Sporadic Alzheimer’s disease (sAD) unfolds over decades as genetic and environmental factors amplify neuroinflammatory signaling, accelerating pathology in the aging brain. Preventive treatments therefore hinge on pinpointing the earliest inflammation-driven triggers of disease. Traditional mouse models fall short because rodent immune responses and cortical architecture diverge markedly from those of humans (King, 2018). Aging rhesus macaques, by contrast, share human-like association cortices and immune pathways and spontaneously develop tau and amyloid pathology, synaptic loss, and cognitive decline, including age-related activation of critical glial cell-types such as microglia and astrocytes (Arnsten et al., 2021; Paspalas et al., 2018; Beckman et al., 2021; Beckman et al., 2019). Crucially, they allow direct study of the initial, soluble phase of hyperphosphorylated tau (pTau) — including pT217Tau — that is neurotoxic and capable of propagating pathology across cortical networks (Barthelemy et al., 2020; Palmqvist et al., 2020; Datta et al., 2024). These early-stage, soluble pTau species degrade almost immediately post-mortem in humans (Matsuo et al., 1994; Wang et al., 2015), limiting research to rare biopsy samples, whereas they can be sampled ex vivo from macaque brain and plasma and visualized with high-resolution imaging (Datta et al., 2024). Recent macaque studies reveal that age-related inflammatory cascades disrupt calcium homeostasis, drive tau hyperphosphorylation, and spur Aβ production; conversely, anti-inflammatory and calcium-modulating therapies markedly lessen pathology, pointing to viable strategies for early intervention (Datta and Arnsten, 2025; Arnsten et al., 2019; Arnsten et al., 2021; Datta and Arnsten, 2025). Because macaques are naturally homozygous for ApoE-ε4, insights from this model are especially pertinent for developing treatments tailored to ApoE-ε4 carriers, who have responded poorly to current Aβ- and tau-directed antibodies (van Dyck et al., 2023; Congdon et al., 2023), underscoring the need for alternative, upstream therapeutic strategies in appropriate model systems.

AD is defined by two key neuropathological features: extracellular amyloid-β (Aβ) plaques and intracellular neurofibrillary tangles (NFTs) composed of hyperphosphorylated tau. These processes are interconnected—Aβ oligomers can promote tau phosphorylation (Um et al., 2013), while phosphorylated tau aggregates may, in turn, enhance Aβ production, creating a self-reinforcing pathological loop (Arnsten et al., 2021; Paspalas et al., 2018). Notably, cognitive decline in AD correlates more strongly with the presence of NFTs than with Aβ plaques (Nelson et al., 2012), highlighting the importance of understanding how tau phosphorylation develops in aging association cortices.

Tau, a microtubule-associated protein encoded by MAPT, is generated in six splice variants that differ by the presence of 0, 1, or 2 N-terminal inserts (0 N, 1 N, 2 N) and by having either three or four microtubule-binding repeats (3R or 4R) at the C-terminal end. Structurally, tau contains an N-terminal projection domain, a proline-rich segment, the microtubule-binding region (with its 3 or 4 tandem repeats), and a short C-terminal tail (Fitzpatrick et al., 2017; Mandelkow and Mandelkow, 1998). Under normal conditions, tau stabilizes microtubules, but under pathological conditions, tau protein undergoes an extensive array of post-translational modifications: numerous kinases heavily phosphorylate residues in the proline-rich region; additional regulation comes from acetylation, methylation, ubiquitination, and proteolytic truncation. Collectively, these modifications disrupt tau’s normal microtubule-stabilizing functions and heighten its propensity to aggregate in neurodegenerative diseases (Fitzpatrick et al., 2017; Mandelkow and Mandelkow, 1998; Rauch et al., 2020; Rissman et al., 2004), ultimately forming fibrils within dendrites that eventually accumulate in the soma as NFTs. This pathological process is accompanied by neuronal death via autophagic degeneration, leaving behind characteristic “ghost tangles” (Arnsten et al., 2021).

In individuals with sAD, cortical tau pathology initially emerges in layer II of the transentorhinal and entorhinal cortices (ERC), corresponding to Braak Stages I–II (Hyman et al., 1984; Braak et al., 2011). From there, it spreads to interconnected limbic and association cortical areas, as well as the hippocampus, during Braak Stages III–IV. The layer II cell islands of the ERC serve as a critical hub, channeling input from widespread association cortices into the hippocampus to support new memory formation (Hyman et al., 1984). This region’s anatomical importance makes it especially vulnerable as a site where tau pathology begins to seed and propagate through cognitive and memory circuits (Kaufman et al., 2018). As the disease progresses, tau aggregates are found in regions such as the dorsolateral prefrontal cortex (dlPFC), which is essential for abstract reasoning, working memory, and executive functions—changes that correlate with cognitive impairment (Giannakopoulos et al., 2003). By Braak Stage V, tau pathology becomes widespread in association cortices but does not affect primary sensory areas, such as visual and auditory cortices, until the final stage of the disease (Braak Stage VI) (Braak et al., 2011; Lewis et al., 1987). This spatial progression of pathology mirrors the clinical course of AD, beginning with recent memory loss and expanding to broader cognitive decline and long-term memory impairment, while primary sensory-motor functions are largely preserved until late stages.

Amyloid beta (Aβ) peptides are generated through the sequential cleavage of amyloid precursor protein (APP) by β-secretase followed by γ-secretase (LaFerla et al., 2007). This process is accelerated within endosomes that contain β-secretase (Grynspan et al., 1997), and may be further intensified by the presence of the ApoE-ε4 genotype. In contrast, when APP is localized to the plasma membrane, it is more commonly processed by α-secretase, which leads to its degradation rather than Aβ production (LaFerla et al., 2007). Because Aβ is primarily released into the extracellular space, it is likely more readily detected in cerebrospinal fluid (CSF) and plasma compared to phosphorylated tau (pTau). As extracellular Aβ accumulates, monomers begin to aggregate into oligomers, then protofibrils, and eventually form insoluble fibrillar plaques. Amyloid pathology first emerges in the association areas of the temporal neocortex and later spreads to other regions of the neocortex (Braak and Braak, 1991). As AD advances, Aβ plaques also accumulate in subcortical structures such as the striatum and thalamus [reviewed in Braak and Del Trecidi (2015)]. This distribution pattern aligns with the idea that Aβ is released from the axons of neurons affected by pTau, with higher-order association cortices potentially contributing Aβ to their downstream projection targets (Braak and Del Trecidi, 2015; Braak and Del Tredici, 2015).

Comprehensive anatomical studies have shown that aging rhesus macaques exhibit a qualitatively similar pattern and progression of tau pathology to that seen in humans, across subcellular, cellular, and regional levels (Arnsten et al., 2019; Arnsten et al., 2021). For instance, the spatial and temporal development of tau pathology in rhesus macaque’s echoes that of sAD in humans—beginning in the layer II cell islands of the ERC, later appearing in the dlPFC, but notably sparing V1 cortices. In both species, tau phosphorylation originates in the distal dendrites and dendritic spines and gradually advances into the neuronal soma (Paspalas et al., 2018; Datta and Arnsten, 2025; Braak and Del Trecidi, 2015; Braak and Braak, 1991; Braak and Del Tredici, 2011). This pattern is especially evident with early, soluble phosphorylated tau epitopes such as pS214Tau and pT217Tau in rhesus macaques at the nanoscale-level with high spatial resolution ultrastructural studies (Paspalas et al., 2018; Datta et al., 2024; Carlyle et al., 2014). In fact, elevated plasma levels of pT217Tau are now considered a promising early biomarker that predicts future AD (Palmqvist et al., 2020; Barthélemy et al., 2024; Mendes et al., 2024; Pandey et al., 2025), and the macaque model provides key insights not possible in other species. Research in aging rhesus monkeys has revealed that early-stage, soluble pS214Tau and pT217Tau can traffic between neurons, particularly near or within glutamatergic synapses, potentially enabling the spread of tau pathology across networks of higher-order cortical glutamatergic neurons (Paspalas et al., 2018; Datta et al., 2024; Carlyle et al., 2014). This synaptic tau propagation has been observed in layer II of the ERC in middle-aged rhesus monkeys and in layer III of dlPFC in older animals (Paspalas et al., 2018; Datta et al., 2024; Carlyle et al., 2014). Similar tau “seeding” behavior has been documented in human sAD brain tissue, where the ERC demonstrated the highest efficiency in transmitting tau pathology (Kaufman et al., 2018). Mouse models have also provided evidence for tau propagation between neurons (Gibbons et al., 2019). However, the rhesus monkey studies offer direct visualization of this process at the nanoscale and suggest a plausible mechanism by which tau pathology may spread through association cortical circuits at very early stages of the disease, providing a remarkable opportunity to understand how this intersects with other pathological sequelae, e.g., inflammatory cascades (see below).

In human AD, affected neurons ultimately succumb to autophagic degeneration, leaving behind characteristic “ghost tangles” [reviewed in Arnsten et al. (2021)]. Correspondingly, our research has identified autophagic vacuolar degeneration in dendrites of aged rhesus macaques—specifically in layer II of the ERC in “early-aged” animals, and in layer III of the dlPFC in “late-aged” animals—mimicking the degenerative trajectory observed in human sAD (Paspalas et al., 2018; Datta et al., 2024). Strikingly, extremely aged rhesus macaques develop classic NFTs in the ERC, composed of paired helical filaments identical in size and helical frequency to those found in human sAD, and immunolabeled by the diagnostic AT8 antibody (pS202/pT205Tau) (Paspalas et al., 2018). These animals also show significant impairments in recognition memory (Paspalas et al., 2018), paralleling early cognitive symptoms in humans.

Age-related structural changes in rhesus macaques also align with human pathology. Notably, there are selective decreases in dendritic spine density in the dlPFC, but not in V1 (Young et al., 2014), consistent with the pattern of synaptic loss seen in human AD (DeKosky and Scheff, 1990). In addition, aged rhesus macaques naturally develop amyloid plaques with similar morphology and dimensions to those in humans (Uno and Walker, 1993; Mufson et al., 1994). Beyond amyloid and tau pathology, these monkeys also display a range of AD-like neurodegenerative features, including mitochondrial dysfunction, inflammatory pathway activation, microglial engulfment of synapses, synaptic degeneration, argyrophilic deposits, buildup of late-phase lysosomes, and dystrophic neurites (Paspalas et al., 2018; Datta et al., 2024; Datta and Arnsten, 2025). Rhesus macaques of extreme age also exhibit pronounced recognition memory deficits as well as impaired executive functions (Arnsten et al., 2021; Rapp and Amaral, 1989). These pathological similarities to human sAD provide construct validity in using non-human primate models to investigate the etiology of inflammatory dysfunction in neurodegenerative diseases.

While non-human primate (NHP) models offer unique advantages for studying the earliest etiological events of sAD—particularly those involving higher-order association cortices, primate-specific immune signaling, and soluble tau species—they also present practical and conceptual limitations (Defelipe, 2011; Hardingham et al., 2018; Hill and Walsh, 2005; Preuss, 2012; Sousa et al., 2017). NHP studies are resource-intensive, have lower throughput, and offer limited opportunities for large-scale genetic manipulation, which can constrain mechanistic dissection of complex gene–gene and gene–environment interactions (Defelipe, 2011; Hardingham et al., 2018; Hill and Walsh, 2005; Preuss, 2012; Sousa et al., 2017). In contrast, mouse models provide unparalleled genetic tractability, enabling precise manipulation of individual inflammatory, metabolic, and synaptic pathways and rapid testing of causal relationships. Rodent systems are therefore especially powerful for probing multifactorial genetic risk, screening therapeutic targets, and defining molecular cascades underlying neuroinflammation. However, key aspects of sAD—including association cortex architecture, microglial transcriptional states, neuroinflammatory mechanisms, neuromodulatory control of calcium signaling, and the emergence of soluble tau pathology—are either absent or fundamentally different in rodents. Accordingly, we view rodent and NHP models as complementary rather than competing systems: mouse models are ideally suited for mechanistic discovery and hypothesis generation, whereas aging NHPs are essential for validating disease-relevant mechanisms and therapeutic strategies in a translationally faithful neuroimmune and cortical context.

The primate cortex exhibits a steep hierarchical, lattice-like organization (Magrou et al., 2024). Distinct functional and structural differences emerge as one moves from the primary sensory areas through sensory association regions, up to higher-order cognitive association cortices, and to limbic cortices involved with emotion. As schematically illustrated in Figure 1A, computational studies have shown that neurons across this hierarchy operate on progressively longer timescales, meaning that a neuron’s current activity is increasingly shaped by its past activity—supporting processes like sensory integration, working memory, and sustained emotional states (Murray et al., 2014; Arnsten et al., 2021; Monosov et al., 2020).

These functional gradients are mirrored by anatomical and molecular features. Pyramidal neurons in higher-order regions exhibit greater dendritic spine density than those in lower regions such as primary visual cortex [V1; Elston et al. (2001) and Elston et al. (2011)], and increased expression of calcium-related genes, including GRIN2B (which encodes the slow-closing, high-calcium-flux NMDA receptor subunit GluN2B), and CALB1 (calbindin), a marker of elevated intracellular calcium handling (Figure 1B) (Burt et al., 2018). In macaques, hierarchical expression of calbindin is driven by pyramidal neurons, not interneurons, consistent with their expanding synaptic inputs across the cortical hierarchy (Figure 1C) (Kondo et al., 1999), including further expansion of calbindin expression in pyramidal cells of the limbic cortices (Joyce et al., 2020). Calcium signaling is often increased by cAMP-PKA signaling (Arige and Yule, 2022), and proteomic data from human brain shows increasing expression of PDE4D and GRM3 (mGluR3)—both key regulators of cAMP–PKA signaling—along the hierarchy, with the highest levels in the dlPFC and lowest in V1 (Carlyle et al., 2017). Similarly, there is an expansion of stress-related genes in layer III pyramidal cells across the cortical hierarchy, including increases in dopamine D1 receptors (DRD1) and the gene encoding the “master stress peptide,” PACAP (Pituitary Adenylate Cyclase-Activating Polypeptide) (Bian et al., 2024), both of which increase cAMP and/or calcium signaling. There is also a hierarchical increase in the gene encoding the SK3 potassium channel (Enwright et al., 2022; Arion et al., 2023), which is opened by calcium, and causes calcium to reduce neuronal firing in the dlPFC (Datta et al., 2024). Inflammatory pathways and cell-types also show gradients across cortical hierarchy. Transcriptomic profiling across 15 brain regions in rhesus macaques shows regionally graded age-associated gene expression changes, especially in immune and neurodegeneration-related genes, particularly affecting association cortices like the dlPFC (Chiou et al., 2022). Notably, the pattern of calbindin expression across pyramidal neurons aligns strikingly with the regions most affected by tau pathology and neurodegeneration in AD (Arnsten et al., 2021).

Interestingly, this hierarchical pattern for calcium signaling and inflammatory mechanisms has also expanded during primate evolution. In humans, there is a marked increase in both dendritic spine density and GRIN2B expression compared to simpler primates (Elston et al., 2001, 2011; Muntané et al., 2015). These species-specific features are highly relevant to AD, as many neuroinflammatory processes show expanded or altered expression in humans relative to rodents (Kodamullil et al., 2017), and mice have a less differentiated cortical hierarchy (Magrou et al., 2024; Gilman et al., 2017). For example, microglia in rodents typically exhibit pronounced state transitions, cytokine levels, and morphological changes following acute insults; whereas microglia in primates exhibit more gradual, region-specific alterations with more subtle microglial process velocity, dystrophic morphologies, and chronic low-level cytokine increases during aging [reviewed in Edler et al. (2021)]. In addition, cross-species single-cell analysis reveals distinct microglial gene modules in primates, including unique complement pathway and inflammatory-related expression patterns not seen in rodents (Geirsdottir et al., 2019). As described below, inflammatory pathways affecting NMDAR neurotransmission and mGluR3 regulation of cAMP-calcium signaling are also expanded in primates. Thus, primate models are especially valuable for studying early disease mechanisms that are absent or poorly represented in rodent brains—particularly in excitatory neurons and glial cell-types in higher-order cortical regions.

The sustained firing of dlPFC neurons is needed to maintain information in working memory across a delay when sensory stimulation is no longer available. These dlPFC “Delay cells” are thought to reside in layer III, and exhibit unusual neurotransmission and neuromodulation required for flexible but sustained neuronal firing. dlPFC Delay cells rely heavily on NMDAR neurotransmission (Figure 2A), including receptors with GluN2B as well as GluN2A subunits within the post-synaptic density (PSD). As mentioned above, NMDARs containing GluN2B conduct particularly large calcium currents. NMDAR neurotransmission can only occur when the PSD is depolarized, ejecting Mg²⁺ from the NMDAR ion pore (Nowak et al., 1984). In a typical glutamate synapse, glutamate stimulation of AMPARs supply these permissive excitatory actions (Bekkers and Stevens, 1989). However, for layer III dlPFC Delay cells, the permissive excitation is supplied by acetylcholine, including by nicotinic a7-receptors (nic-a7R) in the glutamate PSD (Figure 2A) (Yang et al., 2013). Nic-a7R also flux calcium into the cell, which may be helpful in maintaining a depolarized PSD (Yang et al., 2013). Calcium is also contributed by L-type voltage-gated channels such as Cav_1.2 (Datta et al., 2024), and by internal calcium release from the smooth endoplasmic reticulum (SER), which is called the spine apparatus within dendritic spines (Figure 2A) (Datta et al., 2024). cAMP-PKA signaling drives calcium release from the SER and through Cav_1.2, and calcium in turn drives more cAMP production, thus creating feedforward cAMP-calcium signaling (seen in more detail in Figure 3). Under healthy conditions, feedforward cAMP-calcium signaling in layer III dlPFC pyramidal cells is tightly regulated by calbindin buffering of calcium, phosphodiesterase (PDE4) catabolism of cAMP, and α2A-AR and mGluR3 inhibition of cAMP production (Figure 2A). Unlike in rodents where mGluR3 are predominately presynaptic, in primate layer III dlPFC they are post-synaptic on dendritic spines, where they inhibit cAMP drive on calcium release (Jin et al., 2018; Jin et al., 2017). mGluR3 are stimulated not only by glutamate, but by NAAG, which is co-released with glutamate and is selective for mGluR3 (Figure 2A). NAAG greatly increases dlPFC Delay cell firing, emphasizing the power of this mechanism in primates (Jin et al., 2018; Jin et al., 2017).

Under conditions of uncontrollable stress, high levels of catecholamine release in the dlPFC drives feedforward cAMP-calcium signaling to open nearby K⁺ channels to rapidly weaken network connections and reduce Delay cell firing (Figure 2B) (Datta and Arnsten, 2019; Woo et al., 2021; Datta and Arnsten, 2018). High levels of norepinephrine release during stress engage low affinity, α1-AR and β1-AR; the latter activate nearby Cav_1.2 currents, similar to stress actions in heart (Datta et al., 2024; Datta et al., 2019). High levels of cytosolic calcium in turn open SK channels, PKA opens KCNQ2 channels, and cAMP opens HCN-Slack channels on spines, reducing the efficacy of NMDAR synapses, decreasing Delay cell firing and switching control of behavior to more primitive circuits that are strengthened by high levels of norepinephrine release (Datta and Arnsten, 2019; Ramos et al., 2003; Vijayraghavan et al., 2007). In healthy subjects, regulation, e.g., by PDE4s, restores dlPFC connectivity, and these rapid changes in network connectivity are a “signature of flexibility” that promote coordination of arousal state with cognitive state, a mechanism termed Dynamic Network Connectivity (Arnsten et al., 2012; Arnsten et al., 2022). However, with chronic stress and/or inflammation, PFC connections are lost. In this context, it is noteworthy that stress is a risk factor for future AD (Datta and Arnsten, 2025; Arnsten et al., 2019; Arnsten and Datta, 2024; Arnsten et al., 2025; Bathla et al., 2023).

It is not known whether this “signature of flexibility” is common to other association cortical synapses, but it is highly relevant to the etiology of AD that layer II of the ERC, which are the first cortical cells to show tau pathology, exhibit this signature on spines and at excitatory synapses on dendrites of excitatory neurons (Datta et al., 2023). However, the layer II cell islands in ERC have sparse expression for calbindin, even when young and healthy (Beall and Lewis, 1992). The expression of magnified calcium signaling in the absence of protective factors may contribute to vulnerability to tau pathology.

Inflammation weakens layer III dlPFC connections on dendritic spines by blocking neurotransmission, and by dysregulating the stress response (Figure 2C). Under inflammatory conditions, microglia metabolize tryptophan to kynurenine, which can be further metabolized to kynurenic acid (KYNA; Figure 2C) or quinolinic acid (Badawy, 2017). Kynurenine is also generated by the peripheral immune system and is actively taken up from blood into brain (Stone and Williams, 2023). Much of the research in this field has focused on the excitotoxic effects of quinolinic acid, which stimulates NMDAR, with KYNA referred to as the protective metabolite, as it blocks NMDAR (Stone and Addae, 2002). Although this may be true under conditions of excess glutamate, such as during stroke, KYNA’s actions may be detrimental to cognition under conditions of normal or reduced glutamate actions. It is noteworthy that KYNA blocks nicotinic-a7R as well as NMDAR, the two receptors most needed for dlPFC neurotransmission (Albuquerque and Schwarcz, 2013). Recent research shows that KYNA markedly reduces the dlPFC Delay cell firing in macaque dlPFC needed for working memory (Figure 2C), and conversely, that inhibiting the production of KYNA can restore neuronal firing and working memory performance in aged macaques (Yang et al., 2024). Thus, elevated KYNA in dlPFC is detrimental to higher cognitive function, and this is especially true in primates. KYNA is metabolized from kynurenine by KAT II, and transcriptomic analyses show a great expansion in the gene encoding KAT II from mice to primates, including extensive expression within neurons in macaque and human dlPFC (Yang et al., 2024). This may reflect the parallel expansion of NMDAR-GluN2B expression in primate dlPFC (Muntané et al., 2015), and helps to explain why so many neuroinflammatory disorders, e.g., long-COVID, are associated with dlPFC cognitive impairment (Vanderlind et al., 2021). Interestingly, KYNA activates IDO, the enzyme that metabolizes tryptophan to kynurenine, and thus sustains its own production (Badawy, 2023). This may contribute to the prolonged nature of many neuroinflammatory disorders, such as long-COVID (Cysique et al., 2023). KYNA is elevated in the early stages of AD (Almulla et al., 2022), and increased plasma kynurenine correlates with measures of Aβ and neurofilament light chain assays of degeneration (Chatterjee et al., 2019). It is noteworthy that sustained exposure to KYNA in vitro causes synapse loss (Orhan et al., 2025), and disorders associated with elevated KYNA such as schizophrenia (Schwarcz et al., 2001; Kindler et al., 2020) and AD (Widner et al., 2000) are associated with synapse loss in the dlPFC (DeKosky and Scheff, 1990; Glantz and Lewis, 2000).

Advanced age and/or inflammation also causes dysregulation of stress response in primate dlPFC, with loss of calbindin, PDE4A and PDE4D and reduced mGluR3 regulation of cAMP-calcium signaling (Figure 2C) (Datta and Arnsten, 2025). Inflammation induces microglia and astrocytes to generate and release GCPII (glutamate carboxypeptidase II) which catabolizes NAAG, the endogenous ligand for mGluR3, thus dysregulating feedforward cAMP-calcium-K⁺ channel signaling (Figure 2C) (Datta and Arnsten, 2025). Conversely, inhibiting GCPII greatly enhances Delay cell firing and improves working memory in aged macaques (Yang et al., 2022). GCPII activity is also related to tau pathology, and GCPII inhibitors may have therapeutic potential.

With sustained inflammation, dysregulated cAMP-calcium signaling leads to tau and amyloid pathology and autophagic degeneration, as summarized in Figures 3, 4 [reviewed in Datta and Arnsten (2025), Arnsten et al. (2019), Arnsten et al. (2021), and Arnsten et al. (2025)]. Very high levels of cytosolic calcium activate calpain-2, which cleaves and disinhibits a number of culprits known to drive AD pathology. Calpain-2 cleaves off the regulatory end of GSK3β, a major kinase involved with tau hyperphosphorylation, and it cleaves p35 to p25, which activates cdk5, another key kinase in tau hyperphosphorylation. Activated cdk5 also increases β-secretase cleavage of APP to Aβ₄₂, thus increasing amyloid pathology. Ab₄₂ and the AICD peptide cleaved from APP both increase internal calcium release, and Ab₄₂ can create calcium pores in the plasma membrane, further driving pathology. Calpain-2 also cleaves and activates Hsp_70.1 which drives autophagic degeneration, the manner by which neurons die in AD. Thus, inflammation can drive multiple aspects of AD pathology via calcium dysregulation in dlPFC, and in other vulnerable neurons.

As described above, mGluR3 have a new, regulatory role in primate association cortex that is especially important to cognition and cognitive disorders. Genetic studies emphasize the importance of mGluR3 and GCPII signaling to human cognition. For example, a loss-of-function in GRM3 encoding mGluR3 is a risk factor for schizophrenia (Arnsten and Wang, 2020), and a gain-of-function alteration in FOLH1, which leads to excessive levels of GCPII, is associated with impaired cognitive abilities in humans (Zink et al., 2020). Research in macaques helps to explain why this signaling pathway has such importance to human cognition.

In contrast to rodents where mGluR3 are primarily presynaptic and inhibit glutamate release (Woo et al., 2022), research in the rhesus monkey dlPFC has revealed that mGluR3 are postsynaptic on dendritic spines, and they play a key role in regulating cAMP drive on calcium-K⁺ channel signaling (Figure 2A), thus maintaining the strength of excitatory connections involved in working memory (Jin et al., 2018; Jin et al., 2017; Yang et al., 2022; Arnsten and Wang, 2020). This spatial configuration with mGluR3 immunolabeling in postsynaptic compartments in dendritic spines and shafts is also expressed in macaque ERC layer II microcircuits, that are especially vulnerable in sAD (Datta et al., 2023).

Within dendritic spines, mGluR3s are predominantly localized on the spine membrane in close proximity to the spine apparatus—a specialized extension of the SER that regulates intracellular calcium dynamics within dendritic spines (Jin et al., 2018; Jin et al., 2017; Datta et al., 2023). This spatial arrangement places mGluR3 in an ideal position to modulate local calcium signaling. As described above, elevated cAMP-PKA activity can stimulate calcium release from the spine apparatus, which in turn promotes further cAMP production, creating a feedforward loop of calcium–cAMP–PKA signaling (Figure 3). PKA signaling also enhances calcium influx through NMDA receptors and L-type voltage-gated calcium channels like Cav_1.2, leading to cytosolic calcium accumulation (Datta et al., 2024; Wang et al., 2013) (Figure 3). The physiological contributions of mGluR3s have been demonstrated through iontophoretic application of NAAG (the endogenous mGluR3 agonist) or GCPII inhibitors directly onto dlPFC neurons in behaving monkeys (Yang et al., 2022; Arnsten and Wang, 2020). Both treatments significantly increased task-related neuronal firing by suppressing the cAMP–PKA–K⁺ channel pathway. A dose-dependent relationship was observed, where higher GCPII activity (and thus lower NAAG availability) led to reduced neuronal firing, underscoring the regulatory role of this signaling pathway (Yang et al., 2022; Arnsten and Wang, 2020). Overall, under normal physiological conditions NAAG–mGluR3 signaling serves a protective role by suppressing excessive cAMP and calcium signaling, thereby enhancing the connectivity of primate higher cortical circuits, fundamentally differing from the inhibitory presynaptic role of mGluR3 in rodents.

In the aging primate brain, the GCPII-NAAG-mGluR3 pathway becomes compromised due to the upregulation of GCPII, which degrades NAAG (Bathla et al., 2023; Arnsten and Wang, 2020). Pharmacological inhibition of GCPII has emerged as a promising strategy to restore mGluR3 function and suppress pathogenic inflammatory and calcium signaling. Acute administration of GCPII inhibitors in aged macaques has been shown to restore neuronal firing in the dlPFC and improve working memory performance, providing direct functional evidence for therapeutic benefit. Supporting these findings, rodent studies demonstrate that GCPII inhibition enhances spatial memory and object recognition (Datta et al., 2021; Olszewski et al., 2017), reinforcing its cognitive benefits across species. Notably, the orally bioavailable GCPII inhibitor 2-MPPA is particularly well suited for translational use due to its favorable side effect profile, making it viable for long-term preventive administration in at-risk individuals.

In aged rhesus macaques, increased GCPII activity has been strongly correlated with elevated levels of pT217Tau, a pathological marker associated with early AD, suggesting that inflammation-induced loss of mGluR3 signaling may directly promote tau pathology (Bathla et al., 2023). Chronic treatment with 2-MPPA in aging rhesus macaques has led to a significant reduction in both GCPII activity and pT217Tau levels in the dlPFC and ERC, two brain regions highly vulnerable to early AD pathology (Bathla et al., 2023). Additionally, decreases in pT217Tau were also observed in blood plasma, indicating the potential of this biomarker for tracking treatment response non-invasively even in non-human primates (Bathla et al., 2023). These findings highlight the therapeutic promise of targeting GCPII to preserve mGluR3 signaling, regulate intracellular calcium, and ultimately protect neural circuits from tau-mediated degeneration (Datta and Arnsten, 2025). Given the conserved biology between macaques and humans in this signaling pathway, aging rhesus macaques represent a powerful translational model for evaluating early-stage AD interventions (Datta and Arnsten, 2025). The accumulated data strongly supports further investigation of 2-MPPA and related GCPII inhibitors as viable preventative treatments for sAD, particularly in individuals with inflammation-related risk factors.

The current therapeutic landscape targeting neuro-immune interactions in sAD reflects a growing recognition that inflammatory signaling, microglial state transitions, and neuronal calcium dysregulation are deeply interconnected drivers of disease progression. Many emerging therapeutic strategies—while often framed around immune modulation or protein clearance—intersect mechanistically with calcium- and cAMP-dependent pathways that regulate synaptic plasticity, tau phosphorylation, and neuronal survival. Thus, immune-targeted therapies may exert downstream benefits by indirectly restoring calcium homeostasis in vulnerable association cortices.

Several anti-inflammatory and metabolic agents currently under investigation act upstream of calcium dysregulation by suppressing cytokine-driven kinase signaling. For example, NE3107, a small molecule inhibitor of the NF-κB/ERK axis, reduces MAPK activation and pro-inflammatory cytokines including TNFα, IFNγ, IL-1α, and TGF-β, while also enhancing insulin signaling (Haroon et al., 2024; Reading et al., 2021). Because cytokine-activated kinases such as ERK and PKA can potentiate calcium release from intracellular stores and amplify cAMP–calcium feedforward signaling, dampening these pathways may indirectly stabilize calcium dynamics in pyramidal neurons. Similarly, semaglutide, a GLP-1 receptor agonist originally developed for metabolic disease, has been associated with reduced dementia risk in diabetic populations (Norgaard et al., 2022), although recent Phase 3 clinical trials have yielded negative results. GLP-1 signaling has been shown to modulate neuroinflammation, mitochondrial function, and calcium handling, suggesting that metabolic–immune therapies may converge on shared calcium-regulatory mechanisms.

Microglia-focused biologics further highlight the intersection between immune pathways and neuronal calcium vulnerability. AL002c, a monoclonal IgG1 antibody acting as a TREM2 agonist, promotes microglial state transitions associated with phagocytosis and plaque containment (Wang et al., 2020). In animal models carrying the R47H TREM2 variant, chronic AL002c treatment reduced filamentous amyloid plaques, neuritic dystrophy, and microglial inflammatory responses while improving behavioral outcomes (Wang et al., 2020). Importantly, microglial encapsulation of plaques—the so-called “microglial barrier”—limits the diffusion of inflammatory mediators, reactive oxygen species, and synaptotoxic factors that can destabilize neuronal calcium signaling (Condello et al., 2018). Thus, TREM2-based strategies may indirectly protect synapses and dendritic calcium nanodomains by shaping the inflammatory microenvironment. Related immune pathways, including the complement cascade, further link neuroinflammation to synaptic and calcium-dependent pathology. Complement proteins such as C1q and C3 are upregulated with aging and AD and contribute to aberrant synaptic pruning. Excessive complement activation can weaken synaptic integrity and increase neuronal calcium load, thereby sensitizing circuits to tau phosphorylation and degeneration. Although complement-targeting therapies are still largely preclinical in AD, these mechanisms align closely with models in which inflammatory weakening of synaptic and calcium-regulatory systems precedes overt neurodegeneration.

Peripheral immune modulation also intersects with central neuroimmune–calcium pathways. For example, daratumumab, an FDA-approved anti-CD38 antibody, modulates CD38(+) CD8(+) T cells. Single-cell immune profiling in AD patients has revealed an expansion of CD8(+) effector memory T cells that negatively correlates with cognitive performance (Gate et al., 2020). Because peripheral immune activation can influence central cytokine levels, oxidative stress, and microglial reactivity, targeting these pathways may further reduce inflammatory amplification of neuronal calcium signaling.

Finally, our work on GCPII inhibition provides a direct mechanistic bridge between immune activation and calcium dysregulation. Inflammatory upregulation of GCPII degrades NAAG and weakens mGluR3-mediated suppression of cAMP–calcium signaling in primate association cortex, thereby promoting tau pathology. Therapeutic strategies that restore neuromodulatory control of calcium signaling may therefore complement microglial- and cytokine-focused interventions. Collectively, these emerging therapies suggest that successful disease modification in sAD may require coordinated targeting of immune pathways and the calcium–cAMP signaling cascades through which inflammation exerts its most deleterious effects on vulnerable cortical circuits.

Neuroinflammation-focused therapeutic strategies in AD are rapidly advancing, with efforts ranging from microglial modulation to cytokine regulation and synaptic protection. Although some clinical approaches have faced setbacks, the therapeutic pipeline continues to expand, underscoring the centrality of immune mechanisms in disease progression. Importantly, NHPs provide a critical bridge between rodent models and humans for the development of these therapies. Unlike rodents, NHPs naturally develop AD-like pathology, including amyloid and tau accumulation within the same association cortices affected in humans, and they exhibit age-related neuroinflammation that closely parallels human disease. Future work integrating human iPSC-based models across ApoE genotypes, as well as emerging humanized ApoE non-human primate approaches, will be important for mechanistic testing of neuro-immune hypotheses. Moreover, NHPs share highly conserved immune gene expression profiles and microglial responses, while supporting translational biomarker validation such as plasma and CSF pT217Tau. Their unique convergence of immune and neural features, combined with cognitive complexity, makes NHPs indispensable for testing neuroimmune-targeted interventions and predicting both efficacy and safety in humans.