Executive function and soluble urokinase-type plasminogen activator receptor (suPAR): a longitudinal study of midlife adults

Abstract

Executive function (EF) declines with age. Declines in EF are associated with adverse outcomes, but there is substantial inter-individual variability in rate and magnitude of decline. Evidence suggests that systemic inflammation may contribute to declines in EF. Extant studies are largely cross-sectional and examine markers of inflammation that are not specific to the immune system and influenced by acute contemporaneous factors. We examined a newer marker of chronic inflammation, soluble urokinase Plasminogen Activator Receptor (suPAR) alongside more traditional markers, IL-6 and CRP. We hypothesized that higher baseline and greater increases in suPAR across an 8–19 (median 17) year period of midlife would associate with larger age-related declines in EF. Data were drawn from the Adult Health and Behavior (AHAB) study (n=599). Participants (55.5% female, 86.2% white, mean age 45 years at W1 and 60 years at W2) completed a neuropsychological battery and blood draw at both waves. EF was comprised of the Trail Making Test, the STROOP Test, and Matrix Reasoning. Univariate and Bivariate Latent change score models (LCSM) examined whether baseline and change in circulating levels of inflammatory mediators were associated with change in EF from W1 to W2. Baseline inflammation was not associated with change in EF. Increases in suPAR were significantly related to concomitant declines in EF (= −0.315; p <.05). No similar effects were observed for IL6 or CRP. Further research should examine suPAR as potential marker of chronic systemic inflammation and its relationship to cognitive aging.

1. Introduction

Cognitive decline occurs in several domains during mid-to-later life (Schaie, 1994). Executive function is one domain of cognition that is known to decline with age (Hedden and Gabrieli, 2004; Salthouse, 2009). Executive function is a broad construct that subsumes several related abilities, including goal-oriented behaviors, shifting attention, working memory, and inhibition, or the ability to control impulses and certain behaviors (Miyake et al., 2000; Zelazo et al., 2004). Declines in executive function have been associated with risk for frailty (Gross et al., 2016; Hughes et al., 2020), falls (Kearney et al., 2013), and mortality (Amirian et al., 2010; Hall et al., 2009). However, there is substantial variability between individuals in the magnitude and rate of decline in executive function (Salthouse, 2009). These differences could be influenced by many factors, such as genetics (Gustavson et al., 2023; Lessov-Schlaggar et al., 2007; Reas et al., 2019), behavioral factors (Gomes Gonçalves et al., 2023; Lin et al., 2022; Naderi et al., 2019), psychosocial stressors (Lindert et al., 2021; Xingjia Cui et al., 2007; Zainal and Newman, 2022) and environmental exposures (Goodwill et al., 2018; Kulick et al., 2020). Recently, systemic chronic inflammation has been considered as a plausible biological contributor to mid- to later-life declines in executive function; however, evidence for this possibility has been mixed.

To elaborate, a 2022 meta-analysis by Fard and colleagues investigating the relationship between circulating levels of the inflammatory markers interleukin-6 (IL-6) and C-reactive protein (CRP) and executive function, amongst other cognitive domains, in healthy older adults found inconsistent evidence for the relationship between inflammatory markers and executive function across nine studies. Four of the studies found that higher levels of circulating IL-6 or CRP were associated with lower levels of executive function or declines in executive function (Marioni et al., 2009; Tegeler et al., 2016; Trollor et al., 2012; Zheng and Xie, 2018). Two studies found support for a cross-sectional association between higher circulating IL-6 and CRP and lower executive function, but not for higher circulating levels being associated with decline in executive function over follow-ups ranging from 1 to 4.6-years (Rafnsson et al., 2007; Schram et al., 2007a). The three remaining studies found no relationship between circulating IL-6 or CRP and executive function (Arfanakis et al., 2013; Dik et al., 2005; Palta et al., 2015).

Discrepancies in the human literature could be partly explained by methodological differences. These differences include variability in study design (cross-sectional or longitudinal), small sample sizes, and differences in how executive function was operationalized and measured. Collectively, the latter may all plausibly contribute to the heterogeneity observed in the meta-analytic review (Fard et al., 2022). The operationalization and measurement of executive function might be particularly impactful because different functional abilities are assessed across studies, and correlations between different assessments of executive function can be quite low (Ettenhofer et al., 2006; Miyake et al., 2000). For this reason, utilizing a latent measure of executive function instead of individual tests may better capture the range of functions subsumed by executive function, because latent measures retain the shared variance of the underlying construct of interest without keeping unrelated variance.

Another limitation of prior studies is the single assessment of inflammatory markers that are subject to diurnal variation (IL-6; Nilsonne et al., 2016) and impacted by contemporaneous factors and conditions near to testing, such as infection, recent physical activity, medication use, etc. (IL-6 and CRP; Pedersen et al., 2001; Pepys, 1995; Steptoe et al., 2007; van Leeuwen et al., 2009). For this reason, attention has turned to a newer circulating marker of inflammation, soluble urokinase plasminogen activator receptor (suPAR), that is more temporally stable, shows no circadian variability, and appears less impacted by contemporaneous factors (Marsland, 2021; Rasmussen et al., 2021b; Thunø et al., 2009). The stability of suPAR, in addition to its correlation with risk factors and outcomes of systemic chronic inflammation, supports its utility for examining changes in systemic chronic inflammation over time. Cumulative evidence suggests that suPAR exhibits a test–retest reliability of ICC = 0.91 over 4 months, and has a correlation coefficient of r = 0.55–0.58 over 5–7 years (Haupt et al., 2019; Rasmussen et al., 2021a; Rasmussen et al., 2021b). suPAR has been shown to be associated with neurocognitive functioning in midlife, where higher suPAR was associated with lower IQ both cross-sectionally (at age 45) and longitudinally (steeper declines in cognitive functioning from childhood to adulthood; Rasmussen et al., 2021a). suPAR has also been shown to be elevated in depression (Murphy et al., 2024), which is a risk factor for cognitive decline (Mehta et al., 2022). Moreover, suPAR was shown to outperform IL-6 and CRP in its consistency or magnitude of association with psychosocial stressors, such as childhood victimization, adult stressful life events, or loneliness and social isolation (Bourassa et al., 2021; Matthews et al., 2024; Rasmussen et al., 2020).

The current study aims to address unresolved questions about systemic chronic inflammation and cognitive aging by employing a longitudinal design, utilizing a latent measure of executive function, a relatively large sample, and assessing suPAR, a more temporally stable and reliable marker of immune activation. Comparisons of suPAR were also made against conventional markers of inflammation, namely IL-6 and CRP, to further examine the consistency of relationships between these systemic chronic inflammation indicators and executive function over time. Specifically, the current study aimed to assess whether increases in suPAR, IL-6, or CRP predict decline in executive function across a ~ 14-year period of midlife. Given that adiposity is a source of markers of systemic inflammation (Anty et al., 2006; Bastard et al., 1999; Haupt et al., 2014; Kern et al., 2001), a secondary aim was to consider the possibility that adiposity contributes to any observed relationships between inflammation and executive function.

2. Methods

2.1. Participants and procedures

Study data were drawn from the University of Pittsburgh Adult Health and Behavior (AHAB) study. The AHAB study provides a registry of behavioral and biological measurements, on midlife community volunteers recruited via mass-mail solicitation from communities of southwestern Pennsylvania (principally Allegheny County; Manuck et al., 2011). There are two study cohorts (AHAB-1 and AHAB-2) that were each assessed on two occasions, Wave 1 (W1) and Wave 2 (W2). AHAB-1 W1 data were collected between 2001–2005, and AHAB-2 W1 data were collected between 2008–2011. Data for W2 were collected from both cohorts beginning in 2017 and ending in February 2024. All demographics for both waves are listed in Table 1. Data were collected to gather information about psychiatric history, demographics, social relationships, cognition, disease risk, medical history, and measurements of biomarkers were taken to assess cardiometabolic risk factors. At W1, these measures were assessed over four visits at the Behavioral Physiology Lab at the University of Pittsburgh. At W2, these measures were reassessed over six visits at the Behavioral Physiology Lab at the University of Pittsburgh. Participants recruited into both AHAB cohorts were 30–54 years of age at W1, with no clinical history of atherosclerotic cardiovascular disease, chronic kidney or liver disease, cancer treatment in the preceding year, or major neurologic disorders, schizophrenia, or other psychotic illness. Other AHAB study exclusion criteria at W1 included pregnancy and use of insulin, glucocorticoid, antiarrhythmic, psychotropic, or prescription weight loss medications, systolic/diastolic blood pressure > 180/110 mmHg (AHAB-1) or > 160/100 mmHg (AHAB-2), alcohol consumption > 25 drinks per week (AHAB-1) or > 35 week (AHAB-2). The AHAB-2 cohort had additional exclusion criteria, including use of diabetic medications, antihypertensives, lipid lowering drugs, fish oil supplements, and participation in shift work. AHAB-2 participants were all employed > 25 h per week. Informed consent was obtained in accordance with approved protocols and guidelines of the University of Pittsburgh Institutional Review Board. There were no further exclusion criteria at W2. Participants were 40–71 (M = 60) years of age at W2 data collection.

Table 1.

Descriptive information of analytic sample.

	W1 (N = 599)	W2 (N = 599)
Age in Years
Mean (SD)	45.2 (6.44)	60.0 (7.22)
Time Between Visits (Years)
Mean (SD)	–	14.3 (2.83)
Sex
Male	266 (44.4 %)	–
Female	333 (55.6 %)	–
Race
White	516 (86.1 %)	–
Non-White	83 (13.9 %)	–
Years of Education
GED or HS Diploma	43 (7.2 %)	–
Tech Training	30 (5.0 %)	–
Some College	67 (11.2 %)	–
College Degree or Higher	459 (76.6 %)	–
Smoking Status
Smoking	110 (18.3 %)	75 (12.5%)
Nonsmoking	489 (81.6 %)	524 (87.5 %)
EF Composite
Mean (SD)	46.6 (22.8)	39.2 (28.8)***
suPAR
Mean (SD)	2.37 (0.84)	2.60 (0.79)***
CRP
Mean (SD)	1.57 (1.88)	1.94 (2.09)***
IL6
Mean (SD)	1.54 (1.47)	2.11 (1.55)***
BMI
Mean (SD)	26.79 (5.46)	28.25 (6.02)***

Note. suPAR, is measured in ng/mL, CRP is measured in mg/L, and IL-6 is measured in pg/mL; − = N/A or remained the same. Executive function composite was created by adding up raw Matrix Reasoning, STROOP, and Trails B-A scores (multiplied by −1). GED = General Education Development; EF = executive function; suPAR = soluble urokinase plasminogen activator receptor; CRP = C-reactive protein; IL-6 = interleukin-6; BMI = body mass index

^*p < 0.05.

^**p < 0.01.

^***p < 0.001.

2.2. Measures

2.2.1. Demographic variables

All demographic measures: age, sex, race, smoking status, and education, were collected via self-report questionnaires. All descriptive statistics of the sample are in Table 1.

2.2.2. Body mass index (BMI)

Height (in) and weight (lb.) were measured at both waves, and BMI was calculated with the following formula: $\mathrm{w}\mathrm{e}\mathrm{i}\mathrm{g}\mathrm{h}\mathrm{t}(\mathrm{l}\mathrm{b}.)/{[\mathrm{h}\mathrm{e}\mathrm{i}\mathrm{g}\mathrm{h}\mathrm{t}\left(\mathrm{i}\mathrm{n}\right)]}^2\times 703$.

2.2.3. Executive function

A battery of neuropsychological tests was administered at both waves. Three measures of executive function, the Trail Making Test (a measure of task-switching), Matrix Reasoning (a measure of concept formation; from the Wechsler Abbreviated Scale of Intelligence), and the STROOP Task (a measure of response inhibition), were administered. The Trail Making Test includes two timed parts. In part A, participants are asked to connect ascending numbers (1–20) in order as quickly as possible. In part B, the participants must connect alternating numbers and letters (1-A-2-B) as quickly as possible. The score is the difference in time taken between part A and Part B (Trails B-A), with higher values reflecting poorer performance. Scores were multiplied by negative 1 so that larger scores would reflect better performance. For Matrix Reasoning, participants are shown a sequence of colors or pictures with one missing and are asked to pick the missing element in 30 s. Participants can be shown up to 35 stimuli, with testing discontinued after four wrong answers on five consecutive items. Raw scores ranged from 0 to 35 with higher scores reflecting better performance. For the STROOP Task, participants completed three conditions. Condition three was used to operationalize response inhibition. In this condition participants list the ink color of words printed in an incongruous color (e.g., the word ‘red’ printed in blue ink). The number of correct responses read aloud in 45 s for condition three was utilized, with higher scores indicating better performance.

2.2.4. Inflammatory markers

Four blue citrate tubes (4 mL each) of blood for the analysis of inflammatory markers were collected from participants at both waves. These samples were then spun down to plasma, aliquoted, and stored in freezers at −80 °C until ready to be processed in batches. Plasma levels of IL-6 and suPAR were measured with ELISA kits (Human IL-6 Quantikine HS ELISA made by R&D Systems, cat.# HS600B, and suPARnostic AUTO Flex ELISA made by ViroGates, respectively), and run according to the manufacturers’ directions. To assess reliability, all samples were run in duplicate, and an average of the measures was taken. Samples with a CV > 15 % were rerun. For suPAR, the assay standard range is 0.4 ng/mL to 16 ng/mL. The inter-assay average coefficient of variation (CV) was < 10 %, and the intra-assay CV was < 5 %. Paired samples (from W1 and W2) for each participant were analyzed on the same plate. For IL-6, the assay standard range is 0.156 to 10 pg/mL. The inter-assay average CV was < 5 %, and the intra-assay CV was < 6 %. Samples from W1 were analyzed after data collection, as were samples at W2, meaning that the samples from W1 and W2 were analyzed for IL-6 11–17 years apart. Analysis of CRP has been reported previously (Marsland et al., 2015). CRP was measured at the University of Vermont’s Laboratory of Clinical Biochemistry Research with the BNII nephelometer from Dade Behring utilizing a particle enhanced immunonephelometric assay. More information can be found at: https://www.siemens-healthineers.com/en-us/plasma-protein/systems/bn-ii-system. As with IL-6, samples from W1 and W2 were analyzed at separate time points, and results are reported in mg/L.

2.3. Statistical analyses

All statistical analyses were performed in R (R Core Team, 2022) and R Studio (Version 2022.12.0 + 353). Latent Change Score Models (LCSMs) were created using the ‘lavaan’ package (Rosseel, 2012). For all neuropsychological tests, invalid scores were taken out (e.g., individuals who were color-blind had no STROOP Color-Word Score, N = 17). Scores on the cognitive measures were winsorized at 3 SDs from the mean (n = 6 at W1 and n = 10 at W2 for Matrix Reasoning; n = 10 at W1 and n = 9 at W2 for Trails B-A; n = 0 at both timepoints for STROOP). Age at W2 and sex (male = 0, female = 1) were added as covariates in the LCSMs because they are both associated with variability in markers of inflammation. Time between visits ranged from 8 to 19 (median 17) years and was also added as a covariate. Age and time between visits was centered. Because 86 % of the sample was white, race was dichotomized (white = 0, nonwhite = 1). All immune markers were log-transformed to approximate a normal distribution. At W1, 32 individuals had levels of CRP higher than 10 mg/L or IL-6 higher than 10 pg/mL, indicating an acute infection, such as a cold, and were removed from analyses. At W2, 35 values were removed for the same reason. There are currently no established clinical cutoffs for suPAR, and thus all values for suPAR were retained. Immune markers, age, and time between visits were then mean centered for interpretability. Correlation coefficients for inflammatory markers at W1 and W2 were estimated to assess their temporal stability. Paired t-tests were used to examine differences in means for inflammatory markers and BMI from W1 to W2.

Before LCSMs were estimated, longitudinal measurement invariance of latent executive function, as defined by STROOP, Trails, and Matrix Reasoning scores, was tested using configural, metric, and scalar models and comparing them with chi-square difference tests, change in Comparative Fit Index (CFI) values, and change in McDonald’s non-centrality index (NCI). Measurement invariance confirmatory factor analyses were conducted with increasing numbers of constraints (i.e., on factor loadings, then intercepts) to ensure that the executive function construct was measured in a consistent fashion across time. Specifically, a configural model was estimated to ensure that the form of the model was the same between W1 and W2. In this study, this would mean that the observed STROOP, Trails, and Matrix Reasoning scores are adequate estimators of latent executive function at W1 and W2. When moving from configural to metric, the only addition to the model was that factor loadings from each indicator (test of executive function) to the latent variable are constrained, which gives information about how each indicator relates to the latent variable at both waves. Finally, when moving from metric to scalar, the only addition to the model is that intercepts for each indicator are constrained. This indicates whether someone with the same level of executive function at W1 and W2 would have the same expected value on a given indicator (Clark and Donnellan, 2021). Because these models are nested within each other, they can be compared using chi-square difference tests. A nonsignificant chi-square indicates measurement invariance, suggesting that the construct was similar at both time points. However, due to chi-square difference tests being sensitive to sample size (i.e., chi-square tests are likely to be significant in large samples even when differences in factor loadings or intercepts are trivial), we also report the change in CFI values, along with change in McDonald’s NCI. A change in CFI equal to or less than 0.01 indicates measurement invariance, while, for this specific model, a change in McDonald’s NCI less than 0.0065 would indicate measurement invariance (Meade et al., 2008).

An unconditional model was estimated first, where change in executive function from W1 to W2 was examined without covariates. Latent executive function variables for W1 and W2 were defined and estimated from the three tests of executive function at the corresponding wave. Next, latent executive function at W1 and latent executive function at W2 were used to estimate a new latent variable representing change in executive function (Δ EF). To investigate the relationship between baseline inflammation and change in executive function, univariate LCSMs were fitted. In these models, Δ EF was regressed onto each inflammatory marker at baseline, adjusting for covariates. For the univariate LCSMs, there were four models, one to investigate inflammatory markers separately and one model with all inflammatory markers predicting Δ EF. To investigate the correlation between change in inflammation and change in executive function, bivariate latent change scores models were fitted. Executive function was treated the same as in the univariate models, but in the bivariate models, inflammation at W1 and W2 were used to estimate a latent variable representing change in inflammation. Separate models were estimated for each inflammatory marker. Correlations between change in inflammation and change in executive function were estimated in each model. To investigate how BMI might contribute to the relationship between inflammation and executive function, change in BMI was then added into the LCSMs as a covariate. Change in BMI was calculated by taking W2 BMI-W1 BMI so that increases in BMI would be positive and decreases would be negative. Change in BMI was then mean centered. All models were fitted using maximum likelihood estimation. For all models, chi-square value and p-value, root mean square error of approximation (RMSEA) and RMSEA 95 % CI values, CFI, and Tucker–Lewis index (TLI) are reported as measures of model fit.

Sensitivity analyses were run removing individuals who self-reported autoimmune disorders and/or taking steroids or other medicines that affect circulating levels of inflammatory markers from analyses to ensure that no medical conditions or medications influenced the results. Specifically, participants who reported one or more of the following were true for them were removed in the sensitivity analyses: a history of hepatitis B or C, history of rheumatoid arthritis or history of autoimmune “connective tissue” disorders (lupus, psoriatic arthritis, Sjorgren’s syndrome (SICCA), scleroderma, polymyositis or dermatomyositis, graves, Hashimoto’s, Chronic Inflammatory Demyelinating Polyneuropathy (CIDP), polymyalgia rheumatica), use of inhaled corticosteroids, taking any other form of steroids, taking any immunosuppressants, taking any sedating antihistamines, taking any cold or allergy remedy, and taking any antibiotics. In another model, smoking status at W2 was added as a covariate (nonsmoker = 0, smoker = 1) because smoking has been associated with elevated levels of suPAR (Haupt et al., 2014; Haupt et al., 2019), IL-6 (Yanbaeva et al., 2007), and CRP (Yanbaeva et al., 2007). A post-hoc sensitivity analysis was run to ensure that using a partially invariant measurement model of executive functioning was not impacting results. To do so, the partial measurement invariance model described below was compared to a fully invariant model (i.e., all indicator factor loadings and intercepts constrained to be equal across time) to ensure that the results were robust across these models.

3. Results

3.1. Sample characteristics

Descriptive statistics are shown in Table 1. Participants (N = 599) were 60 years old on average (SD 7.22), 55.6 % female, 86.1 % white, and 76.6 % had a college degree or higher. All inflammatory markers increased from W1 to W2. Average time between visits ranged from 8–19 years with a mean of 14.3 years.

3.2. Correlations for immune markers over follow-up

The correlation coefficient for log-transformed IL-6 from W1 to W2 was the lowest at follow up (r = 0.20), while both log-transformed CRP (r = 0.56) and log-transformed suPAR (r = 0.61) were moderately correlated over time. However, because W1 and W2 samples were analyzed years apart for IL-6 and CRP, but concurrently for suPAR, we are not interpreting the strength of the correlations for markers.

3.3. Longitudinal measurement invariance

A configural model was supported and showed excellent fit (χ2(5) = 5.017, p = 0.414, RMSEA = 0.002, 95 % CI: 0.000, 0.057, CFI = 1.00, TLI = 1.00), indicating that the observed STROOP, Trails, and Matrix Reasoning scores were underlain by a single latent factor at both time points. When configural and metric models were compared, there was a significant change in chi-square (Δ χ2 = 13.508, p < 0.01). Change in CFI was nonsignificant (Δ CFI = 0.01), but change in McDonald’s NCI was significant (Δ McDonald’s NCI = 0.009; Meade et al., 2008). After investigating which factor loading varied from W1 to W2, we found that the Trails B-A loading had to be freed to achieve partial weak invariance. This allowed the Trails B-A loading to vary across time, rather than being constrained to be equal at W1 and W2.

Nested metric and scalar models were also compared, and the scalar model showed significantly worse fit on all indices (Δ χ2 = 31.424, p < 0.001, Δ CFI = 0.022, Δ McDonald’s NCI = 0.025). After investigating which intercept varied from W1 to W2, partial scalar invariance was achieved by freeing the Matrix Reasoning intercept. This allowed the Matrix Reasoning intercept to vary across time, rather than being constrained to be equal at W1 and W2. Since strong partial measurement invariance was achieved, this measurement model was used in our primary analyses.

3.4. Unconditional change and change adjusting for covariates

Results for an unconditional partially invariant change model, reflecting change in executive function over time without any covariates included in the LCSM, showed executive function declined from W1 to W2 (unstandardized ΔEF = −0.774, 95 % CI: [−0.998, −0.552]; standardized ΔEF = −0.924, p < 0.001).

3.5. Univariate latent change score analyses

The partially invariant univariate LCSMs all showed excellent fit before adjusting for age, sex, and time between visits (fit indices in Supplementary Table 1). While fit indices were worse after adding covariates, they were retained to ensure parameters between inflammatory markers and change in executive function were not biased by factors associated with inflammation. Because the latent change score functions as an intercept in the model (i.e., it reflects the average degree of change in executive function when all covariates are zero; Grimm et al., 2016), the change score with covariates in the model (gender, age, time between visits) reflects the average degree of change in executive function for average-aged males and the average amount of time between visits.

Age and time between visits significantly influenced the decline in executive function such that higher ages at baseline and longer times between visits led to steeper declines in executive function in the sample. Males showed steeper declines in executive function than females, but this was not significant. The relations between age, sex, and time between visits and change in executive function were consistent across models, and thus we focus on results for the inflammatory markers (models are depicted in Supplementary Figs. 1–3). The partially invariant model for IL-6 at W1 predicting change in executive function showed poor to adequate fit indices (All univariate LCSM fit indices shown in Table 2; Hu and and Bentler, 1999). IL-6 was not significantly related to change in executive function (β = −0.083, p = 0.410). The partially invariant model for CRP at W1 predicting change in executive function showed reasonable fit, but CRP was not significantly related to change in executive function (β = −0.121, p = 0.203). The partially invariant model for suPAR at W1 predicting change in executive function also showed reasonable fit. suPAR was also not significantly related to change in executive function (β = −0.137, p = 0.19). When all biomarkers were entered simultaneously as predictors of change in executive function, the partially invariant model showed reasonable fit. IL-6, CRP, and suPAR were all weakly and non-significantly related to change in executive function (β = −0.089, β = −0.127, and β = −0.110 respectively, all ps > 0.2) with CRP showing the strongest relation.

Table 2.

Fit indices for univariate latent change score models.

Model	χ2	χ2 p-value	RMSEA	RMSEA CI	CFI	TLI	df	n
IL-6	122.682	0<.001	0.073	0.060–0.087	0.937	0.909	26	577
CRP	93.628	0<.001	0.067	0.053–0.082	0.953	0.929	26	578
suPAR	74.427	0<.001	0.061	0.045–0.077	0.958	0.937	26	501
All	94.159	0<.001	0.061	0.046–0.075	0.946	0.920	34	482

Note. χ² = Chi-Square value; RMSEA = Root Mean Square Error of Approximation; RMSEA CI = Root Mean Square Error of Approximation 95 % Confidence Interval; CFI = Comparative Fit Index; TLI = Tucker-Lewis Index; df = Degrees of Freedom; n = number of observations used for analysis; suPAR = soluble urokinase plasminogen activator receptor; CRP = C-reactive protein; IL-6 = interleukin-6.

3.6. Bivariate latent change score analyses

Similar to the partially invariant univariate LCSMs, the partially invariant bivariate LCSMs all showed excellent fit before adjusting for age, sex, and time between visits (fit indices shown in Supplementary Table 2; models for IL-6 and CRP depicted in Supplementary Figs. 4 & 5). After adjustment, all partially invariant bivariate change score models had poor to adequate model fit (all fit indices for bivariate LCSMs shown in Table 3). In all partially invariant bivariate LCSMs, age (β’s = −0.52 to −0.49, p’s < 0.001) and time between visits (β’s = −0.38 to −0.32, p’s < 0.01) were significantly related to decline in executive function, such that older participants had greater decline in executive function, and longer time between visits was related to greater decline in executive function. Longitudinal correlational change in IL-6 with executive function showed poor fit, and change in IL-6 was negatively correlated with change in executive function (r = −0.095), but this effect was not significant (p = 0.50). The bivariate latent change score model examining longitudinal correlational change between CRP and executive function showed poor to adequate fit. Change in CRP correlated positively with change in executive function (r = 0.147), but this was not significant (p = 0.29). Longitudinal correlational change in suPAR predicting change in executive function showed poor fit. suPAR correlated negatively with executive function over time and this was significant (Cov = −0.030, 95 % CI: [−0.059, −0.002], r = −0.314, p < 0.05; model depicted in Fig. 1).

Table 3.

Fit indices for bivariate latent change score models.

Model	χ2	χ2 p-value	RMSEA	RMSEA CI	CFI	TLI	df	n
IL-6	141.469	0<.001	0.070	0.058–0.082	0.933	0.903	36	548
CRP	95.931	0<.001	0.053	0.040–0.066	0.965	0.949	36	569
suPAR	179.418	0<.001	0.082	0.070–0.094	0.920	0.884	36	487

Note. χ² = Chi-Square value; RMSEA = Root Mean Square Error of Approximation; RMSEA CI = Root Mean Square Error of Approximation 95 % Confidence Interval; CFI = Comparative Fit Index; TLI = Tucker-Lewis Index; df = Degrees of Freedom; n = number of observations used for analysis; suPAR = soluble urokinase plasminogen activator receptor; CRP = C-reactive protein; IL-6 = interleukin-6.

Fig. 1.

Bivariate latent change score model showing correlation between change in suPAR and change in executive functioning. Note. Means and covariates (age, sex, and time between visits) are omitted for visual clarity. ΔS = change in soluble urokinase plasminogen activator receptor (suPAR); EF = executive functioning; TMT = Trails Making Test; ST = STROOP Color-Word Test; MR = Matrix Reasoning. *p < .05, **p < .01, ***p < .001.

3.7. BMI analyses

Because none of the partially invariant univariate LCSMs were significant, BMI was only added as a covariate to the bivariate LCSMs. On average, BMI increased by 1.48 points between W1 and W2. After adjusting for change in BMI, longitudinal correlational change between IL-6 and executive function was largely unaffected (r = −0.095 to r = −0.094) and was still nonsignificant (p = 0.53). BMI significantly predicted change in IL-6 (p < 0.001). In the CRP bivariate LCSM, after adjusting for change in BMI, the longitudinal correlational change between CRP and executive function went from r = 0.147 to r = 0.140 and was still nonsignificant (p = 0.34). BMI significantly predicted change in CRP (p < 0.001). In the suPAR bivariate LCSM, after adjusting for change in BMI, longitudinal correlational change between suPAR and executive function went from r = −0.314 to r = −0.312 and remained significant (p < 0.05). Change in BMI did not significantly predict change in suPAR (p = 0.90).

3.8. Sensitivity analyses

Since the only significant predictor of decline in executive function was increases in suPAR, we only report sensitivity analyses for the model with suPAR as a covariate. After removing individuals deemed immune ineligible at W1 or W2 (N = 122) from the analyses, fit indices remained consistent, and longitudinal increases in suPAR were still significantly related to declines in executive function (Cov = −0.030, 95 % CI [−0.061, −0.002], r = −0.321, p < 0.05).

When smoking status was added into the model, fit indices remained about the same for the longitudinal relationship between suPAR and declines in executive function, albeit with a p just above 0.05 (Cov = −0.028, 95 % CI: [−0.055, 0.000], r = −0.290, p = 0.051). For our sensitivity analyses surrounding measurement invariance, the change in EF was examined when assuming strong measurement invariance. Specifically, we constrained factor loadings and intercepts of the three EF indicators to be equal across W1 and W2, and the estimated change in EF from W1 to W2 (unstandardized ΔEF = −0.828; 95 % CI [−1.070, −0.586]; standardized ΔEF = −0.892, p < 0.001) did not substantively differ from the change estimated under partial measurement invariance (unstandardized ΔEF = −0.774, 95 % CI [−0.998, −0.552]; standardized ΔEF = −0.924, p < 0.001). These results highlight that the observed decrease in EF between W1 and W2 is largely uninfluenced by measurement artifacts.

4. Discussion

The primary aim of the current study was to investigate whether elevated levels of inflammatory markers relate to decline in executive function across a ~ 14-year period of midlife. Given the long follow-up period in a sample of older adults, we expected to see decline in executive function, which was found. In separate models, we found that baseline circulating levels of the inflammatory markers IL-6, CRP, and suPAR were not related to future decline in executive function. When investigating change in inflammation, change in IL-6 and change in CRP were not correlated with decline in executive function. However, increases in suPAR across the follow-up period were significantly correlated with concomitant decreases in executive function. All findings were independent of age, sex, and time between visits, and the inverse relationship between change in suPAR and change executive function remained significant in sensitivity analyses removing individuals with medical disorders and/or taking medications that impact circulation levels of inflammatory mediators. After adjusting for smoking status, the relationship between change in suPAR and change in executive function was attenuated and no longer statistically significant, although the magnitude of covariance remained about the same. A secondary aim of the present study was to examine how change in BMI might contribute to this relationship, as adipose tissue secretes IL-6 (Coppack, 2001; Kern et al., 2001) and is associated with circulating levels of the above inflammatory markers (Anty et al., 2006; Bastard et al., 1999; Haupt et al., 2014; Kern et al., 2001; Rasmussen et al., 2021b). Change in BMI significantly predicted change in IL-6 and change in CRP in the bivariate LCSMs, but was unrelated to change in suPAR and did not change the magnitude of the relationship between change in suPAR and change in executive function. Overall, the finding that increases in suPAR are correlated with decreases in executive function supports the idea that systemic chronic inflammation is related to the decline in executive function that accompanies aging.

Much of the evidence in support of a biological pathway between systemic chronic inflammation and cognition comes from nonhuman rodent studies. For example, in transgenic rodent models, in which mice have been genetically modified to overproduce IL-6, TNF-α, and IL-12, the overexpression of cytokines leads to neurodegeneration and declines in cognition (Reviewed in Campbell et al., 2010; Heyser et al., 1997). Conversely, transgenic mice deficient in the cytokine IL-6 show less neurodegeneration and retain cognitive function when compared to wildtype mice (Dugan et al., 2009; Reynolds et al., 2024). This work in nonhuman rodent studies has also elucidated some of the mechanisms by which inflammation may affect cognition. The delivery of the proinflammatory cytokine IL-1 β to the hippocampus has been shown to reduce long-term potentiation (LTP), a biochemical process crucial to learning and the formation of memory (Bellinger et al., 1993). Further, the cytokines IL-6, TNF-α, and IL-1 β have all been shown to modulate cognitive function through mechanisms affecting synaptic plasticity, LTP, and neurogenesis (McAfoose and Baune, 2009). At physiological levels, these cytokines have been shown to support cognition, but higher levels of these cytokines result in tissue damage and decline of cognition (McAfoose and Baune, 2009).

The present study investigated the inflammatory markers IL-6, CRP, and suPAR because they are plausibly related to executive function. IL-6 is one of the markers consistently studied in the animal literature and has been shown to mediate biochemical processes underlying cognition (For review see McAfoose and Baune, 2009). Furthermore, when considering the pathways by which peripheral inflammation might relate to inflammation in the central nervous system (CNS), IL-6 is one of the few cytokines that has direct transport mechanisms that allow for IL-6 in the periphery to directly signal the CNS and promote CNS inflammation (Maier and Watkins, 1998). CRP is an acute phase protein, produced by the liver in response to the immune system encountering bacteria or from injury and trauma (El Hassan et al., 1990; Samols et al., 2002), and IL-6 is a mediator of this process (Samols et al., 2002). Thus, higher levels of circulating IL-6 can initiate higher levels of CRP via the liver. Higher levels of CRP are associated with immune activation, which is one reason why it is frequently used as a marker of systemic inflammation and could be related to decline in executive function. We also included a novel examination of a newer immune biomarker, suPAR, which is thought to be a general marker of and contributor to chronic innate immune activation.

suPAR is cleaved from monocytes, neutrophils, and macrophages in response to activation of the immune system, as well as from senescent cells during aging (Amor et al., 2020; Coppé et al., 2010), with higher levels reflecting higher immune activation and disease (Thunø et al., 2009). In this respect, it is likely that suPAR captures a broader range of inflammatory- and immune-related activities than IL-6 or CRP (Rasmussen et al., 2021b; Smith and Marshall, 2010). Moreover, there is evidence that suPAR itself is proinflammatory, although how much of an active role suPAR plays in immune activation and the underlying mechanisms is still under investigation. suPAR is composed of three primary domains, with the first domain being connected to the second and third by a linker region (Fazioli et al., 1997). When the first domain is cleaved off, the remaining second and third domains have chemotactic properties, playing a role in the recruitment of immune cells (Rasmussen et al., 2021b; Thunø et al., 2009). When all three domains of suPAR are intact, dimerized, and bound to urokinase-type plasminogen activator (uPA) and vitronectin, this complex can activate uPA pathways leading to proinflammatory activity (primarily affecting cell migration and adhesion; Hamada et al., 2024; Rasmussen et al., 2021b; Sidenius et al., 2002; Thunø et al., 2009). It is also theorized that suPAR may impair neutrophil efferocytosis, which, if confirmed, would implicate suPAR in sustaining the inflammatory response (Park et al., 2009; Rasmussen et al., 2021b). Taken together, this evidence suggests that suPAR reflects immune activation and also may promote inflammatory responses through multiple pathways. Both of these factors make suPAR a potentially relevant biomarker in relation to decline in executive function.

While evidence from nonhuman rodent studies offers experimental support for a link between inflammation and executive function, and IL-6 and CRP both have biologically plausible ties to cognitive function, evidence in human studies is less consistent. Some inconsistencies may come from methodological considerations mentioned previously. For example, patterns of results tend to differ between cross-sectional and longitudinal studies. Of ten studies that investigated cross-sectional relationships between IL and 6 and/or CRP with executive function, nine found a relationship between higher circulating levels and worse executive function (Marsland et al., 2006; Mooijaart et al., 2013; Rafnsson et al., 2007; Schram et al., 2007a; Schram et al., 2007b; Tegeler et al., 2016; Trollor et al., 2012; Vintimilla et al., 2019; Wersching et al., 2010) and only one did not (Teoh et al., 2023). Of twelve studies that examined longitudinal relationships, ten investigated levels of IL-6 or CRP at one timepoint predicting decline in executive function, and the results were reversed. Nine of the ten studies found no relationship between baseline IL-6 or CRP and decline in executive function (Beydoun et al., 2018; Dik et al., 2005; Marioni et al., 2009; Palta et al., 2015; Rafnsson et al., 2007; Schram et al., 2007b; Teoh et al., 2023; Walker et al., 2019; West et al., 2020), while one found that higher levels of CRP predicted decline in executive function (Bourassa and Sbarra, 2017). There were two studies that investigated the relationship between change in CRP and change in executive function and both found that increases in CRP were related to declines in executive function (Bahorik et al., 2024; Zheng and Xie, 2018). This suggests that study design impacts observed results, with cross-sectional studies and studies examining parallel changes in inflammation and executive function more likely to find a significant relationship than studies examining one time-point of inflammation predicting change. Our finding that baseline inflammation was not related to decline in executive function is in-line with most of the findings from the above studies. Our finding that change in CRP was not related to decline in executive function is in contrast to the other two studies that did find a relationship between increasing CRP and declines in executive function, but the length of follow-up in these studies was between 5–10 years later, whereas the average follow-up time for the present study was 14.3 years, which could be one explanation for the discrepancy.

Another methodological consideration is which inflammatory markers are operationalized as systemic chronic inflammation. IL-6 and CRP are assessed as markers of systemic chronic inflammation in the literature, and these markers may be of biological relevance for cognition function, as discussed above. However, CRP is an acute phase protein, and IL-6 is a cytokine and mediator of the acute phase response, meaning that both proteins respond to factors unrelated to long-term cognition. Many factors cause acute fluctuations in CRP and IL-6 including: infections and acute illness (Pepys, 1995; Vollmer-Conna et al., 2004), exercise (Kasapis and Thompson, 2005; Pedersen et al., 2001), duration of sleep (Meier-Ewert et al., 2004; van Leeuwen et al., 2009), current levels of adiposity (Brooks et al., 2010; Park et al., 2005; Qi et al., 2007), and acute stress (Steptoe et al., 2007). Meta-analytic evidence suggests that IL-6 and CRP show correlation coefficients of r = 0.385 and r = 0.355 over 5 years, respectively, suggesting low stability over time (Walsh et al., 2023). Thus, while these markers are proinflammatory and theoretically tied to mechanisms relating to cognitive decline, they can change rapidly in response to many different stimuli and are less reliable markers of stable differences that characterize individuals across longer periods of time, ~ 14 years in the current study. Of note, suPAR is more temporally stable than CRP and IL-6, and is less influenced by acute contemporaneous events (Rasmussen et al., 2021b). The present study found that CRP and suPAR were correlated more highly across timepoints than IL-6, however, we were unable to meaningfully compare these results as IL-6 and CRP were analyzed when they were collected at W1 and at W2, while suPAR from W1 was analyzed at the same time as W2, which prevented changes in the assays over time (e.g., differences in antibodies) from influencing results. We recommend that future studies investigate the temporal stability of suPAR in healthy adults, as most of the literature examining its stability has come from clinical samples (Huttunen et al., 2011; Sier et al., 1999; Thunø et al., 2009) or from very small samples (Riisbro et al., 2001). The few studies that have been done in larger, healthy samples do indicate that suPAR is relatively stable over time (Haupt et al., 2019; Rasmussen et al., 2021a).

The secondary aim of the present study was to investigate whether change in BMI might contribute to the relationship between change in inflammation and change in executive function. There is meta-analytic evidence that executive function is lower in obese and overweight individuals compared to individuals with a normal BMI (Yang et al., 2018), and executive function improves over time after weight-loss through bariatric surgery or calorie restriction (Veronese et al., 2017). In relation to this evidence and the association between BMI and inflammation, we hypothesized that BMI would reduce the magnitude of the relationship between inflammation and executive function. Results showed that increase in BMI across the two study waves accompanied an increase in IL-6 and CRP, but was not related to change in suPAR, which is in line with most studies that have found weak or no correlations between BMI and suPAR (Fagerberg et al., 2017; Rasmussen et al., 2019; Rasmussen et al., 2021b). Further, change in BMI did not alter the relationship between suPAR and executive function. Given this finding, along with the age of the sample, and the long follow up period, it is possible that change in suPAR is tapping into aging-related processes, such as aging of the immune system. Aging of the immune system is associated with so-called inflammaging, i.e., age-related increase in chronic, low grade inflammation with higher circulating levels of inflammatory markers (Bruunsgaard et al., 2001; Castelo-Branco and Soveral, 2014; Shaw et al., 2010) and immunosenescence (Lee et al., 2022). In support of this idea, there is evidence that elevated suPAR has been associated with accelerated aging at midlife (Rasmussen et al., 2021a) and that suPAR is produced by senescent cells (Amor et al., 2020; Coppé et al., 2010). Age and time between visits were adjusted for in the analyses and found to be significantly related to decline in executive function. Older individuals and longer follow-up times were both related to greater declines in executive function. However, increases in suPAR were significantly correlated to decline in executive function over and above age and time between visits. This suggests that, while age could be a variable influencing both executive function and suPAR, change in suPAR additionally reflects, and may contribute to, changes in immune activation which are thought to be causally related to declines in cognition, as outlined above.

In sensitivity analyses, the association between increases in suPAR and decreases in executive function was not influenced by removing individuals taking medications, or with medical conditions, that have been found to influence the immune system. However, adjusting for smoking status did attenuate the relationship and it was no longer significant. Smoking is associated with higher levels of suPAR (Rasmussen et al., 2021b) and has been associated with increased cellular senescence in various cell types (Nyunoya et al., 2006; Tsuji et al., 2004). Furthermore, smoking has widespread effects on both innate and adaptive immunity that could lead to increases in systemic chronic inflammation, such as increased expression of toll-like receptors and exposure to reactive oxygen species, though the exact cellular mechanisms between smoking and systemic chronic inflammation are yet to be elucidated (For review see Qiu et al., 2017). The finding that smoking attenuated the relationship between suPAR and executive function suggests that smoking may be one potential pathway by which suPAR might relate to declines in executive function.

There are a number of limitations and strengths to the current study. The main limitations pertain to the demographics of the sample, which was mostly white and highly educated. This reduces the generalizability of the study. Additionally, suPAR is a newer biomarker, and these results will need to be replicated in other samples. However, some of the strengths of the current study include investigating change in both inflammatory markers and executive function, rather than investigating one time point of inflammation. More studies that investigate how change in inflammation is associated with change in executive function, and other cognitive domains, are needed to better understand biological pathways that may be intervened on to slow cognitive decline or stratify individuals into risk categories for cognitive decline. Another strength is the utilization of a latent score of executive function derived from the shared variance of three separate tests, which may provide a more reliable index of executive function than individual tests of more specific executive function skills.

5. Conclusion

The findings from the present study indicate that increases in a stable marker of systemic chronic inflammation and immune activation are correlated with declines in executive function over a long follow-up period in relatively healthy, older adults. The current study also supports the idea that suPAR is a putative marker of systemic chronic inflammation and adds to literature by finding it may have implications for cognitive aging.

Appendix A. Supplementary data

Supplementary data to this article can be found online at https://doi.org/10.1016/j.bbi.2025.06.030.

Data availability

Data will be made available on request.