance with dietary habits reflecting MedDiet. We scored the questions 0 (did not meet criteria) or 1 (achieved criteria), resulting in a maximum score of 17, with higher scores indicating greater adherence to the MedDiet. This scale is an updated version of the 14-item MedDiet adherence questionnaire [14-MEDAS], previously validated by our group [66]. Both scales share most items. The updated version [17-MEDAS] added a few additional items and an energy-reduced MedDiet framework. We have published the detailed questions that comprise the 17-MEDAS score [61]. Subsequently, the score was categorized into categorical variables: one based on the mean of the population (less than or higher than 8 points) and another categorized into three approximate tertiles: low (0–6) medium (7–9), and high adherence (10–17 points).

We selected the Beck Depression Inventory (BDI-II) [67], which has been validated in Spain [68], to assess depressive symptoms at baseline. This is one of the most widely used and well-validated self-report measures of depressive symptoms in adolescents and adults (from 13 to 80 years), providing a current measurement of depression-related symptoms. The BDI-II consists of a 21-item questionnaire utilizing a Likert-type response scale for each item ranging from 0 (not present) to 3 (all of the time). The sum of all responses yields the total score, which spans from zero to 63. Higher total scores indicate more severe depressive symptomatology. There are some standardized cut-offs to classify individuals depending on the total score reached: minimal depression (0–13), mild depression (14–19), moderate depression (20–28), and severe depression (29–63) [69].

We assessed individual variability in diurnal preference or chronotype using the widely used morningness–eveningness questionnaire [70]. The MEQ is considered an international gold standard measure of chronotype and is one of the most predominantly used measures in chronobiology. The MEQ comprises 19 items that pertain to preferred wake-up and bedtimes, as well as daily activity schedules. A higher total score indicated morningness. We used the MEQ score as a continuous variable for some analysis, taking into account the potential differences in the cut-off points reported in different studies [71]. The originally proposed MEQ cut-offs for chronotypes were as follows: definitely morning type (70–86 points), moderately morning type (59–69 points), intermediate type (42–58 points), moderately evening type (31–41 points), and definitely evening type (19–30 points). We also grouped extreme and moderate morning and evening types for further analysis (less than 42 were evening types and more than 58 were morning types).

Genomic DNA was isolated from blood. We measured the quantity of double-stranded DNA using PicoGreen (Invitrogen Corporation, Carlsbad, CA, USA). As previously reported [61], we analyzed DNA methylation using the Infinium HumanMethylationEPIC BeadChip (850 K) array (Illumina Inc., San Diego, CA, USA). We additionally processed the arrays at the Human Genomics Facility at Erasmus MC in Rotterdam, the Netherlands, which included bisulfite conversion using the Zymo EZ-96 DNA Methylation Kit (Zymo Research, Irvine, CA, USA), hybridization, and scanning using an Illumina HiScan machine. We also conducted quality control procedures to assess the quality and reliability of the DNA methylation data obtained using the R packages Minfi (release 3.18), Meffil (release 1.1.1), and ewastools (release 1.7.2) [72,73,74]. We successfully performed quality control on a total of 414 samples, after which we computed DNA methylation clocks. We carried out data normalization [75,76], including functional normalization and normal-exponential out-of-band (NOOB) correction, and obtained beta values for the corresponding CpG sites. DunedinPACE scores were calculated from the beta matrices using the DunedinPACE R package as described by Belsky et al. [29]. Although our main aim was to analyze the depression associations with accelerated aging using the DunedinPACE biomarker, considering that it is a novel biomarker and that previous studies analyzed second-generation biomarkers such as the GrimAge epigenetic clock and age acceleration (AgeAccelGrim) [25], we also computed this biomarker from the beta values according to the algorithm proposed by Lu et al. [25] using the online Horvath epigenetic age calculator [24].

Before testing the study’s hypotheses, we completed the initial screening and cleaning of the data. We implemented chi-square tests to analyze proportions. Students’ t-tests and ANOVA tests were applied to compare crude means of continuous variables. We analyzed the associations between the factor of interest and the dependent variables (DBI-II-score, morningness–eveningness, or age acceleration) using multivariable regression models sequentially adjusted for potential confounders. When indicated, models were sequentially adjusted as follows: model 1, unadjusted; model 2, adjusted for sex, age, and diabetes; model 3, additionally adjusted for BMI, smoking, physical activity, sleep duration, and education. Furthermore, depending on the statistical analysis, models were additionally adjusted for depression diagnosis, for chronotype, and/or for adherence to the MedDiet. For the study of interactions between the chronotype and adherence to the MedDiet we used multivariable hierarchical regression models including the main effects and the interaction terms and additionally adjusted for the above-mentioned covariates. For accelerated aging, we focused our analysis on the Dunedin-PACE score as the dependent variable, and we fitted models sequentially adjusted for the potential confounders as indicated. Additionally, and only for comparative purposes with other studies, we analyzed the AgeAccelGrim biomarker as the dependent variable. We used general linear models for continuous variables as the dependent variable. We used multivariable logistic regression models for dichotomous variables as outcomes. We estimated the adjusted means for the continuous variables from the corresponding multivariate corrected models. We conducted analyses for the entire population, stratifying them based on sex. We performed statistical analyses using IBM SPSS Statistics version 26.0, NY, and specific tools in R. All tests were two-tailed, and p-values < 0.05 were considered statistically significant for these associations.

At baseline, Table 1 displays the demographic, lifestyle, and clinical characteristics of the study participants (n = 465) by sex. This represented the whole sample size of participants recruited in the PREDIMED-Plus-Valencia study, which included both men (n = 198) and women (n = 267). They all had metabolic syndrome, and the mean age was 65 years (SD: 5 years). The prevalence of obesity was high, and there were no statistically significant differences between men and women (p = 0.651 for BMI). Similarly, we did not observe any statistically significant differences in the prevalence of diabetes based on sex (p = 0.753). We also did not find statistically significant differences in adherence to the MedDiet by sex (p = 0.022). However, we did find statistically significant differences between men and women (p < 0.05) in the level of education (predominantly up to primary education for women), in tobacco smoking (less consumption among women), and in physical activity (more sedentary among women). Regarding the prevalence of previously diagnosed depression, it was 8% of the total population, and the average DBI-score was 9 points. (SE: 0.2). While 77% of participants reported “minimal depressive symptoms”, 15.5% of the participants reported “mild depressive symptoms”, 5.2% “moderate”, and 1.5%, “severe depressive symptoms”.

For all these variables of depression and the level of depressive symptomatology, we have found very significant differences by sex, with values always higher in women (p < 0.05). However, for diagnosed depression, we additionally gathered the age of onset for any lifetime depressive disorder. In this population, women were not likely to have an earlier onset of depression compared to males, as no statistically significant differences were observed in the mean age at onset of depression between men (51.8; SE: 2.5 years) and women (50.3; SE: 2.5 years); p = 0.751. When we analyzed sleep duration, it was shorter in women than in men (p < 0.05). Similarly, the average score of their MEQ questionnaire was also lower (p < 0.05), indicating a more evening chronotype. Chronotype data were collected from only 426 participants who provided complete responses to all items on the MEQ, which was administered only at the Valencia site.

First, we tested the internal consistency of the BDI-II scale in this population and obtained an excellent result. We found that our participants had a Cronbach’s alpha of alpha = 0.87. Second, we analyzed the association between the main demographic, clinical, and lifestyle characteristics and depressive symptoms in both men and women. Table 2 presents the results of a multivariable regression model, adjusted for the other variables.

Sex was strongly associated with the severity of depressive symptoms (p < 0.001). Women had an average regression coefficient 3.23 (SE: 0.68) points higher than men, even after the adjustment of the other covariates. Similarly, having a previous medical diagnosis of depression was associated with higher levels of depressive symptoms (beta: 3.12; SE: 1.04; p = 0.003) in the multivariable model taking into account the adjustment for sex. Type 2 diabetes was also significantly associated with higher levels of depressive symptoms (1.64: SE: 0.06 points in diabetic subjects compared to non-diabetics; p = 0.005). We did not find a statistically significant association between education, smoking, physical activity, BMI, and age and depressive symptoms after the multivariable adjustment, including sex. Interestingly, sleep duration was inversely associated with depressive symptoms (p = 0.010). Although we measured sleep duration both on the workdays and on the free days, due to the high correlation of these variables and to prevent multicollinearity, we only included one of them (sleep duration on the workdays) in the fitted model. Further, we estimated the association of the above-analyzed variables with diagnosed depression in a multivariable logistic regression model including all of them. In this model, only sex was significantly associated with lifetime clinical diagnosis of depression (OR = 3.11; 95% CI: 1.28–7.61; p = 0.01 for women versus men).

The total scores of the MEQ in this sample ranged from 26 to 76 points, with higher values indicating a morningness chronotype. The internal consistency of the MEQ in the whole sample was good (Cronbach’s alpha = 0.727). We derived the five categories for the chronotypes considering the MEQ-score cut-offs indicated in the Methods. Thus, in the whole sample, the prevalence of the five categories for chronotypes were as follows: definitely morning type (3.5%), moderately morning type (33.3%), intermediate type (58.9%), moderately evening type (3.6%), and definitely evening type (0.7%). We detected a strong association between the morningness–eveningness scores and the depressive symptoms according to the BDI inventory (Figure 1). Mean BDI scores for each chronotype were 4.8 ± 0.8; 8.6 ± 0.8; 9.2 ± 0.4; 11.2 ± 2.4, and 12.3 ± 4.9 points for the definitely morning, moderately morning, intermediate, moderately evening, and definitely evening types, respectively (p-trend = 0.005). This association remained statistically significant after adjustment for sex, age, and diabetes (p = 0.032) and even after additional adjustment for BMI, smoking, physical activity, sleep duration, and education (p = 0.035).

We also grouped extreme and moderate morning and evening types for further analysis and created a three-category variable (evening type, intermediate type, and morning type). This variable was also strongly associated with levels of depressive symptoms. Thus, the prevalence of severe depression was 11.1%, 1.6%, and 0.0% in subjects with the evening, intermediate, and morning chronotypes, respectively, whereas the prevalence of minimal depression was 61.1%, 77.7%, and 81.5% in subjects with the evening, intermediate, and morning chronotypes, respectively (p < 0.047). The stratified analysis by sex for current depressive symptoms indicated similar trends in both men and women. Additionally, we examined an association between a lifetime diagnosis of depression and the MEQ-score examined as a continuous variable. Initially, we tested the association in the entire population, including both men and women. In the unadjusted model, the MEQ score exhibited a significant association (beta = −2.92 ± 1.40 points; p = 0.023) among patients with diagnosed depression compared to those without a diagnosis. Nonetheless, following adjustments for sex, age, and diabetes, the statistical significance was diminished (p = 0.052). The stratified model by sex revealed a stronger and statistically significant association in women (p = 0.011) compared to men (p = 0.551). Figure 2 shows the association between the MEQ score (as a continuous variable) and previously diagnosed depression in women. This model (adjusted for age and diabetes) indicates that women with a prior diagnosis of depression scored −4.25 ± 1.70 points lower on the MEQ compared to those without such a diagnosis. This association remained statistically significant following further adjustments for BMI, smoking, physical activity, sleep duration, and education (p = 0.008).

First, we analyzed the association between adherence to the MedDiet as a continuous variable and depressive symptoms in both men and women together. We used the 17-MEDAS score, where a higher score indicates greater adherence. In all the models used, adherence to the MedDiet had a highly significant inverse association with depressive symptoms (beta = −0.34 ± 0.10 points decrease in depressive symptoms per 1-point increase in adherence; p = 0.001 in a model adjusted for sex, age, and diabetes). This result remained statistically significant even after additional adjustment for BMI, smoking, physical activity, sleep duration, and education (beta = 0.35 ± 0.10: p = 0.001). In the sex-stratified analysis, the findings of the inverse association were comparable in both men and women but more pronounced in women. Figure 3 illustrates the association between the TDI-II score and adherence to the MedDiet, categorized into three levels: low, medium, and high, as detailed in the methods section. Once again, our model adjusted for sex, age, diabetes, BMI, smoking, physical activity, sleep duration, and education revealed a strong inverse association with depressive symptoms (p = 0.002).

When we looked at the association between adherence to MedDiet (as continuous) and having been diagnosed with depression at some point in the life, we did not find a statistically significant p-value for the whole population (p = 0.142 for the fully adjusted model). Likewise, we did not find a statistically significant association with depression diagnosis in either men (p = 0.531) or women (p = 0.056).

After examining the separate effect of chronotype and adherence to MedDiet on depressive symptoms, now we have analyzed the joint effects including both variables in multivariate regression models. Table 3 presents the results of the statistical models analyzing these joint effects (beta ± SE and p-values). Models 1, 2, and 3 have the same sequential fit as those used for the separate effects.

The chronotype and adherence to the MedDiet variables both remained statistically significant in all adjusted models, each being an independent predictor of depressive symptomatology. In the fully adjusted model (model 3), both variables presented statistically significant additive effect, such that a morning chronotype was associated with lower depressive symptoms (−0.89 ± 0.45; p = 0.046) for each category of chronotype advancement towards the morning. At the same time, greater adherence to the MedDiet would also contribute to a decrease in depressive symptoms (−0.29 ± 0.11 points; p = 0.007) for each point increase in adherence. After comparing the main effects, another regression model was additionally adjusted including the interaction terms between chronotype and MedDiet. Nonetheless, we did not see statistically significant interactions, suggesting an absence of heterogeneity in the additive joint effects in the three models fitted.

As indicated in the Methods, we calculated the DunedinPACE scores as our selected biomarker for age acceleration. It is a novel blood DNA methylation biomarker that measures the rate of biological aging from a single blood sample and indicates the rate at which an individual ages in comparison to the standard rate of one year of aging for each calendar year. Values above 1 indicate “faster” biological aging, whereas values below 1 indicate “slower” biological aging. DNA methylation data and the subsequent DunedinPACE biomarker estimations were available for 414 participants. In this sample, DunedinPACE scores ranged from 0.66 to 1.53. First, we analyzed the association between depressive symptoms (BDI-II score) and the DunedinPACE biomarker in both men and women. However, after adjusting for sex, age, and diabetes, we obtained results that were borderline significant (p = 0.078). Then, we did a separate analysis for men and women and found a statistically significant association between higher BDI-II scores and faster aging in women (beta = 0.003 ± 0.001; p = 0.040). No statistically significant results were obtained in men (p = 0.905) in the same regression model adjusted for sex, age, and diabetes. Further, we tested the association between the severity of depressive symptoms and the DunedinPACE biomarker for age acceleration in both men and women. In women (p = 0.026), but not in men (p = 0.289), we obtained a statistically significant association in the model adjusted for age and diabetes (Figure 4). This association remained statistically significant after additional adjustment for BMI, smoking, physical activity, sleep duration, and education.

Moreover, when additional adjustments for the chronotype and adherence to the MedDiet were considered, the association between higher severity of depressive symptoms and accelerated aging (DunedinPACE biomarker) remained statistically significant (p = 0.027). Finally, with comparative purposes, we analyzed the association between depressive symptoms and age acceleration using the classic GrimAge epigenetic clock and age acceleration (AgeAccelGrim) as described in the Methods. For this biomarker, we observed a similar trend consisting of an increase in age acceleration associated with the severity of depressive symptoms, but results did not reach statistical significance (p = 0.375 in the whole population and p = 0.132 in women).