Lifespan Trajectories of White Matter Changes in Rhesus Monkeys

Abstract

Progress in neurodevelopmental brain research has been achieved through the use of animal models. Such models not only help understanding biological changes that govern brain development, maturation and aging, but are also essential for identifying possible mechanisms of neurodevelopmental and age-related chronic disorders, and to evaluate possible interventions with potential relevance to human disease. Genetic relationship of rhesus monkeys to humans makes those animals a great candidate for such models. With the typical lifespan of 25 years, they undergo cognitive maturation and aging that is similar to this observed in humans. Quantitative structural neuroimaging has been proposed as one of the candidate in vivo biomarkers for tracking white matter brain maturation and aging. While lifespan trajectories of white matter changes have been mapped in humans, such knowledge is not available for nonhuman primates. Here, we analyze and model lifespan trajectories of white matter microstructure using in vivo diffusion imaging in a sample of 44 rhesus monkeys. We report quantitative parameters (including slopes and peaks) of lifespan trajectories for 8 individual white matter tracts. We show different trajectories for cellular and extracellular microstructural imaging components that are associated with white matter maturation and aging, and discuss similarities and differences between those in humans and rhesus monkeys, the importance of our findings, and future directions for the field.

Significance Statement: Quantitative structural neuroimaging has been proposed as one of the candidate in vivo biomarkers for tracking brain maturation and aging. While lifespan trajectories of structural white matter changes have been mapped in humans, such knowledge is not available for rhesus monkeys. We present here results of the analysis and modeling of the lifespan trajectories of white matter microstructure using in vivo diffusion imaging in a sample of 44 rhesus monkeys (age 4–27). We report and anatomically map lifespan changes related to cellular and extracellular microstructural components that are associated with white matter maturation and aging.

Introduction

The ability to identify and monitor brain changes occurring during development, maturation, and aging is fundamental to our understanding of brain function and our ability to differentiate between changes associated with normative development from those induced or alleviated by a certain procedure, drug, or pathology. Historically, gray matter was the primary focus in developmental brain research. However, advances in in vivo imaging over the past few decades have brought methodologies that allow for quantification of white matter integrity (Holland et al. 1986; Basser et al. 1994) and for investigating the involvement of white matter in various neurodevelopmental disorders. This has placed white matter at the “center stage” of research on brain development (Kochunov et al. 2014).

Lifespan trajectories of white matter throughout the course of maturation and aging have been recently mapped in humans through in vivo neuroimaging (Giedd et al. 1999). Studies show that white matter volume slowly expands between birth and 20 years of age, after which, it gradually decreases with age (Giedd et al. 1999; Knickmeyer et al. 2008). Greater insight into the anatomical specificity of lifespan changes and their neurobiological correlates has recently been made possible with the introduction of diffusion tensor imaging (DTI). To date, several DTI studies conducted in large human populations across the lifespan (Lebel et al. 2010, 2012; Yeatman et al. 2012, Westlye et al. 2010, Kochunov et al. 2011, 2012) were able to reveal lifespan trajectories for fractional anisotropy (FA), the most common DTI metric of white matter microstructure frequently associated with myelin integrity (Beaulieu 2002). These studies show that FA in white matter gradually increases until it reaches a peak at 25–30 years of age, after which it slowly decreases (Lebel et al. 2010; Kochunov et al. 2011; Lebel and Beaulieu 2011; Yeatman et al. 2012). However FA has also been shown to be modulated by changes in tract coherence (Beaulieu 2002), fiber geometry (Whitford et al. 2011) axonal size, or edema and inflammation (Alexander et al. 2007). In humans, age-related increases in FA during maturation have been linked to increased myelin content and greater fiber packing (Beaulieu 2002), whereas reductions in FA have been associated with the breakdown of axonal myelin (Marner et al. 2003) and chronic neuroinflammation in aging (Godbout and Johnson 2006; Ownby 2010). Yet, the neurobiological specificity of these imaging measures is still limited as there are few studies, which aim to validate the biological underpinnings of these microstructural imaging measures.

To date, postmortem investigations still remain the gold standard for the characterization of microstructural changes related to physiological and pathological processes affecting the brain. However, such investigations in humans are often limited by the heterogeneity of the studied population, small sample sizes, and a focus on very localized brain regions. Therefore, animal models can serve as very attractive alternatives for a number of reasons. First, the ability to regulate environmental conditions minimizes the impact of confounding environmental variables and, second, animals tend to have less anatomical variability of brain regions across populations compared with humans, thereby limiting intersubject heterogeneity. Lastly, it is easier to collect complete behavioral, imaging, and brain tissue data from animal models because there is little to no attrition and data can be collected within a predefined period of time. For these reasons, animal models can serve a very powerful role in translational research, particularly when investigating longitudinal trajectories of structural brain changes.

One of the most widely used animal models for the investigation of both normal brain development and neurodevelopmental disorders, such as autism, Attention Deficit Hyperactivity Disorder (ADHD), or schizophrenia, is the Rhesus macaque (Austad 1997; Machado and Bachevalier 2003; Machado et al. 2009). The average lifespan of a rhesus monkey is 25 years, with some living as long as 40 years (Dyke et al. 1986; Tigges et al. 1988; Finch and Austad 2012), suggesting that there is a ratio of rhesus to human years of about 1:3. It has been argued, however, that this ratio might be closer to 1:4 in earlier stages of development and additionally complicated by the sex-differences in reaching developmental milestones. Studies show that female rhesus monkeys reach puberty around 30–40 months (Wilson et al. 2013) and can have offsprings from 3 years of age (Bercovitch et al. 2000) whereas males reach puberty slightly later at 4 years of age (Dixson and Nevison 1997; Bercovitch et al. 2003) but most are not reproductively successful until after 8 years of age or when they reach adult size (Bercovitch et al. 2000). Therefore, the 1:3 ratio is only an approximation, as it does not map directly to lifespan trajectories of other physiological systems, such as the reproductive system, where fertility periods in monkeys are much longer than in humans, spanning from 3 to mid 20s (Walker & Herndon 2008). On the other hand, Rhesus monkeys lifespan trajectories of cognitive changes are similar to these observed in humans (Bachevalier 1993; Herndon et al. 1997), such that cognitive abilities tend to be more greatly affected with increased age (Herndon et al. 1997). Interestingly, recent histological studies show that the aging process in rhesus monkeys correlates more strongly with an overall reduction in white matter volume as opposed to gray matter volume (Peters and Rosene 2003; Wisco et al. 2008) or neuronal number (Peters et al. 1998). In fact, gray matter volume and neuronal number exhibit little or no change with aging (Lai et al. 1995; Peters et al. 1998; Giannaris and Rosene 2012). These findings directly contrast the human literature, which primarily describes a gradual decrease of both cortical gray and white matter volumes as a function of age (Giedd et al. 1999; Nitin Gogtay 2010).

With respect to the brain development phase, postmortem histological studies in the macaque show that synaptic density increases rapidly in the first several months after birth throughout the cortex, after which time the number of synapses begins to decline (Rakic et al. 1986). While in the motor cortex synapse reduction occurs until 10 years of age (Zecevic et al. 1989), within the prefrontal cortex there is an extended plateau in synaptic density with a gradual decrease beginning around puberty (Bourgeois et al. 1994). In a structural MRI study conducted by Knickmeyer et al. (2010) on 37 rhesus monkeys (20 males and 17 females) ranging from 10 to 64 months of age, the authors reported that unlike cortical maturation patterns in humans, the GM volumes within the macaque did not show a postpubertal decline in most cortical regions (occipital, parietal, and temporal limbic areas). The notable exception was the prefrontal cortex, suggesting that these are early maturing areas in the macaque. While the authors did not observe any significant maturational change in total GM after 1–2 years of age in the rhesus macaque, WM maturation was observed to be more similar to that in humans, such that WM volumes increased through puberty and into adulthood. In a separate cross-sectional diffusion tensor imaging study (Shi et al. 2013) characterizing macaque brain neurodevelopment between 1 and 6 years of age in 25 healthy rhesus monkeys (14 males and 11 females), the authors observed substantial increase in FA and a decrease in radial diffusivity (RD) for white matter.

Due to the similarities in cognitive maturation and aging trajectories between humans and rhesus monkeys, and the relative lack of changes in gray matter volume beyond very early maturation phases, it could be suggested that age-related increases as well as decreases in cognitive abilities may be more associated with the changes in white matter rather than gray matter. However, there are a limited number of longitudinal studies in monkeys that look at lifespan trajectories of white matter. If we are to understand the translational value of the Rhesus monkey model as a valuable alternative to human postmortem studies, it is necessary to increase our current understanding of the similarities and differences between the lifespan trajectories of each species.

Presently, lifespan trajectories of structural brain changes in Rhesus monkeys have not yet been established. In this study, we utilize diffusion magnetic resonance (dMR) data from 44 rhesus monkeys that span a wide range of ages, from 4 to 27 (corresponding to 12–81 in humans), which allows us to describe lifespan trajectories of FA for the entire brain, as well as for specific white matter structures (tracts). In addition, since FA has been shown to be sensitive, but not biologically specific, we complement FA lifespan trajectories with the trajectories of 2 measures derived from a “free-water” model (Pasternak et al. 2009): the FA of the tissue (FA-t) and the fractional volume of extracellular free water (FW). The FW model extends DTI by accounting for a component of freely diffusing water that can only originate from extracellular spaces. The FA-t parameter eliminates the contribution of the extracellular FW, and is therefore more specific than FA to changes that occur in the brain tissue itself in the vicinity of cellular membranes. At the same time, the FW parameter can quantify other types of changes, such as atrophy and neuroinflammation that mainly affect the extracellular space (see the review by Pasternak et al. 2016). The FW model and associated diffusion indices have recently received a lot of attention, and have been demonstrated to be sensitive to white matter pathology in several diseases, including schizophrenia (Pasternak et al. 2012; Lyall et al. 2017) depression (Bergamino et al. 2016) Parkinson’s disease (Burciu et al. 2016; Planetta et al. 2016), and Alzheimer’s disease (Maier-Hein et al. 2015). We then compared the findings in our analysis to lifespan trajectories that have been described in humans (Lebel et al. 2012). This will provide lifespan trajectory models of biologically specific in vivo measures of white matter microstructure, which, in turn, will enable better utilization of rhesus monkeys as animal model for studying developmental brain pathology.

Materials and Methods

Demographics

All monkeys in this study were part of larger program investigating age-related cognitive impairment. The monkeys were all selected from the colonies of the Yerkes National Primate Research Center at Emory University (http://www.yerkes.emory.edu). All animals are singly housed in individual cages that were located in a colony room. Full health records were available for all animals and were examined to verify that all animals had known birth dates and were free from any experimental manipulations or chronic illnesses that may affect brain function. In addition all animals were given clinical MRI scans to ensure there was no occult brain damage. In this study, MRI scans were collected from a total of 44 monkeys, 26 females and 18 males. The average age was 15.14 years old (females 17.58 and males 11.61) and ranged in age from 4 to 27 years. Hence, the sample would be approximately equivalent to studying humans between 12 and 81 years of age.

MRI Acquisition

MRI scans were collected on all animals as described below. For image acquisition, monkeys were anesthetized with a mixture of ketamine (10 mg/kg) and xylazine (0.10 mg/kg) and placed into an MRI compatible stereotactic machine that fixed their head in the standard coronal plane (Saunders et al. 1990). During scanning, heart rate and respiration were monitored, muscle tone was assessed, and additional half doses of anesthetic were administered to maintain anesthesia and avoid movement artifact. Imaging data was acquired on 2 different 3 T scanners, one located at the Martinos’ Center, Massachusetts General Hospital (MGH) (Siemens Allegra), the other in the Boston University (BU) Medical Center (Phillips Achieva). Overall, 23 animals were scanned at BU and 21 animals at MGH, 26 females (14 at BU and 12 at MGH), and 18 males (9 at BU and 9 at MGH). All monkeys received 3D T1-weighted high-resolution anatomical scan, which was used for data registration and distortion correction, and diffusion-weighted (DT-MRI) scan. For the Siemens scanner: the MP-RAGE: relaxation time (TR) = 2530 ms, echo time (TE) = 3.36 ms, inversion time (TI) = 1100 ms, flip angle = 7°, 128 coronal slices with 0.78 mm thickness (no gap), data matrix = 256 mm × 256 mm, field of view (FOV) = 200 mm × 200 mm, bandwidth = 200 Hz/pixel and a total imaging acquisition time of 8 min. The DT-MRI data were constructed by obtaining a 7-shot acquisition (b = 0 s/mm², and 6 diffusion spatial encoding directions) with: TR = 6600 ms, TE = 77 ms, flip angle = 90°, 60 coronal slices with 1.3 mm thickness (no gap), data matrix = 64 mm × 64 mm, FOV = 83 mm × 83 mm, bandwidth = 2005 Hz/pixel, diffusion sensitivity b = 600 s/mm², a total imaging acquisition time of 9 min and 35 s. For the Philips scanner: the MP-RAGE: TR = 7.09 ms, TE = 3.17 ms, flip angle = 8°, 130 coronal slices with 0.6 mm thickness (no gap), data matrix = 256 mm × 256 mm, FOV = 153 mm × 153 mm, bandwidth = 241 Hz/pixel and a total imaging acquisition time of 8 min. The DT-MRI data were constructed by obtaining a 16-shot acquisition (b = 0 s/mm², and 12 diffusion spatial encoding directions) with: TR = 14500 ms, TE = 53 ms, flip angle = 90°, 80 coronal slices with 1 mm thickness (no gap), data matrix = 64 mm × 63 mm, FOV = 63 mm × 63 mm, bandwidth = 1602 Hz/pixel, diffusion sensitivity b = 1000 s/mm², a total imaging acquisition time of 5 min and 23 s.

Data Processing

The data was preprocessed using slicer 4 software (http://www.slicer.org), (Fedorov et al. 2012) and manually/visually checked for quality. Out of 65 available datasets, 17 were deemed unusable because of excessive motion, aliasing or very low signal-to-noise ratio (SNR) and 4 were later removed from analysis due to registration errors. This procedure was followed by motion and eddy current distortion correction and skull-stripping using Functional Magnetic Resonance Imaging of the Brain (FMRIB) Software Library (FSL) software (http://www.fmrib.ox.ac.uk/fsl) (Smith 2002). Brain masks were then manually edited by a trained research assistant. Next, since the DTI data was spatially distorted, an additional step was introduced into the procedure that involved echo planar imaging (EPI) distortion correction. Since T2-weighted images were not originally acquired as part of the data acquisition, the T2-weighted (T2W) skull stripped images were created by intensity reversal of T1-weighted (T1W) scan using in-house Matlab scripts. These generated T2W scans were then linearly registered to the DTI image in order to bring T2W image to diffusion space. The DTI image was then nonlinearly warped in one dimension of diffusion space to align with the T2W scan in the best-suited direction using Advanced Normalization Tools (ANTs) registration (Avants et al. 2014) to create the best possible distortion-corrected diffusion image. Following these preprocessing steps, FA was estimated using least square fit Slicer procedure. FW and the FA of the tissue (FAt) were also calculated for each voxel using in house software (Pasternak et al. 2009) (see Supplemental material for an axial section of the FA, FAt and FW maps). Diffusion data was then subjected to 2 separate analytic pipelines: tract-based spatial statistics (TBSS) (Smith et al. 2006) for whole brain voxel-wise analysis of DTI measures; FA, FAt, and FW, and ROI analysis for tract-specific analysis of the same DTI measures. In the human developmental literature, FA is most sensitive but nonspecific diffusion measure and is usually termed an index of white matter “integrity,” FW reflects extracellular water content, which can occur due to either neuroinflammation or cell loss, while FAt has been suggested to be sensitive to any changes that occur in the vicinity of the tissue, such as myelin changes or axonal growth and degeneration.

Data Analysis-Whole Brain

To calculate whole brain diffusion measures, we utilized a standard TBSS pipeline. Several studies have successfully used TBSS FSL tool for whole brain analysis of rhesus monkey diffusion data (Chen et al. 2013; Howell et al. 2014; Meng et al. 2014, 2017). We modified the original TBSS scripts by omitting registration of all cases into the MNI space. Instead, we selected one of the animals to be used as a template for registration. This way, the resolution of the final registered animal data was the same as the original resolution, instead of the 1 × 1 × 1 mm³ used for the MNI template. We have previously used this approach for TBSS analysis in rats (Kikinis et al. 2017). This method is designed to nonlinearly register FA, FW, and FAt maps of each individual subject to a preselected template (i.e., the most representative subject), and create a “white matter skeleton,” on which diffusion parameters are projected. Then diffusion parameters for each skeleton (FA, FAt, and FW) are averaged, which results in one value per whole brain skeleton.

Data Analysis-Region of Interest

To calculate diffusion measures for each of the anatomical regions, diffusion-weighted images of each individual subject were normalized using ANTs registration to a preselected template (i.e., the most representative subject). The Center for Morphometric Analysis (CMA) atlas (Makris et al. 2010) (Fig. 2), was nonlinearly registered to the template space, using ANTs. Diffusion measures (FA, FAt, and FW) were first computed separately for the left and the right for all the tracts except CC and anterior commissure, and then, to match the human studies, left and right values were averaged. Eight of the largest white matter structures (fiber tracts) were identified using an atlas, that is, superior longitudinal fasciculus (SLF) II, III, cingulum bundle (CB), anterior commisure (AC), corpus callosum (CC), anterior and posterior limbs of the internal capsule (ALIC and PLIC).

Figure 2.

Plots for rates of the annual percentage change in indices related to FA, FAt, and FW (note the largest rate of increase of diffusion indices during the development and maturation period—1% per year on average for the FA, 1.5% for the FAt, and 0.6% for the FW).

Statistical Analysis

FA, FW, and FAt measures were harmonized to account for scanner variation in the data because the MR images were acquired on 2 different scanners. Similar to using a statistical covariate method, we used a more explicit computation to decouple the data from the effect of aging in order to determine the scanner variation. More specifically, we normalized the DTI-derived FA from both sites by applying a linear shift to the group average FA from the BU site. A similar approach was taken for the FW model derived parameters of FW and FAt. This normalization ensured that there was no group difference between the 2 sites and normalization was done for each ROI separately. Such a data normalization scheme has been used elsewhere previously in larger multisite studies (Pohl et al. 2016). Diffusion measures for whole brain or each ROI tract were averaged across the subjects from the age of 5–10 for both BU and MGH data. The scaling factor between these 2 datasets was then calculated and multiplied with all the diffusion measures for BU data to remove the scanner effects. These DTI measures computed for the whole brain and each tract-specific ROI were then fit to Poisson curves according to the following equation FA (or FAt or FW) = C + A × age × exp^{−B × age} (Lebel et al. 2012) to model the age-related changes. Parameters were estimated for the best fit and were assessed for significance using F-tests. The standard error of each estimated fitting parameter was determined and the derivative of the best-fit equation was used to calculate the age at which FA/FAt reaches its maximum or FW reaches its minimum: FA_peak (or FAt_peak or FW_min) = 1/B (Lebel et al. 2012). The error in this peak, or minimum age, was computed by recalculating the DTI measure by replacing B with B + standard error of B (Lebel et al. 2012). The percent change in FA/FAt and FW from (1) age 4 years to peak age and (2) peak age to 27 years, was analyzed for each tract. An opposite trend in FW values was also determined by calculating percent decrease in FW (1) from 4 years to minimum age and (2) increase in FW from minimum age to 27 years. In order to investigate the change in DTI measures per year for the whole brain, the rate of FA/FAt increase/decrease and the rate of FW decrease/increase, as well as percent rate of change in FA/FAt/FW with age were computed.

DTI measures were also fit to quadratic curves using the equation: FA (or FAt or FW) = A × age + B × age² + C (Lebel et al. 2012). Poisson curve was deemed to be a better and more robust fit than quadratic fit to model changes of DTI measures across the age span. This was determined after comparing the errors in peak/minimum age estimates for both curves for the entire brain (standard errors for the quadratic vs. Poisson fits for FA: 12.8 vs. 1; for FW: 12.3 vs. 0.9; for FAt: 12.7 vs. 1.2). Similar to Lebel et al. (2012), standard error in peak/minimum age estimates was computed using equations derived from the delta method separately for both Poisson and Quadratic models (Oehlert 1992). For the Poisson model: peak/min Age = 1/B, where, b is the exponential fitting parameter in the Poisson equation. The error in peak/minimum age estimates for Poisson model was calculated using an equation derived from delta method: SE = sqrt (−(1/B²) × variance(B) × (1/B²)) (Oehlert 1992). For the Quadratic model: peak/min Age = −A/(2 × B), where A and B are the fitting parameter estimates of quadratic model: DTI measure = A×age + B×age² + C. Error in peak/min age estimates was again calculated using delta method: SE = sqrt (−(1/(2 × B)) × variance(A) × −(1/(2 × B)) + A/(2 × (B²)) × variance(B) × A/(2 × (B²)) + −(1/(2 × B)) × covariance(A,B) × A/(2 × (B²))).

Results

We modeled the lifespan trajectories of FA and the 2 measures derived from the FW model: FW and FAt. The FA, FAt, and FW measures were obtained from cross-sectional in vivo measurements of 44 rhesus monkeys ages between 4 and 27 years (mean 14.97). For each animal, we obtained separate measures: (1) FA, FAt, and FW averaged over the entire white matter skeleton and (2) FA, FAt and FW averaged over each of the 8 major white matter tracts.

Whole brain FA analysis, which represents the estimate of a global trend of age-related brain changes, shows that the rhesus monkey exhibits an asymmetric lifespan trajectory characterized by a relatively rapid increase within the first decade of life, reaching a peak at the age of 11.3 years (SE = 1), followed by gradual decrease of FA values (Fig. 1). The average increase of FA during the development and maturation period, (i.e., before the peak), is 1% per year, with the fastest rate of change being 3% a year in the youngest period. The average decrease of the FA during the “aging” period, (i.e., following the peak), is 0.2% per year (Fig. 2). The lifespan trajectory of FAt is similar to FA with the peak reached at the age of 11.2, but with a more accentuated steeper average annual increase of 1.5% per year before the peak and the fastest rate of change being 6% during the youngest period. Similar to FA, FAt exhibits an average of 0.2% decrease after the peak. Finally, the lifespan trajectory of FW gradually decreases initially (by 0.6% per year on average) until the age of 10.9 years (SE = 0.9), after which it increases on average by 0.1% per year.

Figure 1.

Lifespan Trajectories of changes of DWI indices (FA = fractional anisotropy, FAt = FA corrected for free water, free water = FW) with age. Each red dot represents diffusion measure averaged over the entire brain skeleton for each animal. The Blue line represents the best fit (Poisson), with the black vertical line representing peak of the curve (note the FA peak at the age of 11.3, FAt at the age of 11.2, and the FW at the age of 10.9).

Following the whole brain analysis, specific anatomical structures (fiber tracts shown in Fig. 3) were extracted after superimposition of the CMA anatomical atlas of white matter tracts (ROIs of the CMA atlas sampled from 3 midbrain locations are shown in Fig. 4) onto the individual diffusion image. This subsequent analysis, similar to the human literature, has revealed significant differences between speed and slope of maturational trajectories among separate fiber tracts (Fig. 5). Table 1 contains detailed information about peaks (in years) for each individual structure, as well as the rates of increase and decrease of diffusion parameters in percentages. There are apparent differences in maturational FA and FAt trajectories between individual tracts. For example, the CB demonstrates a maturational trajectory that is most similar to the overall whole brain white matter trajectory of FAt change, such that it reaches a peak around 12 years of age and exhibits the most dynamic increase (as a function of development and maturation) of 12% before the peak, as well as decrease (as a function of degeneration and aging), by 12% overall after the peak. PLIC as well as CC express similar lifespan trajectory shape, and peak around the same age. ALIC is the tract that maturates earliest, as it peaks at 6.6 years, and it also exhibits fastest decline with age (18% overall post peak). SLF II, SLF III and AF are the tracts that peak earlier than the average for the white matter (9.9, 9.9, and 9.7 years, respectively), while the AC is the only tract that does not follow common trajectory, instead the FAt continues increasing throughout the entire lifespan. It is also worth noting, that the highest values of the FAt (0.7) are reached in the CC, followed by the ALIC and PLIC, which are in both humans and NHP large and well myelinated fiber tracts. As for the trajectories of extracellular FW measures, their shapes for the individual tracts are quite similar to each other, all exhibiting significant, gradual increase with age of about 21–65% overall (with the exception of AF, which does not show the increase). Since FW has not been investigated in the context of maturation and aging in humans, the Supplemental material includes a figure, depicting lifespan trajectories of changes for the MD measure over the entire skeleton, as well as for each of the individual tracts.

Figure 3.

Diffusion tractography was used to reconstruct 8 major white matter fiber bundles of rhesus monkey brain measured in this study. Colors correspond to the 3D reconstructions of each structure.

Figure 4.

Three representative coronal sections (with levels indicated on the sagittal scout) of the rhesus monkey brain demonstrating location and extent of the white matter regions of interest used for the tract measurements. ROI used are part of the Center for Morphometric Analysis (CMA) Rhesus White Matter Atlas.

Figure 5.

Schematic of the age trajectories of DWI indices for white matter bundles. Each color represents individual anatomical structure- white matter fascicle. Each trajectory represents a best statistical fit (Poisson).

Table 1.

Peaks, their standard errors and the lifespan percent changes of DWI indices for each white matter bundle estimated using Poisson fit

Tracts	FA				FW				FAt
	Peak (years)	SE	Overall % increase before peak age	Overall % decrease after peak age	Peak (years)	SE	Overall % decrease before peak age	Overall % increase after peak age	Peak (years)	SE	Overall % increase before peak age	Overall % decrease after peak age
CC	10.8	1.1	12	15	11.6	0.6	28	44	9.1	2.1	3	8
CB	11.8	0.7	15	14	8.3	4.1	7	24	11.8	0.8	12	12
SLF II	9.3	1.3	4	9	6.4	7.5	2	22	9.9	1.1	5	8
ALIC	6.9	1.4	3	21	8.6	1.8	13	46	6.6	1.6	2	18
PLIC	11.3	1.5	8	9	9.4	1.3	22	65	10.9	1.8	4	6
SLF III	9.9	1.2	8	15	7.3	5.7	4	21	9.9	1.4	8	4
AC	12.9	7.3	4	3	7.8	12.3	10	46	5.9	9	−0.5	−6
AF	9.2	1.9	4	9	20.7	16.3	24	2	9.7	1.9	4	9

Discussion

To the best of our knowledge, this is the first diffusion imaging study that characterizes in vivo lifespan trajectories of white matter changes that occur in early childhood to late adulthood in nonhuman primates. Our study reveals that the overall trajectory of white matter changes, as measured by most common diffusion index FA, mirrors that observed in humans (Lebel and Beaulieu 2011). FA peak around the age of 11.3 in rhesus monkeys, corresponds to age of 34 in humans, and is consistent (if not a little higher) than the human reports of FA trajectories (Lebel et al. 2010; Lebel and Beaulieu 2011; Yeatman et al. 2014). This further suggests that the nonhuman primate model of human white matter maturation and aging is valid for rhesus monkeys.

By using 2 independent analytic approaches to white matter analysis, that is, a whole brain, skeletonized analysis, as well as a tract-specific analysis, we provide both generalized and anatomically specific trajectories of lifespan changes in white matter. In fact, tract-specific profiles may be useful in acquiring a better understanding of the variance in cognitive performance during maturation and aging. While the FA trajectory establishes the comparability with previous human studies, it is important to remember that FA has been shown to be modulated by changes in myelination, but also tract coherence (Beaulieu 2002), fiber geometry (Whitford et al. 2011) axonal size, or edema and inflammation (Alexander et al. 2007). Thus by applying another, more specific model for diffusion imaging (“free-water” model), we provide separate, but complementary, information about lifespan changes taking place in both the cellular and extracellular compartments. With the FW model, we demonstrate that the cellular and extracellular compartments are characterized by different time-courses, possibly tapping into 2 distinct biological processes associated with white matter maturation and aging. The FAt trajectory, which is proposed to be more specific to changes in tissue (cellular) integrity, reaches its peak at the age of 11.2 years (corresponding to human 33 years), suggesting that the changes in cellular (axon and myelin) size and density might be driving the biological process of maturation during the first 4 decades of life in humans. The FAt measure appears to be more sensitive to cellular changes during this period than the FA, that is, while they both peak around the same time (11.2 years for FAt vs. 11.3 years for FA), FA increases at the rate of 3% per year initially, comparing to 6% per year for the FAt. The FW measure seems to be the least affected by both maturation (less than 1% increase annually) and aging (0.1% annually) (Fig. 2). In fact, both FA and FAt measures also do not exhibit marked changes during the aging phase, both decreasing only at the rate of 0.2% per year.

Rhesus monkeys have been used predominantly as animal models for understanding the nature of anatomical, neurobiological, and neurocognitive dysfunctions in various human neuropsychiatric diseases, as well as for testing various interventions in preclinical trials (Roth et al. 2004). The majority of such experiments, however, either ignores the effect of aging, or assumes that this effect is linear and attempt to include age as a covariate in statistical analyses. Recent neuroimaging research have established that the lifespan trajectories of white matter changes in humans are nonlinear (Lebel et al. 2010, 2012; Yeatman et al. 2012, Westlye et al. 2010, Kochunov et al. 2011, 2012). Our study indicates that this is also the case for nonhuman primates, and suggests that the assumptions about linearity of age-related changes on white matter should be revisited in any statistical analyses that utilizes either longitudinal or cross-sectional designs with a wide age range.

The similarity of the rhesus white matter trajectories to those of human confirms previous assumptions that the age-related effects in rhesus monkeys are proportional to those in humans, when multiplied by the factor of 3. The “aging,” or descending, part of the lifespan FA trajectory is in with agreement with previous volumetric studies that show the aging process in rhesus monkeys correlates with an overall, gradual reduction in white matter volume (Peters and Rosene 2003; Wisco et al. 2008). Interestingly, those studies also show little or no changes in gray matter volume (beyond very early phase (Knickmeyer et al., 2010)), placing it in contrast with human literature, which describes a gradual cortical volume decrease as a function of increased age after arriving at the maturational peak (Giedd et al. 1999; Gogtay 2010). Although conducting an in-depth analysis of the relationship between gray matter and white matter during development and aging was beyond the scope of this work, the issue is extremely interesting, and further studies are needed to better understand differential neurobiological changes appearing in gray and white matter during maturation and aging in humans and nonhuman primates.

As discussed above, age-related white matter alterations in both humans and rhesus monkeys have been associated with axonal growth and myelination on one hand, and breakdown of axonal myelin (Marner et al. 2003), as well as neuroinflammation (Godbout and Johnson 2006; Ownby 2010) on the other. Thus, attempts to separately model changes affecting cells from those related to extracellular spaces (as done here) can potentially improve our understanding about the time-course of neurobiological processes associated with maturation and aging. FAt, representing the cellular component of the diffusion MR signal, seems to closely follow previously reported trajectories of cognitive abilities such that it increases sharply during adolescence into the third, or even fourth, decade of life, after which it gradually declines (Park and Reuter-Lorenz 2009). Therefore, based on our results, FAt appears to serve as a potentially useful imaging correlate of cognitive maturation and cognitive aging. FW, which represents the extracellular diffusion signal, exhibits a relatively flat trajectory both before and after it reaches its peak. Our results thus demonstrate that brain maturation is primarily associated with changes in the “cellular” diffusion signal, which is most likely influenced by the growth of axons and myelination. This process varies regionally, with ALIC, PLIC and CC, that is, tracts containing the largest, most myelinated axons of the brain, reaching highest FAt values. The CB matures later than other tracts, whereas the ALIC shows the largest degree of deterioration with aging. The AC not appear to follow the common trajectory of other tracts in this study, that is, FAt increases throughout the entire lifespan. Interestingly, while overall, FW shows very little change in extracellular compartment, we do observe variations in this trajectory by tract, with ALIC, PLIC, and CC demonstrating relatively large increases of FW during aging. Those later changes are most likely driven by documented biological aging processes such as chronic neuroinflammation, expansions of the Virhov-Robin spaces, and loss of cellular processes (Kumar et al. 2007; Chen et al. 2011; Franceschi and Campisi 2014).

While neuroimaging features such as FA, FAt, or the FW can serve as indirect representations of biological mechanisms and events that influence brain maturation and aging, the neurobiological specificity of MRI and DMRI remains limited. The MR signal is averaged over several cubic millimeters (the resolution of single “voxel”). The amount of “free” and bound water, as well as the macromolecules within the voxel, can influence this signal. Since brain tissue is comprised of cell bodies (neuronal and non-neuronal), axons, dendrites, synapses, myelin, vasculature, and extracellular space, it is important to consider these elements when interpreting imaging data because these components are far too small to be differentiated in vivo in MRI data available today. Therefore, it is important to stress that all of the measures utilized in this study should be considered indirect surrogates of underlying neurobiological changes occurring in an aging nonhuman primate brain. In addition to the relative lack of specificity of neuroimaging biomarkers, the direct comparison of human and rhesus monkey data requires caution. It is important to notice that the size (i.e., volume) of the rhesus monkey’s brain is approximately 8 times smaller than this of human. To account for such size differences, experiments directly comparing imaging results between humans and rhesus monkeys would have to use 8 times higher spatial resolution, which is not currently achievable in large clinical scanners where in vivo experiments are currently being conducted. The data available, thus far, in monkeys is therefore more affected by partial volume effects than data collected in humans, making comparisons between anatomical structures or biological compositions of voxels in monkeys and humans difficult. Additional potential differences between human and monkey scans can stem from the fact that animals are scanned under general anesthesia. Recent studies suggest that extracellular water volume can expand during sleep and anesthesia (Xie et al. 2013) which might be possibly affecting diffusion signal. Another limitation of the present study is that the diffusion data was acquired on 2 different scanners over the span of 10 years. Despite collecting large samples on each scanner and normalizing data using 5 monkeys of the same age scanned at both scanners, there is still the chance of a between-scanner bias or within-scanner signal drift. Additionally, our sample sizes are too small and distributions between scanners too skewed for us to investigate sex effects. Thus, while we show information regarding both scanner and sex for our population in the Supplementary Material, our results should be treated with caution.

In summary, our data provides the first in vivo benchmark for microstructural changes in rhesus monkeys across the lifespan. We present and discuss local specificity and longitudinal trajectories of 3 measures derived from dMRI data, one nonspecific measure of white matter tract “integrity” (FA), and 2 more biologically specific, microstructural components of white matter maturation and aging: cellular- related to myelin development and breakdown (FAt), and extracellular, related to atrophy and neuroinflammation (FW). Such normative data should prove useful in applying to animal models to better our understanding of anatomy and biology of brain changes associated with neurodegenerative and neurodevelopmental diseases. A greater knowledge of normative trajectories of white matter along the lifespan should be also extremely useful in monitoring effects of pharmacological, surgical, or cognitive therapies in preclinical trials in nonhuman primate models. Future work should focus on creating large and flexible normative databases that would combine neuroimaging methodologies, immunohistology, and neurocognitive performance into integrative profiles of brain development, maturation and aging of rhesus monkeys. Our study constitutes a first step in this direction.

Funding

Grants from the National Institutes of Health (R01AG042512, RO1AG043640, P01AG000001, R01MH102377, K24MH110807, R01MH112748, R21AT008865, R01MH097979, and R01MH111917).