Species-specific allostatic load indices fail to identify predictors or health outcomes of higher allostatic load in zoo-housed chimpanzees (Pan troglodytes) and bonobos (Pan paniscus)

Introduction: While the potential for using allostatic load to better understand animal wellbeing is well recognized in the zoo and conservation community, existing applications of allostatic load in wildlife have not been as promising as hoped given the robust results seen in human research.

Methods: Through a retrospective study of zoo-housed chimpanzees and bonobos, we 1) developed a species-specific allostatic load index (ALI), 2) analyzed whether allostatic load scores were associated with multiple predictor variables (e.g., age, sex, origin) or predicted health outcomes, and 3) compared ALIs with another multi-biomarker method, Mahalanobis distance (D_M).

Results: Only one predictor variable showed a significant association with allostatic load, with male chimpanzees having lower allostatic load than females. In contrast, older age predicted D_M in chimpanzees, male bonobos had significantly lower D_M than females, and zoo-born individuals of both species had significantly lower D_M than wild-caught conspecifics. Age and sex alone better predicted all-cause morbidity and cardiac disease compared to models containing ALIs, and models for mortality risk were not substantially improved by the inclusion of an ALI. ALIs were compared to D_M using Akaike’s information criterion (AIC) for chimpanzees and AIC adjusted for small sample sizes (AICc) for bonobos. In both species, D_M models had the lowest AIC/AICc for all health outcomes compared to the ALI models, indicating improved performance of D_M over allostatic load for predicting health outcomes.

Discussion: Overall, ALIs poorly measure wellbeing and health outcomes in chimpanzees and bonobos; therefore, other multi-biomarker methods like D_M may be more useful in nonhuman primates and other taxa.

Introduction

The potential for using allostatic load to better understand animal wellbeing is well recognized in the zoo and conservation community, with mentions in more than 500 articles and attempted applications in dozens of species (Seeley et al., 2022). Briefly, allostatic load is the physiological dysregulation that occurs due to experiencing repeated acute and/or chronic stressors over the lifespan (McEwen and Stellar, 1993). Nearly three decades ago, allostatic load was first operationalized in humans with the development of an allostatic load index (ALI; Seeman et al., 1997). ALIs combine biomarkers from multiple somatic systems (e.g., neuroendocrine, immune, cardiovascular) to calculate a single score that estimates the amount of allostatic load an individual has accumulated relative to others in the same sample, with higher scores indicating greater physiological dysregulation.

ALIs may be well suited as physiological indicators of wellbeing in animals, particularly because they reflect long-term functioning rather than acute stressors. Higher allostatic load in humans can result from environmental and psychosocial stress, and is associated with early life adversity (Horan and Widom, 2015; Thayer et al., 2017), low socioeconomic position (Johnson et al., 2017; Guidi et al., 2020), and a higher number of stressful events (Glei et al., 2007; Siew et al., 2023). Thus, applying ALIs in animals could provide a direct measurement of the impacts of early life conditions (e.g., how animals are reared), social dynamics (e.g., position within the dominance hierarchy), and stressors over the lifespan (e.g., immobilizations for veterinary exams, transfers). Additionally, higher allostatic load in humans has been associated with a number of morbidities also seen in animals in human care, including cardiac disease, hypertension, diabetes, arthritis, cancer, periodontal disease, parasitic infection, and frailty, as well as mortality risk (for review, see Juster et al., 2010; Beckie, 2012; Guidi et al., 2020; Parker et al., 2022). As such, allostatic load may help identify individuals at greatest risk of poor health outcomes, which could allow for preemptive care and earlier diagnosis (Edes et al., 2018a; Wolfe and Edes, 2023), a goal in human research as well (Carbone et al., 2022).

The biomarker suites used for ALIs are typically composed of a combination of primary and secondary mediators. Primary mediators include acute response biomarkers like stress-related hormones and inflammatory responses (e.g., glucocorticoids, catecholamines, cytokines), while secondary mediators reflect cumulative changes over the long-term and are more closely related to disease risk (e.g., lipids, blood pressure, insulin resistance; McCaffery et al., 2012). Dozens of different biomarkers have been used in human ALIs (Juster et al., 2010; Carbone et al., 2022; Beese and Graves, 2022), but there is no “gold standard” agreement as to which are best to include (Johnson et al., 2017; Carbone et al., 2022; McCrory et al., 2023; Duong et al., 2017). While there have been attempts in recent years to determine the best scoring methods once a biomarker suite has been selected (for review, see Carbone et al., 2022), there are minimal efforts to come to a consensus on which biomarkers should be included, although most agree that ALIs should include both primary and secondary mediators (Johnson et al., 2017). Unfortunately, while this flexibility indicates the robustness of allostatic load as a concept and suggests many of the biomarkers routinely measured by zoos could be useful for measuring allostatic load, the lack of an agreed upon ALI can present a considerable roadblock for applications in wildlife. For most wildlife species, far fewer biomarkers have been validated or are routinely measured, which complicates decisions on exactly which biomarkers to measure, and appropriate tissue sample types (e.g., serum) are rare and, when they do exist, often low in volume. Despite these considerable barriers, ALIs have been applied in a handful of species with mixed results, including rats (McCreary et al., 2019), giraffe (Beer et al., 2023), polar bears (Teman et al., 2025), ring-tailed lemurs (Seeley et al., 2021), western lowland gorillas (e.g., Edes et al., 2018b, 2023), and chimpanzees and bonobos (Edes et al., 2023).

Through a retrospective study of zoo-housed chimpanzees and bonobos, the objectives of this study were three-fold. First, we developed a species-specific ALI by identifying a subset of biomarkers from a larger dataset using a feature-selection approach, building upon previous attempts (Edes et al., 2021). Second, we analyzed whether those allostatic load scores were predicted by variables commonly associated with allostatic load in humans (e.g., age, sex, conditions during early life) or were associated with all-cause morbidity, cardiac disease, and mortality risk in chimpanzees and bonobos. We predicted that species-specific ALIs would result in better predictions of future health risk. Finally, we compared ALIs with another multi-biomarker method, Mahalanobis distance (D_M). Similar to allostatic load, D_M also estimates physiological dysregulation using a suite of biomarkers by calculating a score per individual relative to a reference population (Cohen et al., 2013, 2014). Overall, our goal was to understand whether multi-biomarker methods, and species-specific ALIs in particular, are viable options for evaluating animal wellbeing in chimpanzees and bonobos.

Methods

Subjects

We obtained serum samples collected during previous immobilizations from chimpanzees (238 samples from 162 unique individuals) and bonobos (44 samples from 38 unique individuals) housed at 22 North American zoological institutions. Consistent with our previous work (Edes et al., 2023, 2023), samples collected from individuals younger than 7 years old were excluded from the dataset to reduce variation in biomarker concentrations due to growth and development. Following this, for any individuals with multiple samples remaining, we used the following criteria to rank samples for inclusion: 1) samples from routine immobilizations were selected over those collected during non-routine immobilizations (e.g., diagnosis or treatment of acute illness or injury); 2) if multiple samples were from routine immobilizations, the one with the fewest missing biomarkers was selected, and, finally, 3) if the same number of biomarkers were available for each sample, the one collected at the animal’s oldest age was selected (e.g., a sample collected at 17 years old would be selected over a sample collected at 15 years old). This resulted in a dataset with 148 chimpanzees (aged 7–79 years at sample collection, $\overline{x}$= 26.98, SD = 12.75, 60 males) and 33 bonobos (aged 7–48 years at sample collection, $\overline{x}$= 22.15, SD = 11.50, 18 males). All participating zoos were accredited by the Association of Zoos and Aquariums and approved the study. This research adheres to the American Society of Primatologists’ Principles for the Ethical Treatment of Primates.

Predictors and outcomes of higher allostatic load

We used a combination of husbandry and medical records provided by each institution as well as ZIMS (Species360 Zoological Information Management System (ZIMS), 2021) to retrospectively collate data on potential predictors and outcomes of higher allostatic load for each individual. Potential predictors of higher allostatic load for all animals included age, sex, the number of stressful events prior to the sample collection date, parity (i.e., nulliparous vs. parous), origin (i.e., wild-caught or zoo-born), and rearing for zoo-born individuals (i.e., mother- vs. nursery-reared). Parity was only analyzed for chimpanzees, as all bonobo females were parous. When analyzing differences in rearing type, we excluded six peer-reared chimpanzees due to small sample size. Stressful events included immobilizations, transfers between holding institutions, new wounds or injuries, and pregnancies for females. Whenever possible, the number of stressful events was counted from the date of birth up to the date of sample collection. We note that retrospective data are incomplete for some individuals that were transferred from their birth institution and/or are older, as records are not always transferred alongside animals. Outcomes of higher allostatic load were chronic all-cause morbidity (e.g., osteoarthritis, neoplasia, obesity), cardiac disease, and all-cause mortality. Cardiac disease was analyzed as a subset of all-cause morbidity due to the high prevalence in zoo-housed great apes (Lowenstine et al., 2016; Murphy et al., 2018; Strong et al., 2018). All-cause morbidity, cardiac disease, and all-cause mortality were coded as dichotomous variables (disease: 0 = no diagnosed condition, 1 = diagnosed condition; mortality: 0 = alive, 1 = deceased). Husbandry and medical records were not provided for seven bonobos, so these individuals were excluded from analyses of all variables except those obtainable from ZIMS, including age, sex, number of transfers, and mortality risk.

Biomarker assays

As our first objective was to identify a subset of biomarkers from a broader array for a species-specific ALI, we assayed 17 different biomarkers from the serum samples: albumin, α-1-acid glycoprotein (AGP), bilirubin, cortisol, creatinine, dehydroepiandrosterone-sulfate (DHEA-S), fructosamine, glucose, high-density lipoprotein (HDL), immunoglobulin A (IgA), insulin, interleukin-6 (IL-6), low-density lipoprotein (LDL), serum amyloid A (SAA), total cholesterol (TC), triglycerides (TG), and tumor necrosis factor-α (TNF-α). Serum samples were collected between 1984 and 2019, and cryopreserved at -80 °C until analysis in 2020-2023. All assays were conducted by A.N.E. or by an endocrinology technician within J.L.B.’s laboratory.

We used solid-phase enzyme immunoassays (EIAs) to measure cortisol, DHEA-S, IgA, insulin, TNF-α, and SAA. For EIAs, all samples were analyzed in duplicate with coefficients of variation (CVs) maintained below 10% and inter-assay CVs maintained below 15% for high and low concentration controls. Cortisol was measured using the Smithsonian National Zoo and Conservation Biology Institute’s in-house double antibody EIA, as previously described (Edes et al., 2023), and IgA was measured using the Saint Louis Zoo’s in-house EIA, also as previously described (Edes et al., 2025). These EIAs were validated biochemically for measuring cortisol and IgA in chimpanzee and bonobo serum through parallelism and matrix interference (or spike-and-recovery) assessments, and subsequent regression analyses to assess linearity of displacement curves and whether the observed-background concentrations were predicted by the expected concentration, respectively. We considered a biomarker validated when it demonstrated dilutional linearity with an R² ≥ 0.90 and recovery between 80-120%. For cortisol, serial 2-fold dilutions of serum yielded a displacement curve parallel to the standard curve (chimpanzee: y = 1.019x – 2.114, R² = 0.998, F_1,4 = 1753.968, p< 0.001; bonobo: y = 0.986x + 1.449, R² = 0.996, F_1,6 = 1566.448, p< 0.001). There was no evidence of matrix interference for cortisol, as addition of appropriately diluted serum (chimpanzee 1:150; bonobo 1:100) to assay standards did not alter the amount observed (chimpanzee: y = 1.183x – 13.972, R² = 0.979, F_1,7 = 321.043, p< 0.001; bonobo: y = 1.060x – 11.286, R² = 0.992, F_1,7 = 849.689, p< 0.001). For IgA, serial 2-fold dilutions of serum yielded a displacement curve parallel to the standard curve (chimpanzee: y = 1.109x + 0.003, R² = 0.984, F_1,6 = 367.737, p< 0.001; bonobo: y = 1.127x + 0.079, R² = 0.993, F_1,6 = 869.620, p< 0.001). Unfortunately, we observed evidence of matrix interference for IgA, as the addition of appropriately diluted serum (chimpanzee: 1:5,000,000; bonobo: 1:10,000,000) to assay standards altered the amount observed, averaging 162% recovery for chimpanzees and 208% for bonobos. Although this overestimation was somewhat linear (chimpanzee: y = 1.093 + 1.218, R² = 0.998, F_1,6 = 3458.210, p< 0.001; bonobo: y = 1.150 + 1.384, R² = 0.999, F_1,6 = 7821.877, p > 0.001), inconsistent results from samples analyzed repeatedly suggests this matrix interference prevented accurate IgA quantification and the decision was made to exclude IgA from consideration for the ALI.

We analyzed DHEA-S, insulin, IL-6, TNF-α, and SAA using commercially available EIAs and performed the assays according to manufacturer instructions. Assay sensitivity for DHEA-S (K054, Arbor Assays, Michigan, USA) ranged from 96-60,000pg/ml. Serial 5-fold dilutions of serum yielded a displacement curve parallel to the standard curve (chimpanzee: y = 0.975x + 8.855, R² = 0.994, F_1,3 = 525.601, p< 0.001; bonobo: y = 0.985x + 11.015, R² = 0.997, F_1,3 = 983.034, p< 0.001). There was no evidence of matrix interference, as addition of appropriately diluted serum (chimpanzee 1:250; bonobo 1:200) to assay standards did not alter the amount observed (chimpanzee: y = 0.915x + 653.511, R² = 0.995, F_1,3 = 566.412, P< 0.001; bonobo: y = 0.860x – 402.019, R² = 0.999, F_1,3 = 3812.135, P< 0.001). The ultrasensitive EIA for insulin (10-1132-01, Mercodia, North Carolina, USA) had a range of 0.15-20.0mU/l. Serum samples were mostly undiluted for analysis on the insulin EIA, with some diluted 1:2 in assay diluent where necessary. There was no evidence of matrix interference, as addition of undiluted serum to assay standards did not alter the amount observed (chimpanzee: y = 0.962x – 0.182, R² = 0.995, F_1,3 = 597.498, p< 0.001; bonobo: y = 0.968x – 0.108, R² = 0.998, F_1,3 = 1668.326, p< 0.001). The assay range for IL-6 and TNF-α (IL-6: HS600C, TNF-α: HSTA00E, R and D Systems, Minnesota, USA) was 0.20-10.0pg/ml. Both IL-6 and TNF-α were biochemically validated using linearity (IL-6: chimpanzee 96.9%, bonobo 107.9%; TNF-α: chimpanzee: 102.7%; bonobo: 112.9%) and spike and recovery assessment (IL-6: chimpanzee: 99.6%; bonobo: 105.7%; TNF-α: chimpanzee: 85.6%; bonobo: 89.8%) within the range of dilutions used (IL-6: undiluted to 1:30; TNF-α: undiluted to 1:5). The SAA EIA (KHA0011, Invitrogen, Massachusetts, USA) had a range of 9.4-600.0ng/ml. While we were able to successfully biochemically validate the SAA assay for bonobos, we were unable to do so for chimpanzees (Edes et al., 2023). As such, SAA was excluded from further consideration for the ALI.

Glucose, fructosamine, AGP, albumin, TC, TG, HDL, LDL, bilirubin, and creatinine were quantified on a RX Daytona automated clinical chemistry analyzer (Randox Laboratories-US, Ltd., Kearneysville, WV, USA). Commercially available reagents, calibrators, and two-level controls were purchased from Randox Laboratories-US, Ltd. (Kearneysville, WV, USA) for each biomarker (Edes et al., 2023). Serum was run undiluted. The analyzer was subject to routine quality control measurements throughout the study, with normal and elevated controls for each analyte maintained within two standard deviations of the respective target value. Most bilirubin readings fell below the assay detection limit (83.1% of chimpanzees, 63.6% of bonobos). Due to limited reagent quantity available, AGP was able to be assayed on nearly all of the bonobo dataset but only a third of the chimpanzee samples, and of the samples assayed many fell below the assay range (46.9% of chimpanzees assayed, 83.9% of bonobos assayed). Given the limited data available for bilirubin and AGP, we excluded these two biomarkers from further consideration for the ALI.

Determining a species-specific allostatic load index

Different combinations of biomarkers may best predict individual health outcomes. To identify this suite of biomarkers in chimpanzees, we used elastic net regression we used elastic net regression (Zou and Hastie 2005). Elastic net balances the advantages of LASSO and ridge regression such that less informative features are penalized and co-linear data can be more easily modeled. This approach requires complete data for all predictor biomarkers; thus, we chose to exclude three biomarkers (fructosamine, HDL, and LDL) due to a high frequency of missing data. We also stratified biomarker values by sampling modality (i.e., routine vs. non-routine) to assess differences in values by modality. Overall, distributions for non-routine and unknown samples largely fell within the range of routine samples (Figure 1); thus, we chose to include all samples to estimate the penalty parameter and model health outcomes.

Figure 1

Figure 1. Biomarker distributions in chimpanzees by reason for immobilization. DHEA-S, dehydroepiandrosterone-sulfate; IL-6, interleukin-6; TNF-α, tumor necrosis factor-α.

We split the chimpanzee dataset into 70:30 training and testing sets for all three health outcomes: all-cause morbidity, cardiac disease, and all-cause mortality. We estimated lambda in the training data using the CV function of the “glmnet” package (Friedman et al., 2010; Tay et al., 2023), where outcomes were modeled as a function of the biomarkers with a binomial error structure and α = 0.5. We used five data folds and area under the curve (AUC) for measuring loss given the binary measurement for all three health outcomes. We ran 100 iterations per outcome and calculated the median lambda: 0.113 for all-cause morbidity, 0.207 for cardiac disease, and 0.233 for all-cause mortality risk. We used these values to model outcomes in the testing data using the glmnet function. Small sample size prevented the use of elastic net regression to determine a species-specific ALI for bonobos. Given that these species are sister taxa, we therefore applied the chimpanzee ALI to our bonobo dataset.

We investigated ALIs made with three different biomarker suites. Our initial models predicted cardiac disease (median AUC = 0.619) and all-cause mortality (median AUC = 0.588) modestly well, but the prediction of all-cause morbidity (median AUC = 0.508) was nearly equivalent to random chance. Results varied slightly when the elastic net regression was performed using the dataset with all samples versus the dataset with samples collected during routine immobilizations only using lambdas estimated on the full training dataset (Table 1). We combined the six biomarkers identified as potentially important predictors of health outcomes when using elastic net regression with either the routine-only or all samples dataset into one ALI (ALI-6). ALI-6 included albumin, cortisol, creatinine, glucose, IL-6, and TC. A 10-biomarker suite was used for calculating D_M (see sub-section on Mahalanobis distance), so we also constructed an ALI-10 with the same biomarkers. In addition to the biomarkers from ALI-6, ALI-10 also included DHEA-S, insulin, TNF-α, and TG. Finally, as ALIs may be more robust when more biomarkers are included (Liu et al., 2021), we also created an ALI using the full 13-biomarker dataset (ALI-13), including HDL, LDL, and fructosamine in addition to the other 10.

Table 1

Table 1. Elastic net regression predictors and betas for health outcomes in chimpanzees when using all samples and routine-only samples.

Calculating allostatic load scores

Allostatic load scores were calculated for each biomarker suite (ALI-6, ALI-10, and ALI-13) by determining a high-risk threshold, or cut-point, for each biomarker (Table 2) and then counting the number of biomarkers beyond that threshold for each individual. We calculated high-risk thresholds using the original quartile methodology (Seeman et al., 1997), with values in the fourth quartile or first quartile defined as high-risk; the quartile designated as high-risk for the six biomarkers identified by elastic net regression depended on whether the regression results indicated a positive (fourth quartile; albumin, cortisol, creatinine, IL-6) or negative (first quartile; glucose, TC) relationship with the health outcomes (Table 2). Following standard practices in human allostatic load research, the fourth quartile was defined as high-risk for all additional biomarkers in ALI-10 and ALI-13 except DHEA-S and HDL, for which the first quartile was defined as high-risk. Allostatic load scores could range from 0, indicating no biomarkers were within a high-risk quartile, up to the total number of biomarkers included in the ALI, indicating all biomarkers were within a high-risk quartile. For individuals with missing values for some biomarkers, those biomarkers were not counted toward the allostatic load score (i.e., if an animal was missing two of six biomarkers, its maximum allostatic load score would be 4).

Table 2

Table 2. Biomarker cut-points by ALI for calculating allostatic load in chimpanzees and bonobos.

Calculating Mahalanobis distance

Mahalanobis distance (D_M) is a multivariate distance metric calculated by measuring the distance between a point and the mean of a reference distribution. This approach is particularly useful for biomarker data because the correlations between biomarkers are controlled, such that biomarker combinations with high variances contribute less to the distance. Thus, higher scores reflect greater physiological dysregulation as compared to a reference population (Cohen et al., 2013, 2014). Previous research on multiple primate species has demonstrated that closely related species can serve as reference populations for one another (Dansereau et al., 2019). Therefore, we elected to use the chimpanzee reference population for both the chimpanzee and bonobo datasets given the small size of the latter. Typically, healthy non-geriatric individuals without any missing values are selected for the reference population (Cohen et al., 2013, 2014). As there were missing values for fructosamine, HDL, and LDL for 56.1, 14.2, and 18.2% of the chimpanzee dataset, respectively, we removed these three biomarkers from the dataset to retain the largest reference population possible. The final list of biomarkers used to calculate D_M included cortisol, DHEA-S, insulin, IL-6, TNF-α, TG, TC, albumin, glucose, and creatinine. A previous study on sanctuary-housed chimpanzees selected non-overweight individuals aged 15–20 years for their reference population (Cole et al., 2024). Similarly, we selected chimpanzees aged 15–25 years with samples from routine immobilizations that had not been diagnosed with any chronic conditions for the reference population (n = 23). Biomarkers with a non-normal distribution in the reference population were log transformed for the entire dataset, and then biomarker values for all individuals were z-scored to the reference sample (Dansereau et al., 2019; Cole et al., 2024). D_M scores were then calculated separately for chimpanzees and bonobos using the MDmiss function in the “modi” package (Hulliger et al., 2023) in R (v.4.3.2, R Core Team, 2024), which accounts for missing observations. The output of the MDmiss function is squared, so we took the square root to obtain the final D_M scores. We analyzed both raw and log transformed D_M scores (logD_M) for consistency with previous research (e.g., Cohen et al., 2013, 2014; Cole et al., 2024).

Quantitative analyses

Analyses of scores estimated using the ALIs and D_M are outlined in Table 3. As allostatic load scores were calculated as a count variable, generalized linear models (GLMs) with a Poisson distribution and a log-link function were used to analyze whether allostatic load in chimpanzees or bonobos was predicted by age, sex, stressful events, parity, origin, or rearing history. Binomial GLMs with logit links were used to assess if allostatic load predicted all-cause morbidity and cardiac disease. Whether allostatic load predicted all-cause mortality risk was analyzed using Cox proportional hazards models based on time from sample collection to date of death (censored animals were still alive as of December 5, 2024). For all three health outcomes, we first assessed the relationships with age and sex and, if either or both were significant predictors of a specific outcome, they were included in the models alongside allostatic load. Model fit was quantified using both the proportion of variance explained using Nagelkerke’s pseudo-R² and the relative model fit using Akaike’s information criterion (AIC) for chimpanzees and AIC corrected for small sample size (AICc) for bonobos (Burnham and Anderson, 2002; Symonds and Moussalli, 2011); models with Δ AIC/AICc ≤ 2 were considered equivalent to the best fit model (Harrison et al., 2018; Richards et al., 2011; Arnold, 2010; Symonds and Moussalli, 2011). Analyses were repeated for each ALI (ALI-6, ALI-10, and ALI-13) and were conducted separately by species. Allostatic load scores did not vary based on whether the samples were initially collected during routine or non-routine immobilizations (GLM; results not shown), so the sampling modality was not included as a covariate in analyses. We analyzed data with and without zoo included as a random effect; models performed better and results did not vary based on its inclusion, so results are presented without zoo included as a random effect. D_M and logD_M were analyzed using the same model structure as the ALIs (Table 3), using linear models instead of GLMs to assess possible predictors of D_M and logD_M. As with the ALI analyses, reason for immobilization was not included as a covariate and zoo was not included as a random effect because model results did not vary and were not improved by their inclusion. All analyses described throughout this section were run in R (v.4.3.2, R Core Team, 2024) using Colab notebooks and R Studio (v.2024.04.2 + 764). GLMs were analyzed using the “lme4” package (Bates et al., 2024), R² was estimated using the “performance” package (Lüdecke et al., 2025), all-cause mortality risk was estimated using the “survival” package (Therneau et al., 2022), and data were visualized using the "ggplot2" package (Wickham et al., 2018).

Table 3

Table 3. Structure of analyses for allostatic load (AL) and Mahalanobis distance (D_M) scores.

Data and code availability

Data are proprietary to each cooperating institution and available upon reasonable request from the first author with approval from the cooperating institution. A non-archived version of all code used to analyze the data is available on GitHub (https://github.com/brandcm/Pan_biomarkers).

Results

Allostatic load

For ALI-6, allostatic load scores ranged from 0-4 ($\overline{x}$ = 1.48, SD = 0.98) for chimpanzees and 0-5 ($\overline{x}$ = 1.58, SD = 1.25) for bonobos. For ALI-10, allostatic load scores ranged from 0–7 for both chimpanzees ($\overline{x}$ = 2.44, SD = 1.45) and bonobos ($\overline{x}$ = 2.61, SD = 1.56). For ALI-13, allostatic load scores ranged from 0-8 ($\overline{x}$ = 2.98, SD = 1.72) for chimpanzees and 1-8 ($\overline{x}$ = 3.42, SD = 1.71) for bonobos. The distribution of each ALI by species is presented in Figure 2.

Figure 2

Figure 2. Distribution of allostatic load scores by allostatic load index for chimpanzees (top) and bonobos (bottom).

Model estimates, p-values, 95% confident intervals, and Nagelkerke’s pseudo-R² for analyses of predictor variables for each ALI are presented by species in Table 4. Sex, but not age, significantly predicted allostatic load in chimpanzees, with males having significantly lower allostatic load than females when using ALI-10 and ALI-13 (Figure 3). Allostatic load was not predicted by age or sex in bonobos using any ALI. Total stressful events experienced ranged from 1–156 per chimpanzee ($\overline{x}$ = 34.0, SD = 32.6) and 1-87 ($\overline{x}$ = 16.3, SD = 19.4) per bonobo; each chimpanzee experienced 0–5 transfers ($\overline{x}$ = 1.2, SD = 1.1), 0–147 wounding events ($\overline{x}$ = 26.3, SD = 30.1), and 1–42 immobilizations ($\overline{x}$ = 8.9, SD = 8.7), and each bonobo experienced 0–3 transfers ($\overline{x}$ = 1.3, SD = 1.0), 1–45 wounding events ($\overline{x}$ = 11.5, SD = 12.2), and 1–44 immobilizations ($\overline{x}$ = 8.5, SD = 10.2). There were 10 nulliparous and 32 parous (1–9 births each) female chimpanzees (there were no pregnancy data for 46 females), and seven parous bonobo females (2–6 births each; there were no pregnancy data for eight females). Whether analyzed in total or by type, stressful events did not predict allostatic load using any ALI for chimpanzees or bonobos. In chimpanzees, parity did not significantly predict allostatic load; this variable could not be analyzed in bonobos due to no known nulliparous females. There were 30 wild-caught, 103 zoo-born, and 15 unknown origin individuals in the chimpanzee dataset, and 8 wild-caught, 22 zoo-born, and 3 unknown origin individuals in the bonobo dataset. There were no significant differences by origin (i.e., wild-caught vs. zoo-born) for either chimpanzees or bonobos. Rearing history did not significantly predict allostatic load in chimpanzees (mother-reared: n = 70; nursery-reared: n = 27) or bonobos (mother-reared: n = 19; nursery-reared: n = 3).

Table 4

Table 4. Associations of predictor variables with allostatic load in zoo-housed chimpanzees and bonobos; bold values are significant at p ≤ 0.05.

Figure 3

Figure 3. Significant sex differences in allostatic load scores were observed in chimpanzees when using ALI-10 and ALI-13.

In the chimpanzee dataset, 82 (55.4%) individuals had at least one known morbidity, 50 (33.8%) had cardiac disease, and 57 (38.5%) were deceased on December 5, 2024. For chimpanzees, age but not sex was associated with all-cause morbidity and cardiac disease, so age was retained in models with allostatic load for predicting these health outcomes (Table 5). Allostatic load scores did not significantly predict all-cause morbidity or cardiac disease for chimpanzees. All models for all-cause morbidity and cardiac disease had ΔAIC ≤ 2, suggesting the models were equivalent to one another, and the pseudo-R² values suggest models with ALIs explain the same amount of variation in each health outcome as the model with just age and sex. Both age and sex significantly predicted all-cause mortality risk in chimpanzees, explaining approximately 20% of the variation in risk, and so were retained in models alongside allostatic load for predicting this health outcome (Table 5). Although allostatic load was not a significant predictor within any models, the model containing ALI-6 had the best fit for predicting all-cause mortality risk in chimpanzees; when age and sex were included in the model, each 1-point increase in allostatic load was associated with a 26% increase in all-cause mortality risk for chimpanzees when using ALI-6. However, this model was within ΔAIC ≤ 2 of the model containing only sex and age, suggesting it was equivalent to the simpler model.

Table 5

Table 5. Predictions of all-cause morbidity, cardiac disease, and all-cause mortality risk in zoo-housed chimpanzees and bonobos using allostatic load and age and/or sex; the best-fit model, as determined by the lowest AIC (chimpanzees) or AICc (bonobos) score, is in bold.

In the bonobo dataset, 6 (18.2%) individuals had at least one known morbidity, 5 (15.2%) had cardiac disease, and 9 (27.3%) were deceased on December 5, 2024. For bonobos, age but not sex predicted all-cause morbidity, cardiac disease, and all-cause mortality risk, with models explaining between one-quarter and one-half of the variation in these outcomes (Table 5); age was retained in models alongside allostatic load for all three health outcomes. Irrespective of the biomarker suite used, allostatic load scores did not significantly predict all-cause morbidity, cardiac disease, or all-cause mortality risk for bonobos. Models containing allostatic load scores from ALI-13 were the best fit models for predicting all-cause morbidity and cardiac disease for bonobos, but they were within ΔAICc ≤ 2, suggesting those models were equivalent with the model containing only sex and age. The best fit model for all-cause mortality risk in bonobos was the model with only age and sex.

Mahalanobis distance (D_M)

D_M ranged from 2.0-32.3 ($\overline{x}$ = 5.4, SD = 4.1) for chimpanzees and 2.6-7.2 ($\overline{x}$ = 4.2, SD = 1.0) for bonobos, and the distribution was positively skewed for both species (Figure 4). Model estimates, p-values, 95% confident intervals, and adjusted R² for analyses of predictor variables with logD_M are presented by species in Table 6. In chimpanzees, logD_M was significantly predicted by age (Figure 5), with age explaining approximately 8% of the variation in logD_M, but did not differ by sex. Age did not significantly predict logD_M in bonobos, but males had significantly lower logD_M than females (Figure 5), with sex explaining 12% of the variation in logD_M. The count of stressful events, by type and total, did not predict logDM in chimpanzees or bonobos. For both chimpanzees and bonobos, individuals born in zoos had significantly lower logD_M than wild-born conspecifics (Figure 5), with origin explaining 4% and 14% of variation in logD_M respectively, but there was no difference between mother- and nursery-reared zoo-born individuals.

Figure 4

Figure 4. Distribution of Mahalanobis distance (D_M) scores for chimpanzees and bonobos.

Table 6

Table 6. Associations of predictor variables with log Mahalanobis distance (D_M) in zoo-housed chimpanzees and bonobos; bold values are significant at p ≤ 0.05.

As with the allostatic load analyses, age was included for chimpanzees and bonobos in models predicting all three health outcomes, and sex was also included in mortality risk analyses for chimpanzees (Table 7). In chimpanzees, D_M significantly predicted all-cause morbidity, and the model that included D_M had the lowest AIC and explained more variation in the outcome. D_M did not predict all-cause morbidity in bonobos, with the model containing D_M having the same AICc as the model with age and sex only, although both models explained approximately half of the variation in this outcome as estimated using pseudo-R². Cardiac disease was not predicted by D_M for either chimpanzees or bonobos, and the models with just age and sex had the lowest AIC/AICc, explaining just 3% of the variation in cardiac disease for chimpanzees but 25% for bonobos. Finally, D_M significantly predicted all-cause mortality risk in chimpanzees, with a 15% increase in mortality risk with every 1-point increase in D_M, and the model with D_M had the best fit according to AIC and R². For bonobos, all-cause mortality risk was predicted by D_M, with every 1-point increase in D_M associated with a 190% increase in mortality risk, and the model containing D_M had the lowest AICc but a similar R². Results for all health outcomes were similar when using logD_M (not shown).

Table 7

Table 7. Predictions of all-cause morbidity, cardiac disease, and all-cause mortality risk in zoo-housed chimpanzees and bonobos using Mahalanobis distance (D_M) and age and/or sex; the best-fit model, as determined by the lowest AIC (chimpanzees) or AICc (bonobos) score, is in bold.

Figure 5

Figure 5. Associations of age and log transformed Mahalanobis distance (logD_M) in chimpanzees (top left), sex differences between raw D_M scores in bonobos (top right), and differences in raw D_M scores by origin in chimpanzees and bonobos (bottom).

Discussion

It has been posited that measuring allostatic load will be a useful tool in monitoring the wellbeing of animals in human care. To address this suggestion, we designed a retrospective study of zoo-housed chimpanzees and bonobos with three objectives. First, we wanted to test a statistical method for narrowing down a larger list of biomarkers into a species-specific ALI. Secondly, using the allostatic load scores from that species-specific ALI, we wanted to explore potential predictors and outcomes of higher allostatic load. Lastly, we compared allostatic load with another multi-biomarker method, D_M, to determine the better approach for measuring physiological dysregulation to predict health outcomes in chimpanzees and bonobos. Overall, our goal was to understand whether ALIs are useful for studying animal wellbeing from a physiological perspective.

Allostatic load results

We used elastic net regression to narrow down a large list of biomarkers into a 6-biomarker ALI specific to chimpanzees. This species-specific ALI was compared to an ALI made from a 10-biomarker suite to match our analyses of D_M as well as a 13-biomarker suite containing all possible biomarkers. The same ALIs were applied to our much smaller bonobo dataset.

Irrespective of how the ALIs were constructed, variables often related to physiological dysregulation mostly failed to predict allostatic load in chimpanzees and bonobos. Despite the natural increase in physiological dysregulation with age, allostatic load was not predicted by age in either chimpanzees or bonobos. A positive association between age and allostatic load has been reported in dozens of human studies (e.g., Crimmins et al., 2003; Mauss et al., 2016; Honkalampi et al., 2024; Robertson and Watts, 2016). These results are partially inconsistent with what we observed in our first ALI developed for these species, wherein we observed a positive relationship between age and allostatic load for chimpanzees but not bonobos (Edes et al., 2023). When using ALI-10 and ALI-13, male chimpanzees had significantly lower allostatic load than females; no sex differences in allostatic load were observed for bonobos. Sex differences have been reported in some human studies (e.g., lower allostatic load in men than women: Kusano et al., 2016; Lateef et al., 2020; higher allostatic load in men than women: Robertson and Watts, 2016; Kezios et al., 2022), but not others (e.g., Brooks et al., 2014; Bailey et al., 2025; Cave et al., 2020). Early studies in gorillas showed males had higher allostatic load than females (Edes et al., 2016b, 2016a), but our more recent analyses showed no sex difference in gorillas, chimpanzees, and bonobos (Edes et al., 2018b, 2023). In one study, allostatic load was found to be higher in men than women when scores were estimated using pooled cut-points but not sex-specific cut-points (Kezios et al., 2022). As we used pooled cut-points for this study but not our first ALI in chimpanzees and bonobos, that may help explain the sex differences observed herein for chimpanzees. Allostatic load was not associated with sex and age in ring-tailed lemurs (Seeley et al., 2021) or giraffe (Beer et al., 2023). In polar bears, adult females without cubs showed an inverse relationship between age and allostatic load, but these variables were not significantly associated in adult females with young cubs or in adult males, and no sex differences were observed (Teman et al., 2025).

Stressful events, whether considered by type or cumulatively, were not associated with allostatic load using any ALIs for chimpanzees or bonobos. With an ALI developed for gorillas, chimpanzees, and bonobos, we previously observed a positive association between cumulative stressful events in chimpanzees (Edes et al., 2023), and our earliest work in gorillas showed positive relationships with cumulative stressful events as well (Edes et al., 2016b, 2018a). Stressful events have been associated with allostatic load in humans (Glei et al., 2007; Siew et al., 2023) and giraffe (Beer et al., 2023), and cumulative stress burden predicted allostatic load in rats (McCreary et al., 2019), but similar relationships were not observed in ring-tailed lemurs (Seeley et al., 2021). Our failure to detect relationships between stressful events and allostatic load may be due, at least in part, to having incomplete historical records, as described in the Methods, resulting in too few stressors being counted for many individuals. In support of this possibility, stressful event records were complete across the lifespan for giraffe, where a positive association with allostatic load was observed (Beer et al., 2023), but not for ring-tailed lemurs, where a significant association was not observed (Seeley et al., 2021).

Despite the aversive experience of being captured at a young age and taken into human care, allostatic load did not vary by origin for either species, nor were there significant differences between mother- and nursery-reared zoo-born conspecifics. In humans, there is substantial evidence that conditions early in life can impact health outcomes later in life (Lacagnina, 2020). Accordingly, in humans, early life adversity is associated with higher allostatic load in adulthood (Danese and McEwen, 2012; Horan and Widom, 2015; Thayer et al., 2017). Although work in gorillas showed higher allostatic load in wild-caught individuals, an effect potentially mediated by sex differences (Edes et al., 2016b, 2016a, 2018b; Edes et al., 2023), similar results were not observed in chimpanzees or bonobos (Edes et al., 2023). Consistent with the results herein, differences between mother- and nursery-reared zoo-born conspecifics were not significant in our previous work in gorillas, chimpanzees, and bonobos (Edes et al., 2016a, 2018b; Edes et al., 2023). Differences in origin or rearing have not been analyzed in other allostatic load research in animals, although there is evidence that origin and rearing can have lasting impacts. For example, rearing and origin have been shown to impact behavior in both chimpanzees (Clay et al., 2023; Pascual et al., 2023) and bonobos (Laméris et al., 2020). Effects of rearing are not just behavioral, with altered cortisol (Sanchez, 2006; Barr et al., 2004), basal norepinephrine (Pryce et al., 2004), and blood pressure (Pryce et al., 2004) observed in alternatively reared primates. Orphaned African elephants (Loxodonta africana) have higher fecal glucocorticoids and fecal T3 but no differences in body condition scores (Chusyd et al., 2025), showing that early life adversity impacts physiology in other taxonomic groups as well, although these differences in early life may not translate to greater physiological dysregulation in adulthood.

Allostatic load scores from the three ALIs also largely failed to predict health outcomes in chimpanzees and bonobos. All-cause morbidity and cardiac disease were predicted by age for both chimpanzees and bonobos. Allostatic load was not significantly associated with all-cause morbidity or cardiac disease for either species; while ALIs were technically included in the top model set for each outcome, they were within Δ AIC/AICc ≤ 2 of the models with just age and sex, suggesting the inclusion of allostatic load scores did not substantially improve predictions. Allostatic load did not significantly predict all-cause mortality risk for bonobos, although a trend was observed for chimpanzees, with a 1-point increase associated with a 26% increase in mortality risk when using ALI-6, and the model with age, sex, and ALI-6 was the best-fit model of those tested. However, again, all models were contained within the top model set, suggesting the inclusion of allostatic load did not substantially improve predictions of mortality risk in chimpanzees. In humans, higher allostatic load is associated with a number of health outcomes, including conditions present in this dataset such as arthritis, periodontal disease, and cardiac disease (for review, see Edes and Crews 2017; Juster et al., 2010; Beckie, 2012; Guidi et al., 2020). Higher allostatic load also predicts mortality risk in humans (for review, see Parker et al., 2022). These results are consistent with our previous research showing weak relationships with health outcomes in great apes (Edes et al., 2018b, 2020, 2023; Edes et al., 2021). Higher allostatic load in rats was associated with reduced neuronal density in the prefrontal cortex, indicating increased risk of poor health outcomes (McCreary et al., 2019), but most demographic groups of polar bears showed no relationship of allostatic load with body condition, which could indicate health issues, or with onshore habitat use, which could result in increased pathogen exposure (Teman et al., 2025).

Why allostatic load fails as a measure of physiological dysregulation in great apes

When considered altogether, the weak and inconsistent results with both predictors and outcomes of higher allostatic load suggest these ALIs fail to adequately measure or represent multi-systemic physiological dysregulation in chimpanzees and bonobos. While it is possible that applying the chimpanzee-specific ALI to bonobos could be part of the issue, the fact that it also worked poorly in chimpanzees suggests the problem lies elsewhere. One possible reason for this failure is how we calculated allostatic load scores once the biomarker suites were determined. For example, results from the elastic net regression indicated glucose and TC had an inverse relationship with health outcomes in chimpanzees, while albumin had a positive relationship, meaning they were designated as high-risk at the first and fourth quartiles, respectively; this is opposite of how they are traditionally scored in human research and, for albumin and glucose, inconsistent with our previous analyses on individual biomarker associations with health outcomes in chimpanzees and bonobos (Edes et al., 2023). However, aside from this difference, we followed standard practices and our methods are comparable to those used in human studies. A recent study exploring 14 different methods for calculating allostatic load found that using sex-specific cut-points marginally improved performance over pooled cut-points (McLoughlin et al., 2020). While we did not use sex-specific cut-points herein, we did in our first ALI applied to chimpanzees and bonobos and observed similar results (Edes et al., 2023), suggesting this is not the reason behind the ALIs not working as predicted. Furthermore, while some have suggested that using high-risk quartiles to determine cut-points, as was done for the first ALI (Seeman et al., 1997), is not the best way to represent physiological dysregulation, it remains the most popular method in human research and multiple studies comparing it to a variety of other calculation methods have shown it provides similar results (Li et al., 2019; Berger et al., 2018; McLoughlin et al., 2020; Kezios et al., 2022). Finally, it is possible that we did not have large enough biomarker suites, as ALIs calculated using the traditional high-risk quartile method are considered more stable when more biomarkers are included (Liu et al., 2021), but there are no rules for how many biomarkers should be included, “streamlined” ALIs containing as few as five have been used successfully in human research (Mauss et al., 2015, 2016; McCrory et al., 2023), and our ALIs ranged in size from 6–13 biomarkers with minimal differences in results.

Given that our methods for calculating allostatic load scores from the ALIs are comparable to those used in humans, it is currently unclear whether there are factors limiting our ability to apply ALIs to this dataset specifically or if the problem lies with the application of ALIs in great apes more broadly. It has been suggested that focusing biomarker selection on those that have the strongest associations with health outcomes of interest, as we did with the elastic net regression, can result in an ALI that is not representative of all major systems (Li and Sanon Rosemberg, 2020). As discussed above, the elastic net regression results did indicate some relationships between biomarkers and health outcomes that were unexpected based on human research and our previous work in these species (e.g., albumin, glucose). However, all of our ALIs included both primary and secondary mediators of allostatic load (McCaffery et al., 2012; Johnson et al., 2017) and, even for ALI-6, our biomarker suites were representative of numerous physiological systems, including the neuroendocrine, metabolic (both lipid and glucose metabolism), and immune systems (Beese and Graves, 2022; Carbone et al., 2022; Juster et al., 2010; Li and Sanon Rosemberg, 2020; Liu et al., 2021). While we lack representation of some systems, such as cardiovascular and zoometric biomarkers, studies on humans using databases that lack neuroendocrine biomarkers, which are primary mediators thought to be intrinsic to allostatic load (Johnson et al., 2017), still show robust associations with health outcomes (e.g. Liu et al., 2021; Polick et al., 2024; Sabbah et al., 2008). We previously attempted to refine biomarker selection in gorillas using other methods. After our preliminary results in gorillas from one zoo (Edes et al., 2016b, 2016a) failed to replicate in a broader sample (Edes et al., 2018b), we attempted to refine our early ALIs using a combination of forward stepwise regression and independent associations with each outcome of interest (Edes et al., 2021). This attempt was similarly unsuccessful, and further exploratory analyses incorporating lipid markers showed limited improvements over our initial ALIs (Edes et al., 2020).

While our previous work to refine biomarker selection was done in gorillas, a comparison of the results from the first ALI in chimpanzees and bonobos (Edes et al., 2023) with the results from the species-specific ALIs generated herein shows that we have again failed to improve this measure. Given that the other multi-biomarker method outperformed allostatic load (see next section), allostatic load methods may not be suitable for wildlife. Regarding biomarker suitability, while we always encourage additional biomarker discovery, our biomarker suite represented numerous physiological systems and we have no reason to suspect that biomarkers commonly used in ALIs applied to humans would not work in our closest relatives. Instead, we suggest that the fact that all of our serum samples were collected from individuals under anesthesia may contribute to our issues in successfully measuring allostatic load in great apes. Anesthesia can impact circulating levels of many biomarkers. For example, ketamine, a common anesthetic agent, has been shown to suppress the release of albumin (Bennett et al., 1992; Kim et al., 2005), IL-6 (De Kock et al., 2013; Walker et al., 2015), and TNF-α (De Kock et al., 2013; Walker et al., 2015). Effects on other biomarkers included herein have mixed reports, with some studies suggesting no changes due to anesthesia while others suggest either increases or decreases, including in cortisol, insulin, glucose, HDL, TC, and TG (Bentson et al., 2003; Castro et al., 1981; Crockett et al., 2000; Kemnitz and Kraemer, 1982; Lehmann et al., 1997; Adaramoye et al., 2013; Perumal et al., 2007; Saranteas et al., 2005; Yoshida et al., 1986). Furthermore, some individuals were darted for immobilization while others voluntarily presented for injection, which has been shown to impact levels of multiple biomarkers related to acute stress in chimpanzees (Lambeth et al., 2006). While allostatic load is meant to be a long-term measure of physiological dysregulation, many of the biomarkers included in ALIs respond to acute stressors. The minimal differences we observed in individual biomarkers between samples collected for routine vs. non-routine purposes may be partially explained by the animals being immobilized; anesthesia could minimize or mask differences that we might otherwise expect based on reason for immobilization (e.g., higher levels of glucocorticoids following immobilization to treat wounding vs. hand injection for a routine veterinary exam). There also was no statistical difference between allostatic load scores calculated from samples obtained during routine vs. non-routine immobilizations (data not shown). Unfortunately, as not many great apes are trained for voluntary sample collection, use of anesthesia is a substantial limitation that we are unlikely to be able to address in the near future. In other allostatic load studies in animals, samples were collected voluntarily from only giraffe (Beer et al., 2023), with anesthesia used for wild polar bears (Teman et al., 2025) and manual restraint for ring-tailed lemurs (Seeley et al., 2021).

Comparing allostatic load and Mahalanobis distance

Whether certain variables could predict physiological dysregulation was also explored using D_M. We observed a positive association between D_M and age in chimpanzees, lower D_M in male bonobos compared to females, and higher D_M in wild-caught than zoo-born conspecifics in both species. These results are inconsistent with the allostatic load results observed herein, where the only significant association was between sex and ALI-10 and -13 in chimpanzees. As an increase in physiological dysregulation is expected with age (Crimmins et al., 2003; Cohen et al., 2013; Milot et al., 2014), it has been suggested that a biomarker suite for D_M can be validated by showing an increase in scores with age (Arbeev et al., 2016). Subsequently, these results suggest that D_M is more reflective of physiological dysregulation, at least in chimpanzees. Multiple studies in humans have shown a positive association between age and D_M (Cohen et al., 2013, 2014; Milot et al., 2014; Arbeev et al., 2016; Kraft et al., 2020). Similarly, Dansereau et al. (2019) reported higher D_M with age across nine of 10 primate species, including chimpanzees, and Cole et al. (2024) also reported a positive association between age and D_M in chimpanzees. In chimpanzees, sex differences in overall D_M were not observed in one study, although D_M increased more slowly over time in males compared to females (Dansereau et al., 2019), while another study found lower D_M in males but no difference in the rate of increase with age (Cole et al., 2024). Although we did not observe differences by sex for chimpanzees, lower D_M in males would be logical considering the social structure and dynamics of chimpanzee troops, wherein males are dominant over females (Gruber and Clay, 2016; Parish and De Waal, 2000), but following that same logic, it is unexpected that bonobo males would have lower D_M than females, as observed herein, when females are dominant over males (Gruber and Clay, 2016; Parish and De Waal, 2000). Differences in environment may contribute to variation in sex differences from what is expected, as the samples herein were from zoo-housed individuals, while the other studies in chimpanzees have been from research laboratories and sanctuaries. Similarly, our initial work in gorillas showed lower allostatic load in males, which we hypothesized may be a result of different environments; while males are dominant, in the wild they also must defend their group from other males and could be expected to have increased stress, but without those challenges present in zoos, they likely face fewer stressors, potentially resulting in lower allostatic load (Edes et al., 2016a).

In chimpanzees, D_M significantly predicted all-cause morbidity and all-cause mortality risk, and in bonobos, D_M significantly predicted all-cause mortality risk. Higher D_M also has been associated with chronic diseases and increased mortality risk in humans (Cohen et al., 2014, 2015; Milot et al., 2014; Li et al., 2015; Arbeev et al., 2016; Cohen et al., 2018). Our results showing a significant relationship between D_M and all-cause mortality risk in chimpanzees also are consistent with those observed by Dansereau et al. (2019). However, we observed no associations between D_M and cardiac disease, although one has been reported in humans (Milot et al., 2014). When comparing ALIs to D_M using AIC/AICc, ALI models for all health outcomes in both species had higher values than D_M models, indicating worse performance. Additionally, models containing ALIs were within Δ AIC/AICc ≤ 2 of the models containing sex and age only, suggesting the latter were equivalent to the models containing allostatic load scores and that including measures of allostatic load did not improve predictions. The amount of variation explained by D_M was also higher than that explained by allostatic load for all-cause morbidity and mortality risk in chimpanzees.

Given the similarities between allostatic load and D_M scores, it could be suggested that they are simply discrete vs. continuous methods of estimating physiological dysregulation. For example, Egorov et al. (2020) used D_M scoring methods to develop a continuous measure of allostatic load in humans. Although the methods have their similarities and can certainly be used to address similar questions, there are some key differences between how they are scored or biomarkers are considered that are important to keep in mind and likely play a role in the improved performance of D_M over allostatic load in chimpanzees and bonobos. First, we selected our biomarkers based on their association with the three health outcomes of interest as assessed via elastic net regression. This is a common approach when developing ALIs for humans, and it has been cautioned that doing this can result in non-representative biomarker suites (Li and Sanon Rosemberg, 2020). While this was not an issue with our ALIs, studies have shown that biomarker suites for calculating D_M do not need to be chosen based on a priori hypotheses about their roles in health outcomes (Cohen et al., 2015) and often that the biomarkers included in the suites poorly predict the outcomes of interest when analyzed individually (Arbeev et al., 2016; Kraft et al., 2020; Cohen et al., 2014, 2015). Indeed, research in humans has found that D_M is robust to the biomarker suite chosen, with consistent results observed even using very different biomarker combinations (Cohen et al., 2014, 2015). A similar result was observed in shorebirds, where the relationship between D_M and body condition score was not driven by individual biomarkers (Milot et al., 2014). This is not dissimilar from studies of allostatic load in humans, which have also shown that ALIs are better than individual biomarkers (e.g., Levine and Crimmins, 2014; Robertson et al., 2017; Castagné et al., 2018), and the same has been shown in giraffe (Beer et al., 2023). Notably, when using elastic net regression, the associations between individual biomarkers and health outcomes were few and weak in this dataset as well. This suggests multi-biomarker methods may still be superior for predicting risk of poor health outcomes in wildlife compared to individual biomarkers.

Once a biomarker suite is selected, how they are incorporated into a multi-biomarker method also likely impacts how well that method performs. Estimating allostatic load uses the distribution of each biomarker across the entire dataset to determine high-risk cut-points or thresholds, while D_M scores are based on similarity to a smaller subset of the data, referred to as the reference population, that is typically non-geriatric and healthy. These methodological differences may increase the likelihood for geriatric and less healthy individuals to have higher scores of physiological dysregulation when using D_M compared to ALIs. Furthermore, the information we can obtain from biomarkers tends to be “highly redundant, non-linear, and complex” (Arbeev et al., 2016) and, relatedly, dichotomizing continuous variables like biomarkers may dilute relationships (Johnson et al., 2019). Calculating allostatic load typically requires dichotomizing biomarkers by designating either the highest or lowest values as high-risk, whereas D_M (Cohen et al., 2013) allows for continuous, non-linear relationships between each biomarker and the score. Using D_M also removes the need for a researcher to decide which end of the distribution is high-risk; while the decision seems (and may be) fairly obvious for some biomarkers, other biomarkers, like cortisol (Badanes et al., 2011; Heim et al., 2000; Raison and Miller, 2003) and total cholesterol (Harris et al., 1992; Kronmal et al., 1993), have been shown to become dysregulated at both high and low levels. Although we have previously addressed this by using split two-tailed quartiles (top and bottom 12.5% of the distribution, rather than top or bottom 25%) for our ALIs (e.g., Edes et al., 2018b, 2023), that approach may not be sufficient. Additionally, ALIs do not take multicollinearity between variables into account, whereas D_M accounts for the correlations between biomarkers (De Maesschalck et al., 2000; Cohen et al., 2013; Li et al., 2015). Unsupervised dimensionality reduction methods, such as principal component analysis (PCA), may also prove helpful to characterize biomarker patterns while reducing multicollinearity. While ALIs were developed for and are largely used in geriatric human populations, where health differences may be more stark and likely to be captured by methods like ALIs, D_M can be used at younger ages (Cohen et al., 2014). As most research on wildlife is not limited to geriatric groups, a method that can measure physiological dysregulation at earlier ages is likely to be useful.

Conclusion

With limitations in using individual biomarkers for studying animal wellbeing, interest in allostatic load, a multi-biomarker method, has grown over the years. For reasons that are unclear, the flexibility with which allostatic load can be estimated in humans while still showing strong relationships with a wide variety of variables does not appear to extend to wildlife. While a handful of studies in wildlife have created multi-biomarker ALIs and calculated allostatic load scores, few have observed the associations that would be expected based on the wealth of research on humans. Our results suggest that D_M may be a better multi-biomarker alternative for studying animal wellbeing from a physiological perspective. Given that D_M is robust to varying biomarker suites and, like allostatic load, does not require clinical reference ranges to implement, which do not exist for many biomarkers across species, it may be especially well-suited to research in wildlife and retrospective analysis of existing datasets. Of course, more work is needed to determine if D_M truly estimates physiological dysregulation in wildlife, is sensitive to variation in predictors and conditions that impact physiology over the lifespan, and can predict risk of poor health outcomes in the future well enough to be a useful tool for measuring animal wellbeing.

Data availability statement

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation and with approval from the cooperating zoological institution(s).

Ethics statement

The animal studies were approved by each participating institution, which reviewed the research proposal and approved the inclusion of records from their individuals. The studies were conducted in accordance with the local legislation and institutional requirements. Written informed consent was obtained from the institutions for the participation of their animals in this study.

Author contributions

AE: Conceptualization, Data curation, Formal analysis, Funding acquisition, Methodology, Project administration, Resources, Validation, Visualization, Writing – original draft, Writing – review & editing. CB: Formal analysis, Methodology, Visualization, Writing – original draft, Writing – review & editing. JB: Funding acquisition, Project administration, Resources, Supervision, Writing – review & editing. KE: Data curation, Funding acquisition, Methodology, Project administration, Resources, Supervision, Validation, Writing – review & editing.

Funding

The author(s) declare financial support was received for the research and/or publication of this article. This research was supported by funding from an Association of Zoos and Aquariums Conservation Grants Fund award (#19-1578), a Columbus Zoo and Aquarium Conservation Fund award, a Sacramento Zoo Conservation Committee Small Grants award, Friends of the National Zoo, and the Endocrinology Research Laboratory at the Smithsonian Conservation Biology Institute. The automated clinical chemistry analyzer was supported by funds from the Smithsonian Women’s Committee.

Acknowledgments

This research would not have been possible without the assistance of veterinarians, keepers, and staff at the 22 participating zoological institutions: ABQ BioPark, Chattanooga Zoo, Cincinnati Zoo, Columbus Zoo and Aquarium, Dallas Zoo, Detroit Zoological Society, Jacksonville Zoo, Los Angeles Zoo, Lincoln Park Zoo, Lion Country Safari Park, Little Rock Zoo, Maryland Zoo, Memphis Zoo, Milwaukee County Zoo, North Carolina Zoo, Oklahoma City Zoo, Oregon Zoo, Sacramento Zoo, Saint Louis Zoo, Sedgwick County Zoo, Tulsa Zoo, and Zoo Knoxville. We thank Matthew Krcmarik and Steve Paris for their assistance in the laboratory. We also thank research assistants Drew Arbogast, Myra Brooks, Rachael Figueroa, Erica Greene, Stephanie Kaufman, Megan Lucas, and Daniel Neale for their help collating project data. Finally, we thank two reviewers for their helpful feedback in improving this manuscript.

Conflict of interest

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Generative AI statement

The author(s) declare that no Generative AI was used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

References

Adaramoye O. A., Akinwonmi O., and Akanni O. (2013). Effects of propofol, a sedative-hypnotic drug, on the lipid profile, antioxidant indices, and cardiovascular marker enzymes in wistar rats. ISRN Pharmacol. 2013, 230261. doi: 10.1155/2013/230261

PubMed Abstract | Crossref Full Text | Google Scholar

Arbeev K. G., Cohen A. A., Arbeeva L. S., Milot E., Stallard E., Kulminski A. M., et al. (2016). Optimal versus realized trajectories of physiological dysregulation in aging and their relation to sex-specific mortality risk. Front. Public Health 4, 3. doi: 10.3389/fpubh.2016.00003

PubMed Abstract | Crossref Full Text | Google Scholar

Arnold T. W. (2010). Uninformative parameters and model selection using Akaike’s information criterion. J. Wildlife Manage. 74, 1175–1178. doi: 10.2193/2009-367

Crossref Full Text | Google Scholar

Badanes L. S., Watamura S. E., and Hankin B. L. (2011). Hypocortisolism as a potential marker of allostatic load in children: associations with family risk and internalizing disorders. Dev. Psychopathol. 23, 881–896. doi: 10.1017/S095457941100037X

PubMed Abstract | Crossref Full Text | Google Scholar

Bailey A., Payton A., Fleming J., Rager J. E., and Jaspers I. (2025). A novel approach for measuring allostatic load highlights differences in stress burdens due to race, sex and smoking status. PloS One 20, e0323788. doi: 10.1371/journal.pone.0323788

PubMed Abstract | Crossref Full Text | Google Scholar

Barr C. S., Newman T. K., Shannon C., Parker C., Dvoskin R. L., Becker M. L., et al. (2004). Rearing condition and rh5-HTTLPR interact to influence limbic-hypothalamic-pituitary-adrenal axis response to stress in infant macaques. Biol. Psychiatry 55, 733–738. doi: 10.1016/j.biopsych.2003.12.008

PubMed Abstract | Crossref Full Text | Google Scholar

Bates D., Maechler M., Bolker B., Walker S., Bojesen Christensen R. H., Singmann H., et al. (2024). Lme4: linear mixed-effects models using “Eigen” and S4 (R Package Version). Available online at: https://cran.r-project.org/package=lme4 (Accessed November 2, 2025).

Google Scholar

Beckie T. M. (2012). A systematic review of allostatic load, health, and health disparities. Biol. Res. Nurs. 14, 311–346. doi: 10.1177/1099800412455688

PubMed Abstract | Crossref Full Text | Google Scholar

Beer H. N., Karr L. K., Shrader T. C., and Yates D. T. (2023). Allostatic load index effectively measures chronic stress status in zoo-housed giraffes. J. Zoological Botanical Gardens 4, 623–636. doi: 10.3390/jzbg4030044

Crossref Full Text | Google Scholar

Beese S. and Graves J. M. (2022). Allostatic load measurement: A systematic review of reviews, database inventory, and considerations for neighborhood research. Int. J. Environ. Res. Public Health 19, 17006. doi: 10.3390/ijerph192417006

PubMed Abstract | Crossref Full Text | Google Scholar

Bennett J. S., Gossett K. A., McCarthy M. P., and Simpson E. D. (1992). Effects of ketamine hydrochloride on serum biochemical and hematologic variables in rhesus monkeys (Macaca mulatta). Veterinary Clin. Pathol. / Am. Soc. Veterinary Clin. Pathol. 21, 15–18. doi: 10.1111/j.1939-165X.1992.tb00576.x

PubMed Abstract | Crossref Full Text | Google Scholar

Bentson K. L., Capitanio J. P., and Mendoza S. P. (2003). Cortisol responses to immobilization with telazol or ketamine in baboons (Papio cynocephalus/anubis) and rhesus macaques (Macaca mulatta). J. Medial Primatology 32, 148–160. doi: 10.1034/j.1600-0684.2003.00018.x

PubMed Abstract | Crossref Full Text | Google Scholar

Berger M., Juster R.-P., Westphal S., Amminger G.P., Bogerts B., Schiltz K., et al. (2018). Allostatic load is associated with psychotic symptoms and decreases with antipsychotic treatment in patients with schizophrenia and first-episode psychosis. Psychoneuroendocrinology 90, 35–42. doi: 10.1016/j.psyneuen.2018.02.001

PubMed Abstract | Crossref Full Text | Google Scholar

Brooks K. P., Gruenewald T., Karlamangla A., Hu P., Koretz B., and Seeman T. E. (2014). Social relationships and allostatic load in the MIDUS study. Health Psychol. 33, 1373–1381. doi: 10.1037/a0034528

PubMed Abstract | Crossref Full Text | Google Scholar

Burnham K. P. and Anderson D. R. (Eds.) (2002). Model Selection and Multimodel Inference: A Practical Information-Theoretic Approach (New York, NY: Springer).

Google Scholar

Carbone J. T., Clift J., and Alexander N. (2022). Measuring allostatic load: approaches and limitations to algorithm creation. J. Psychosomatic Res. 163, 111050. doi: 10.1016/j.jpsychores.2022.111050

PubMed Abstract | Crossref Full Text | Google Scholar

Castagné R., Garès V., Karimi M., Chadeau-Hyam M., Vineis P., Delpierre C., et al. (2018). Allostatic load and subsequent all-cause mortality: which biological markers drive the relationship? Findings from a UK birth cohort. Eur. J. Epidemiol. 33, 441–458. doi: 10.1007/s10654-018-0364-1

PubMed Abstract | Crossref Full Text | Google Scholar

Castro M. I., Rose J., Green W., Lehner N., Peterson D., and Taub D. (1981). Ketamine-HCI as a suitable anesthetic for endocrine, metabolic, and cardiovascular studies in Macaca fascicularis monkeys. Proc. Soc. Exp. Biol. Med. Soc. Exp. Biol. Med. 168, 389–394. doi: 10.3181/00379727-168-41292

PubMed Abstract | Crossref Full Text | Google Scholar

Cave L., Cooper M. N., Zubrick S. R., and Shepherd C. C.J. (2020). Racial discrimination and allostatic load among first nations Australians: A nationally representative cross-sectional study. BMC Public Health 20, 1881. doi: 10.1186/s12889-020-09978-7

PubMed Abstract | Crossref Full Text | Google Scholar

Chusyd D. E., Brown J. L., Paris S., Boisseau N., Mwaanga W., Kasongo M., et al. (2025). Fecal glucocorticoid metabolite and T3 profiles of orphaned elephants differ from non-orphaned elephants in Zambia. PeerJ 13, e19122. doi: 10.7717/peerj.19122

PubMed Abstract | Crossref Full Text | Google Scholar

Clay A. W., Ross S. R., Lambeth S., Vazquez M., Breaux S., Pietsch R., et al. (2023). Chimpanzees (Pan troglodytes) in U.S. Zoos, sanctuaries, and research facilities: A survey-based comparison of species-typical behaviors. Animals 13, 251. doi: 10.3390/ani13020251

PubMed Abstract | Crossref Full Text | Google Scholar

Cohen A. A., Legault V., Li Q., Fried L. P., and Ferrucci L. (2018). Men sustain higher dysregulation levels than women without becoming frail. Journals Gerontology. Ser. A Biol. Sci. Med. Sci. 73, 175–184. doi: 10.1093/gerona/glx146

PubMed Abstract | Crossref Full Text | Google Scholar

Cohen A. A., Li Q., Milot E., Leroux M., Faucher S., Morissette-Thomas V., et al. (2015). Statistical distance as a measure of physiological dysregulation is largely robust to variation in its biomarker composition. PloS One 10, e0122541. doi: 10.1371/journal.pone.0122541

PubMed Abstract | Crossref Full Text | Google Scholar

Cohen A. A., Milot E., Li Q., Legault V., Fried L. P., and Ferrucci L. (2014). Cross-population validation of statistical distance as a measure of physiological dysregulation during aging. Exp. Gerontology 57, 203–210. doi: 10.1016/j.exger.2014.04.016

PubMed Abstract | Crossref Full Text | Google Scholar

Cohen A. A., Milot E., Yong J., Seplaki C. L., Fülöp T., Bandeen-Roche K., et al. (2013). A novel statistical approach shows evidence for multi-system physiological dysregulation during aging. Mech. Ageing Dev. 134, 110–117. doi: 10.1016/j.mad.2013.01.004

PubMed Abstract | Crossref Full Text | Google Scholar

Cole M. F., Barnes P., Monroe I. G., Rukundo J., Thompson M. E., and Rosati A. G. (2024). Age-related physiological dysregulation progresses slowly in semi-free-ranging chimpanzees. Evolution Medicine Public Health 12, 129–142. doi: 10.1093/emph/eoae010

PubMed Abstract | Crossref Full Text | Google Scholar

Crimmins E. M., Johnston M., Hayward M., and Seeman T. E. (2003). Age differences in allostatic load: an index of physiological dysregulation. Exp. Gerontology 38, 731–734. doi: 10.1016/S0531-5565(03)00099-8

PubMed Abstract | Crossref Full Text | Google Scholar

Crockett C. M., Shimoji M., and Bowden D. M. (2000). Behavior, appetite, and urinary cortisol responses by adult female pigtailed macaques to cage size, cage level, room change, and ketamine sedation. Am. J. Primatology 52, 63–80. doi: 10.1002/1098-2345(200010)52:2<63::AID-AJP1>3.0.CO;2-K

PubMed Abstract | Crossref Full Text | Google Scholar

Danese A. and McEwen B. S. (2012). Adverse childhood experiences, allostasis, allostatic load, and age-related disease. Physiol. Behav. 106, 29–39. doi: 10.1016/j.physbeh.2011.08.019

PubMed Abstract | Crossref Full Text | Google Scholar

Dansereau G., Wey T. W., Legault V., Brunet M. A., Kemnitz J. W., Ferrucci L., et al. (2019). Conservation of physiological dysregulation signatures of aging across primates. Aging Cell 18, e12925. doi: 10.1111/acel.12925

PubMed Abstract | Crossref Full Text | Google Scholar

De Kock M., Loix S., and Lavand’homme P. (2013). Ketamine and peripheral inflammation. CNS Neurosci. Ther. 19, 403–410. doi: 10.1111/cns.12104

PubMed Abstract | Crossref Full Text | Google Scholar

De Maesschalck R., Jouan-Rimbaud D., and Massart D. L. (2000). The Mahalanobis distance. Chemometrics Intelligent Lab. Systems: Int. J. Sponsored by Chemometrics Soc. 50, 1–18. doi: 10.1016/S0169-7439(99)00047-7

Crossref Full Text | Google Scholar

Duong M. T., Bingham B. A., Aldana P. C., Chung S. T., and Sumner A. E. (2017). Variation in the calculation of allostatic load scores: 21 examples from NHANES. J. Racial Ethnic Health Disparities 4, 455–461. doi: 10.1007/s40615-016-0246-8

PubMed Abstract | Crossref Full Text | Google Scholar

Edes A. N., Baskir E., Riepl E., Musgrave K. E., Edwards K. L., Kozlowski C. P., et al. (2025). Things are not always what they seem: combining multiple approaches to understand possible impacts of visitors on a zoo-housed francois langur. Zoo Biol., 70016. doi: 10.1002/zoo.70016

PubMed Abstract | Crossref Full Text | Google Scholar

Edes A. N., Brown J. L., and Edwards K. L. (2023a). Evaluating individual biomarkers for predicting health risks in zoo-housed chimpanzees (Pan troglodytes) and bonobos (Pan paniscus). Am. J. Primatology 85, e23457. doi: 10.1002/ajp.23457

PubMed Abstract | Crossref Full Text | Google Scholar

Edes A. N., Edwards K. L., Wolfe B. A., Brown J. L., and Crews D. E. (2020). Allostatic load indices with cholesterol and triglycerides predict disease and mortality risk in zoo-housed western lowland gorillas (Gorilla gorilla gorilla). biomark. Insights 15, 1–13. doi: 10.1177/1177271920914585

PubMed Abstract | Crossref Full Text | Google Scholar

Edes A. N., Edwards K. L., Zimmerman D., Jourdan B., Crews D. E., Wolfe B. A., et al. (2023b). Comparing predictors and outcomes of higher allostatic load across zoo-housed African great apes. J. Zoological Botanical Gardens 4, 158–175. doi: 10.3390/jzbg4010016

Crossref Full Text | Google Scholar

Edes A. N., Wolfe B. A., and Crews D. E. (2016a). Rearing history and allostatic load in adult western lowland gorillas (Gorilla gorilla gorilla) in human care. Zoo Biol. 35, 167–173. doi: 10.1002/zoo.21270

PubMed Abstract | Crossref Full Text | Google Scholar

Edes A. N., Wolfe B. A., and Crews D. E. (2016b). Assessing stress in zoo-housed western lowland gorillas (Gorilla gorilla gorilla) using allostatic load. Int. J. Primatology 37, 241–259. doi: 10.1007/s10764-016-9899-8

Crossref Full Text | Google Scholar

Edes A. N. and Crews D. E. (2017). Allostatic load and biological anthropology. Am J Phys Anthropol 162 Suppl 63, 44–70.

Google Scholar

Edes A. N., Wolfe B. A., and Crews D. E. (2018a). Evaluating allostatic load: A new approach to measuring long-term stress in wildlife. J. Zoo Wildlife Med. 49, 272–282. doi: 10.1638/2016-0070.1

PubMed Abstract | Crossref Full Text | Google Scholar

Edes A. N., Wolfe B. A., and Crews D. E. (2018b). The first multi-zoo application of an allostatic load index to western lowland gorillas (Gorilla gorilla gorilla). Gen. Comp. Endocrinol. 266, 135–149. doi: 10.1016/j.ygcen.2018.05.006

PubMed Abstract | Crossref Full Text | Google Scholar

Edes A. N., Wolfe B. A., and Crews D. E. (2021). Testing a method to improve predictions of disease and mortality risk in western lowland gorillas (Gorilla gorilla gorilla) using allostatic load. Stress 24, 76–86. doi: 10.1080/10253890.2020.1748003

PubMed Abstract | Crossref Full Text | Google Scholar

Egorov A. I., Griffin S. M., Converse R. R., Styles J. N., Klein E., Scott J., et al. (2020). Greater tree cover near residence is associated with reduced allostatic load in residents of central north carolina. Environ. Res. 186, 109435. doi: 10.1016/j.envres.2020.109435

PubMed Abstract | Crossref Full Text | Google Scholar

Friedman J., Hastie T., and Tibshirani R. (2010). Regularization paths for generalized linear models via coordinate descent. J. Stat. Software 33, 1–22. doi: 10.18637/jss.v033.i01

PubMed Abstract | Crossref Full Text | Google Scholar

Glei D. A., Goldman N., Chuang Y.-L., and Weinstein M. (2007). Do chronic stressors lead to physiological dysregulation? Testing the theory of allostatic load. Psychosomatic Med. 69, 769–776. doi: 10.1097/PSY.0b013e318157cba6

PubMed Abstract | Crossref Full Text | Google Scholar

Gruber T. and Clay Z. (2016). A comparison between bonobos and chimpanzees: A review and update. Evolutionary Anthropology 25, 239–252. doi: 10.1002/evan.21501

PubMed Abstract | Crossref Full Text | Google Scholar

Guidi J., Lucente M., Sonino N., and Fava G. A. (2020). Allostatic load and its impact on health: A systematic review. Psychother. Psychosomatics. 9, 11–27. doi: 10.1159/000510696

PubMed Abstract | Crossref Full Text | Google Scholar

Harris T., Feldman J. J., Kleinman J. C., Ettinger W. H. Jr., Makuc D. M., and Schatzkin A. G. (1992). The low cholesterol-mortality association in a national cohort. J. Clin. Epidemiol. 45, 595–601. doi: 10.1016/0895-4356(92)90131-6

PubMed Abstract | Crossref Full Text | Google Scholar

Harrison X. A., Donaldson L., Correa-Cano M. E., Evans J., Fisher D. N., Goodwin C. E.D., et al. (2018). A brief introduction to mixed effects modelling and multi-model inference in ecology. PeerJ 6, e4794. doi: 10.7717/peerj.4794

PubMed Abstract | Crossref Full Text | Google Scholar

Heim C., Ehlert U., and Hellhammer D. H. (2000). The potential role of hypocortisolism in the pathophysiology of stress-related bodily disorders. Psychoneuroendocrinology 25, 1–35. doi: 10.1016/S0306-4530(99)00035-9

PubMed Abstract | Crossref Full Text | Google Scholar

Honkalampi K., Kraav S.-L., Kerr P., Juster R.-P., Virtanen M., Hintsa T., et al. (2024). Associations of allostatic load with sociodemographic factors, depressive symptoms, lifestyle, and health characteristics in a large general population-based sample. J. Affect. Disord. 350, 784–791. doi: 10.1016/j.jad.2024.01.189

PubMed Abstract | Crossref Full Text | Google Scholar

Horan J. M. and Widom C. S. (2015). From childhood maltreatment to allostatic load in adulthood: the role of social support. Child Maltreatment 20, 229–239. doi: 10.1177/1077559515597063

PubMed Abstract | Crossref Full Text | Google Scholar

Hulliger B., Sterchi M., and Schoch T. (2023). Modi: Multivariate Outlier Detection and Imputation for Incomplete Survey Data (version 0.1.2) (R Package Version). Available online at: https://cran.r-project.org/package=modi (Accessed April 14, 2025).

Google Scholar

Johnson S. C., Cavallaro F. L., and Leon D. A. (2017). A systematic review of allostatic load in relation to socioeconomic position: poor fidelity and major inconsistencies in biomarkers employed. Soc. Sci. Med. 192, 66–73. doi: 10.1016/j.socscimed.2017.09.025

PubMed Abstract | Crossref Full Text | Google Scholar

Johnson A. J., Dudley W. N., Wideman L., and Schulz M. (2019). Physiological risk profiles and allostatic load: using latent profile analysis to examine socioeconomic differences in physiological patterns of risk. Eur. J. Environ. Public Health 3 (2), em0029. doi: 10.29333/ejeph/5870

Crossref Full Text | Google Scholar

Juster R.-P., McEwen B. S., and Lupien S. J. (2010). Allostatic load biomarkers of chronic stress and impact on health and cognition. Neurosci. Biobehav. Rev. 35, 2–16. doi: 10.1016/j.neubiorev.2009.10.002

PubMed Abstract | Crossref Full Text | Google Scholar

Kemnitz J. W. and Kraemer G. W. (1982). Assessment of glucoregulation in rhesus monkeys sedated with ketamine. Am. J. Primatology 3, 201–210. doi: 10.1002/ajp.1350030118

PubMed Abstract | Crossref Full Text | Google Scholar

Kezios K. L., Suglia S. F., Doyle D. M., Susser E., Bradwin G., Cirillo P., et al. (2022). Comparing different operationalizations of allostatic load measured in mid-life and their patterning by race and cumulative life course socioeconomic status. Psychoneuroendocrinology 139, 105689. doi: 10.1016/j.psyneuen.2022.105689

PubMed Abstract | Crossref Full Text | Google Scholar

Kim C.-Y., Lee H.-S., Han S.-C., Heo J.-D., Kwon M.-S., Ha C.-S., et al. (2005). Hematological and serum biochemical values in cynomolgus monkeys anesthetized with ketamine hydrochloride. J. Med. Primatology 34, 96–100. doi: 10.1111/j.1600-0684.2005.00097.x

PubMed Abstract | Crossref Full Text | Google Scholar

Kraft T. S., Stieglitz J., Trumble B. C., Garcia A. R., Kaplan H., and Gurven M. (2020). Multi-system physiological dysregulation and ageing in a subsistence population. Philos. Trans. R. Soc. London. Ser. B Biol. Sci. 375, 20190610. doi: 10.1098/rstb.2019.0610

PubMed Abstract | Crossref Full Text | Google Scholar

Kronmal R. A., Cain K. C., Ye Z., and Omenn G. S. (1993). Total serum cholesterol levels and mortality risk as a function of age: A report based on the Framingham data. Arch. Internal Med. 153, 1065–1073. doi: 10.1001/archinte.1993.00410090025004

PubMed Abstract | Crossref Full Text | Google Scholar

Kusano Y., Crews D. E., Iwamoto A., Sone Y., Aoyagi K., Maeda T., et al. (2016). Allostatic load differs by sex and diet, but not age in older Japanese from the goto islands. Ann. Hum. Biol. 43, 34–41. doi: 10.3109/03014460.2015.1013985

PubMed Abstract | Crossref Full Text | Google Scholar

Lacagnina S. (2020). The developmental origins of health and disease (DOHaD). Am. J. Lifestyle Med. 14, 47–50. doi: 10.1177/1559827619879694

PubMed Abstract | Crossref Full Text | Google Scholar

Lambeth S. P., Hau J., Perlman J. E., Martino M., and Schapiro S. J. (2006). Positive reinforcement training affects hematologic and serum chemistry values in captive chimpanzees (Pan troglodytes). Am. J. Primatology 68, 245–256. doi: 10.1002/ajp.20148

PubMed Abstract | Crossref Full Text | Google Scholar

Laméris D. W., Staes N., Salas M., Matthyssen S., Verspeek J., and Stevens J. M.G. (2020). The influence of sex, rearing history, and personality on abnormal behaviour in zoo-housed bonobos (Pan paniscus). Appl. Anim. Behav. Sci. 234, 105178. doi: 10.1016/j.applanim.2020.105178

Crossref Full Text | Google Scholar

Lateef S. S., Najafi M. A., Dey A. K., Batool M., Abdelrahman K. M., Uceda D. E., et al. (2020). Relationship between chronic stress-related neural activity, physiological dysregulation and coronary artery disease in psoriasis: findings from a longitudinal observational cohort study. Atherosclerosis 310, 37–44. doi: 10.1016/j.atherosclerosis.2020.07.012

PubMed Abstract | Crossref Full Text | Google Scholar

Lehmann R., Wagner J. L., Fernandez L. A., Bourgoignie J. J., Ricordi C., Alejandro R., et al. (1997). Effects of ketamine sedation on glucose clearance, insulin secretion and counterregulatory hormone production in baboons (Papio hamadryas). J. Med. Primatology 26, 312–321. doi: 10.1111/j.1600-0684.1997.tb00060.x

PubMed Abstract | Crossref Full Text | Google Scholar

Levine M. E. and Crimmins E. M. (2014). A comparison of methods for assessing mortality risk. Am. J. Hum. Biol. 26, 768–776. doi: 10.1002/ajhb.22595

PubMed Abstract | Crossref Full Text | Google Scholar

Li Y., Rosemberg M.-A. S, Dalton V. K., Lee S. J., and Seng J. S. (2019). Exploring the optimal allostatic load scoring method in women of reproductive age. J. Advanced Nurs. 75, 2548–2558. doi: 10.1111/jan.14014

PubMed Abstract | Crossref Full Text | Google Scholar

Li Y. and Sanon Rosemberg M.-A. (2020). The promise of allostatic load rests upon strategic operationalization, scoring, and targeted interventions. Psychoneuroendocrinology 123, 104877. doi: 10.1016/j.psyneuen.2020.104877

PubMed Abstract | Crossref Full Text | Google Scholar

Li Q., Wang S., Milot E., Bergeron P., Ferrucci L., Fried L. P., et al. (2015). Homeostatic dysregulation proceeds in parallel in multiple physiological systems. Aging Cell 14, 1103–1112. doi: 10.1111/acel.12402

PubMed Abstract | Crossref Full Text | Google Scholar

Liu S. H., Juster R.-P., Dams-O’Connor K., and Spicer J. (2021). Allostatic load scoring using item response theory. Compr. Psychoneuroendocrinology 5, 100025. doi: 10.1016/j.cpnec.2020.100025

PubMed Abstract | Crossref Full Text | Google Scholar

Lowenstine L. J., McManamon R., and Terio K. A. (2016). Comparative pathology of aging great apes: bonobos, chimpanzees, gorillas, and orangutans. Veterinary Pathol. 53, 250–276. doi: 10.1177/0300985815612154

PubMed Abstract | Crossref Full Text | Google Scholar

Lüdecke D., Makowski D., Ben-Shachar M. S., Patil I., Waggoner P., Wiernik B. M., et al. (2025). Performance: Assessment of Regression Models Performance. Available online at: https://CRAN.R-project.org/package=performance (Accessed November 2, 2025).

Google Scholar

Mauss D., Jarczok M. N., and Fischer J. E. (2016). The streamlined allostatic load index: A replication of study results. Stress 19, 553–558. doi: 10.1080/10253890.2016.1219718

PubMed Abstract | Crossref Full Text | Google Scholar

Mauss D., Li J., Schmidt B., Angerer P., and Jarczok M. N. (2015). Measuring allostatic load in the workforce: A systematic review. Ind. Health 53, 5–20. doi: 10.2486/indhealth.2014-0122

PubMed Abstract | Crossref Full Text | Google Scholar

McCaffery J. M., Marsland A. L., Strohacker K., Muldoon M. F., and Manuck S. B. (2012). Factor structure underlying components of allostatic load. PloS One 7, e47246. doi: 10.1371/journal.pone.0047246

PubMed Abstract | Crossref Full Text | Google Scholar

McCreary J.K., Erickson Z. T., Paxman E., Kiss D., Montina T., Olson D. M., et al. (2019). The rat cumulative allostatic load measure (rCALM): A new translational assessment of the burden of stress. Environ. Epigenet. 5, 1–10. doi: 10.1093/eep/dvz005

PubMed Abstract | Crossref Full Text | Google Scholar

McCrory C., McLoughlin S., Layte R., NiCheallaigh C., O’Halloran A. M., Barros H., et al. (2023). Towards a consensus definition of allostatic load: A multi-cohort, multi-system, multi-biomarker individual participant data (IPD) meta-analysis. Psychoneuroendocrinology 153, 106117. doi: 10.1016/j.psyneuen.2023.106117

PubMed Abstract | Crossref Full Text | Google Scholar

McEwen B. S. and Stellar E. (1993). Stress and the individual: mechanisms leading to disease. Arch. Internal Med. 153, 2093–2101. doi: 10.1001/archinte.1993.00410180039004

Crossref Full Text | Google Scholar

McLoughlin S., Kenny R. A., and McCrory C. (2020). Does the choice of allostatic load scoring algorithm matter for predicting age-related health outcomes? Psychoneuroendocrinology 120, 104789. doi: 10.1016/j.psyneuen.2020.104789

PubMed Abstract | Crossref Full Text | Google Scholar

Milot E., Cohen A. A., Vézina F., Buehler D. M., Matson K. D., and Piersma T. (2014a). A novel integrative method for measuring body condition in ecological studies based on physiological dysregulation. Methods Ecol. Evol. 5, 146–155. doi: 10.1111/2041-210X.12145

Crossref Full Text | Google Scholar

Milot E., Morissette-Thomas V., Li Q., Fried L. P., Ferrucci L., and Cohen A. A. (2014b). Trajectories of physiological dysregulation predicts mortality and health outcomes in a consistent manner across three populations. Mech. Ageing Dev. 141–142, 56–63. doi: 10.1016/j.mad.2014.10.001

PubMed Abstract | Crossref Full Text | Google Scholar

Murphy H. W., Danforth M. D., and Clyde V. L. (2018). The great ape heart project. Int. Zoo Yearbook 52, 103–112. doi: 10.1111/izy.12180

Crossref Full Text | Google Scholar

Parish A. R. and De Waal F. B.M. (2000). The other ‘Closest living relative’: how bonobos (Pan paniscus) challenge traditional assumptions about females, dominance, intra- and intersexual interactions, and hominid evolution. Ann. New York Acad. Sci. 907, 97–113. doi: 10.1111/j.1749-6632.2000.tb06618.x

PubMed Abstract | Crossref Full Text | Google Scholar

Parker H. W., Abreu A. M., Sullivan M. C., and Vadiveloo M. K. (2022). Allostatic load and mortality: A systematic review and meta-analysis. Am. J. Prev. Med. 63, 131–140. doi: 10.1016/j.amepre.2022.02.003

PubMed Abstract | Crossref Full Text | Google Scholar

Pascual A., Kalcher-Sommersguter E., Riba D., and Crailsheim D. (2023). Long-term assessment of captive chimpanzees: influence of social group composition, seasonality and biographic background. Animals 13, 424. doi: 10.3390/ani13030424

PubMed Abstract | Crossref Full Text | Google Scholar

Perumal N., Ramasamy V., Kumar M. J. M., and Majumdar S. S. (2007). Effects of ketamine and thiopentone anesthesia on serum lipid parameters in adult bonnet monkeys (Macaca radiata). J. Am. Assoc. Lab. Anim. Science: JAALAS 46, 21–23.

PubMed Abstract | Google Scholar

Polick C. S., Harris-Gersten M. L., Dennis P. A., Noonan D., Hastings S. N., Calhoun P. S., et al. (2024). Allostatic load, morbidity, and mortality among older adults: A multi-wave analysis from the national health and aging trends study. J. Appl. Gerontology: Off. J. South. Gerontological Soc. 43, 7334648241230010.

PubMed Abstract | Google Scholar

Pryce C. R., Dettling A. C., Spengler M., Schnell C. R., and Feldon J. (2004). Deprivation of parenting disrupts development of homeostatic and reward systems in marmoset monkey offspring. Biol. Psychiatry 56, 72–79. doi: 10.1016/j.biopsych.2004.05.002

PubMed Abstract | Crossref Full Text | Google Scholar

Raison C. L. and Miller A. H. (2003). When not enough is too much: the role of insufficient glucocorticoid signaling in the pathophysiology of stress-related disorders. Am. J. Psychiatry 160, 1554–1565. doi: 10.1176/appi.ajp.160.9.1554

PubMed Abstract | Crossref Full Text | Google Scholar

R Core Team (2024). R: A language and environment for statistical computing (version 4.4.1) (Vienna, Austria: R Foundation for Statistical Computing).

Google Scholar

Richards S. A., Whittingham M. J., and Stephens P. A. (2011). Model selection and model averaging in behavioural ecology: the utility of the IT-AIC framework. Behav. Ecol. Sociobiology 65, 77–89. doi: 10.1007/s00265-010-1035-8

Crossref Full Text | Google Scholar

Robertson T., Beveridge G., and Bromley C. (2017). Allostatic load as a predictor of all-cause and cause-specific mortality in the general population: evidence from the Scottish health survey. PloS One 12, e0183297. doi: 10.1371/journal.pone.0183297

PubMed Abstract | Crossref Full Text | Google Scholar

Robertson T. and Watts E. (2016). The importance of age, sex and place in understanding socioeconomic inequalities in allostatic load: evidence from the Scottish health survey, (2008-2011). BMC Public Health 16, 126. doi: 10.1186/s12889-016-2796-4

PubMed Abstract | Crossref Full Text | Google Scholar

Sabbah W., Watt R.G., Sheiham A., and Tsakos G. (2008). Effects of allostatic load on the social gradient in ischaemic heart disease and periodontal disease: evidence from the Third National Health and Nutrition Examination Survey. J. Epidemiol. Community Health 62, 415–420.

PubMed Abstract | Google Scholar

Sanchez M. M. (2006). The impact of early adverse care on HPA axis development: nonhuman primate models. Hormones Behav. 50, 623–631. doi: 10.1016/j.yhbeh.2006.06.012

PubMed Abstract | Crossref Full Text | Google Scholar

Saranteas T., Zotos N., Lolis E., Stranomiti J., Mourouzis C., Chantzi C., et al. (2005). Mechanisms of ketamine action on lipid metabolism in rats. Eur. J. Anaesthesiology 22, 222–226. doi: 10.1097/00003643-200503000-00010

PubMed Abstract | Crossref Full Text | Google Scholar

Seeley K. E., Proudfoot K. L., and Edes A. N. (2022). The application of allostasis and allostatic load in animal species: A scoping review. PloS One 17, e0273838. doi: 10.1371/journal.pone.0273838

PubMed Abstract | Crossref Full Text | Google Scholar

Seeley K. E., Proudfoot K. L., Wolfe B., and Crews D. E. (2021). Assessing allostatic load in ring-tailed lemurs (Lemur catta). Animals: Open Access J. MDPI 11, 3074. doi: 10.3390/ani11113074

PubMed Abstract | Crossref Full Text | Google Scholar

Seeman T. E., Singer B. H., Rowe J. W., Horwitz R. I., and McEwen B. S. (1997). Price of adaptation - allostatic load and its health consequences. Arch. Internal Med. 157, 2259–2268. doi: 10.1001/archinte.1997.00440400111013

Crossref Full Text | Google Scholar

Siew R. V. K., Bowe S. J., Turner A. I., Sarnyai Z., Nilsson C. J., Shaw J. E., et al. (2023). The role of combined modifiable lifestyle behaviors in the longitudinal association between stressful life events and allostatic load in Australian adults. Psychoneuroendocrinology 149, 106021. doi: 10.1016/j.psyneuen.2022.106021

PubMed Abstract | Crossref Full Text | Google Scholar

Species360 Zoological Information Management System (ZIMS) (2021). Available online at: https://zims.Species360.org (Accessed December 6, 2024).

Google Scholar

Strong V. J., Martin M., Redrobe S., White K., and Baiker K. (2018). A retrospective review of great ape cardiovascular disease epidemiology and pathology. Int. Zoo Yearbook 52, 113–125. doi: 10.1111/izy.12193

Crossref Full Text | Google Scholar

Symonds M. R.E. and Moussalli A. (2011). A brief guide to model selection, multimodel inference and model averaging in behavioural ecology using Akaike’s information criterion. Behav. Ecol. Sociobiology 65, 13–21. doi: 10.1007/s00265-010-1037-6

Crossref Full Text | Google Scholar

Tay J.K., Narasimhan B., and Hastie T. (2023). Elastic net regularization paths for all generalized linear models. J. Stat. Software 106, 1. doi: 10.18637/jss.v106.i01

PubMed Abstract | Crossref Full Text | Google Scholar

Teman S. J., Atwood T. C., Converse S. J., Fry T. L., and Laidre K. L. (2025). Measuring polar bear health using allostatic load. Conserv. Physiol. 13, coaf013. doi: 10.1093/conphys/coaf013

PubMed Abstract | Crossref Full Text | Google Scholar

Thayer Z., Barbosa-Leiker C., McDonell M., Nelson L., Buchwald D., and Manson S. (2017). Early life trauma, post-traumatic stress disorder, and allostatic load in a sample of American Indian adults. Am. J. Hum. Biol. 29, 1–10. doi: 10.1002/ajhb.22943

PubMed Abstract | Crossref Full Text | Google Scholar

Therneau T. M., Lumley T., Atkinson E., and Crowson C. (2022). Survival: Survival Analysis (version 3.3-1) (R Package Version). Available online at: https://cran.r-project.org/package=lme4 (Accessed November 2, 2025).

Google Scholar

Walker A. J., Foley B. M., Sutor S. L., McGillivray J. A., Frye M. A., and Tye S. J. (2015). Peripheral Proinflammatory markers associated with ketamine response in a preclinical model of antidepressant-resistance. Behav. Brain Res. 293, 198–202. doi: 10.1016/j.bbr.2015.07.026

PubMed Abstract | Crossref Full Text | Google Scholar

Wickham H., Chang W., Henry L., Pedersen T. L., Takahashi K., Wilke C., et al. (2018). Ggplot2: Create Elegant Data Visualisations Using the Grammar of Graphics (R Package Version 3.1.0). Available online at: https://cran.r-project.org/package=ggplot2 (Accessed October 29, 2025).

Google Scholar

Wolfe B. A. and Edes A. N. (2023). “Allostatic Load as a Measure of Animal Stress and Health Risk,” in Fowler’s Zoo and Wild Animal Medicine: Current Therapy, vol. 10 . Eds. Eric Miller R., Calle P. P., and Lamberski N. (Elsevier Inc, St. Louis, Missouri), 261–266.

Google Scholar

Yoshida T., Suzuki K., and Shimizu T. (1986). The effects of ketamine anesthesia on hematological and serum biochemical values in female cynomolgus monkeys (Macaca fascicularis). Exp. Animlas 35, 455–461. doi: 10.1538/expanim1978.35.4_455

PubMed Abstract | Crossref Full Text | Google Scholar

Zou H. and Hastie T. (2005). Regularization and variable selection via the elastic net. J. R. Stat. Society. Ser. B Stat. Method. 67, 301–320. doi: 10.1111/j.1467-9868.2005.00503.x

Crossref Full Text | Google Scholar