Abstract
Background:
Recent studies have identified plasma metabolites associated with cognitive decline and Alzheimer’s disease; however, little research on this topic has been conducted in Latinos, especially Puerto Ricans.
Objective:
This study aims to add to the growing body of metabolomics research in Latinos to better understand and improve the health of this population.
Methods:
We assessed the association between plasma metabolites and global cognition over 12 years of follow-up in 736 participants of the Boston Puerto Rican Health Study (BPRHS). Metabolites were measured with untargeted metabolomic profiling (Metabolon, Inc) at baseline. We used covariable adjusted linear mixed models (LMM) with a metabolite * time interaction term to identify metabolites (of 621 measured) associated with ∼12 years cognitive trajectory.
Results:
We observed strong inverse associations between medium-chain fatty acids, caproic acid, and the dicarboxylic acids, azelaic and sebacic acid, and global cognition. N-formylphenylalanine, a tyrosine pathway metabolite, was associated with improvement in cognitive trajectory.
Conclusions:
The metabolites identified in this study are generally consistent with prior literature and highlight a role medium chain fatty acid and tyrosine metabolism in cognitive decline.
INTRODUCTION
Dementia impacts more than 50 million people globally and results in economic costs estimated 1 trillion dollars annually [1]. The prevalence of dementia is projected to triple to 150 million people by 2050 [1]. Prior studies have demonstrated substantial racial and ethnic disparities in the risk of dementia, as well as in its underlying risk factors [2]. Latinos in the United States experience a disproportionate burden of dementia: they are at approximately double the risk of Alzheimer’s disease and related dementia (ADRD), compared to non-Hispanic Whites [3, 4]. Among Latinos, Puerto Rican adults have been shown to have an increased likelihood of cognitive impairment of Mexican Americans [3].
Several studies to date have attempted to identify blood metabolites associated with AD risk [5–12]. It is essential to identify biomarkers associated with cognitive change in order to enable earlier diagnosis and potential future treatment for AD/ADRD. Few prior studies have included diverse participants, particularly Latinos or used an untargeted approach, which would allow discovery of new associations, and the results have been conflicting [13, 14]. In a recent analysis within the Study of Latinos-Investigation of Neurocognitive Aging (SOL-INCA), several metabolites, including a number of phosphatidylcholines, were identified as key to cognitive decline [15]. In prior cross-sectional work within the Boston Puerto Rican Health Study, we observed that carbohydrates such as glucose, and amino acids such as N-acetylisoleucine and tyramine O-sulfate were inversely and β-cryptoxanthin positively associated with cognitive function [16] in Puerto Ricans, which was also confirmed in SOL-INCA [17]. In this study, we sought to examine whether similar associations would be observed longitudinally between the baseline metabolome and cognitive function trajectories over ∼12 years and 3 waves of follow-up among a cohort of Boston Area Puerto Ricans.
METHODS
The BPRHS is an ongoing, longitudinal cohort study of Puerto Rican older adults living in the Boston area, initially enrolled in 2004–2008 [18]. Participants, aged between 45 and 75 years at baseline, were originally identified based on the year 2000 Census, from Boston area blocks densely populated by Hispanic individuals (∼80%), contacted via major Puerto Rican events in the Boston-metropolitan area (9%), or through media and personal referral [18]. Those unable to answer study questions due to serious health conditions, who reported plans to move away from the area, or with a Mini-Mental State Examination (MMSE) score≤10 were excluded. Follow-up visits occurred an average of 2.2, 6.2, and 12.7 years after baseline and global cognition was assessed at the baseline, second, and fourth visits.
Demographic and socio-economic status questionnaires were designed based on the National Health and Nutrition Examination Survey (NHANES III) [19], the Hispanic Health and Nutrition Examination Survey (HHANES) [19], and the National Health Interview Survey Supplement on Aging (NHIS) [20]. Participants were asked about their level of education and household income. Dietary intake was assessed with a food frequency questionnaire (FFQ) designed for and previously used in this population [21]. Physical activity was assessed using a modified Paffenbarger questionnaire [22]. Participants underwent anthropometric and blood pressure measurements and provided a blood sample at each visit. Blood samples were drawn after a 12-h fast and immediately brought to the Human Nutrition Research Center on Aging at Tufts University in coolers with dry ice. They were cooled to 4°C and separated within 2 h in a refrigerated centrifuge. Plasma aliquots were saved in 1 mL cryogenic, screw-cap tubes, and stored at –80°C. The human subjects research in this study was approved by the University of Massachusetts, Lowell, Institutional Review Board.
Of 1,500 participants at baseline, 736 participants had metabolomics data. Of these, 637 had complete cognitive data at baseline, 589 at the second follow-up, and 335 at the fourth follow-up. Figure 1 outlines the details of the participant selection process.
Participant flowchart for BPRHS metabolomics cognitive study.
Assessment of cognition
A series of culturally appropriate cognitive tests, standardized using a U.S. Spanish-speaking population [23], was administered in Spanish or English (98% Spanish) at baseline, and the second, and fourth follow-ups by trained research assistants [24] under the supervision of a clinical neuropsychologist (T.S.). These tests include: 1) Mini-Mental State Examination (MMSE, general cognition) [25]; 16-word list learning test [23] with 2) word list learning (sum of words recalled over 5 attempts), 3) word recognition and 4) percentage retention (# of words recalled after a delay relative to # of correct responses on the fifth learning trial); 5) digit span forward and backward (working memory) [23]; 6) Stroop test (executive function) [23]; 7) verbal fluency (naming as many words as possible starting with a given letter) [23]; 8) clock drawing [26]; and 9) figure copying (visuo-spatial function and organization) [27]. As done previously [24], a global cognitive score (GCS) was calculated as the mean of the z-scores of each of the following cognitive measures: MMSE, word list learning, recognition, percentage retention, Stroop, letter fluency, digit span forward and backward, clock drawing, and weighted figure copying.
Assessment of metabolites
At baseline, participants provided fasting blood samples, which were subsequently stored at -80 C until profiling. Metabolomic profiling was performed by Metabolon (Metabolon, Inc., Morrisville, NC) using previously described proprietary procedures [28]. Briefly, the Metabolon platform uses liquid chromatography-MS/MS methods with positive ion and negative ion modes (Waters ACQUITY ultra-performance liquid chromatography; Thermo Scientific Q-Exactive high resolution/accurate mass spectrometer interfaced with a heated electrospray ionization source and Orbitrap mass analyzer operated at 35,000 mass resolution). Four modes of sampling were used to quantify metabolites: 1) acidic positive ion (hydrophilic molecules); 2) acidic positive ion (hydrophobic molecules); 3) basic negative ion; and 4) negative ionization from eluent of a HILIC column. Raw data were extracted, and metabolite peaks were identified, using more than 3,300 commercially available purified molecules as reference. Injection order was random and internal quality control samples were included. For each metabolite, relative metabolite concentration is reported as a normalized area under the curve. From the data provided by Metabolon, a total of 1,303 metabolites were identified, of which 943 could be assigned a chemical annotation; 360 were unknown. Analyses were restricted to the annotated metabolites (n = 943) and excluded xenobiotics (n = 229) to focus conclusions on biological function. Analyses were further restricted to metabolites available for ≥80% of participants, and missing values were imputed for individual metabolites as 50% of the minimum value of the respective metabolite across the other participants with non-missing values [16]. After imputation, metabolite values were log-transformed and Pareto scaled [29]. Multiple methods are available for imputation [30] and scaling [31], and although a number of authors have considered and compared these different methods, there is no gold standard agreed or recommended by the community as evidenced by the range of methods observed across the metabolomic epidemiology literature. A large number of papers both from our own labs and from others have previously used the 50% of the lowest value and Pareto scaling in highly cited peer reviewed publications [32–38]. Furthermore, within our own lab we have conducted comparisons of different imputation and scaling methods within our own data and found that the choice of method makes little difference to the key findings and overall conclusions. After data processing, this analysis included 621 metabolites.
Covariates
We adjusted all models for time, age, education, body mass index (BMI), smoker, physical activity score, APOE, Mediterranean diet score, medication use [proton pump inhibitors (PPI), non-steroidal anti-inflammatory drugs (NSAIDs), and metformin] and depression (Center for Epidemiological Studies of Depression (CESD) Score). Participants provided information on age (continuous) and education level (< 8th grade, 9th–12th grade, college/graduate school). BMI was treated as a continuous covariate, calculated using weight (kg) divided by height (m) squared. Frequency and history of smoking were assessed at all time points and categorized as current, former, or never. A physical activity score was calculated as the sum of hours spent on activities in a typical 24-h day (heavy, moderate, light, or sedentary activity, sleeping) multiplied by weighting factors that parallel the rate of oxygen consumption associated with each activity, and treated as a continuous covariate. APOE4 was assessed via genome-wide association testing. Mediterranean diet adherence score was computed using FFQ data and treated as a continuous covariate in the analyses, with a range of 1 (low adherence) to 9 (high adherence) [39]. To account for potential pharmacological effects of medications, we adjusted for baseline use PPI, metformin, and NSAIDs.
Statistical analysis
Metabolites associated with 12-year cognitive trajectory
We used linear mixed models with metabolite concentrations at baseline, and covariates at baseline, second, and fourth follow-up cognitive outcomes, with participant as a random effect to identify metabolites significantly associated with cognitive trajectory over the 3 follow-up cycles (mean follow-up duration 12.7 years). Primary analyses, including interaction with time, were conducted with log-transformed and Pareto scaled metabolites as continuous exposures. Analyses were adjusted for time (years since baseline), age, sex, education, BMI, smoking status, physical activity score, APOE, Mediterranean diet adherence, and use of medications (PPI, metformin, and NSAIDs) at baseline. To examine the impact of baseline metabolite concentration on cognitive trajectory and to identify metabolites associated with beneficial or detrimental cognitive trajectory, we included a metabolite * time (years since baseline) interaction term to our fully adjusted LMM.
All p values were corrected for multiple comparisons using the Benjamini-Hochberg False Discovery Rate (FDR), with P-FDR < = 0.05 for metabolite * time terms considered significantly associated with cognitive trajectory. For significantly associated metabolites (P-FDR < 0.05), we further modeled and then visualized relationships between baseline tertile of metabolite concentration and cognitive trajectory.
In sensitivity analyses, we examined, for metabolites that met the P-FDR < = 0.05 threshold in our study, whether the association between diet on cognitive function was mediated by each of the metabolites. Diet was modeled as the Mediterranean diet score at baseline, and each metabolite (of those associated with cognitive decline at P-FDR< = 0.05), independently was assessed for mediation. We used the mediate package in R to conduct the mediation analyses, individually for each of the significant metabolites.
RESULTS
The characteristics of our study participants as well as the comparison of participants selected versus not for metabolomic profiling in the BPRHS shown in Table 1. Overall, our sample is 100% US Based Puerto Rican, predominantly female, with low levels of education, high BMI, high proportion of diabetic and hypertensive participants. We observed overall declines in GCS scores over the ∼12 years of follow-up. GCS scores in the BPHRS decline over the ∼12 years of follow-up.
Baseline characteristics of 637 BPHRS participants with metabolomic profiles compared to those without metabolomic profiles
Metabolites associated with 12-year cognitive trajectory
We identified five metabolites significantly associated with cognitive trajectory (P-FDR < = 0.05) (Table 2) in the BPRHS. Of these, four were associated with worse cognitive trajectory (caproate (6 : 0), azelate (C9-DC), sebacate (C10-DC), and 4-methoxyphenol sulfate) and one with better cognitive trajectory (N-formylphenylalanine). Cognitive trajectory according to metabolite tertile for the metabolites significantly associated with GCS is presented in Fig. 2.
Metabolites associated with 12-y cognitive trajectory in the BPRHS
Top Metabolites Associated with 12-year Global Cognitive Trajectory in BPRHS. Results based on a linear mixed model (LMM) adjusted for time (years since baseline), age, education, BMI, smoker, physical activity score, APOE, Mediterranean diet score, medication use (PPI, NSAID, and metformin), CESD, and metabolite * time. Continuous covariates were centered in analyses. For each metabolite with P-FDR less than or equal to 0.05, association between metabolite tertiles and cognitive trajectory was assessed via LMM. *Metabolites with FDR < 0.05 were plotted.
Figure 3 illustrates the inter-correlations between the metabolites associated with cognitive trajectory in this study. As expected, we observed strong correlations between the top cognition-detrimental metabolites, caproate, azelate, and sebacate. Generally, metabolites associated with better cognitive trajectory were positively correlated with each other, and inversely correlated with the metabolites that were inversely associated with cognitive trajectory (Fig. 3).
Inter-relationships of top cognition-associated metabolites in BPRHS. Spearman correlation coefficients between the top cognition-associated metabolites identified in this study.
In sensitivity analyses examining mediation, none of the metabolites identified in this study mediated the impact of Mediterranean diet score on cognition (Supplementary Table 1).
DISCUSSION
This analysis identified several potential metabolite predictors of cognitive trajectory in Puerto Rican older adults. Medium chain and dicarboxylic fatty acids, caproate, azelate and sebacate, and the tyrosine metabolite 4-methoxyphenol sulfate were associated with more rapid decline, while the tyrosine metabolite N-formylphenylalanine was associated with improvement in global cognitive function over time.
In this study medium chain and dicarboxylic fatty acids, caproate, azelate, and sebacate were associated with cognitive decline. Prior studies have reported similar relationships between these dicarboxylic acids and brain health. A recent study found that urinary azelaic acid (azelate (C9-DC)) was elevated in participants with major depressive or bipolar disorders, compared to controls [40]. Urinary azelaic acid and sebacic acid (sebacate (C10-DC)) were also significantly elevated in AD patients, relative to cognitively healthy controls. The same study also found a negative correlation with C7-C10 dicarboxylic acids, which include sebacic and azelaic acids, and hippocampal volume [41]. Dicarboxylic acids are formed from oxidative breakdown of unsaturated fatty acids and the increase in oxidative stress associated with AD has been hypothesized to alter dicarboxylic acid formation from long-chain monounsaturated as well as polyunsaturated fatty acids [41]. Oxidative damage of brain lipids, leading to brain tissue loss among AD or otherwise cognitively declining patients, would also lead to an increase in urinary excretion of oxidized dicarboxylic acids [41]. Interestingly, sebebacate and azelate were among the metabolites positively associated with lead (Pb) exposure in the Normative Aging Study. Pb, a well-known neurotoxin, a heavy metal that has been associated with adverse cognitive outcomes for centuries [42].
N-formylphenylalanine, a metabolite in the tyrosine pathway, was associated with better global cognitive function trajectory in our study. Tyrosine is essential for the production of several neurotransmitters including dopamine, and prior work has reported positive associations between serum tyrosine and cognition among patients with Lewy body dementia [43]. Interestingly, another tyrosine metabolite, the amino acid 4-methoxyphenolsulfate, as associated with worse cognitive trajectory, potentially suggests a role of disruption of tyrosine metabolism in cognitive decline.
Strengths of our study include a focus on an under-studied unique population of Boston-Area Puerto Ricans with comprehensive covariate assessment and untargeted metabolomic assessment on a relatively large sample. Our study has several limitations. A key limitation is the loss to follow-up over the three study waves (Fig. 1). Loss to follow-up is likely a consequence of the socio-economic challenges experienced by study population. The BPRHS cohort is undertaking efforts to track reasons for loss to follow-up; these efforts are ongoing and not fully completed at the time of submission. We have thus far confirmed death in 101 of the 1,502 participants at baseline. Other reasons for loss to follow-up include reported (not yet confirmed) death (8), health reason (40), moved out of state (36), not enough compensation (8), process too long (15), and too busy (20). We are continuing to track participants in order to better understand reasons for loss to follow-up in the BPRHS. Another limitation is that, due to cost constraints, only approximately half of the BPRHS participants had metabolomic profiling performed. The BPRHS has plans to expand metabolomic profiling in future years. Finally, future research should be undertaken to model the inter-dependencies between metabolites, such as via pathway analyses.
A healthy diet has been associated with reduced AD risk [44] and better cognitive function, including in our study. We, however, did not see mediation of dietary pattern effect by any of the metabolites identified in this study, suggesting that these metabolites may be coming from sources other than diet. Further research is needed to identify these sources and see how they can be modified to help with cognition.
In summary, this investigation of the metabolome and cognitive trajectory over 12 years of follow-up in the BPRHS, the first such study in Puerto Ricans, identified several metabolites, including dicarboxylic acids, medium chain fatty acids, and tyrosine metabolites as associated with cognitive trajectories over time. Additional research on the role of the metabolome in Latinos, are needed to confirm these observed associations.
AUTHOR CONTRIBUTIONS
Scott Gordon (Data curation; Formal analysis; Writing – review & editing); Jong Soo Lee (Writing – review & editing); Tammy M. Scott (Conceptualization; Writing – review & editing); Shilpa Bhupathiraju (Data curation; Writing – review & editing); Jose Ordovas (Writing – review & editing); Rachel S. Kelly (Conceptualization; Visualization; Writing – review & editing); Katherine L. Tucker (Funding acquisition; Project administration; Supervision; Writing – review & editing); Natalia Palacios (Conceptualization; Funding acquisition; Project administration; Supervision; Writing – original draft; Writing – review & editing).
Footnotes
ACKNOWLEDGMENTS
The authors would like to thank Esther Jennings for assistance with this manuscript.
FUNDING
This work was supported by the National Institutes of Health (P01 AG023394, P50 HL105185, R01 AG055948, RF1 AG075922, K01 DK107804 to N.P., and K01 HL146980 to R.S.K.).
CONFLICT OF INTEREST
The authors have conflicts of interest to report.
DATA AVAILABILITY
The data supporting the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.
