Spatial statistics and point pattern analysis reveal lifespan trajectories of microglial density and clustering in the primate hippocampus

Abstract

The hippocampus undergoes substantial structural remodeling across the lifespan, yet how its resident immune cells, microglia, reorganize during development and aging remains poorly understood in primates. We quantified microglial density and spatial organization across hippocampal subfields in 15 female rhesus macaques spanning late gestation (gestational day 140) to advanced age (32.4 years) using tissue from the MacBrain Resource Center. Animals were grouped as perinatal (GD140–7 days postnatal), postnatal (2.5–6 months), juvenile/adult (11.4 months–9.8 years), and aged (18.7–32.4 years). Microglial density exhibited a U-shaped trajectory, declining during early postnatal life before increasing through adulthood and aging, with the dentate gyrus showing the greatest age-related change. Nearest-neighbor distance followed an inverse pattern, indicating maximal microglial dispersion during the postnatal period. Ripley’s H-function analysis revealed age-dependent alterations in microglial territorial organization, including an expanded exclusion radius in aged animals. Tile-based spatial analyses further identified increased microglial clustering within CA1 in the aged group. In the oldest animal, regions of microglial clustering corresponded with amyloid-β and phosphorylated tau immunoreactivity. Together, these findings demonstrate dynamic lifespan-dependent remodeling of microglial spatial organization and establish a quantitative framework for studying neuroimmune architecture in the aging primate hippocampus.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1038/s41598-026-60014-x.

Introduction

Microglia are the resident immune cells of the central nervous system, playing essential roles in brain development, homeostasis, and adaptive function across the lifespan. Derived from early myeloid progenitors, microglia colonize the embryonic brain and mature during postnatal development in parallel with synaptogenesis and neural circuit refinement^1,2. In the adult brain, microglia remain dynamic, remodeling their morphology and spatial organization to support synaptic pruning, myelin maintenance, and adaptive neuroimmune signaling through the regulated release of cytokines and signaling molecules, such as IL-1 and ATP, respectively^3,4. Together, these actions position microglia as key regulators of brain development, plasticity and resilience.

Microglial number, distribution, and morphology vary widely across the mammalian brain and are influenced by age, sex, species, and environmental context^5–7. Functional microglial states, including pro-inflammatory and homeostatic phenotypes, are reflected in biochemical signatures such as surface marker expression, cytokine profiles, and metabolic programs^8,9. Importantly, microglial functional state transitions may accompany changes in proliferation, migration, and spatial organization including territorial coverage and local aggregation. For example, during inflammatory signaling, increased microglial density and clustering around synapses, myelinated tracts, or sites of injury may occur^4,10–12. Microglia additionally exhibit region-specific spatial patterning that reflects local circuit demands and hence may contribute to differential vulnerability to neurodegenerative disease^11,13,14.

Despite substantial advances enabled by genetic targeting and in vivo imaging in rodent models, the spatial organization of microglia in the primate brain remains comparatively understudied, limiting our understanding of how neuroimmune architecture changes across the primate lifespan^15,16. Here, we focus on the hippocampus, a structure essential for learning/memory and highly vulnerable to neurodegeneration^17,18, which exhibits pronounced sexual dimorphism with estrogen shown to directly influence microglial number and signaling within subregions as the dentate gyrus (DG)^19,20. Using archival tissue from the female rhesus macaque grouped into four age groups: perinatal (GD140-7 days postnatal), postnatal (2.5-6 months), juvenile/adult (11.4 months-9.8 years), aged (n = 3, 18.7–32.4 years), we developed an automated, quantitative image analysis platform based on IBA1 (ionized calcium-binding adaptor molecule 1) immunolabeling to quantify microglial density, spacing, territorial organization and local clustering across hippocampal fields (Fig. 1). Our findings establish a spatiotemporal framework for understanding microglial organization in the primate hippocampus and provide a foundation for future studies examining how microglial population changes contribute to hippocampal function including learning and memory across the lifespan, with implications for human neurodegenerative disease.

Fig. 1.

Experimental workflow for microglial analysis across the rhesus macaque hippocampus. (A) Hippocampal sections (50 μm thickness) from female macaques (MacBrain Resource Center, Collection 6, https://macbraingallery.yale.edu/collection6/) were sampled along the rostro-caudal axis (− 13.95 to − 20.7 mm anterior to Bregma). IBA1 immunostained sections (in SVS format) were downloaded and parcellated into 12 hippocampal subfields (DG, CA3, CA2, CA1) and their layers. All images were extracted as ome.tiff in analysis software QuPath and cell bodies were detected using QuPath software with watershed segmentation following image preprocessing. Centroid coordinates (X, Y µm) were extracted for each detected cell and analyzed using R packages lme4 and spatstat to compute and compare overall hippocampal volume (mm³), microglia density (cells/mm²), microglia nearest-neighbor distance (NND, µm), and H(r)_max using point pattern analysis (PPA) quantifying spatial clustering via the H-function. (B) Image processing and analysis pipeline. Image deconvolution was applied based on median stain intensity from representative sections from macaque samples, followed by applying a median radius filter (3 μm). Image threshold (0.4) and watershed detection was performed in QuPath.

Results

A quantitative framework for anatomically resolved analysis of hippocampal microglia

Microglia undergo dynamic changes across development and aging that shape synaptic pruning, circuit maintenance and neural repair⁷. However, how these immune cells organize spatially across the extended primate lifespan, especially within hippocampal subfields vulnerable to age-related decline, is less understood. To address this, we developed a quantitative workflow combining anatomical parcellation, automated cell detection, and spatial analysis of IBA1-immunolabeled microglia in archival hippocampal tissue spanning late gestation through old age (Fig. 1A). Using whole-slide images from the Yale MacBrain Resource Center (MBRC) collection 6 (Fig. 2A), microglial density, nearest-neighbor distance (NND) and spatial organization were analyzed.

Fig. 2.

Data structure and hippocampal parcellation. (A) Hierarchical experimental design included 15 female rhesus macaques distributed across four age groups: perinatal (n = 4 brains), postnatal (n = 4), juvenile/adulthood (n = 4), and aged (n = 3). Individual brain ID [B], number of brains [n_B] and section counts [number of sections, n_S] are shown. Each brain contributed 5–9 sections spanning the rostro-caudal axis. Analysis included 12 hippocampal regions per section, and centroid coordinates (X, Y µm) were extracted for each detected microglial cell. (B) Representative coronal sections illustrating hippocampal architecture and anatomy across multiple histological stains used for precise anatomical parcellation of subfield and layer boundaries. Parcellation stains include Nissl, CR (calretinin), MBP (myelin basic protein), SMI-32 (neurofilament), CB (calbindin), and AChE (acetylcholinesterase). (C) IBA1 immunostaining in a representative section showing parcellation of hippocampal subfields. Dashed black lines delineate boundaries between CA1, CA2, CA3, and DG regions, with layer-specific annotations (ML: molecular layer; GCL: granule cell layer; Hil: hilus; Rad: radiatum; LacMol: lacunosum-moleculare; OrPyr: oriens/pyramidale). (D) Microglial density (cells/mm²) distribution for each of the 12 hippocampal regions across the entire dataset. Each data point represents a brain (n = 15). Overall, the DG Hil shows the largest median microglia density across all samples, and the CA2 LacMol shows the lowest median density. Box boundaries represent the first and third quartiles (IQR) and horizontal lines indicate median values.

To enable anatomically resolved analysis, adjacent sections stained with Nissl, SMI-32, calretinin (CR), calbindin (CB) and acetylcholinesterase (AChE) from every brain were used to delineate hippocampal subfields and layers, serving as references for cytoarchitectural boundaries, laminar organization and subfield-specific neurochemical signatures (Fig. 2B). Specifically, cornu Ammonis 1–3 (CA1-3) regions were delineated into stratum oriens/pyramidale (OrPyr); stratum radiatum (Rad); and stratum lacunosum-moleculare (LacMol). The dentate gyrus (DG) subfield was parcellated into the polymorphic layer also known as the hilus (Hil), granule cell layer (GCL), and molecular layer (ML) based on neuronal cell density, myelination patterns and differential expression of calcium-binding proteins. This approach yielded a total of 12 anatomically distinct hippocampal regions of interest (ROIs) for analysis (Fig. 2C). Across our dataset, a total of 1308 hippocampal ROIs were analyzed (Table 1), enabling systematic tracking of microglia organization in individual macaques and age groups during key periods of hippocampal development, maturation, and aging.

Table 1.

Study groups and sample characteristics.

Brain	Age	Classification
B99	GD140	Group I: perinatal 0-1 month
B114	GD149
B91	0 d.
B64	7 d.
B66	2.5 mo.	Group II: postnatal 1–12 months
B74	2.5 mo.
B89	6 mo.
B94	6 mo.
B83	11.4 mo.	Group III: juvenile and adult 1–3 years (juvenile) 5–15 years (adult)
B92	1 year.
B72	8 year.
B79	9.8 year.
B124	18.7 year.	Group IV: aged  ≥ 18 years
B68	19 year.
B95	32.4 year.

Female rhesus macaques (n = 15 brains) from the Yale MBRC Collection 6 were grouped into four age groups: Group I (fetal/perinatal, n = 4 brains, gestational day 140 to 7 days postnatal), Group II (early postnatal, n = 4, 2.5-6 months), Group III (adolescence/adulthood, n = 4, 11.4 months-9.8 years), and Group IV (aged/menopause, n = 3, 18.7–32.4 years). GD = gestational day; d. = days; mo. = months; yr. = years.

Microglial detection was performed using an automated QuPath pipeline that was optimized for this analysis (Fig. 1B). To assess detection accuracy, 30 randomly selected 250 × 250 μm tiles were quantified by two independent raters and compared with automated counts. QuPath detections showed strong agreement with manual quantification (Pearson R = 0.88, p < 0.001; Figure S1A). Bland–Altman analysis confirmed strong agreement between automated cell detection by the QuPath pipeline and manual validation, revealing a mean bias of − 1.9 cells and 95% limits of agreement from − 9.7 to 5.8 cells (Figure S1B).

Across all individual macaques in our dataset, microglia density varied substantially among hippocampal regions and layers (Fig. 2D) with the DG Hil showing the highest density (~ 600 cells/mm²) followed by CA3 OrPyr. Conversely, other regions such as CA2 LacMol and CA1 Rad displayed nearly half (~ 300 cells/mm²) the density of DG Hil.

Hippocampal volume, which we calculated using the Cavalieri principle, showed an increase from approximately 10mm³ during late gestational and early postnatal days, before reaching almost double the volume during 2.5 months of age and steadily stabilizing, then finally peaking again at ~ 30mm³ between 8 and 18.7 years (Fig. 3A). In the two oldest macaques (19- and 32-year-old), a small decrease in volume was seen, but on average the aged group still exhibited larger hippocampal volume compared to all other groups. Notably, hippocampal volume exhibited relatively low within-group variability across all four age groups (Fig. 3A), suggesting minimal tissue shrinkage and supporting the consistency of tissue preservation across samples. One-way ANOVA identified a significant difference between aged and perinatal animals (p = 0.012; Tables S1–2), while all other pairwise comparisons were not significant (Fig. 3B), suggesting that hippocampal volume reaches near-adult levels by the juvenile period and remain relatively stable through adulthood, a finding consistent with the literature^21–23.

Fig. 3.

Age-related changes in hippocampal volume. (A) Using the Cavalieri principle, total hippocampal volume (mm³) was calculated across all brain sections in order of youngest to oldest. GD = gestational day; d. = days; mo. = months; yr. = years. (B) Mean hippocampal volume (mm³) showed a positive, linear trend across perinatal (n = 4), postnatal (n = 4), juvenile/adult (n = 4) and aged (n = 3) groups (one-way ANOVA). * p < 0.05, **p < 0.01, ****p < 0.0001. Grey bars represent standard error of the mean.

Spatial mapping reveals age-related changes in microglial distribution in the hippocampus

We examined the distribution of IBA1-immunolabeled microglia across hippocampal subfields and layers in all brains spanning the four age groups. Representative microglia regional and laminar patterning within both the DG and CA1 is shown in Fig. 4, with differential distribution across the DG GCL, Hil and ML, and CA1 OrPyr, Rad and LacMol, respectively. The visual increase in hippocampal size across representative images (first panel) is consistent with the volumetric changes described earlier (Fig. 3B).

Fig. 4.

Representative microglial immunostaining across the lifespan. Representative IBA1 and Nissl immunostained images of the hippocampus. First image for each group (left) shows low-magnification views of the entire hippocampus, with changes in overall shape and size. Middle and right images show high-magnification images of DG and CA1 regions, respectively, illustrating layer-specific microglial arrangement. DG layers: Hil (hilus), GCL (granule cell layer), ML (molecular layer). CA1 layers: OrPyr (stratum oriens/pyramidale), Rad (stratum radiatum), LacMol (stratum lacunosum-moleculare).

We generated spatial point-pattern maps to visualize microglial spatial organization across the entire hippocampus without hard anatomical boundaries. As shown in Fig. 5, microglia centroids, shown across the age groups with corresponding Gaussian kernel density estimates (sigma = 200 μm), highlight a local shift in microglial concentration across the hippocampus. These spatial reconstructions reveal shifts in microglial distribution and density within the hippocampus relative to age. For example, the perinatal age group exhibited a high microglial intensity (372 cells/mm²) within the DG Hil, CA3 OrPyr and CA1 OrPyr relative to the other subregions during this period. Overall, average microglial intensity dropped in postnatal stages to 168 cells/mm² but remained relatively high in the DG Hil and CA1 OrPyr. Finally, an increase in average intensity from postnatal stages was seen in both juvenile/adult (308 cells/mm²) and aged (358 cells/mm²) groups.

Fig. 5.

Spatial visualization of microglial distribution. (A) Point pattern maps showing individual microglial centroids, color-coded by hippocampal region for representative brains from each age group (perinatal to aged). Each point represents a single IBA1 labeled cell within the parcellated regions of interest along with its spatial coordinates along the X and Y axis (µm). (B) Density kernels (using spatstat, sigma = 200 μm) with average intensity in cells/mm² represented by color scale. DG layers: Hil (hilus), GCL (granule cell layer), ML (molecular layer). CA layers: OrPyr (stratum oriens/pyramidale), Rad (stratum radiatum), LacMol (stratum lacunosum-moleculare).

Microglial density declines sharply after birth and rebounds with aging

We quantified microglial distribution in all animals across hippocampal subfields and layers using generalized linear mixed models (GLMMs) to analyze density and spacing, with brain included as a random effect to account for dependency (see "Methods"section). This approach enabled an assessment of age- and region-specific effect while accounting for variability across brain samples.

Density was calculated from IBA1 labeled centroid cell counts normalized to ROI area for each animal across the age groups, revealing changes in abundance across hippocampal subregions (Fig. 6; Tables S3-4). Heatmaps were generated across the rostro-caudal axis based on section-level measures (Fig. 6A (representative brain); Figure S4 (all brains)). Consistent with earlier data from density kernel estimates, the postnatal age group displayed the lowest microglial density across all hippocampal subregions, most evident in DG (GCL and ML) and CA2 (OrPyr).

Fig. 6.

Age-related changes in microglial density. (A) Heatmaps showing microglial density (cells/mm²) for each tissue section (rostral to caudal) from representative brains (B64, B66, B72, B68, top to bottom) across the hippocampus. Each bar represents one section, with color intensity reflecting cell density (light gray: 300 cells/mm²; dark gray/black: 900 cells/mm²). X marks in red within the younger brain indicates limited section availability. A representative animal is shown to illustrate rostro-caudal patterning; heatmaps for all subjects are provided in Figure S4. (B) Mean microglial density (cells/mm²) for each age group within the hippocampus. Grey bars represent standard error of the mean. Significant pairwise comparisons are indicated: * p < 0.05, ** p < 0.01, **** p < 0.0001, GLMM using lme4 with Bonferroni adjustment. Perinatal (n = 4), postnatal (n = 4), juvenile/adult (n = 4) and aged (n = 3). DG layers: Hil (hilus), GCL (granule cell layer), ML (molecular layer). CA layers: OrPyr (stratum oriens/pyramidale), Rad (stratum radiatum), LacMol (stratum lacunosum-moleculare).

Overall, our analysis of microglia density across all brains indicated a trend towards a U-shaped trajectory with a consistent drop during the postnatal period (Fig. 6B). Statistical analysis revealed significant changes in the DG, with the GCL showing the strongest changes between perinatal (404 cells/mm²) and postnatal groups (163 cells/mm²; p = 6.8 × 10⁻⁷), juvenile/adulthood stage (271 cells/mm², p = 0.02 vs. postnatal), and aged (373 cells/mm²; p = 4.1 × 10⁻⁵ vs. postnatal; ns vs. juvenile/adult) stages. Similarly, the DG Hil showed a similar, albeit more modest trend with a drop from the perinatal (707 cells/mm²) to postnatal group (423 cells/mm²; p = 0.016). The DG ML also showed a clear pattern of density reduction in the postnatal group (235 cells/mm²) compared to perinatal (419 cells/mm²; p = 0.004), followed by subtle increases in the juvenile/adult group (256 cells/mm², ns) and the aged group (342 cells/mm², ns).

CA3 layers (LacMol, OrPyr, and Rad) showed similar reduction in microglial density, especially between perinatal and postnatal groups, though only CA3 LacMol reached statistical significance, decreasing from 401 cells/mm² in perinatal stages to 246 cells/mm² in postnatal stages (p = 0.026) (Fig. 6B). Likewise, CA2 layers followed a consistent pattern, with only CA2 OrPyr demonstrating significant decline in density from perinatal group (423 cells/mm²), to the postnatal group (243 cells/mm², p = 0.007), then increasing partially in the juvenile/adult stage (353 cells/mm², ns vs. postnatal) before returning to high, near-perinatal levels in the aged group (417 cells/mm², p = 0.020 vs. postnatal) (Fig. 6B).

Finally, most CA1 layers showed modest density changes, with only CA1 OrPyr decreasing significantly from perinatal (458 cells/mm²) to postnatal stages (276 cells/mm², p = 0.018), and subsequent subtle increases in juvenile/adult stages and aging (ns vs. postnatal) (Fig. 6B).

Microglial spacing mirrors density changes across the lifespan

To determine age- and region- related changes in intercellular spacing between individual cells, we used NND, defined as the shortest (Euclidean) distance between microglia centroids, on average. (Fig. 7, Tables S5-6). Across the hippocampus, NND showed an inverse pattern to density, as regions with high density exhibited tight (low) spacing, and lower density regions increased spacing. Section-level NND heatmaps revealed that postnatal group brains display wider spaces between cells, across most regions along the rostro-caudal axis, compared to other age groups (Fig. 7A (representative brain), Figure S5 (all brains)). In the DG GCL, perinatal microglia were tightly packed with an average NND of 35 μm, expanding significantly during postnatal development to 63 μm (p = 1.9 × 10⁻⁸), before contracting to 45 μm in juvenile/adulthood (p = 0.006; postnatal vs.), and returning to near-perinatal spacing in the aged group with a mean NND of 35 μm (p = 8.6 × 10⁻⁸ vs. postnatal). Consistently, the DG ML mirrored this pattern, with early postnatal expansion (32 μm to 43 μm; p = 0.010) followed by modest age-related contraction, while the DG Hil showed a similar trend but no significant differences (Fig. 7B).

Fig. 7.

Age-related changes in microglial spacing by region. (A) Heatmaps showing microglial nearest-neighbor distance (NND; µm) for each tissue section (rostral to caudal) in brains (B64, B66, B72, B68, top to bottom) across the hippocampus. Each bar represents one section, with color intensity reflecting spacing according to the scale (dark purple: 100 μm; yellow: 40 μm). X marks in red within the younger brain indicates limited section availability. A representative animal is shown to illustrate rostro-caudal patterning; heatmaps for all subjects are provided in Figure S5. (B) Mean NND (µm) for each age group within the hippocampus. Grey bars represent standard error of the mean. Significant pairwise comparisons are indicated: * p < 0.05, ** p < 0.01, **** p < 0.0001, GLMM using lme4 with Bonferroni adjustment. Perinatal (n = 4), postnatal (n = 4), juvenile/adult (n = 4) and aged (n = 3). DG layers: Hil (hilus), GCL (granule cell layer), ML (molecular layer). CA layers: OrPyr (stratum oriens/pyramidale), Rad (stratum radiatum), LacMol (stratum lacunosum-moleculare).

CA1-3 fields and their layers displayed comparable inverse density-spacing relationships. Within CA1 OrPyr, NND increased from 30 μm in perinatal stages to 40 μm in the postnatal group (p = 0.019), then contracted to 33 μm in the juvenile/adult group and 29 μm in the aged (p = 0.019 vs. postnatal). CA1 Rad again followed a similar profile, expanding from 37 μm in the perinatal group to 45 μm in postnatal, then decreasing in aged macaques at 32 μm (postnatal vs. aged; p = 0.009) (Fig. 7B). CA2 layers demonstrated similar patterns with significant postnatal expansion, particularly in CA2 OrPyr, where NND increased from 32 μm to 42 μm (p = 0.021) from perinatal to postnatal, only to contract again in the aged group to 30 μm (p = 0.011) (Fig. 7B). CA3 LacMol also showed early postnatal expansion (33 μm to 44 μm; p = 0.018) followed by progressive contraction in the juvenile/adult period to the aged stage, with CA3 Rad displaying a comparable pattern (33 μm to 44 μm to 37 μm to 34 μm respectively; perinatal vs. postnatal: p = 0.023) (Fig. 7B).

Microglial territory size contracts postnatally and re-expands in aged animals

The spatial organization of microglia is actively remodeled in response to local cues and during injury, neuroinflammation, and protein aggregation, where microglia may exhibit pronounced clustering and nodule formation^13,24–27. However, such changes in microglial spacing, territorial organization, and aggregation can also occur during normal development and aging, where they may reflect alterations in tissue surveillance or certain homeostatic demands, especially in regions such as the hippocampus thought to exhibit lifelong neurogenesis²⁸. Therefore, to determine whether age-related changes in microglial organization extend beyond differences in density and NND, we applied Ripley’s H-function analysis to quantify spatial organization across multiple scales^29–31. This approach enabled measurement of regional and local microglial clustering in the hippocampus based on a spatial distribution metric: H(r) > 0 indicating clustering, H(r) = 0 indicating random spacing (i.e., complete spatial randomness (CSR)), and H(r) < 0 indicating dispersion.

First, we applied Ripley’s H-function analysis to whole-hippocampus point patterns using anatomically defined ROIs as observation windows. Across all age groups, H(r) curves exhibited a radius of exclusion marked by a dip in H(r), followed by a transition toward positive values at larger distances, consistent with short-range microglial exclusion and longer-range aggregation seen in the literature^31–33(Fig. 8A). While the overall shape of the H(r) curves was broadly similar across age groups, differences emerged in the spatial scales at which territorial features occurred. Building on existing applications^31–33, we extracted the exclusion radius (rHmin), interpreted as an estimate of average microglial territory size, and the aggregation radius (rHmax), representing the radius of aggregation (Fig. 8B–C). Our data shows that the mean territory size was lowest during the postnatal period (19.5 μm), increased modestly in juvenile/adult animals (21.3 μm), and was highest in the aged animal group (27.8 μm) (Table S8). A one-way ANOVA with Tukey post-hoc revealed a significant increase within aged animals relative to the postnatal group (p = 0.046; Table S7), whereas all other pairwise comparisons were not significant. In contrast, aggregation radius (rHmax) did not show a statistically significant change across the age groups despite a subtle gradual increase from perinatal to juvenile/adult and aged samples (Fig. 8C; Tables S9–10). Together, these findings indicate that lifespan-related changes across the hippocampus are primarily reflected in microglial territorial organization, as detected by Ripley’s analysis, rather than in the characteristic spatial scale of microglial aggregation.

Fig. 8.

Ripley’s H-function analysis reveals age-dependent changes in microglial territorial organization and local clustering. (A) Mean whole-hippocampus H(r) curves for each age group. Shaded regions represent standard error of the mean across brains. Negative H(r) values indicate local dispersion, whereas positive values indicate clustering relative to complete spatial randomness (CSR). (B) Mean exclusion radius (rHmin), interpreted as an estimate of microglial territory size, increases with aging (C) Quantification of the radius of maximal aggregation (rHmax), representing cluster size, across age groups (D) Representative hippocampus observation window for tiled analysis (E) Mean H(r) curves for tiles classified as clustered (solid lines) or non-clustered (dashed lines) within each hippocampal region (F) Representative 200 × 200 μm hippocampal tiles exhibiting clustered and non-clustered microglial spatial arrangements across age groups. Colored points indicate individual microglial centroids. (G) Percentage of clustered tiles within CA1, CA2, CA3 and DG across age groups. Grey bars represent standard error of the mean. Statistical comparisons were performed using one-way ANOVA followed by Tukey’s multiple comparison. Significant pairwise comparisons are indicated: * p < 0.05, ** p < 0.01, **** p < 0.0001. Perinatal (n = 4), postnatal (n = 4), juvenile/adult (n = 4) and aged (n = 3).

Local spatial organization reveals increased microglial clustering in aged CA1

Although whole-hippocampus Ripley analysis captures global patterns of spatial organization, localized changes in microglial distribution may be obscured when large anatomical regions are treated as a single point pattern. To examine spatial organization at a finer scale, the hippocampus was subdivided into 200 × 200 μm tiles and classified as clustered or non-clustered according to the behavior of their H(r) curves relative to CSR (Fig. 8E).

Representative tiles in Fig. 8F reveal distinct spatial signatures for clustered and non-clustered regions across all age groups. Supporting this classification, H(r) curves (Fig. 8E) show that clustered tiles remained positive and well above complete spatial randomness (CSR) across the sampled distance range, whereas non-clustered tiles exhibited consistently negative H(r) values.

To quantify regional differences, the percentage of clustered tiles was calculated for each hippocampal subfield (Fig. 8G; Table S12), and among all regions examined, CA1 exhibited the strongest age-related effect. Specifically, the proportion of clustered CA1 tiles was lowest in postnatal (7.4%) and juvenile/adult (7.7%) groups, almost doubling in aged macaques (14.1%). A similar trend was observed in CA2, however, statistical analysis only revealed significantly higher clustering proportions in aged CA1 compared to postnatal (p = 0.024) and juvenile/adult groups (p = 0.035; Table S11) following multiple comparison correction. In both CA1 and CA2, perinatal animals exhibited intermediate levels of clustering (11 and 10.6%, respectively) that did not differ significantly from other age groups. Despite the large proportion of clustered tiles in the DG across all groups, no significant age-related differences in the percentage of clustered tiles were detected in the DG. Together, these findings indicate that local microglial clustering appears selectively enriched within aged CA1, whereas spatial organization remains comparatively stable in other hippocampal subfields.

Association of microglial clustering with amyloid-β and phosphorylated Tau in an aged female macaque

We examined the relationship between microglial clustering and markers of amyloid-β and tau phosphorylation in the hippocampus of a highly aged female macaque (B95; 32.4 years old). Tau pathology was assessed using anti-tauP217 antibodies, which recognize tau phosphorylated at threonine 217, a marker associated with pathological tau accumulation³⁴, while amyloid-β was detected using the 6E10 antibody. These immunohistochemical markers were available as part of the MacBrain Collection analysis of B95. We therefore investigated whether local amyloid-β and tau immunoreactivity was associated with microglial spatial organization. Given the pronounced microglial clustering observed in the CA1 region of aged animals (Fig. 8G), subsequent analyses focused on this hippocampal subregion.

To assess the spatial relationship between amyloid-β/tau pathology and microglial clustering, anatomically aligned serial hippocampal sections were analyzed as described in the Methods. Quantitative measurements of 6E10 and TauP217 immunoreactivity were mapped to corresponding CA1 tiles containing IBA1-labeled microglia and compared between clustered and non-clustered regions. As shown in Fig. 9B, higher levels of both 6E10 and TauP217 immunoreactivity were associated with regions exhibiting greater microglial clustering. Spearman correlation analyses revealed moderate positive relationships between the proportion of clustered microglial tiles and both 6E10 and TauP217 signal intensities (Table S13), although these correlations did not reach statistical significance (rho = 0.6, p = 0.35; rho = 0.6, p = 0.35, respectively). Notably, while the strength of the association with microglial clustering was similar for 6E10 and TauP217, the slope of the relationship differed, and the degree of correlation appeared to increase along the rostro-caudal axis (sections.  1–5) (Fig. 9B).

Fig. 9.

Exploratory spatial analysis of CA1 microglial clustering in B95. (A) Representative images from B95 showing IBA1-positive microglia, amyloid-β (6E10), and phospho-tau (P217) immunoreactivity. (B) Top panels show section-level relationship between the percentage of clustered CA1 tiles and mean signal intensity of amyloid-β (6E10) and phosphor-tau (P217) immunoreactivity. Numbers denote individual CA1 Sects.  (1–5) along the rostro-caudal axis. Spearman rank correlation analysis demonstrated positive association between clustered tiles and both 6E10 and P217 signals (ρ = 0.60 for both markers). Bottom panels show the distribution of tile-level signal in clustered and non-clustered CA1 tiles. Clustered tiles exhibited significantly greater 6E10 and P217 immunoreactivity than non-clustered tiles (Wilcoxon rank-sum test; ****P < 0.0001, *P < 0.05).

To further examine these relationships, 6E10 and TauP217 intensities were compared between clustered and non-clustered microglial tiles identified using the tiled Ripley’s clustering approach (Fig. 8D–G). Clustered tiles exhibited significantly greater 6E10 immunoreactivity (Wilcoxon rank-sum test, p < 0.0001). Similarly, TauP217 signal intensity was greater in clustered than non-clustered regions (Wilcoxon rank-sum test, p < 0.05). Together, these findings indicate that microglial clustering co-localizes with regions of increased amyloid-β and phosphorylated tau immunoreactivity in the aged CA1.

Discussion

Rhesus macaques offer critical advantages for modeling human neuroimmune aging given their extended lifespan, prolonged postnatal brain development period (~ 3–4 years to reach adult volumes), reproductive cycle and age-related hormonal transitions³⁵. Using this non-human primate model, we provide a population-level analysis of microglial organization in the hippocampus from late gestation to old age. Leveraging the MBRC’s high-resolution IBA1-immunolabeled images^36,37, we identify key patterns of microglial reorganization across age stages, establishing a foundational framework for understanding microglial spatial dynamics in primates (Fig. 10). Three principal findings emerge: (i) microglial density follows a broadly U-shaped trajectory across the lifespan, decreasing postnatally then re-emerging during adulthood and aging. (ii) NND measures inversely track cell density changes, providing a confirmation of microglial abundance analysis and indicates coordinated regulation of spacing; (iii) Spatial point pattern analysis reveals age-related changes in microglial territorial organization at the global hippocampus level and increased local clustering within aged CA1. Together, these findings demonstrate that microglia undergo coordinated remodeling across the primate lifespan and suggest that spatial arrangement may provide a sensitive indicator of age-related neuroimmune adaptation and vulnerability within the hippocampus.

Fig. 10.

A model of age-dependent changes in microglial spatial organization across the lifespan. During development, high microglial density and extensive territorial coverage support synaptic pruning, circuit refinement, and brain maturation. Microglial density peaks during this period. In adulthood, following reproductive maturation, microglial density stabilizes and cells exhibit a relatively uniform spatial distribution as hippocampal volume expands. During aging, microglial density increases and is accompanied by a reorganization of microglial territories, characterized by expanded surveillance domains, reduced spatial uniformity, and increased clustering within vulnerable hippocampal subregions.

Postnatal microglial spatial remodeling coincides with hippocampal circuit refinement

Early postnatal development represents a crucial period of hippocampal circuit formation and refinement that aligns with the peak of synaptogenesis and synaptic pruning in primates, which occurs primarily during the first postnatal year^38,39. During this window, microglia play an important role in circuit development by releasing growth factors and cytokines and actively engulfing synaptic elements through complement- and activity-dependent mechanisms^2,40,41.

Our analysis indicated high microglial density within perinatal brains consistent with the peak time window of microglial proliferation and the smaller hippocampal volume of the neonate brain^21–23. We observe a subsequent sharp reduction in microglial density during the postnatal period (1–12 months) across virtually all hippocampal regions with a particularly significant 60% decrease in the DG GCL. This region is involved in postnatal granule cell neurogenesis and sparse hippocampal coding underlying pattern separation^42–44. Such changes in microglial density may reflect a developmental transition from proliferative and synapse-sculpting microglial functions to a more homeostatic surveillance state as DG circuitry matures and hippocampal volume expands.

Changes in density were accompanied by coordinated alterations in both microglial NND spatial metrics and territorial organization. Across the hippocampus, postnatal animals exhibited the largest NND measure, indicating that neighboring microglia were, on average, farther apart as cell density declined. In contrast, whole-hippocampus Ripley’s H-function analysis identified the smallest microglial exclusion radius during the postnatal period and the largest in the aged group, suggesting that NND and territory size capture distinct aspects of microglial organization, with the former capturing intercellular spacing, and the latter reflecting the exclusion radius and scale of spatial repulsion^31,32.

Microglia territories are actively maintained through a combination of process tiling and contact-mediated repulsion which are modulated by purinergic pathways, including ATP-dependent P2RY12 receptor signaling⁴⁵. These mechanisms help distribute surveillance coverage across tissue and minimize overlap between neighboring cells^3,46. Emerging evidence suggests that these regulatory pathways are impacted by aging. A recent transcriptomic analysis of the aged human brain reported reduced microglial P2RY12 expression, consistent with age-related alterations in microglial surveillance and immune function⁴⁷. Similarly, age related alterations in microglial morphology and behavior have been shown to impair surveillance capacity, with aged microglia displaying smaller and less branched arbors, reduced process motility, and altered responses to extracellular ATP⁴⁸. These findings suggest that microglial soma positioning and functional surveillance coverage become increasingly uncoupled with age. As a result, larger territories defined by cell positioning may not correspond to proportionally greater coverage by microglial processes. This framework may explain the increased territorial organization observed in aged animals and why local changes in exclusion radius do not consistently mirror changes in microglial spacing.

One challenge when applying spatial statistics to biological tissue is that anatomical regions are rarely uniform in shape and size. This is particularly relevant in the hippocampus, where subfields differ in geometry and undergo marked changes across development and aging. Such differences can complicate direct comparisons of spatial point patterns because estimates of clustering may be influenced by the size and boundaries of the observation window. To address this, we complemented whole-hippocampus Ripley’s analysis with a tile-based approach that evaluated local microglial organization within fixed 200 × 200 μm sampling windows. This approach also enabled spatial organization to be examined at a scale more closely aligned with microglial NND and territory formation.

Our tiled analysis revealed that aged animals exhibited a greater proportion of clustered tiles especially within CA1 compared to postnatal and juvenile/adult groups, suggesting that aging may be accompanied by changes in both territorial organization and localized microglial aggregation within specific hippocampal regions. Together, these complementary analyses may explain why aged animals exhibited an increased exclusion radius while simultaneously showing localized regions of increased clustering, highlighting that age-related microglial remodeling occurs across multiple spatial scales.

The biological interpretation of microglial aggregation depends on anatomical context, as clustering is often thought to represent a distinct form of spatial organization associated with coordinated cellular behavior including heightened activity, phagocytosis and cytokine signaling in response to developmental and environmental cues^12,24,25. While age-related changes in local clustering were most evident within CA1, the DG consistently displayed a high level of clustered tiles across all age groups. During brain development, localized microglial assemblies are observed within the DG where they contribute to neurogenesis and circuit maturation⁴⁶. The persistence of microglial clustering may support lifelong structural plasticity and remodeling within the DG²⁸. Together, these observations suggest that local microglial clustering reflects region-specific modes of organization, constituting a persistent feature of the DG while emerging more selectively in the aging CA1.

Regional microglial clustering corresponds with elevated amyloid-β and phosphorylated tau

Although the primary focus of the present study was to characterize age-dependent changes in microglial spatial organization across the macaque lifespan using IBA1 immunolabeling, the analysis of an aged female (B95) provides an illustrative example of how these spatial metrics may relate to markers of neurodegeneration. B95 represents a rare example of an exceptionally aged rhesus macaque (32.4 years), providing an opportunity to examine microglial organization in advanced age. Clustering is suggested to reflect a disruption of homeostatic arrangement potentially arising from local recruitment, proliferation, migration toward pathological stimuli, or the collapse of normal spatial boundaries between neighboring cells^27,33,49. Within CA1, we found that areas exhibiting increased microglial clustering were also associated with higher amyloid-β (6E10) and phosphorylated tau (TauP217) immunoreactivity than non-clustered areas. While these observations are limited to a single aged female within the collection, and are interpreted cautiously, they suggest an interaction between microglial spatial organization and indicators of neuropathological burden.

The hippocampal formation, and interconnected association cortical networks, are recognized as sites of early age-related neurodegenerative pathology^50,51. Aging rhesus macaques naturally develop amyloid-β deposits and progressive phospho-tau pathology that recapitulate key features of the earliest stages of late-onset Alzheimer’s disease, including a pattern of pathology that originates within entorhinal circuits and extends into the association cortex^50,51. In addition, amyloid-β positive regions in aged rhesus macaques have been shown to exhibit pronounced glial pathology, including microgliosis, astrocytosis, and morphological alterations of microglia, even in the absence of overt neurodegeneration⁵². Analysis of the exceptionally aged female macaque B95 revealed an increased proportion of clustered microglial tiles accompanied by elevated 6E10 and phospho-tau (TauP217) immunoreactivity within the CA1 region. Although correlational in nature, these findings suggest that spatial measures of microglial organization may identify local tissue environments enriched for age-related pathology. More broadly, quantitative characterization of microglial spatial architecture provides a framework for relating cellular organization to the distribution of amyloid-β and phosphorylated tau in the aging primate hippocampus and may represent an intermediate neuroimmune phenotype linking normative aging to early neurodegenerative processes.

Implications, limitations and future directions

This study provides a comprehensive quantitative framework for characterizing microglial spatial organization across the hippocampus within non-human primates. By integrating measures of density, intercellular spacing, territorial organization, and local clustering, our findings demonstrate that microglial spatial arrangement is remodeled across the lifespan and varies among hippocampal subregions. These results show that spatial organization captures biological features not readily available from first-order metrics alone, highlighting microglial organization as an additional dimension of neuroimmune regulation.

A particular strength of this work is the application of spatial statistics and point pattern analysis to large-scale histological datasets. While density and morphology remain widely used measures of microglial biology, approaches such as Ripley’s analysis allow microglia to be examined as coordinated cellular populations whose organization may reflect local circuit demands. Our tile-based framework provides a means of assessing local spatial organization while minimizing biases introduced anatomical boundaries. This approach may help generate standardized spatial units that can be readily aligned with other datasets, allowing local measures of cellular organization to be compared with regional gene expression, protein abundance, or connectivity patterns. This creates opportunities to examine how microglial spatial architecture relates to molecular and circuit-level properties of the brain.

Several limitations should be considered. First, the use of NHP archival tissue limited the study to 3–4 animals per age group, and due to the current composition of collection 6, only female animals were included. As more samples are added to the archive, future studies incorporating larger cohorts including males will be important. Second, although CA2 boundaries were defined using convergent cytoarchitectonic and immunohistochemical criteria, future studies incorporating canonical molecular markers such as Purkinje cell protein 4 (PCP4) or regulator of G-protein signaling 14 (RGS14), which were not available in the archive for the present study, will provide a more precise delineation of CA2. Finally, IBA1 is a pan-microglial marker that does not distinguish activation states or exclude infiltrating macrophages. Future studies integrating state-specific markers such as TMEM119, P2RY12 and CD68, together with morphological measurements, will be important for linking microglial spatial patterns to specific phenotypes and functions. Together, such multimodal approaches allow spatial patterning to be directly linked to microglial behavior within hippocampal subregions and may support hippocampal circuit modeling endeavors⁵³.

Tissue source and sample classification

We analyzed paraformaldehyde-fixed brain sections from one hemisphere of 15 female rhesus macaques (Macaca mulatta) spanning late gestation through to old age, obtained from the MBRC collection 6. Animals were grouped into four stages based on age with perinatal (n = 4 brains, GD140-7 days postnatal), postnatal (n = 4, 2.5-6 months), juvenile/adult (n = 4, 11.4 months-9.8 years), aged (n = 3, 18.7–32.4 years) (Fig. 2A). The juvenile/adult group includes 2 prepuberal and 2 adult females. Optimally, these groups would be analyzed separately, as puberty might influence the distribution of microglia. However, given the low number of individuals in each group and the homogeneity of their results across fields (Figure S3), they were pooled together.

The MBRC collection 6 contains multiple immunohistochemical stains within each brain. Therefore, for microglial quantification, we used the pan-microglia marker IBA1, and for anatomical parcellation of hippocampal subfields and layers, we used AChE, MBP, CB, CR, SMI-32, and Nissl (Fig. 2B). All sections were downloaded as whole-slide pyramidal images in SVS-format for all animals and processed in their maximum resolution for parcellation and analysis. No animals were sacrificed for the present study. All materials were already available in the MBRC collection 6. In all cases, animal studies that contributed brain tissue to this collection were at the time conducted in accordance with all federal and state regulations and were reviewed and approved by the Institutional Animal Care and Use Committee (IACUC) of Yale University.

Parcellation of hippocampal subregions

Analysis was performed on series of IBA1-stained 50 μm thick sections taken at 1 mm intervals across mid-rostro-caudal levels of the hippocampus (approximately − 13.95 to -20.7 mm anterior to bregma⁵⁴ as the caudal-most hippocampal sections exhibit reduced field definition and ambiguity in discriminating subfield boundaries. The hippocampal fields were analyzed by strata: CA1–CA3 were divided into OrPyr, Rad, and LacMol, and the DG was parcellated into the Hil, GCL, and ML (Fig. 2C). Parcellation of subfields and layers on IBA1-stained sections was accomplished following established primate hippocampal cytoarchitecture^55,56 and referencing complementary series of sections labeled for IBA1, AChE, MBP, CB, CR, SMI-32 & Nissl within the same brain.

Due to evolving definitions of the CA2 field in the past several decades, CA2 was delineated using a combination of cytoarchitectonic and immunohistochemical criteria. Specifically, CA2 was identified in Nissl-labeled sections by the presence of larger pyramidal cell somas relative to adjacent CA1 and CA3 fields, consistent with previous descriptions in primates⁵⁷. This territory corresponds to the region extending from the distal termination of mossy fiber innervation toward proximal CA1, as described in rodent studies^58,59 and overlaps with the reported distribution of the CA2 marker RSG14 in the macaque hippocampus⁶⁰. To aid boundary identification, adjacent sections labeled for SMI-32 were also examined. Even though SMI-32 is not specific to CA2 and labels both CA2 and CA3 pyramidal neurons in rodents and primates^61–63, a distinct region of stronger dendritic labeling was consistently observed within the territory corresponding to the cytoarchitectonic definition of CA2 across MBRC samples (Fig. 2B). The convergence of pyramidal cell morphology, anatomical position and SMI-32 labeling were therefore used to define CA2 boundaries for subsequent analyses.

For each animal, 5–9 IBA1-stained sections were sampled yielding a total of 1,308 parcellated hippocampal regions across all brains. This multi-marker approach enabled consistent anatomical registration across brains despite differences in hippocampal size and/or microglia organization.

Estimation of hippocampal volume

We used the Cavalieri principle to estimate hippocampal volume stereologically using the sum of all cross-sectional subregion areas (mm²) in all sections (n = 5–9 based on our chosen stereotaxic coordinates (Eq. 1)).

1

Estimated hippocampal volumes were compared across age groups using a one-way analysis of variance (ANOVA) followed by Tukey’s post-hoc. Statistical significance was defined as p < 0.05.

Quantification of microglia density and nearest-neighbor distance

Digital SVS image files were imported into QuPath⁶⁴(v0.5.1) and hippocampal subregions were manually annotated on IBA1-stained sections using the polygon tool to define ROIs.

To ensure consistent IBA1 detection across animals and groups, we calibrated color deconvolution vectors by selecting one representative section per age group and ordering them by staining intensity. We then used the section with median IBA1 intensity to estimate optimal stain vectors via QuPath’s built in algorithm, and these parameters were applied uniformly across all images.

Microglial cell bodies were then detected using QuPath’s watershed-based cell detection with parameters optimized across representative sections from each developmental group: requested pixel size of 0.5 μm, background radius of 10 μm, median filter radius of 3 μm, Gaussian sigma of 3 μm, detection threshold of 0.4 (using the optical density channel) and a cell area minimum of 50 μm². For each detected cell, QuPath extracted centroid coordinates (X, Y in µm) and ROI area (mm²). Cells were assigned to a ROI only if their centroid (geometric center) fell within the ROI boundary, thus preventing double-counting at borders.

To validate our automated microglial detection pipeline, a subset of sections was manually quantified following an approach similar to previous image analysis validation of microglia³¹. One IBA1-stained section from the first available rostro-caudal level was selected from a representative subset of 10 animals spanning all four age groups. Within each section, three tiles were randomly sampled using QuPath’s built-in random tile sampling function, yielding 30 validation tiles in total. Two independent rates manually counted IBA1 immunolabeled microglial cell bodies within each tile, and the average manual count was used as the reference standard. Agreement between automated QuPath detections and manual counts was assessed using Pearson correlation and systematic bias and limits of agreement were further evaluated using Bland-Altman analysis.

Microglial density was calculated as the total number of detected cells divided by ROI area (cells/mm²; (Eq. 2)). Nearest-neighbor distance (NND) was computed as the Euclidean distance from each cell to its closest neighboring microglia within the same ROI using the nndist function from the spatstat package in R (v4.5.2) (Eq. 3) with mean NND calculated per ROI.

  

2

  

3

where NND is the Euclidean distance (shortest path) between a fixed cell centroid (x₁, y₁) and another cell centroid (x₂, y₂).

Microglia density and NND were measured across multiple hippocampal layers and sections from the same animals, resulting in a hierarchical dataset with observations nested within brains. To account for repeated measurements from individual animals while evaluating age- and region- specific effects, generalized linear mixed-effects models were implemented using the lme4 package in R. Intraclass correlation coefficients (ICC) were calculated to quantify the proportion of variance attributable to differences between brains. Adjusted ICC values indicated substantial clustering at the animal level for both density (ICC = 0.61) and NND (ICC = 0.60), demonstrating the observations obtained from the same brain were not independent. Therefore, separate models were fitted for density (cells/mm²) and NND (µm) using the following structure:

where Age group (perinatal, postnatal, juvenile/adult and aged) and Region (12 hippocampal layers comprising CA1-CA3 Pyr, Rad and LacMol, and the DG ML, GCL and Hil) with BrainID included as a random intercept.

Model convergence and fit were assessed for all outcomes (Figure S2). Post-hoc pairwise comparisons between age groups within each region were performed using estimated marginal means (emmeans package) with Bonferroni correction for multiple comparisons. All analyses were conducted in R with significance defined as p < 0.05.

Spatial point pattern analysis

To assess microglial spatial organization beyond density and spacing, we performed spatial point pattern analysis using Ripley’s K-function and its variance-stabilized transformations (L- and H-functions) implemented in the spatstat package in R^29–31. For all analyses, microglial centroids were treated as point patterns and translation edge correction was applied. Ripley’s H-function was used to quantify departures from complete spatial randomness (CSR). Positive H(r) values indicate clustering relative to CSR, and negative values indicate dispersion, while values near zero indicate a random spatial arrangement.

Spatial organization was first evaluated across the entire hippocampus. For each section, a polygon ROI encompassing the whole hippocampus was manually delineated in QuPath, exported as a GeoJson, and used as the observation window for Ripley analysis. The whole-hippocampus approach minimizes potential bias arising from differences in anatomical shape and size among subregions and across age groups, while preserving the native spatial arrangement of microglia. Ripley’s K(r), L(r) and H(r) functions were calculated as described below (Eqs. 4–7).

  

4

where λ is the mean density of cells (centroids), N is the number of cells within distance r, and Pi is the ith cell, and the sum is taken over n cells.

  

5

  

6

  

7

Extraction of spatial organization metrics

To derive biologically interpretable measures of microglial spatial organization, summary metrics were extracted from whole-hippocampus H(r) curves following approaches previously applied to spatial point pattern analysis^31–33. Specifically, we recorded the distance at which the minimum H(r) value occurred (rHmin), representing the radius of maximal dispersion, and the distance at which the maximum H(r) value occurred (representing the radius of maximal clustering). These metrics represent characteristic spatial scales of organization rather than the magnitude of clustering or dispersion.

Interpretation of these distances was guided by framework in existing literature and subsequent biological applications^31,32,52. The distance associated with the minimum of the H(r) curve was used as an estimate of the characteristic exclusion radius or territory size of interacting cells, reflecting the spatial scale over which neighboring cells avoid one another. In microglial spatial analyses, this metric has been interpreted as the average territory radius surrounding individual microglia³¹. Conversely, the distance associated with the maximum of the H(r) curve was used as an estimate of the characteristic radius of cellular aggregation and has been applied to quantify cluster size in microglial populations³³.

For rHmin, the second local minimum of the H(r) curve was selected to avoid the small-distance artifact commonly observed near the origin of Ripley’s functions. Metrics were calculated for each section, averaged within each brain and subsequently compared across age groups.

Tile-based local spatial organization analysis

Hippocampal subregions differ substantially in shape, size and anatomical complexity, making direct comparison of spatial statistics within anatomically defined ROIs challenging. Therefore, to evaluate local microglial organization while minimizing the influence of anatomical boundary effects, we implemented a tile-based spatial analysis strategy.

The entire hippocampus was subdivided into 200 × 200 μm tiles generated within the manually delineated whole-hippocampus GeoJson/polygon (Fig. 8D). Tiles intersecting the hippocampal boundary were constrained to the ROI such that regions outside the anatomical boundary were excluded from analysis. Ripley’s H(r) was calculated independently for each tile over a distance range of 0–100 μm, which was selected based on observed mean NND within the dataset. Only tiles containing at least 15 microglial centroids were included in the analysis to ensure sufficient sampling density for reliable estimation. Following Ripley analysis, each tile was assigned to a hippocampal subregion (CA1, CA2, CA3 or DG) according to the region’s parcellated ROI.

Tiles were classified as clustered or non-clustered according to the behavior of their H(r) curves relative to CSR. A tile was designated as when H(r) exceeded the CSR expectation throughout the sampled range, while all other tiles were classified as non-clustered. For each section, the percentage of clustered tiles was calculated and used as a measure of local microglial aggregation.

Exploratory spatial correlation analysis in B95

We performed additional analysis in the oldest animal (B95) to explore whether local microglial clustering was associated with amyloid-β (6E10) or phospho-tau (P217) immunoreactivity. Serial adjacent sections for IBA1, 6E10, and P217 were manually aligned in QuPath using interactive image alignment⁶⁴, guided by hippocampal anatomical landmarks to maximize correspondence between sections while accounting for minor distortions introduced during tissue processing. Following registration, a CA1 ROI was delineated and subdivided into 200 × 200 μm tiles using the same tiling framework described earlier. For the 6E10 and P217 sections, stain vectors were estimated and color deconvolution was performed in QuPath to separate DAB signal from the counterstain and mean DAB intensity was quantified within each tile and used as a measure of local 6E10 or P217 immunoreactivity. Tile-level measurements of 6E10 and P217 were extracted from the aligned sections and matched to corresponding CA1 tiles from the IBA1 dataset.

Tiles were categorized as clustered or non-clustered according to the criteria described above, and tile-level measurements were compared between clustered and non-clustered tiles using Wilcoxon rank-sum tests. To evaluate spatial-correspondence relationships at a broader scale, the percentage of clustered tiles was calculated for each CA1 section and compared with mean section-level 6E10 and P217 signal intensity using Spearman rank correlation.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

Concept and design (A.N., A.D., N.K.). Financial support (A.D., N.K.). Data acquisition (A.N., C.M., A.D.). Data analysis and interpretation (A.N., J.A., A.D., N.K.). A.N. and N.K. wrote the paper. All authors provided critical editorial revisions of the manuscript and approved the article.

Funding

This study was funded by the National Institutes of Health Grant MH113257 (AD). AN was supported by a George Mason University Presidential Scholarship.

Data availability

All brain sections used in this study are available via the MacBrain Resource Center collection 6 (https://macbraingallery.yale.edu/collection6/).

Declarations

Competing interests

The authors declare no competing interests.