Age-related differences in cerebral morphology and microstructure in rhesus macaques

Abstract

The rhesus macaque presents a promising model for translational research into human brain aging due to this species’ long lifespan and close phylogenic relationship. We conducted a cross-sectional study identifying microstructural and morphological biomarkers of aging in a cohort of 37 healthy animals (18F/19M, aged 5–28 years), using high-resolution T2-weighted (T2w) and diffusion-weighted (DW) images. Using Tensor Based Morphometry, significant age-associated regional brain atrophy was observed in some areas of the frontal and parietal cortex, as well as the striatum. Additionally, age-associated differences in white matter diffusion were observed in several brain regions, including frontal and temporal white matter areas, and regions of the internal capsule and corpus callosum. Taken together, the results demonstrate that morphological and microstructural age-related differences can be disclosed in cortical, striatal, and thalamic regions, as well as in the white matter fiber pathways connecting these areas, using high-resolution DTI and MRI.

1. Introduction

Neurological changes associated with normative aging present clinical challenges and constitute a major risk factor for neurodegenerative diseases including Alzheimer’s Disease and related dementias (ADRD). Unfortunately, few effective screening tools or therapeutic strategies exist. Recent studies suggest that a long-term sequalae of neuroinflammatory and neurodegenerative processes may begin decades before symptom onset in many cases of ADRD ^1–3. For this reason, animal models that recapitulate these multi-factored aging processes, but on a shorter timescale, are critical for identification of biomarkers of normative aging that can serve as future outcome measures and therapeutic targets ⁴. Because of their close phylogenetic relatedness to humans, rhesus macaques exhibit a higher degree of genetic, molecular, and anatomical convergence than many available rodent models. Macaques also exhibit similarities with humans in complex behaviors such as memory, visuospatial attention, and executive function, which are central qualities of age-related cognitive impairment. Taken together, these features make the rhesus macaque an excellent translational animal model of human aging ^4,5.

Magnetic Resonance Imaging (MRI) is a non-invasive technique for measuring brain anatomy and function, frequently used both clinically and in non-human primate (NHP) models. Earlier structural MRI studies in macaques have reported region-specific patterns of age-related gray matter atrophy ^6–8. In particular, aged macaques (>20 years) have been reported to exhibit modest cortical atrophy, when compared to young-adult (<12 years) macaques, in areas of the dorsolateral prefrontal cortex (dlPFC), anterior cingulate cortex (ACC), and area 7 of the parietal cortex (PC) ^6,9–11. Previous studies have also revealed subcortical atrophy in areas of the caudate, putamen, globus pallidus and thalamus but not the hippocampus, minimal evidence of ventricular enlargement, and inconsistent findings regarding white matter volumetric changes ^6–10,12,13. Of particular relevance is that these previous studies have predominantly compared brains of old macaques to young-adult macaques, leaving a gap in our understanding of how these features manifest during middle age (13–19 years), and the timing of their emergence across the lifespan.

A recent paper by Dash and colleagues sought to begin to address this gap by systematically assessing volumetric changes across the entire adult lifespan of the macaque (including middle age), as well as to explore sex-differences across aging ¹⁴. Using T1w MPRAGE data, they confirmed lower volumes with increasing age in the dlPFC, ACC, caudate, putamen, thalamus, and hypothalamus; and increasing lateral ventricular volume. They also observed an inverted U-shaped pattern in many white matter regions including the corpus callosum (body, genu and splenium), frontal white matter, occipital white matter, and the internal capsule. These findings, along with similar human studies ¹⁵, highlight the importance of including middle-aged individuals in the study population. Unexpectedly, several subcortical structures showed trends for greater volumes in older animals, including the globus pallidus, hippocampus and amygdala. The globus pallidus exhibits a similar pattern in humans ¹⁶, and aligns somewhat with reports of relative preservation compared to other subcortical areas reported in monkeys ¹⁰ although age-related decreases in volume of this region have also been observed ⁸. The hippocampus and amygdala, however, often show negative correlations to age in humans ^17,18, making this finding worthy of follow-up investigations.

White matter microstructural changes in aging macaques have also been reported using diffusion tensor imaging (DTI) methods commonly applied in human studies ¹⁹. DTI data obtained in macaques are comparable to observations in human studies and are consistent with well-known anatomical pathways ²⁰. Findings from DTI studies in aging macaques buttress the anatomical MRI findings and indicate that white matter fiber pathways which interconnect the cortico-striatal-thalamic circuit appear particularly vulnerable in very old macaques ^21–25. Indeed, a pattern of region-specific alterations in white matter microstructure has been reported in old monkeys >20 years compared to young-adult monkeys <12 years. Specifically, significant decreases in fractional anisotropy (FA)—a measure of coherence—occur in fiber tracts interconnecting the dlPFC, orbitofrontal cortex (OFC), ACC, PC, striatum and thalamus (namely, the superior longitudinal fasciculus, cingulum, arcuate fasciculus, uncinate fasciculus, genu and splenium of corpus callosum, anterior limb of internal capsule) ^21–25. However, few studies to-date have linked morphological and microstructural changes together in the same animals, and there have been minimal investigation of other associated diffusion metrics such Axial Diffusivity (AD). This gap limits our ability to interpret these lifespan trajectories.

Towards this goal, the current study sought to improve on this cross-sectional imaging work with several advancements. First, our study cohort includes animals from both sexes and across a wide range of ages, including young-adulthood, middle-age, and very old. Second, we queried gray matter morphology with high-resolution T2w imaging, rather than T1w imaging as has been used in most studies to date. T2w imaging has higher sensitivity than T1w imaging to intensity shifts caused by factors such as iron and amyloid deposits, and provides good contrast to delineate GM structures which may be susceptible to the effects of normal aging or neurodegenerative diseases ^26,27. By applying T2w imaging, our goal was to complement and enhance the established findings in macaques derived from T1w data. Third, in parallel we queried white matter diffusion in the same population of animals using DTI to gain further insight into microstructural features associated with the volumetric changes reported in our previous study ¹⁴. Fourth, to further build on our previous work, here we employed consistent scan acquisitions over a shorter time interval (scans were collected over the span of 1 year) with uniform equipment and sequences for all animals and utilized a data-driven voxel-wise approach, rather than atlas-based parcellations, in order to achieve greater sub-regional specificity. Our goal was to better understand typical differences in primate brains across the lifespan, and to develop biomarkers identifying areas and sub-regions vulnerable to pathological aging.

2. Materials and Methods

2.1. Animals

This study included 37 rhesus macaques (Macaca mulatta), balanced for sex (n=18 female, n=19 male). Their average age was 17.5 years, with the mean age of males (17.16 years) being slightly, but not significantly, lower than the mean age of females (17.97 years); t(35)=0.4121, p=0.6828. There was an extensive age spread for both sexes, with females ranging from 5.70–26.08 years, and males ranging from 5.37–28.25 years. Data from individual animals illustrating this distribution are plotted in Fig. 1.

Fig. 1. Study population breakdown.

Scatterplot illustrating the age distribution of females (blue), males (red), and average ages (solid lines), from the n=37 animals included in this study.

While on study, all the animals were socially housed indoors, maintained on a 12-hour light/dark cycle, and provided with ad libitum access to water, chow rations twice daily, and produce rations once daily. Animal health and well-being was monitored continuously by veterinary and technical staff. All of the experimental procedures were approved by the Institutional Animal Care and Use Committee at the Oregon National Primate Research Center (ONPRC) and Oregon Health and Science University (OHSU), and the guidelines defined by the National Research Council’s Guide for the Care and Use of Laboratory Animals ²⁸ were rigorously observed.

2.2. MRI and DTI Acquisition

All the MR images were acquired on a Siemens Prisma whole body 3T scanner (Erlangen, Germany) using a 16-channel pediatric head coil. To prevent motion artifacts and to ensure the safety of the animals, monkeys were sedated initially with ketamine HCl (10 mg/kg IM), and anesthesia was maintained via inhalation of 1–2% isoflurane gas vaporized in 100% oxygen for the duration of the scanning session. Blood oxygenation, blood pressure, body temperature, and heart rate were monitored by trained veterinary staff; and the animals were closely observed for several hours during their recovery post-scan. Two types of MR images were acquired: 3D T2-weighted sampling perfection with application optimized contrasts using different flip angle evolution (SPACE) ²⁹, and diffusion weighted (DW) imaging scans. Acquisition parameters for these scans were identical to those reported previously by ³⁰. Briefly: three 3D SPACE images were acquired for each animal with 0.5-mm isotropic voxels, TE/TR = 385/3200 ms, flip angle = 120°, 160 × 160 × 112 mm FOV. In the same scanning session, a set of DW volumes were acquired with a spin-echo planar imaging (EPI) sequence (TR/TE = 6700 ms/73 ms, GRAPPA factor = 2, echo train length = 52) and 1.0-mm isotropic voxels. Specifically, seven repetitions of DW volumes with 30-directions and b=1000 s/mm² along with six b0 volumes per repetition were collected using an anterior-to-posterior phase-encoding scheme. In addition, six b0 volumes with reversed (posterior-to-anterior) phase encoding direction were also acquired to correct for susceptibility-induced distortions. Total acquisition time for the T2w and DW data was approximately 60 minutes.

2.3. MRI and DTI processing

SPACE and DW images were processed identically to the methods previously described by ³¹. For the SPACE data, each of the three images collected were aligned with ANTS (version 2.1, http://stnava.github.io/ANTs/) using the first acquired image as a reference, and averaged using FSL. Brain mask generation and extraction were completed using the template brain mask from the ONPRC18 T2w head template, which was inversely mapped to the T2w images to create individual brain masks and extract brains in native space ³⁰. Next, intensity bias correction was performed using the N4BiasFieldCorrection tool in the ANTS package ³². Finally, each individual T2w brain image (skull stripped in native space using the brain mask) was aligned to the ONPRC18 T2w brain template using b-spline non-linear transformations with ANTS. The warp fields produced from this registration step represent the deformations from individual space to ONPRC18 template space and were used to calculate log Jacobian determinant maps for voxel-based Tensor Based Morphometry (TBM) analysis, described in Section 2.4 below.

DW images were un-warped using TOPUP (FSL version 6.0, http://fsl.fmrib.ox.ac.uk/fsl/fslwiki/)), denoised (MATLAB script provided by Dr. Sune Jespersen, Aarhus University) ³³, and eddy corrected (FSL). Next, DTIFit was used to fit corrected DW data to a single tensor (DTI) model. The resulting tensor maps were aligned to the ONPRC18 tensor template ³⁰ using b-spline non-linear registrations with the DTI-TK toolkit ^34,35. Parameter maps of Fractional Anisotropy (FA), Mean Diffusivity (MD), Axial Diffusivity (AD), and Radial Diffusivity (RD) were then generated in template space and compared using voxel-wise statistical comparisons, described in section 2.4 below.

2.4. Quantification and statistical analysis

To establish patterns of age-associated parameter change at a sub-ROI level, voxel-based regressions were performed on the log Jacobian determinant maps, and on the FA, MD, AD and RD maps using FSL RANDOMISE. The analyses included both age- and sex- as co-factors in the model, applied threshold free cluster enhancement (TFCE), 5000 permutations with family-wise error (FWE) corrections ³⁶, and were run using gray matter and white matter masks (derived from the ONPRC18 label map) on the TBM and DTI data, respectively. In accordance with the guidelines described by ³⁷, our results highlight all p<0.05 FWE-corrected voxels using opaque overlays, but also include transparent overlays to illustrate areas with sub-threshold effects (where FWE-corrected p>0.05). All visualizations were created using MRIcroGL (version 12.6.3) and include a series of coronal slices across levels of the brain, and a 3D rendering of p<0.05 FWE-corrected voxels in the bottom right corner.

We next calculated the percent of p<0.05 FWE-corrected voxels in each region of interest (ROI) defined in the ONPRC18 label map, to create a summary of the regional distribution of significant voxel-wise changes. FSL tools were used to binarize and threshold FWE-corrected p-value maps (at FWE p<0.05), and AFNI tools (3dcalc) to extract volume information from each. Histograms illustrating the percent volume of p<0.05 FWE-corrected voxels in each ROI were subsequently plotted using GraphPad PRISM.

To follow-up on the voxel-based statistical results, we also conducted analyses on diffusion measures (FA, AD, RD, MD) calculated from ROIs defined by the same label map. Visual inspection of the diffusion-weighted images revealed that, for a subset of animals, voxels within and near brain regions where iron is known to accumulate with age (the substantia nigra and deep cerebellar nuclei) exhibit very low signal intensity due to short T2 and T2* values. Within such voxels, the diffusivity measures would inaccurately appear to be low, due to the image intensity in the diffusion-unweighted (b0) image being near the noise floor, and as a result the measured signal attenuation due to diffusion sensitization would be artifactually low. To ensure that ROI-averaged diffusivity measures were not biased by voxels with low signal-to-noise ratio (SNR), we conducted the ROI analyses using a modified mask designed to exclude low-SNR voxels. To achieve this, all b0 volumes were extracted from the denoised, eddy-corrected diffusion-weighted images and averaged to create mean b0 images using FSL tools. The noise level for each image was estimated as the highest-likelihood value of the Rician-distributed intensity within image regions that were devoid of tissue. A threshold value was set to 5 times the noise level, and only voxels with intensity above this threshold were included in calculations of mean ROI values for all diffusion metrics from WM ROIs from each subject.

For each ROI, the diffusion data were analyzed as a function of age, by considering potential linear and quadratic effects of age, as well as potential sex differences. The diffusion dependent variables, $y_i$, with $y$ representing FA, MD, AD, or RD for each of the $i$ monkeys were fitted to each of the following five expressions:

$$\begin{gathered} y_i={\beta }_0+ag{e}_i\mathrm{*}{\beta }_1+{\epsilon }_i \\ {y}_i={\beta }_0+{\mathrm{s}\mathrm{e}\mathrm{x}}_i\mathrm{*}{\beta }_1+{\epsilon }_i \\ {y}_i={\beta }_0+ag{e}_i\mathrm{*}{\beta }_1+{\mathrm{s}\mathrm{e}\mathrm{x}}_i\mathrm{*}{\beta }_2+ag{e}_i\mathrm{*}{\mathrm{s}\mathrm{e}\mathrm{x}}_i\mathrm{*}{\beta }_3+{\epsilon }_i \\ {y}_i={\beta }_0+ag{e}_i\mathrm{*}{\beta }_1+ag{e}_i^2\mathrm{*}{\beta }_2+{\epsilon }_i \\ {y}_i={\beta }_0+ag{e}_i\mathrm{*}{\beta }_1+{\mathrm{s}\mathrm{e}\mathrm{x}}_i\mathrm{*}{\beta }_2+ag{e}_i\mathrm{*}{\mathrm{s}\mathrm{e}\mathrm{x}}_i\mathrm{*}{\beta }_3+ag{e}_i^2\mathrm{*}{\beta }_4+ \\ ag{e}_i^2\mathrm{*}{\mathrm{s}\mathrm{e}\mathrm{x}}_i\mathrm{*}{\beta }_5+{\epsilon }_i \end{gathered}$$

and the Bayesian Information Criterion (BIC) associated with each expression was determined. The expression with the lowest BIC was selected for each diffusion measure, for each ROI.

3. Results

3.2. Morphological changes in gray matter regions

TBM can be used to describe morphological features of brain structure. In this study, warp fields co-aligning T2w images with ONPRC18 atlas space were used to calculate deformation parameter maps (the log of the Jacobian determinant) indicative of regional patterns of expansion or contraction. Since, in this context, all individuals are compared to a standard, inferences can therefore be made about how that individual differs from the population based on this TBM parameter.

Using a voxel-wise linear regression approach, significant age-associated differences in TBM brain morphology were observed in the cortex and striatum. Parameter maps of FWE p-values are illustrated in Fig. 2A. Our voxel-wise analysis revealed that approximately 5–10% of the total volume within the ventromedial prefrontal, cingulate, and sensory-motor cortices, as well as the caudate, exhibited significant age-associated effects in TBM, as indicated by a negative relationship with the log of the Jacobian determinant. This percentage reflects contraction or regional atrophy within these regions (FWE-p<0.05), and represents the extent (volume) of each region that exhibits significant TBM voxels. Other cortical and subcortical areas had smaller regions of significant (FWE-p<0.05) contraction, with approximately 1–3% of the volume of dorsal prefrontal, inferior temporal, and parietal cortices and the putamen demonstrating contraction (Fig. 2B). Brain wide, only one region showed significant (FEW-p<0.05) increases in the log of the Jacobian determinant across age (the substantia nigra) (Fig. 2B). Including sex as a covariate in the analysis revealed no significant (FWE-p<0.05) sex by age interactions.

Fig. 2. Gray matter areas with significant age-associated changes in morphology.

(A) Coronal slices of the ONPRC18 T2w template are shown across 19 anterior-to-posterior levels. Decreases in each metric are shown in blue, and increases in red. All p<0.05 FWE-corrected voxels are shown using opaque overlays, and transparent overlays are used to illustrate areas with sub-threshold effects as suggested by Taylor, Reynolds, Calhoun, Gonzalez-Castillo, Handwerker, Bandettini, Mejia and Chen ³⁷. The only significant increases were observed in the substantia nigra. For further spatial context, a 3D rendering of p<0.05 FWE-corrected voxels is provided in the bottom right corner. (B) Histograms illustrating the extent of significant decreases in TBM measures for each ROI in the ONPRC18 label map.

3.1. Microstructural changes in white matter regions

Diffusion weighted imaging can be used to infer microstructural features in brain tissue. In this study, we applied the diffusion tensor model to generate high resolution (1 mm isotropic) brain-wide maps of FA, as well as MD, AD, and RD. Using the same voxel-wise linear regression approach that was applied to the TBM data, significant associations between age and each of these metrics were detected in white matter regions throughout the brain.

Analysis of FA maps highlighted a number of white matter regions with significant negative relationships between age and FA. The regions exhibiting these associations were more widespread than associations revealed by TBM. Parameter maps of FA FWE-p-values are illustrated in Fig. 3 and 4A (top row). The results indicated that >30% of the volume of several temporal and frontal white matter areas (superior temporal gyrus, middle temporal gyrus, dorsal prefrontal, inferior frontal, the superior longitudinal and fronto-occipital fasciculus) exhibited age-associated FWE-p<0.05 negative relationships with FA. Significant negative FA associations were also observed in >10% of the volume of the corona radiata, internal capsule, and corpus callosum (all FWE-p<0.05) (Fig. 4B, top row). No voxels with significant (FEW-p<0.05) age-associated positive associations in FA were found brain-wide. Including sex as a covariate in the analysis revealed no significant sex by age interactions.

Fig. 3. White matter areas with significant age-associated decreases in Fractional Anisotropy.

Coronal slices of the ONPRC18 FA template are shown across 19 anterior-to-posterior levels. Decreases in FA are shown in blue, and increases in red. All p<0.05 FWE-corrected voxels are shown using opaque overlays, and transparent overlays are used to illustrate areas with sub-threshold effects as suggested by Taylor, Reynolds, Calhoun, Gonzalez-Castillo, Handwerker, Bandettini, Mejia and Chen ³⁷ . For further spatial context, a 3D rendering of p<0.05 FEW-corrected voxels is provided in the bottom right corner. Despite numerous significant decreases detected in various brain areas, no significant increases were observed.

Fig. 4. Reduced FA is associated with significant decreases in other DTI metrics.

(A) Coronal slices of the ONPRC18 FA, MD, AD, and RD templates are shown across 6 anterior-to-posterior levels. Decreases in each metric are shown in blue, and increases in red. All p<0.05 FEW-corrected voxels are shown using opaque overlays, and transparent overlays are used to illustrate areas with sub-threshold effects as suggested by Taylor, Reynolds, Calhoun, Gonzalez-Castillo, Handwerker, Bandettini, Mejia and Chen ³⁷. (B) Histograms illustrating the extent of significant (p<0.05 FWE-corrected) decreases by ROI for each diffusion metric (FA, MD, AD, RD).

Analysis of the additional diffusion metrics MD, AD, and RD revealed areas of significant (FWE-p<0.05) age-associated negative relationships, and no significant (FWE-p<0.05) positive associations in any metric. Parameter maps of FWE-p-values for each diffusion metric are illustrated in Fig. 4A. Including Sex as a covariate in the analysis revealed no significant sex by age interactions for any of the diffusion metrics (MD, AD, RD). In particular, it is noteworthy that the areas with negative associations with FA appear to be accompanied predominantly by significant negative associations in both MD and AD, with RD remaining mostly unchanged, in the temporal and frontal white matter areas, as well as in the corona radiata, internal capsule and corpus callosum. There were also additional brain regions with areas of significant (FEW-p<0.05) negative age-associations in MD and AD, that were not detected with FA, including the medial longitudinal fasciculus and the anterior commissure (Fig. 4B).

Parallel ROI-based analyses revealed similar patterns of diffusion metric differences across the age-range in our sample (Supplementary Figures S3–S6), broadly echoing the voxel-wise DTI data and illustrating downward trends with increasing age. For FA, across all 42 WM ROIs, our results found that 30 ROIs were best fit with linear models, while curvilinear models fit the remaining 10 ROIs (middle cerebellar peduncle, medial lemniscus, inferior cerebellar peduncle, cerebral peduncle, anterior limb of the internal capsule, posterior limb of the internal capsule, superior fronto-occipital fasciculus, uncinate fasciculus, dorsal prefrontal WM, and ventral prefrontal WM). There were 11 ROIs that exhibited sex differences in FA; while 6 did not show any interaction with age (corticospinal tract, superior cerebellar peduncle, superior cingulum, anterior commissure, and pyramidal tracts), 4 had a linear interaction (pontine crossing tract, perihippocampal cingulum, medial longitudinal fasciculus, and olivocerebellar tracts), and 1 (medial lemniscus) had a curvilinear interaction. For MD, 36 WM ROIs were best fit with linear models, and curvilinear models fit the remaining 6 (medial lemniscus, posterior limb of the internal capsule, anterior corona radiata, ventral prefrontal WM, dorsal posterior coronal radiata, and the medial longitudinal fasciculus). There were 18 ROIs that exhibited sex differences in MD; while 8 did not show any interaction with age (middle cerebellar peduncle, genu of the corpus callosum, body of the corpus callosum, splenium of the corpus callosum, fornix, superior corona radiata, stria terminalis, and dorsal prefrontal WM), 5 had linear interactions (corticospinal tract, cerebral peduncle, anterior limb of the internal capsule, superior fronto-occipital fasciculus, and the anterior cingulum), and 5 had curvilinear interactions (medial lemniscus, posterior limb of the internal capsule, anterior corona radiata, ventral prefrontal WM, dorsal posterior corona radiata, and the medial longitudinal fasciculus). For AD, 39 WM ROIs were best fit with linear models, and only 3 ROIs were best fit with curvilinear models (medial lemniscus, posterior limb of the internal capsule, and the medial longitudinal fasciculus). There were 12 ROIs that exhibited sex differences in AD; while 2 did not show any interaction with age (middle cerebellar peduncle and adjacent amygdala WM), 8 had linear interactions (splenium of the corpus callosum, corticospinal tract, cerebral peduncle, anterior limb of the internal capsule, superior fronto-occipitial fasciculus, dorsal prefrontal WM, ventral prefrontal WM, and the anterior cingulum), and 2 had curvilinear interactions (posterior limb of the internal capsule and the medial longitudinal fasciculus). For RD, 36 ROIs were best fit with linear models, and 6 were best fit with curvilinear models (medial lemniscus, superior longitudinal fasciculus, uncinate fasciculus, ventral prefrontal WM, dorsal posterior corona radiata, and the medial longitudinal fasciculus). There were 20 ROI that exhibited sex difference in RD; while 13 did not show any interaction with age (middle cerebellar peduncle, genu of the corpus callosum, body of the corpus callosum, posterior limb of the internal capsule, anterior corona radiata, superior corona radiata, sagittal striatum, superior frontal-occipital fasciculus, tapetum, anterior commissure, dorsal prefrontal WM, middle temporal gyrus, and adjacent thalamus WM), 4 had linear interactions (pontine crossing tract, corticospinal tract, cerebral peduncle, and the anterior cingulum) and 3 had curvilinear interactions (medial lemninscus, ventral prefrontal WM, and the medial longitudinal fasciculus). These results are summarized in Supplementary Table 1.

4. Discussion

This cross-sectional neuroimaging study evaluated metrics of brain aging in a balanced population of both female and male rhesus macaques ranging in age from 5 to 28 years. We collected the data over a narrow range of scan acquisition dates, using the same equipment configurations and pulse sequence settings for all animals, to enhance uniformity in our acquisitions. Gray matter morphology was assessed with TBM and T2w SPACE imaging, and white matter microstructure was explored using DTI. The high-resolution T2w SPACE data may offer sensitivity to effects of aging, such as iron accumulation, compared with T1w data^26,27. DTI provided a non-invasive method to gain further insight into microstructural features associated with white matter volumetric changes reported in our recent analysis of T1w images ¹⁴. Additionally, across both imaging modalities, our primary analyses involved voxel-wise regression analyses, which are not restricted by predefined areas and may offer greater spatial resolution to detect focal changes compared to the ROI-based methods used in previous NHP studies ^{7–9,12,14,17,18,21,22,24,25,38}.

The results of our TBM analysis revealed areas of age-related atrophy in the ventromedial prefrontal cortex, anterior and posterior cingulate cortex, sensory-motor cortex, and the caudate. Smaller focal regions of the dorsal prefrontal cortex, inferior temporal cortex, parietal cortex and the putamen showed similar age-related patterns of atrophy. These findings are consistent with our recent study in which age-related volumetric declines were observed in frontal and anterior cingulate ROIs, as well as in the caudate and putamen using T1w MPRAGE (Dash et al., 2023); but our T2w data also identified additional regions of age-related atrophy in the sensory-motor cortex, inferior temporal cortex, and parietal cortex. It should be acknowledged that these findings may be attributable to the different contrasts applied between the two studies (T1 vs T2). However, another possibility is that applying TBM to studies of brain aging rather than atlas-based parcellation approaches also contributed to these differences. TBM includes spatial normalization akin to the corrections applied for intracranial volume and generates brain-wide parameter maps at the same resolution of the MRI data that is suitable for voxel-wise statistical analysis, enabling the identification of sub-ROI findings. It is possible that the size of some of the ROIs used in previous studies were too large to capture the relatively focal changes observed here, particularly in regions of the temporal and parietal cortex. Another possibility is that these differences are related to the relative sensitivity of T1w versus T2w imaging, leading to these different interpretations. Supporting this possibility is the observation that our TBM analysis revealed one area with significant age-associated expansion, the substantia nigra. It is likely that increased iron deposition leads to age-associated T2w image contrast differences that contribute to this finding ³⁹. Supporting this are the sub-threshold log Jacobian increases observed in the internal and external globus pallidus (translucent red voxels in Fig. 2A, and translucent red bars in Supplementary Fig. S1); both areas also known to accumulate iron in aged rhesus macaques ³⁹. Taken together, these TBM findings help to complement the T1w results reported by Dash et al. (2023) by shedding additional light on the specific cortical and subcortical sub-regions that may be vulnerable to advancing age in this model.

The DTI results revealed significant microstructural alterations in cerebral white matter. Several white matter tracts showed significant age-associated differences in FA, including frontal and temporal white matter areas, as well regions of the internal capsule and corpus callosum. Age-related decreases in FA are commonly reported in both the rhesus macaque and human literature, with reductions shown in similar areas such as the superior longitudinal fasciculus, corpus callosum, internal capsule and the anterior commissure ^7,21–25,40. Analyses of the additional diffusion metrics AD, RD, and MD reported here provide additional context to this body of work, and reveal corresponding significant age-associated negative relationships, but no significant positive relationships, suggesting that age-associated differences in FA appear to be accompanied predominantly by differences in MD and AD. This is unexpected on two counts. First, white matter degeneration is often associated with increased diffusivity, reflected in increased MD and RD. Such increases have been observed with advancing age in both macaques and humans, typically complementing declines in FA ^7,41–43; instead, our sample appears to have an age-associated pattern of reduced MD. Second, examining AD reveals that there are widespread age-associated differences that correspond to the spatial pattern of FA changes, suggesting that the FA loss may be attributed to a strong influence of AD. It is worth noting that there are sub-threshold RD increases in some of the areas with FA loss (translucent red voxels in Fig. 4A, and translucent red bars in Supplementary Fig. S2), however these are not as extensive as the significant FA voxels, indicating that other diffusion factors, such as AD, are likely a stronger influence on FA in this study. Chen and colleagues reported similar patterns in rhesus macaques, however, RD demonstrated widespread age-related increases and substantial overlap with decreases in FA, while AD and MD exhibited more localized, but considerably overlapping, decreases ⁷. One possible interpretation is that, despite the relatively large sample for NHP studies, the imbalance in our age distribution favoring middle-age and older individuals may contribute to the differing results and interpretations. However, recent human brain charts studies using analogous DTI data have reported similar downward trajectories of diffusion metrics across the lifespan, with increases occurring more in the ~60–80y range⁴⁴. Bearing this in mind, another possible interpretation is that the age-span of animals in this study more closely aligns with the “healthspan” of humans, and therefore may not model the effects reported in geriatric clinical populations.

To expand on the trends reported here, histopathological readouts associated with these in-vivo diffusion findings are a worthwhile pursuit for future studies. In the current study, one possible interpretation is that the observed FA/AD decreases could indicate that axonal breakdown or inflammation is underlying the age-related white matter diffusivity differences. For example, decreases in AD without change in RD may reflect loss of axonal structure without accompanying myelin injury, or proliferation of microglia in white matter regions. These microstructural features could in turn create barriers to water movement and hinder longitudinal diffusivity ^45–47.

In comparison to the DTI findings, which were quite widespread, the morphological age differences in this model detected using TBM were more localized. This lack of robust age-related volumetric atrophy (and associated neuron loss) in specific regions of interest, like the hippocampus, has been reported in previous NHP studies both using MRI and histology ^{9–12,48–56}. Similarly, age-related alterations in glial cell populations within white matter regions have also been reported in histology studies ^57,58. Taken together with the DTI results reported here, these data suggest that microstructural alterations in white matter may represent a more sensitive biomarker of brain aging in this model.

There are, however, several important limitations to consider. First, although balanced for sex, the study had proportionally fewer young-adults than animals in the middle- and old-age ranges. It is possible that there are sex differences that emerge between young-adulthood and middle-age, for which our study population was less equipped to capture. Second, limitations of the voxel-based linear regression analyses must be acknowledged, as some brain regions may exhibit non-linear patterns of growth and decline over the lifespan ¹⁴. This is supported by our supplementary analyses, which revealed that age-related patterns in our sample in a number of WM ROIs were best-fit by curvilinear models. To facilitate more sophisticated data modeling efforts, and to increase the power and impact of this dataset, we will openly share these data via the Primate Data and Resource Exchange (PRIME-DRE)⁵⁹. This data-sharing effort will hopefully enable future work extending the recent structural “brain chart” projects in macaques⁶⁰ that can follow up on the results reported here in a sufficiently powered sample consisting of pooled multi-site data. Furthermore, large-scale longitudinal studies that track these brain metrics in the same individuals over large swaths of the lifespan would be incredibly valuable to more fully address this possibility and to begin to characterize patterns of individual variability, although significant resources are required to support such work.

Taken together, this study broadly suggests that aged macaques undergo morphological and microstructural alterations in cortico-striatal regions, as well as in the white matter fiber pathways that interconnect these areas, that are detectable using high-resolution T2w MRI and DTI. Application of these noninvasive imaging techniques in this highly translational animal model represents a pragmatic approach to studying the etiology of normative and pathological brain changes in elderly humans, thereby helping to lay the foundation for development of effective therapies.

Highlights.

• Age-related morphometric differences observed in cortico-striatal regions.

• Age-related differences in white matter diffusivity were detected.

• White matter differences were more extensive than gray matter differences.