Eighty-three older adults (range 52–89; 61% female) were recruited from the local community as volunteers for a study on healthy aging (i.e., without clinical symptoms of mild cognitive impairment or dementia, or subjective cognitive complaint). Participants were cognitively assessed as part of the study procedure; summary statistics for these tests and participant demographics are described in Table 1. Unadjusted Montreal Cognitive Assessment (MoCA) scores ranged between 20 and 30. There has been some debate in the literature regarding optimal MoCA cutoff scores for distinguishing normal cognition from mild cognitive impairment. In addition to this screening tool, a preponderance of neuropsychological evidence available for each participant was used to determine cognitive status, and ensure the sample excluded those showing clinical symptoms of impairment. Mean scores reported for all neuropsychological measures, excluding MoCA, reflect t-scores calculated from an age-matched normative population. Individuals with a self-reported history of neurological or psychiatric brain disorders, a diagnosis of a neurodegenerative disease, or MRI ineligibility were excluded from the study. Informed consent was obtained from all participants and all study procedures were approved by the University of Florida Institutional Review Board.

Scanning was performed on a 3T Philips Achieva scanner (Philips Healthcare, Best, The Netherlands), using a 32-channel head coil. Single-voxel ¹H spectra PRESS MRS were collected in two voxels, in the anterior and posterior regions (for placement see Figure 1). A T1-weighted anatomical image (magnetization-prepared rapid gradient-echo; repetition time/echo time 5 8 ms/3.7 ms, 1-mm³ isotropic voxels) was acquired for MRS voxel placement and segmentation. GannetCoRegister software (Harris et al., 2015) was used in data collection. Tissue fractions were derived, and an expert rater confirmed appropriate voxel placement based on landmarks in individual subject anatomical space. Parameters were optimized for primary analyses using MEGA-PRESS data collection. Each voxel was 3 × 3 × 3 cm (27 cm³). The PRESS sequence had a repetition time of 2 s and an echo time of 68 ms. Acquisition variables included: 320 transients with editing ON-OFF scans alternating every two transients, a 16-step phase cycle with steps repeated for on and off, 2048 data points acquired at a spectral width of 2 kHz, and variable pulse power and optimized relaxation delays (VAPOR) water suppression at a width of 80 Hz. No fat suppression was used. Editing-OFF spectra were acquired for use in the present dataset, and research has been done to validate this method (Dhamala et al., 2019). Sixteen transients of water-unsuppressed data were also acquired for quantification using the same acquisition variables. Osprey (Oeltzschner et al., 2020) was run for pre-processing, and LC Model spectral analysis software (Provencher, 2001) with the water-unsuppressed acquisition was used to determine absolute values, in mmol per Kg wet weight (mM), of the following metabolites: NAA (N-Acetylaspartate plus N-Acetylaspartylglutamate), Cr (creatine plus phosphocreatine), MI (myo-inositol), Glx (glutamate plus glutamine), and Cho (glycerophosphocholine plus phosphocholine). Water scaling and eddy-current correction settings were set to true; other LCModel settings were kept as default (ppmst = 4.0, ppmend = 0.2). The MRS metabolites measured, along with their proposed roles, are described in Table 2. Metabolite values were excluded based on a corresponding percent standard deviation of 20 or greater (calculated in LC Model) which has been used as a rough criterion for determining acceptable reliability (Provencher, 2021). This resulted in different sample sizes across the different metabolite concentrations obtained.

Once blood samples were acquired, plasma from each sample was separated, aliquoted, and immediately stored at −80°C. Aliquots were used to measure cytokine levels using an xMAP multiplexed bead array. Using the Luminex-100 system (Luminex Corp., Austin, TX, USA), multiple cytokines were simultaneously quantified by capturing them onto antibody-coated, spectrally distinct fluorescent microspheres and measuring fluorescence intensity. In some cases, cytokine measurements fell below a detectable value, and in those cases, the threshold for detectability was used. The cytokines measured in picograms permilliliter (pg/ml), along with their proposed functions, are described in Table 3.

IBM SPSS Statistics version 25 (IBM Corp., Armonk, NY, USA) was used to perform all analyses. Absolute values of metabolites were corrected for cerebrospinal fluid (CSF) by computing [Metabolite/(1-CSF)] for each. This was performed to address the potential contribution of individual differences in tissue fraction within voxels to metabolite quantities. Tissue correction is particularly relevant in an older adult population where age-related cortical atrophy may drive differences in uncorrected metabolite values. Normality was assessed, and cytokine data were log-transformed to assess skewness. All variables included in the models were standardized by z-scoring to facilitate interpretation of the results. Simple bivariate Pearson correlations were performed to explore the relationships between age and each cytokine and cerebral metabolite in the current sample. Cytokine concentrations and age were entered as predictors of MRS concentrations in the anterior and posterior voxels in backward stepwise regression models. Independent analyses were conducted for each MRS concentration in both voxels. MRS metabolites were entered into each model as the dependent variables; independent variables across all models included circulating cytokines, sex, and age.

These values represent the original data prior to the statistical transformations described above. Findings in Table 4 demonstrate that, within this sample of older adults, there was nosubstantial evidence of age relationships across the majority of included neurometabolites and cytokines. The strongest age relationships identified were between frontal NAA concentrations and TNF-α, such that greater age was related to decreased NAA measured in the frontal voxel, and increased circulating TNF-α. Notably, p-values for correlations presented here were not corrected for multiple comparisons, and when corrected no significant age associations survived.

Table 5 shows the results of the backward linear regressions run for anterior voxel data, using cytokine data to predict concentrations of five metabolites. Age did not survive the final model for any metabolites in the frontal voxel. TNF-α was a significant negative predictor for NAA, Cr, and Glx concentrations. IL-1 β was a significant predictor of NAA. This effect was opposite to the hypothesized direction, such that greater circulating IL-1 β was associated with greater frontal NAA. Likewise, pro-inflammatory cytokine IL-5 was a positive predictor of Cr in the frontal voxel. There were no cytokines that survived the final models for MI and Cho.

Table 6 shows the results of the backward linear regressions run for posterior voxel data, using cytokine data to predict concentrations of five metabolites. Variations in sample size were due to the listwise exclusion of missing values. Age did not survive the final model for any metabolites in the parietal voxel. Findings suggest IL-2 is an important predictor for parietal neurochemical concentrations. IL-2 negatively predicted parietal MI, Cho, NAA, and Cr. As a pro-inflammatory cytokine, its negative direction was expected in models predicting NAA and Cr, but not in those predicting MI and Cho. MI concentrations also were positively predicted by IL-10 and IL-8. The positive direction of IL-10 in this model also was unexpected, particularly because IL-10 was a positive predictor of Cr as well. TNF-α positively predicted both Cho and NAA. NAA also was positively predicted by IL-1 β and was negatively predicted by IL-13 and IL-7. The directions of the relationships of IL-1 β and TNF-α in this model were inconsistent with our hypotheses. There were no cytokines that survived the final model for Glx.

The literature is inconsistent regarding expected cerebral metabolite changes in cognitively normal older adults. This inconsistency, though, may be attributable to methodological variability. Findings are not always comparable for a number of reasons, including differences in the procedure for tissue correction, acquisition parameters, and processing and reporting data (Cleeland et al., 2019). Null findings in the present study across the majority of cerebral metabolites contribute to the growing body of literature investigating age-related neurochemical changes. Though no p-values survived correction for multiple comparisons, a larger sample size may result in more pronounced effects. The effects identified by uncorrected criteria are described below. It is possible that age-associated changes are not as clearly detectable within a group of cognitively normal, healthy older adults, though the present sample had a fairly broad age range. NAA has long been regarded as a neuronal marker due to early studies using tumors and cultured cells to demonstrate detectable NAA in neurons but not in other cell types (Nadler and Cooper, 1972; Urenjak et al., 1992). The present data suggest a relationship between frontal NAA and age, such that decreased NAA was related to increased age, which is consistent with prior studies evidencing age-associated decreases in NAA (Brooks et al., 2001; Ding et al., 2016). As a marker of neuronal density, NAA may be associated with measures of structural brain volume. Frontal decreases in NAA parallel established patterns of frontal lobe cortical volume reductions in older adults (Resnick et al., 2003; Pfefferbaum et al., 2013).

Although prior literature supports an association between age and IL-6 (Ershler, 1993; Forsey et al., 2003), data from the present sample were not able to detect that association. One potential explanation may have to do with participants’ medical health relative to those included in prior studies, though objective medical health data were not collected as part of the larger trial and therefore were not available for this analysis. Ridker et al. (2000) found an association between IL-6 and myocardial infarction after controlling for other cardiovascular risk factors, and Ferrucci et al. (2005) showed a much smaller predictive effect of age for IL-6 after cardiovascular factors were taken into account. It is possible that IL-6 has more to do with cardiovascular health than it does with chronological age. While age and cardiovascular health are related, the degree of heterogeneity for health outcomes in aging introduces some implicit variance. As previously stated, this analysis utilized data from a study deliberately targeting non-pathological aging.

An association between increased age and increased TNF-α was found in the present sample. This is consistent with prior literature demonstrating increases in TNF-α production in healthy older, compared with younger, adults (Fagiolo et al., 1993; Zanni et al., 2003). TNF-α has been an important cytokine in prior literature investigating age-related inflammation. TNF-α is a multifunctional, pro-inflammatory cytokine implicated in a variety of disease states, as well as in chronic systemic inflammation. Findings have demonstrated relationships between TNF-α and cognitive impairment and dementia (Bruunsgaard et al., 1999), incidence of cardiovascular events (Cesari et al., 2003), insulin resistance (Borst, 2004), frailty (Michaud et al., 2013), and mortality. A 10-year study by Varadhan et al. (2014) found that TNF-α predicted all-cause mortality in a sample of older adults, and Roubenoff et al. (2003) additionally demonstrated the relationship between TNF-α and mortality in a population-based sample of older adults over a 4-year period (Roubenoff et al., 2003; Varadhan et al., 2014).

The hypotheses for the present analyses were based on the suppositions: (a) that inflammatory changes occur in healthy aging and drive the development of age-associated morbidities, including pathological brain changes; (b) that peripheral and cerebral inflammatory processes may be parallel such that greater pro-inflammatory cytokines circulating peripherally would correspond with greater metabolic indications of cerebral neuroinflammatory processes; and (c) that greater anti-inflammatory cytokines circulating may relate to some compensatory processes that would correspond with greater metabolic indications of brain health. Evidence has supported that increases in circulating pro-inflammatory cytokines are related to worse outcomes for cardiovascular health and neurodegenerative disease (Bruunsgaard et al., 1999; Cesari et al., 2003; Miwa et al., 2016). It was predicted that more of a classically pro-inflammatory marker would negatively predict a marker of neuronal health and density. However, in some cases, the opposite trend was found.

TNF-α appeared to have a particularly important predictive value across metabolites in both voxels. With specific regard to its negative effects on frontal NAA, Cr, and Glx concentrations, the significance of TNF-α is interesting considering the relative strength of its relationship with age. It is noteworthy that findings indicate TNF-α, above and beyond chronological age, was predictive of neurochemical changes that may occur in pathological aging (Olson et al., 2008; Zeydan et al., 2017). One potential interpretation is that these effects are indicative of greater inflammation, explaining variance in neurochemical changes that may otherwise be attributed to aging. Brain changes also occur in preclinical phases of neurodegenerative diseases, before clinical symptoms may be identified. Further, TNF-α has been associated with diseases (e.g., artherosclerosis, type 2 diabetes, sarcopenia) often related to pathological aging, as well as dementia and all-cause mortality (Bruunsgaard et al., 1999, 2000, 2003; Borst, 2004; Bian et al., 2017). Perhaps the observed relationships between increased TNF-α and decreased frontal NAA and Glx specifically represent a risk of pathological aging outcomes in this sample, and chronological age is less relevant. Of course, in a cross-sectional analysis, it is not possible to say for certain whether that may be the case.

The differential effects observed for cytokines are of potential interest. For example, TNF-α negatively predicted frontal NAA but positively predicted NAA in the parietal voxel. Prior literature has established a differential impact of aging on structural brain outcomes (Raz et al., 2005). As neurochemical changes may serve as a precursor to atrophy, it would follow that certain areas are more vulnerable to age-related metabolic changes than others. An investigation of the relationships between cerebral metabolites and both cortical and subcortical brain volumes in the same sample has recently been published and found divergent relationships between frontal and parietal metabolite concentrations and cortical and subcortical brain volumes (Williamson et al., 2021). Additionally, in many cases, cytokines are pleiotropic or act on each other which further complicates the interpretation of seemingly counterintuitive effects. IL-10, for example, has been regarded as anti-inflammatory and considered in the context of promoting longevity (Lio et al., 2002), but also has some pro-inflammatory roles (e.g., enhancing B cell survival and proliferation). In this sample, IL-10 positively predicted a marker of neuroinflammation, and also positively predicted a marker of energy production. Further, research particularly related to mood disorders have identified a positive relationship between inflammation and elevated extracellular glutamate (Haroon and Miller, 2017; Haroon et al., 2017). Present findings identified a negative relationship between TNF-a and Glx; however, models of glutamate excitotoxicity are important to consider in the context of this study which did not control for participants’ mood.

There is not yet the depth of knowledge provided by additional studies on the topic to comment on the mechanism by which they each are associated. Findings suggest inter-related degrees of circulating cytokines and neurochemical concentrations that may not appear straightforward at baseline in a healthy aging sample, but that may have implications for future progression of various health outcomes. The current sample was clinically healthy at the time of assessment. However, it is possible that stratifying participants by cardiovascular health factors or future conversion to dementia would show a clearer picture of the way circulating cytokines and cerebral metbaolites interact. That also may help to inform about the transition from cognitively normal, “healthy” older adult to pathological aging. The nature of the present findings may suggest that differences in circulating cytokines better predict changes related to age-associated pathologies. Inclusion of alternative biological mechanisms of change (e.g., mitochondrial dysfunction, hypometabolism) may be important for characterizing brain changes occurring in non-pathological aging.

The present study had several limitations worth considering which may help guide future research endeavors. First, the study design was cross-sectional, preventing any causal interpretation of the results. It is possible to speculate about the differences between previous studies’ methodologies or samples in the context of inconsistent evidence across these markers; however, longitudinal data would be useful in determining both how cytokines and MRS-measured cerebral metabolites may change and relate to one another in healthy aging. One methodological consideration is the use of absolute values rather than ratios. The present study assessed only absolute values; however, the use of ratios in future research may add valuable information about the nature of neurochemical alterations in aging.

Another limitation may have been the use of absolute values of individual cytokines. An important direction for future work, considering the sometimes counterintuitive effects of cytokines found here, is to investigate the presence of non-linear relationships. Modeling relative rather than absolute concentrations of cytokines also may be helpful in fully understanding their collective impact, as many are pleiotropic and regulate the production or proliferation of other cytokines. Further, future incorporation of other related factors would be helpful in understanding the complex relationships between cytokines and cerebral metabolites in older adults. Some factors of interest may be objective cardiovascular and general medical health measures to provide context for inflammatory markers, as well as diffusion tensor measures of white matter integrity and quantification of central nervous system cytokines in CSF samples. It would be valuable to compare levels of peripheral and central circulating cytokines, and to assess whether central circulating cytokines have more direct relationships to cerebral metabolites. Prior research has identified relationships between CSF chemokines and MRS-measured metabolites in an HIV population (Letendre et al., 2011). It also will be important to assess sex differences in the relationships between neurometabolites and circulating cytokines. Sex differences previously have been reported both in circulating cytokines (Klein and Flanagan, 2016) and MRS-measure metabolites (Grachev and Apkarian, 2000).

The present study introduces relationships between cerebral metabolites and circulating cytokines in a population in which both have implications for the progression of health outcomes, including neurodegenerative disease. The present findings illustrate that these associations are nuanced and complex, but also may serve as preliminary evidence of the interconnectedness of inflammatory markers and neurometabolites in a presently healthy population. This provides a basis for further investigation. It has long been understood that inflammatory markers are important predictors of cognitive decline, dementia, cardiovascular health, and mortality in older adults. Certain pro-inflammatory cytokines have more often been investigated in the context of aging research (e.g., IL-6, TNF-α, IL-1 β). The findings in this sample also provide a basis for allocation of research exploring a broader collection of cytokines in healthy aging, as there are implications of a greater number of inflammatory markers for brain health in this population. Further exploration of this topic is relevant to building more well-informed risk profiles for cognitive decline and dementia, if patterns of circulating cytokines may be useful in predicting neurochemical changes characteristic of preclinical neurodegenerative disease and pathological aging.