Low circulating adropin concentrations identify vulnerability in learning-dependent cognitive performance in aged rhesus macaques

Abstract

Identifying biomarkers that identify vulnerability to age-related cognitive decline is a major priority in aging research. Adropin, a circulating peptide that regulates metabolic and vascular homeostasis, has been associated with cognitive performance in humans, but its relevance across species has remained unclear. Here we report low plasma adropin concentrations associate with poor decision-making in aged rhesus macaques subject to an increasing food choice test paradigm. Animals with higher adropin levels exhibited faster improvement in reaction time and reductions of variability in reaction time, whereas intrinsic performance on simple tasks was preserved. These associations were most apparent under conditions requiring adaptation to novelty or stress, suggesting that adropin may signal cognitive resilience rather than baseline executive capacity. The findings parallel mechanistic data from rodent models linking adropin signaling to mitochondrial function, resistance to oxidative stress, and hippocampal-dependent learning. Together, these results support measurements of circulating adropin as a conserved, translational biomarker of cognitive aging and a potential therapeutic target.

Introduction

The progressive structural and functional deterioration of the brain observed with aging reduces quality of life and increases risk of mortality^1,2. The cerebral cortex, prefrontal cortex, and hippocampus undergo atrophy; white-matter tracts deteriorate; and gray-matter volume declines as neurons and synaptic connections are lost. Aging is also accompanied by progressive deterioration of the blood–brain barrier, increasing permeability to circulating factors influencing neuronal function and cognitive decline³. These changes impair information processing and memory recall and increase vulnerability to neurodegenerative diseases of which Alzheimer’s disease (AD) is most common^1,2,4.

Identifying modifiable, fluid-based biomarkers signaling at-risk individuals is essential for targeting and monitoring interventions designed to preserve brain health in aging populations^5,6. One promising candidate is the peptide hormone adropin. Adropin was originally described as a hepatokine whose functions coordinate glucose homeostasis and mitochondrial function during transitions between fed and fasted states^7–16. In community-dwelling, late-middle-aged and older adults, low circulating adropin concentrations are associated with increased risk of cognitive decline^7,8. Adropin may therefore link systemic metabolic homeostasis to brain aging and risk of developing neurodegenerative disease.

Experimental data from animal and human studies support this possibility. Systemic administration of synthetic adropin peptide improves cognitive performance, preserves blood-brain barrier functions, and enhances recovery from cerebrovascular injury^17–22. However, ENHO/adropin is also highly expressed in the brain, particularly in astrocytes, and is co-expressed with gene clusters enriched for oxidative processes, protein synthesis, and lipid metabolism^7,8,23,24. Collectively, these findings raise the possibility that circulating adropin serves as a biomarker—and potentially a modulator—of brain health during aging.

We previously reported that low plasma adropin concentrations in rhesus macaques (Macaca mulatta) is predictive of weight gain and metabolic dysregulation²³. The California National Primate Research Center (CNPRC) maintains an aging rhesus macaque colony that provides a valuable preclinical model for investigating how environmental, social, and metabolic variables affect cognitive aging^25,26. The species’ larger brain and complex cognitive repertoire permit behavioral assessments more comparable to those used clinically²⁶ than is possible in other animal models. Here, we tested the hypothesis that low plasma adropin concentrations would associate with poor performance on a value-based, learned food-choice task in aging rhesus macaques. For this purpose, we used an appetitive, value-based, choice task with increasing complexity that engages distributed frontostriatal circuits²⁷. Reaction time (RT) and its variability reflect the efficiency and stability of response selection under increasing choice competition, with modulation by learning, motivational, and environmental factors.

Results

The investigation pooled the results from two studies of male and female nondiabetic, aged rhesus macaques. Housing status, age, and sex are shown in Table 1; data used for this analysis are provided as a supplement (Supplementary Data1.xlsx). The Geriatric Behavioral Assessment (GBA) study involved 118 animals with a mean age of 17.3 years. The Geriatric Mobile Assessment (GMA) was a smaller study (n = 57) that included older animals (mean age 19.1 years). A major difference between the two studies was the site of behavioral testing—the GBA group was transferred from their housing environment to a behavioral suite for testing, while the GMA group was tested in their home cages.

Table 1.

Age (years), sex (% female), and housing conditions (indoor, outdoor) of rhesus macaques used for the GBA and GMA studies

Study	Housing	Age:			%female	N
		Mean	Std. Dev.	Range
GBA (Tested in sound-proof suite)	Indoor	17.8 y	2.3 y		78.0%	59
	Outdoor	16.8 y	2.2 y		62.7%	59
	(All)	17.3 y	2.3 y	15-22 y	70.3%	118
GMA (Tested in home cage)	Indoor	22.7 y a	2.5 y	20–31 y	91.2%	57
GMA/GBA	Indoor / Outdoor	20.2 y 16.8 y a	3.4 y 2.2 y	15–31 y	84.5% 62.7%	116 59
Total		19.1 y	3.5 y	--	77.1%	175

^ap < 0.001 compared to GBA total (Mann-Whitney U test).

^bp < 0.001 compared to indoor housing (Mann-Whitney U test).

GMA animals were pair-housed indoors and were assessed between 10 am and 5 pm in the animal’s home cage. GBA animals were assesse during a 25 h relocation to sound-proofed testing suites in cohorts of 2. Long-term housing was split with some outdoors in large social groups and half housed indoors in pair housing.

Improved reaction time with increasing choices

During testing, animals were offered food items with increasing choice numbers (2-, 3-, or 4-choices; 4 trials each) and complexity (the number of preferred vs. non-preferred food items to choose from). The study assessed reaction time (RT) and variability (std. dev.) in RT (RTSD) following food choice presentation during the three choice tests.

We first assessed how increasing the number of choices (#choices) affects the decision-making process. Repeated measures ANOVA showed highly significant effects of #choices on RT and RTSD (p < 0.001). Mean reaction time declined stepwise across the tests (2_RT = 9.6 ± 9.2 s; 3_RT = 8.2 ± 9.2 s; 4_RT = 4.4 ± 5.4 s; p < 0.001 for all paired comparisons) (Fig. 1A). Variability also decreased (6.3 ± 5.3 s; 4.9 ± 5.5 s; 4.0 ± 4.6 s, p < 0.001 for 2_RTSD vs. 3_RTSD or 4_RTSD; p < 0.05 for 3_RTSD vs. 4_RTSD) (Fig. 1B).

Fig. 1. Reduced latency and variability of reaction time with increasing number of choices.

Animals were allowed 30 seconds to make a choice during each trial. Animals proceeded to the next trial regardless of their prior choices. A Latency to make decisions (reaction time, RT) during the 2-, 3-, or 4-choice trials. ***p < 0.001 between the columns indicated using pairwise comparisons. B Variability in decision making time (standard deviation in RT, RTSD) during the 2-, 3-, or 4-choice trials. ***p < 0.001, *p < 0.05 between the columns indicated using pairwise comparisons. C The rate of improvement in RT during the 2-, 3-, and 4-choice trials is positively correlated with mean latency during the 2-choice trials (mean2_RT). Rapid adaptation is indicated by a lower negative value.

To test whether faster decision times affected accuracy, we examined violations—choices in the 4_RT not predicted by earlier preferences. Spearman’s ρ indicated modest inverse correlations between violations and latency (2_RT, ρ = −0.329; 3_RT, ρ = -0.342; 4_RT, ρ = −0.319; all p < 0.001) and between violations and variability (2_RTSD, ρ = −0.177, 3_RTSD, ρ = −0.336; 4_RTSD, ρ = −0.295; all p < 0.001). Thus, quicker reaction times may lead to compromised accuracy. Violations also showed a weak positive association with the rate of improvement in RT (ρ = 0.155, p = 0.05), but not with RTSD (ρ = –0.044). Mean 2_RT latency strongly correlated with 3_RT (ρ = 0.838, p < 0.001) and 4_RT (ρ = 0.658, p < 0.001), with regression slopes decreasing as choice number increased. As predicted based on these observations, mean 2_RT also correlated with the overall rate of decline across sessions (ρ = −0.796, p < 0.001; Fig. 1C). Overall, these findings suggest animals made faster decisions as they became familiar with the task, and that this relationship is particularly evident in animals showing poor reaction time in the initial trials.

Familiarity with testing environment is a major determinant of latency and variability

The GBA includes both indoor- and outdoor-housed animals, whereas the GMA involved only indoor-housed animals tested in their home cages (Table 1). Preliminary repeated-measures ANCOVA, controlling for age, confirmed significant effects of housing (p < 0.001) and study (GBA vs. GMA, p < 0.05), but not sex. Results from the two studies were therefore analyzed separately (Fig. 2). In the GBA, housing was included as an independent variable, and sex had no effect, so between-sex data are not reported. For latency, a two-way repeated-measures ANCOVA (adjusting for age) indicated a significant interaction between choice number and housing (p < 0.01). As anticipated, housing significantly affected latency (p < 0.01): animals housed outdoors had longer mean latencies choice (Fig. 2A) and a delayed decline in latency with increasing choice number (Fig. 2B). Indoor-housed animals showed a stepwise reduction across tests, while animals housed outdoors exhibited improvement mainly in the 4_RT. A similar trend was observed for variability (Fig. 2C, D), although housing and choice effects did not reach statistical significance in the two-way ANCOVA.

Fig. 2. Animals in the GBA study compared to GMA study, and animals housed outdoors compared to indoors, take longer to decide during the 2- to 4-choice trials.

Data shown are estimated marginal means and standard errors adjusted for age at the time of testing. Statistical comparisons used a repeated measures ANCOVA controlling for age at testing. For the GBA study, housing and sex were independent variables. There was no significant effect of sex, and only housing differences are reported. A Average latency during the 2-, 3-, and 4-choice trials for animals in the GBA study housed outdoors or indoors. **p < 0.01. B Comparison of latency between 2-, 3-, and 4-choice trials for animals in the GBA study. Statistical significance determined using pairwise comparisons: a p = 0.01 vs. 2_RT; b p < 0.001 vs. 2_RT, 3_RT; c p < 0.005 vs. 3_RT, 4_RT; d p < 0.01 vs. 4_RT; e p < 0.001 vs. 2_RT, 3_RT. Differences in latency within choice# and between housing are indicated by lines. *p < 0.05, *** p < 0.001. C Average variability of latency during the 2-, 3-, and 4-choice trials for animals in the GBA study housed outdoors or indoors. *** p < 0.01. D Comparison of the variability of latency between 2-, 3-, and 4-choice trials for animals in the GBA study. Mean values are significantly different between all choice#, p < 0.005 (Related-Samples Friedman’s two-way ANOVA followed by pairwise comparisons). E Mean latency during the 2-, 3-, and 4-choice trials for animals in the GMA study. f p < 0.001 vs. 3_RT, 4_RT. F Variability of latency during the 2-, 3-, and 4-choice trials for animals in the GMA study. g p < 0.005 vs. 3_RT, 4_RT.

In the GMA, both latency (reaction time) and variability (reaction time variability) were markedly lower compared to the GBA ( < 3 s compared to 7–15 s). Differences at all choice levels were highly significant (Mann-Whitney U-test, p < 0.001). Within the GMA, choice number again affected both reaction time and variability in reaction time (Fig. 2E, F). A nonparametric test (Friedman’s Two-Way ANOVA by Ranks) confirmed significant differences among choice numbers for both decision time and decision time variability (p < 0.001).

Low plasma adropin concentrations indicate higher risk of cognitive impairment

We next investigated the relationships between plasma adropin concentrations and decision-making success employing multivariate linear regression using two different approaches.

First, we compared decision time and decision time variability across adropin tertiles within each study. In the GBA, repeated-measures ANCOVA (including adropin tertile, sex, and housing) found no main effect of tertile on decision time. When sex was removed as a variable, a significant interaction emerged between choice number and adropin tertile (p < 0.05): animals in the high-adropin tertile showed a clearer stepwise reduction in RT, whereas the low-adropin group improved only modestly (Fig. 3A). A similar interaction was seen for RTSD (p < 0.05) (Fig. 3B). Violations also differed amongst adropin tertiles, although the relationship does not appear to be linear (age-adjusted estimated marginal means ± SE: low = 0.577 ± 0.034, mid = 0.629 ± 0.033, high = 0.497 ± 0.038; p < 0.05 for mid vs. high in pairwise comparison). No significant differences were observed in the GMA (data not shown).

Fig. 3. High plasma Adropin Concentrations Associate with Faster and More Consistent Response Times in Aging Rhesus Macaques.

A and variability of latency (B) are shown between adropin tertiles in the GBA. Data are estimated marginal means and standard errors adjusted for age at the time of testing. Statistical comparisons used a repeated measures ANCOVA controlling for age at testing. Adropin tertile and sex were independent variables. There was no significant effect of sex, and only differences between tertile are reported. Statistical significance determined using pairwise comparisons: a p < 0.005 vs. 2_RT; 3_RT; b p < 0.05 vs. 2_RT/2_RTSD; c p < 0.001 vs. 2_RT, 3_RT; d p < 0.001 between all groups; e p < 0.001 vs. 2_RTSD; f p < 0.05 vs. 2_RTSD.

We then determined whether plasma adropin concentrations correlate with behavioral indices using linear regression. The regression controlled for age, sex, study (GBA, GMA). During the 4-choice test, plasma adropin concentration significantly correlated with both RT (Table 2) and RTSD (Table 3). For RTSD, there was also a significant relationship during the 2-choice test (Table 2). Plasma adropin concentrations correlated with the rate of improvement in reaction time variability (RTSD) (p < 0.001) (Table 3), with a trend for predicting the rate of improvement in reaction time (RT) (p = 0.056) (Table 2). This is consistent with the relationship between plasma adropin concentrations and reaction time showing no relationship with intrinsic performance. Rather, the data suggest the relationship is dependent on habituation to the testing conditions

Table 2.

Linear regression modeling using plasma adropin concentrations to predict changes in reaction time (RT)

Dependent variable:	LOG10_2RT		LOG10_3RT		LOG10_4RT		Slope
Model Summary	R = 0.621 F4,168 = 26.302 p < 0.001		R = 0.557 F4,162 = 18.223 p < 0.001		R = 0.443 F4,164 = 9.748, p < 0.001		R = 0.431 F4,162 = 9.238 p < 0.001
	β	p-value	β	p-value	β	p-value	β	p-value
Log10_Adropin	0.038	NS	−0.105	NS	−0.205	< 0.01	−0.141	(0.056)
Age at testing	0.032	NS	0.014	NS	0.143	NS	0.085	NS
Study (GBA = 0, GMA = 1)	−0.637	< 0.001	−0.518	< 0.01	−0.426	< 0.001	0.357	< 0.001
Sex (F = 1, M = 1)	−0.041	NS	−0.073	NS	−0.102	NS	0.379	NS

The regressions control for age at testing, sex, study, and housing condition. Values shown in bold highlight the relationships between plasma adropin concentrations and cognitive performance.

Table 3.

Linear regression modeling using plasma adropin concentrations to predict behavioral optimization (variability in reaction time, RTSD)

Dependent variable:	LOG10_2RTSD		LOG10_3RTSD		LOG10_4RTSD		Slope
Model Summary	R = 0.617 F4,168 = 25.883 p < 0.001		R = 0.463 F4,163 = 11.095 p < 0.001		R = 0.511 F4,163 = 14.373 p < 0.001		R = 0.443 F4,160 = 9.788 p < 0.001
	β	p-value	β	p-value	β	p-value	β	p-value
Log10_Adropin	0.212	< 0.001	0.093	NS	−0.145	< 0.05	−0.361	< 0.001
Age at testing	0.262	< 0.005	0.020	NS	0.024	NS	−0.204	(0.055)
Study (GBA = 0, GMA = 1)	−0.788	< 0.001	−0.496	< 0.001	−0.468	< 0.001	0.451	< 0.001
Sex (F = 1, M = 1)	0.043	NS	0.058	0.425	-0.065	NS	−0.109	NS

The regressions control for age at testing, sex, study, and housing condition. Values shown in bold highlight the relationships between plasma adropin concentrations and cognitive performance.

We then switched the analysis to determine whether plasma adropin concentrations can be estimated using behavioral measures, aging controlling for age, study, and housing conditions. Here modeling was successful when all data from all animals was pooled (R = 0.460, p < 0.001). The rate of decline in RTSD was a significant (p < 0.001) negative predictor of plasma adropin concentrations (Table 4), suggesting that higher plasma adropin concentrations associate with a more rapid decline. Interestingly, when studies were assessed separately the relationship was only observed in the GBA study, and that removing the GMA improved the correlation (R = 0.544 vs. R = 0.460).

Table 4.

Linear regression modeling using measures of behavioral optimization (rate of the decline in reaction time (RT) and reaction time variability (RTSD)) in the 2-, 3- and 4-choice tests) to predict plasma adropin concentrations

Predicted variable: Log[adropin]	All R = 0.460, F5,159 = 8.516, p < 0.001		GBA R = 0.544, F5,103 = 8.651, p < 0.001		GMA R = 0.235, F4,51 = 0.745, NS
	β	P-value	β	P-value	β	P-value
Slope_RT	−0.149	(0.056)	−0.176	< 0.05	−0.006	NS
Slope_RTSD	−0.358	< 0.001	−0.372	< 0.001	−0.065	NS
Age at testing	0.041	NS	0.139	NS	−0.232	NS
Sex (1 = F, 0 = M)	−0.207	0.005	−0.296	< 0.001	−0.007	NS
Study (GBA = 0, GMA = 1)	0.323	0.005	--	--	--	--
Housing (1 = indoor, 2 = outdoor)	--	--	−0.106	NS	--	--

The regression controls for age at testing, sex, and housing condition. The plasma adropin data used for the analysis were log transformed. In the GBA, differences in housing (indoor, outdoor) were controlled for in the analysis. In the GMA, all animals are housed indoor. Values shown in bold highlight the relationship between cognitive performance and plasma adropin concentration.

No relationship with depressive or motivational behaviors

We next assessed behaviors that could affect performance during the choice test protocol used in the GBA. Depressive behavior (huddling) observed over 2 days in a subset of the GBA animals was not significantly different between adropin tertile (% time spent in observed behavior over 10 min adjusted for sex, housing, and age for the adropin tertiles: low = 5.2 ± 1.9%; mid = 5.5 ± 2.0%; high = 6.5 ± 2.4%; n = 41/ 40 / 26). Latency to interaction and duration for a known toy (Kong) were similar between adropin tertiles (latency in sec. adjusted for sex, housing, and age for the adropin tertiles: low = 43 ± 14 s; mid = 43 ± 14 s; high = 53 ± 16 s; duration of interaction: low = 33 ± 9 s; mid = 36 ± 9 s; high = 52 ± 10 s; n = 42/42/32). Interaction with a puzzle ball was also not significantly different between adropin tertiles (latency in sec. adjusted for sex, housing, and age for the adropin tertiles: low = 14 ± 7 s; mid = 27 ± 7 s; high = 20 ± 8 s; duration of interaction: low = 103 ± 10 s; mid = 97 ± 10 s; high = 93 ± 12 s; n = 42/42/32). An assessment of scores of emotionality over 2 days also indicated no significant differences between low, mid, and high adropin tertiles (−0.108 ± 0.126, −0.049 ± 0.127, and −0.048 ± 0.158). The differences observed in performance in the multi-choice test thus appear to be independent of emotional status or engagement with novelty.

Discussion

This study identifies a strong relationship between circulating adropin concentrations and decision-making performance in aged rhesus macaques. Animals with higher plasma adropin exhibited faster improvement in reaction time and reduced variability across repeated testing, whereas those with lower concentrations demonstrated slower adaptation and greater inconsistency. These findings extend prior work linking adropin to cognitive function in humans^7,8. The results are consistent with adropin serving as a biomarker of resilience in the context of aging-associated cognitive decline.

Our previous studies reported relationships between circulating adropin levels and risk of cognitive decline in late middle-aged to older community-dwelling humans^7,8. In one of these studies, cognitive performance data was recorded over a 5-year period with plasma adropin concentrations measured early in the study predicting risk of decline⁸. The selection criteria for those study participants made it unclear whether the relationships extended to the general population. The clear relationship observed in the present study is therefore important for two reasons. First, the association between circulating adropin and learning ability appears to be a general phenomenon rather than being restricted to at-risk individuals. Second, it suggests the relationship is conserved across primate species. These findings have important implications for further evaluating circulating adropin as a biomarker of cognitive aging and potentially for designing interventions aimed at enhancing adropin signaling in individuals identified as being at risk based on circulating adropin levels.

However, it is important to acknowledge that this study employs a cross-sectional design. Further studies are needed to determine whether plasma adropin concentrations predict cognitive stability or vulnerability to change over time. It is also not known whether plasma adropin concentrations remain stable over time or shows changes with aging that reflect changes in cognitive status.

In the current study, decision-making performance refers to the efficiency and consistency of behavioral choices across increasing task complexity quantified as decision latency, intra-individual variability, and accuracy. These measures parallel constructs widely used in cognitive and experimental psychology, including choice reaction time and intra-individual response variability, which serve as sensitive indices of executive function and cognitive control in aging²⁸. Framing the results within this context highlights the translational relevance of the aging rhesus macaque model for studying mechanisms that underlie age-related declines in cognitive flexibility and processing efficiency.

An important observation is that differences in performance emerged only in the more demanding 4-choice tests, while baseline decision latency in the 2-choice task was unaffected by adropin status. This pattern indicates that intrinsic decision-making ability—likely dependent on intact prefrontal cortical processing—remains preserved in animals with low adropin. However, deficits become evident as task complexity and memory load increase. The association between adropin and improved performance over repeated trials suggests a link to learning and adaptive processes mediated by the hippocampus.

The environmental context also strongly influenced behavioral outcomes. Animals tested in their home cages (GMA study) performed substantially better than those transferred to a behavioral suite (GBA study). The need to adapt to handling, relocation, and novel surroundings likely imposes additional cognitive load and stress, diverting attentional resources from decision processing. Notably, the relationship between adropin and learning was evident only in the GBA cohort, supporting the view that adropin reflects the ability to maintain cognitive performance under environmental stress or distraction rather than intrinsic learning capacity. The distinction between home-cage and relocated testing environments has practical implications for experimental design, emphasizing the impact of situational stressors on task performance in aged primates.

Classical models such as Hick’s law predict that reaction time increases with the number of response alternatives, however exceptions have been noted²⁹. Notably, large numbers of trials can affect the outcome. This study included repeated trials within each choice condition. Within-session practice and familiarization could therefore influence performance. The apparent improvement in reaction time and its variability with increasing choice number may reflect conditioning across repeated testing.

These findings also carry translational relevance for preclinical testing of candidate drugs aimed at improving cognition during aging. The narrower range of response variability observed in home-cage testing (GMA) suggests that this approach may be less sensitive for detecting treatment effects. In contrast, testing paradigms that include mild environmental challenges, such as relocation (GBA), may better expose interindividual variability in resilience and learning, thereby enhancing detection of therapeutic benefits. Interestingly, this effect was observed even though animals in the GBA study were, on average, five years younger than those in the GMA cohort, underscoring the influence of environmental context on behavioral outcomes.

The association between low plasma adropin and impaired learning performance may indicate elevated risk of cognitive aging or accelerated biological aging more generally. Our previous work in macaques demonstrated that low adropin correlates with increased adiposity and metabolic dysregulation in midlife animals (10–15 years old)²³. The current findings suggest that these metabolic differences extend into later life as vulnerability in cognitive domains. Further studies are needed to determine whether the relationship is specific to cognition or reflects broader age-related frailty. Importantly, diabetic animals were excluded from the present cohort, suggesting that the observed effects are not secondary to overt metabolic disease.

The underlying pathology driving this association remains unclear. However, results from rodent models provides some context. Expression of adropin (ENHO) declines with age in the brain of mice and rats, coinciding with increased markers of oxidative stress and endothelial dysfunction^21,30. Experimental preservation of adropin expression—either through transgenic overexpression or peptide supplementation—maintains spatial learning and memory and mitigates vascular and mitochondrial impairments in aged animals¹⁷. These findings suggest that adropin is not merely a biomarker but may exert protective effects on neural and vascular function. Understanding how declining adropin signaling contributes to age-related cognitive decline could therefore inform therapeutic strategies for promoting brain health in aging populations.

At the systems level, the hepatic ENHO/adropin module is tightly connected to skeletal-muscle gene networks regulating mitochondrial metabolism and protein synthesis²⁴. Across tissues, adropin expression correlates with pathways controlling glycolysis, fatty-acid oxidation, circadian rhythms, stress responses, and insulin action—processes critical for energy homeostasis and resilience to metabolic stress. These observations align with evidence from both mouse and primate studies linking adropin to circadian and metabolic regulation, peripheral and cerebral vascular function, substrate selection, and oxidative capacity^{9–13,15,18–20,23,31–37}. Such coordination across organ systems supports the view that adropin could act as a systemic integrator of metabolic and cognitive health.

With respect to how circadian time affects plasma adropin concentration, it is important to note that the time of blood sampling could be a confounding variable. For the GBA study samples were collected at the same time, however for the GMA study blood sampling occurred between 9 AM and 5 PM. This is unlikely to affect the primary conclusions drawn from the current study, as most of the variability in RT and RTSD was observed in the GBA study (Fig. 2).

Whether circulating, liver-derived adropin influences brain function directly or indirectly remains an open question. Potential indirect mechanisms include modulation of glucose or cholesterol metabolism, both of which affect neuronal energetics and membrane integrity. However, the brain itself expresses high levels of ENHO/adropin, with single-cell RNA-seq data showing enrichment in astrocytes^7,17,23. In cultured human astrocytes, adropin expression co-varies with aging-sensitive genes involved in oxidative metabolism and protein synthesis⁷. Bulk RNA-seq data similarly link adropin expression to mitochondrial pathways and to genes associated with cholesterol transport and dementia risk, including APOE, SLC2A4RG, APOJ, and CYP46A1^17,24. Although causality remains to be established, experimental evidence from mice indicates that administration of exogenous adropin reverses mild age-related cognitive decline and improves recovery after cerebrovascular injury. Collectively, these results support a model in which adropin signaling contributes to neural resilience by maintaining metabolic flexibility and cellular homeostasis in the aging brain.

Methods

Aged rhesus macaques

The behavioral and morphometric data reported here were collected at the California National Primate Research Center (CNPRC) in Davis, California. The CNPRC is a nonhuman primate research facility associated with the University of California, Davis. All animal procedures complied with the standards outlined in the National Institutes of Health guidelines for the care and use of animals and were approved by the Institutional Animal Care and Use Committee at UC Davis.

The goals of the two studies were similar but differed in subject ages and handling of the animals during testing. During Geriatric Biobehavioral Assessment (GBA) with relocation, animals aged 15 y and 19–20 y underwent non-invasive observations and tasks to identify variation in domains of cognition, emotion, social and motor function during a 25 h relocation to a dedicated sound-proofed testing suite. Animals are housed on a 12:12 light cycle. Animals are tested in cohorts of 2. Observers are blinded to animals’ age. The long-term housing the animals originate from before testing was split, with half housed outdoors in large social groups and half housed indoors in pair housing.

During Geriatric Mobile Assessment (GMA), animals aged 20–33 years were assessed in the animal’s indoor living cage. All GMA animals were pair-housed indoors. GMA assessments are conducted between 10 am and 5 pm in the animal’s home cage, and by the same observer. Assessments take 2 h total. At the end of testing, animals remained in their home cages.

The GBA and GMA studies used a battery of tests to assess social, locomotor, and cognitive abilities. Tests were strategically ordered to optimize performance. Unless otherwise stated, GMA tests are conducted in the same fashion as GBA tests. The current study tested the hypothesis that plasma adropin concentrations would correlate with decision-making and reaction time, so only the decision-making test was examined. For the present study, we selected a decision-making task and non-mnemonic observations related to emotional state for comparison.

Decision making and reaction time

When making decisions, reaction time increases and predictability of choices decrease with geriatric aging^38,39. The current study tested reaction time and predictability of choices in animals during three choice tests. During these three tests, animals are offered food items with increasing choice numbers and complexity. Food items were presented on a white platform tray raised to eye level. Twelve trials are presented for each test, with 0-4 preferred food items (i.e., high sugar or salt content like fruits, popcorn) presented beside 0-3 non-preferred food items (low sugar content – vegetables, chow). After the reaction time to make a choice (latency in seconds) is recorded, further choices are not allowed. Preference trees are generated from selections during 2- and 3- choice tests. Choice predictability is tested during the 4-choice test and “violations” are scored when animals make unpredicted selections during each of twelve trials. Reaction times for simple and complex decisions, interindividual variability in reaction time with interference, and the proportion of decision-making violations are calculated. Animals were allowed 30 seconds to make a choice, and then the trial ended. Animals proceeded to the next trial regardless of their prior choices.

Depression

At the beginning of assessment, depressive behavior (hunched posture, disengagement from environment) is recorded live from wireless video connection from outside the animal room^40–42. Depressive behavior is most likely to occur when animals have been undisturbed by humans, so this assessment is conducted first with minimal exposure to the human experimenters leading up to it. Outcome measures reflect durations and frequencies of depressive behavior during a five-minute trial: previous studies at CNPRC show that adults housed indoors exhibit a median of 57 s in the hunched posture per five-minute trial⁴¹. Depressive behavior frequencies (number of bouts of the hunched posture) and durations (number of seconds during two five-minute trials on Day 1 and again on Day 2) are used as outcome measures.

Novelty response

During GBA, animals were presented with four novel objects (Kong toy, hotel bell, rubber snake, puzzle ball with ~25 0.5 × 2.0 strips of soluble paper). During GMA, animals were presented with two novel objects (Kong toy, and puzzle ball with ~25 strips of 0.5 × 2.0 in strips of soluble paper). In all cases, latency to interact with the object, retrieve a preferred food item in front of object, and durations of interaction are recorded⁴³.

Plasma adropin measurements

Plasma adropin concentrations were measured in triplicate samples using an enzyme immunoassay kit from Phoenix Pharmaceuticals, Inc. (cat. no. EK-032–35) following the manufacturer’s protocol. For the GBA, Samples were collected at 10:45 AM on Day 1 of testing. For GMA samples, plasma was collected during veterinary health examinations, which are conducted between 9 AM and 5 PM. GMA samples were collected from 212 days prior to cognitive testing and up to 152 days after testing. Average timing was 55 days prior to testing. Controlling for the interval between blood sampling and behavioral testing did not affect the outcome of the analysis.

Statistical analysis

Data analysis used IBM SPSS Statistics version 29 .0.0.0. The effect of increasing number of choices was assessed using ANOVA with repeated measures (increasing #choices), followed by the sequential addition of independent variables to test specific hypotheses (sex, housing, study, adropin-tertile). Further analysis controlled for variables known that could affect cognitive performance (age, long-term housing outside of testing (ie, indoor pair housing or outdoor large social groups), site of testing). Non-continuous variables were coded when used as covariates (sex, M = 0, F = 1; housing code, 1 = indoors, 2 = outdoors; study). For ANCOVA, noncontinuous (binary) covariates were controlled for by inclusion as independent variables. Age at the time of testing was used as a covariate. Pairwise comparisons using Bonferroni correction was used to determine significance when comparing 3 data sets.

For non-mnemonic assessments over 2 days, repeated measures ANCOVA was used to test for differences between adropin tertile. Non-continuous variables were coded and used as covariates (sex, M = 0, F = 1; housing code, 1 = indoors, 2 = outdoors; study). Age at the time of testing was used as a covariate.

The relationship between plasma adropin concentration and decision-making capability was assessed using multivariate linear regression as described in the results section. Subsequent analysis grouped animals into tertiles ranked by plasma adropin values. Plasma adropin data from both studies (GBA, GMA) was used to rank animals by plasma adropin concentrations within sex. Animals in the low adropin-tertile had values < 33^rd percentile; animals in the high adropin-tertile had values > 67^th percentile. Log-transformed data was used for the statistical analysis.

Author contributions

E.L.K. and J.H.M. have oversight of the original study. A.A.B., E.L.K., and P.J.H. conceived this project. E.L.K. designed assessments and supervised experiments and data analysis. L.C.C., A.M.C., A.D.C., R.S., and G.B.D. assisted with animal assessments. A.P.B. and N.D. processed data for analysis. J.L.G. measured plasma adropin concentrations. A.A.B. analyzed the data, prepared figures and tables, and prepared the first draft of manuscript. E.L.K. and P.J.H. edited the manuscript. All authors reviewed the manuscript.

Data Availability

Source data are provided as a supplementary data file .

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s41514-026-00426-1.