This study utilized tissue from eight male rhesus monkeys (Macaca mulatta) that were part of previous studies on normal aging (Table 1). The monkeys were obtained from national primate research centers and all had known birth dates and complete medical records. Monkeys were housed at the Laboratory Animal Science Center at Boston University Medical Center (BUMC), which is fully accredited by the Association for the Assessment and Accreditation of Laboratory Animal Care (AAALAC). All procedures were performed according to protocols reviewed and approved by the BUMC Institutional Animal Care and Use Committee. All monkeys underwent a battery of behavioral tests to assess cognitive function (Herndon et al., 1997; Moss et al., 1997; Moore et al., 2006). Once testing was completed and prior to perfusion, all monkeys received magnetic resonance imaging (MRI) scans to ensure that there was no occult brain damage.

Monkeys were tranquilized with ketamine (10 mg/kg, intramuscular), and deeply anesthetized with sodium pentobarbital (15 mg/kg, intravenous to effect), before euthanasia by exsanguination during transcardial perfusion of the brain. All monkeys were perfused with 4% Krebs buffer at 4°C, followed by 4% paraformaldehyde in phosphate buffer (0.1 M, pH 7.4) at 37°C. Brains were blocked in situ, in the coronal plane just behind the splenium, removed from the skull, weighed and post-fixed overnight in 4% paraformaldehyde for no more than 18 h, and transferred to cryoprotectant solution to eliminate freezing artifact (Rosene et al., 1986). Blocks were flash frozen at −75°C and stored at −80°C until they were cut on a microtome. The cutting protocol consisted of a repeated series of coronal sections, eight of which were 30 μm thick and one 60 μm thick, resulting in a spacing of 300 μm for sections within each series. Then, one half-series of sections from these animals was used for the current study. Within this half series, we used 3–4 sections from each animal for each field, FC, Par, and TE (Figure 1), corresponding to sections from Bregma: +8.10, +0.05, −5.85, and −10.35 mm (Paxinos et al., 2000).

Other tissue was available from a previous study (Rockland and Nayyar, 2012) in adult macaque monkeys, which had been histochemically reacted for NADPH- diaphorase. This served as a useful confirmation for soma size and shape.

Sections from each of the eight subjects were thawed together and rinsed three times for 5 min in 0.05 M Tris-buffered saline (TBS) to wash off the cryoprotectant. All sections were then processed at the same time and in the same reagents to minimize procedural variability. To quench endogenous peroxidases, sections were incubated for 30 min in 0.05 M TBS containing 1% hydrogen peroxide. After three 5 min washes in 0.05 M TBS, sections were incubated for 1 h in a blocking solution of 10% Normal Goat Serum (NGS) and 0.4% Triton-X in 0.05 M TBS. The sections were incubated for 48 h at 4°C with gentle agitation in a mouse anti-Neu-N IgG (1:10,000; MAB377, Chemicon, Temecula, CA, USA) in 0.05 M TBS, containing 2% NGS, and 0.1% Triton-X. Following incubation with the primary antibody, the sections were washed three times for 5 min in 0.05 M TBS with 2% NGS and 0.1% Triton-X, followed by a 2 h incubation with the secondary antibody (goat anti-mouse, 1:600; Vector, Burlingame, CA, USA) in 0.05 M TBS containing 2% NGS, and 0.4% Triton-X. The sections were then washed three times for 5 min in buffer and subsequently incubated with an avidin biotinylated horseradish peroxidase complex for 1 h. The sections were washed again with three 5 min washes in 0.05 M TBS. All sections were incubated together for 7 min in sodium acetate containing 0.55 mM 3–3′-diaminobenzidine (DAB; Sigma, St. Louis, MO, USA) and 0.01% H₂O₂. The sections were washed three times for 5 min each in 0.05 M TBS and mounted onto gelatin-coated slides, air dried, and cover-slipped with Permount mounting medium (Thermo Fisher Scientific, Waltham, MA, USA).

Designed-based stereological analysis necessitates defining boundaries of a region of interest (ROI) for systematic random sampling from equally spaced sections throughout the structure of interest. Importantly, estimates of volume or cell population in a ROI are dependent on delineation of boundaries. This requirement, while often feasible for gray matter measurements, is less so for white matter since there is no map equivalent to Brodmann areas for white matter and delineating ROIs in relation to overlying gray matter would be at best approximate. Accordingly, we adopted a broad regional approach, similar to García-Marin et al. (2010), where WNN density was assessed according to frontal, parietal, or temporal regions (Figure 1).

Another issue in counting WMNs is the need to distinguish the boundary between superficial white matter and layer 6, and between superficial and deep white matter (Figure 2). The boundary with layer 6 can be estimated by a fall-off in cell density, and a change in soma morphology, where there is a decreased proportion of pyramidal neurons. Care was taken to avoid zones where presumptive WMNs could be confused with tangentially sectioned layer 6. There are fewer guidelines for the designation of deep white matter, except for the continued decline in cell density and the greater distance from the overlying gray matter.

To estimate density and regional distribution of WMNs, Neu-N labeled sections were digitized and sampling boxes overlain (Figure 1), in a semi-systematic approach similar to that of Xu et al. (2011), Kinney et al. (2012), and Yang et al. (2011). In specific, 3–4 slides per region (FC, Par, and TE) were digitized as whole brain hemisphere photomontages, using the 10× objective on an E600 Nikon microscope equipped with Turboscan Montaging system (Objective Imaging, UK). Using ImageJ (version 1.47, NIH, Bethesda, MD, USA) software to process all images identically, ROIs for superficial and deep WM were superimposed with sampling boxes (400 × 400 μm in superficial; 800 × 200 μm in deep WM). This approach yielded 20–30 counting boxes per region (Figure 1) for each subject; that is, a total number of 1275 unbiased counting boxes. Density of neurons in superficial and deep WM was then calculated by counting the number of neurons per total area sampled. An experimenter blind to the age of the monkeys and identity of the ROIs performed the counting from the acquired images. The advantage of the Turboscan Montaging system is that the person performing the counts can zoom in at high-resolution to the overlaid boxes for counting the cells. As a control for accuracy, density measurement counts were replicated in 1 or 2 sections from a subset of three different animals. This procedure resulted in an Interclass Correlation Coefficient of 0.93.

In addition, for two brains, stereological methods were also applied, using a standard sampling grid (400 × 400 μm). This method was less suitable since: (1) a large subsample of boxes was not parallel to the pia surface, contained a portion of layer 6 neurons, and thus were discarded; and (2) superficial and deep white-matter were also “straddled” by large portions of the grid. When neurons from layer 6 were excluded from what was intended as “systematic random sampling, ” the number of superficial and deep WMNs was within 10% of the sampling method used for the whole cohort of animals.

High-resolution images acquired with Turboscan from four monkeys (two young adult and two aged) were used for the morphometric analysis. A distinct advantage of the Turboscan system is that each tile acquired is at its best depth of focus, thus producing images as similar as possible to a “maximum projection” (our sections, while cut at 30 μm, typically dry to about 12–18 μm in thickness after processing and dehydration). Table 2 shows the numbers of WMNs in each strata that were used for quantification of morphometrics. Using ImageJ, the high-resolution images from each ROI were first converted to an 8-bit image; and an experimenter blind to region and to age of the animal, thresholded the 8-bit image and then converted it to a binary image. The experimenter also had an original photomicrograph to monitor the threshold intensity based on single neuron profiles. Three measures were used in the morphometric analysis: Soma Area, Soma Perimeter, and Soma Circularity. Circularity [4π (area/perimeter²)] was a measure of how closely a given soma shape matches a perfect circle and ranges from 0, an infinitely elongated polygon, to 1, a perfect circle.

For the analysis of density of WMNs, as a first step we tested for normality and homoscedasticity of the distribution of WMNs in each zone using Shapiro-Wilk test for normality. The Shapiro-Wilk test of normality for the distribution of data showed that the counts in each strata are not normally distributed (skewed distribution, Figure 2). Specifically, Krustal-Wallis test (non-parametric counterpart to dependent t-test) was used to determine differences in numbers of neurons in superficial vs. deep strata. To determine if age-related differences existed in these strata, a Mann-Whitney U-test (non-parametric counterpart to independent sample t-test) was used. Finally, a Friedmann’s test (non-parametric counterpart to ANOVA) was used to determine overall density differences in each ROI. This was followed up with Bonferroni correction for multiple comparisons.

The morphometric characteristics were also tested for normality. Since these met the necessary assumption of normality, parametric statistical tests were used. A 2 (young vs. old) × 2 (superficial vs. deep) × region (FC, Par, TE) multivariate ANOVA with Bonferroni correction for post hoc analysis was used. The data were analyzed using the Statistical Package for Social Sciences software (SPSS, version 19.0, Chicago, IL, USA). All results are reported as mean ± SEM, unless stated and alpha was set at 0.05.

WMNs were visualized by immunohistochemistry (IHC) for Neu-N (Figures 1, 2). Importantly, Neu-N allows small neurons to be clearly distinguished from similar sized glia, a distinction that is often difficult in standard cellular Nissl stains. Three zones of WMNs were distinguished; namely: (1) putative WMNs immediately below but possibly intermingled with layer 6; (2) those more clearly in the superficial WM, defined as a 400 μm zone below determined layer 6; and (3) those in the deep WM, defined as more than 500 μm below layer 6. These boundaries were adopted as convenient, but are subject to further refinement. For example, the “deep” WM in a gyral context may be better viewed as the merging of two zones of superficial WM. In contrast with the relatively distinct subplate layer (“layer 7”) in rodents, the superficial WM in NHP is, typically broken or scalloped, with small cell-sparse gaps (about 150 μm across; Figure 2). In determining the border between layer 6 and the superficial WM, we took account of diminished cell density in the superficial WM and a trend toward more horizontally oriented somata for superficial WMNs (Figure 2).

A Krustal-Wallis test showed that there are significantly less WMNs in the deep zone (Median = 9.5), compared to the superficial zone (Median = 47.5; Z = −3.657, p < 0.05; Figure 3A). While a significant difference in the number of WMNs in each strata was observed, a Mann-Whitney U test (non-parametric counterpart to independent sample t-test) showed no significant age differences between the young (n = 4) and older monkeys (n = 4) in the overall counts in either strata (Figure 3B).

As shown in Figure 4 the number of superficial Neu-N-immunoreactive (ir) WMNs per 0.16 mm² was determined as about 40 in the superficial, and about 10 neurons for the deep WM ROIs. This was consistent for all three regions. For all three, there was no significant difference between young and old animals. Across the three regions, the density varied slightly (36–45) in the superficial and 8–12 in the deep (zone), with a trend for lower density in the frontal region of the older monkeys (Figure 4).

Finally, since there were no significant age differences in the density of WMNs between the young and old monkeys, we pooled the density measurements of the young and old group for analyzing the regional distribution. A Friedmann’s test (non-parametric alternative to ANOVA) showed significant differences in the density of deep WMNs in the three different regions (FC, TE, and, Par) [λ²₍₈₎ = 7, p < 0.05] but not in the superficial zone. Post hoc analysis with Bonferroni correction for multiple comparisons showed significantly more deep WMNs in TE (12.6 ± 2.30) compared to FC (8.8 ± 1.10) and Par (8.8 ± 1.02; Figure 5).

While no significant age-related differences were observed in cell density, there were several morphometric differences that were statistically significant. A 2 (young vs. old) × 2 (superficial vs. deep) × 3 (FR, Par, and TE) multivariate ANOVA with Bonferroni adjustment for multiple comparisons showed a main effect of age, on three morphometric measurements. Soma area [F_(1,4012) = 5.77, p < 0.05] and perimeter [F_(1,4012) = 3.992, p < 0.05] were larger in young animals compared to the soma size from older monkeys. Soma shape was less circular [F_(1,4012) = 6.856, p < 0.05] in young vs. older monkeys.

The ANOVA also showed a main effect of strata, where the three morphometric measures were statistically different in superficial compared to the deep strata. Overall, somata in the superficial zone are larger in area [F_(1,4012) = 930.50, p < 0.05] and perimeter [F_(1,3276) = 450.211, p < 0.05] but less circular [F_(1,3276) = 19.67, p < 0.05]). A main effect of ROI was also observed; namely, regional analysis of soma morphology showed differences in soma area [F_(2,4012) = 15.18, p < 0.05], perimeter [F_(2,4012) = 3.96, p < 0.05] and circularity [F_(2,4012) = 6.82, p < 0.05] across the three regions (FR, Par, TE).

For soma area and perimeter (Figures 6A–D) the interaction between the factors (age, strata, and ROI) and multiple comparisons with Bonferroni adjustments revealed that soma size in aged animals was significantly smaller in FR and Par, but larger in TE in the superficial zone; this age difference in soma size was not observed in the deep zone. In terms of circularity, somata from older monkeys in FR and Par were more circular in superficial zone but no age differences were observed in TE. In the deep zone, no age difference was observed in circularity (Figures 6E,F, and Table 3). Examples of circularity differences are illustrated in Figure 7.

Heterogeneity of soma size and shape, as reported for the Neu-N material (above), is more clearly evident in the Golgi-like images produced by NADPH diaphorase (Figure 8). These reveal a mix of small granular, fusiform, multipolar, or pyramidal shapes. Soma size varies from a diameter of 5–10 μm for small granular neurons to 25–40 μm in the longest axis of fusiform neurons.

Quantification of WMNs has been undertaken in normal (García-Marin et al., 2010) and schizophrenic (Joshi et al., 2012; reviewed in Connor et al., 2011) postmortem human tissues; and WMN density has been investigated in a developmental series in humans and macaque monkeys (Fung et al., 2011).

Quantitative results can be expected to vary across studies due to differences in sampling protocols and variability in the designated ROIs, but in fact are overall in range. Thus, García-Marin et al. (2010), specify that superficial WMNs are four times as numerous as those in the deep white matter (2169 neurons/mm³ vs. 525 neurons/mm³). Using a counting box of 0.47 × 0.47 mm, Eastwood and Harrison (2003) report for temporal cortex of human controls an average per box number of 28 superficial WMNs and 10 deep WMNs (their Table 2). One study in the frontal lobe of vervet monkey reports 255 ± 14.4 neurons/mm² (Burke et al., 2009).

A third result of this study is the demonstration of regional diversity. In our sample, the distribution of WMNs was similar in temporal and parietal zones, but lower in the frontal zone. Other investigators have reported regional variability. An early study by Smiley et al. (1998), demonstrated that WMNs, as identified by immunoreactivity for M2-muscarinic receptors, were evenly distributed throughout subdivisions of macaque monkey cortex except for being more scant in primary visual cortex. More recently, García-Marin et al. (2010), found there to be a lower density of WMNs in the temporal region of human cortex, as compared with frontal or cingulate cortex; and Meyer et al. (1992) also report a higher density of neurons in areas 11 and 4 in human.

Factors that could contribute to regional diversity of cell number, cell types, or age-related vulnerabilities of WMNs include differences in developmental origin (Ma et al., 2013), in afferent or efferent connectivity, and in surrounding white matter environment.

Beyond the general descriptors of “superficial” vs. “deep” or sulcal and gyral locations, criteria are lacking for finer subdivisions of cortical white matter. Within this general reference frame, WMN soma appear not to have any regular distribution pattern, although distinctive dendritic patterns are readily apparent, especially in the Golgi-like NADPH-d preparations (Barbaresi et al., 2014). In our material, we noted some regularity in soma distribution, where two or three neurons were in close apposition (touching or <10 μm edge-to-edge). Possibly similar clusters of neurons were observed in a postmortem study of WMNs from patients with focal epilepsy, but were thought to be pathological malformations (Loup et al., 2009).

At a different scale, the superficial white matter in NHP, in contrast to the typically distinct, narrow subplate in rodents (Kanold and Luhmann, 2010) was often scalloped and broken, consisting of small interspersed neuron-free and neuron-dense patches.