Emerging roles of granzymes in neurodegeneration and neuroinflammation: mechanistic insights and therapeutic opportunities

Abstract

Neurodegenerative diseases are increasingly recognized as disorders shaped not only by intrinsic neuronal vulnerability, but also by chronic neuroinflammation mediated by maladaptive neuroimmune signaling. Granzymes, a family of serine proteases classically studied for their cytotoxic roles in anti-viral and anti-tumor immunity, are emerging as important mediators of central nervous system (CNS) pathology. In addition to their canonical intracellular functions, granzymes can act extracellularly to cleave the extracellular matrix (ECM), activate cell-surface receptors, disrupt epithelial barrier function, amplify inflammatory cascades, and alter glial and neuronal responses to injury. In this review, we synthesize current knowledge on the roles of Granzyme A (GzmA), Granzyme B (GzmB), Granzyme H (GzmH), and Granzyme K (GzmK) in neurodegeneration and neuroinflammation across diverse CNS disease and injury contexts, such as Alzheimer’s disease (AD), multiple sclerosis (MS), stroke, spinal cord injury (SCI), and age-related macular degeneration (AMD). GzmB is the most extensively characterized, with evidence supporting both intracellular neurotoxicity and extracellular pathogenic functions mediated through protease-activated receptor signaling, ECM cleavage, outer blood–retina barrier disruption, angiogenesis, fibrosis, and chronic inflammation. GzmA is implicated in tau proteolysis and structural destabilization of neurons and astrocytes, while GzmK has emerged as a context-dependent regulator of neuroinflammation through PAR-1 activation, microglial modulation, and complement cascade activation. GzmH remains the least understood but may contribute to nerve injury through mechanisms that are only beginning to be defined. We also discuss endogenous and pharmacological granzyme inhibition, highlighting the therapeutic promise of selective extracellular granzyme targeting, particularly for GzmB, while emphasizing the current lack of selective inhibitors for GzmA, GzmK, and GzmH. Collectively, these findings position granzymes as underappreciated neuroimmune effectors and potential therapeutic targets in neurodegenerative diseases and CNS injury.

Introduction

Neurodegenerative diseases, such as Alzheimer’s disease (AD), multiple sclerosis (MS), and age-related macular degeneration (AMD), represent a growing global health burden, with prevalence rising alongside aging populations [58, 127]. These disorders are generally characterized by abnormal protein aggregation, region-specific vulnerability within the central nervous system (CNS), and distinct neurological symptoms [138]. Broadly, these symptoms fall into three domains: cognitive impairments such as memory loss and executive dysfunction in AD [138], motor dysfunction including tremors and spasticity in MS [83], and sensory deficits such as progressive vision loss in AMD [136] and changes in visual perception in AD [2] and MS [124]. Although symptomatic treatments exist, none address the root cause: progressive neurodegeneration, underscoring the urgent need to identify common pathogenic drivers across CNS diseases.

Neurodegeneration can be driven by various pathogenetic processes, including oxidative stress, metabolic dysregulation, proteostasis, and aging [22, 43, 116, 138]. However, neuroinflammation has emerged as a unifying theme across diverse neurodegenerative diseases, acting not only as a consequence of neuronal injury but also as a driver of disease progression [104, 143]. Within the CNS, microglia serve as the resident immune cells and play a dual role in maintaining homeostasis and driving pathology [138]. While they are essential for synaptic plasticity and debris clearance [32], chronic microglial activation contributes to sustained neuroinflammation, releasing cytokines and reactive oxygen species (ROS) that exacerbate neuronal vulnerability [67]. Although the CNS was once considered “immune privileged”, it is increasingly recognized that peripheral lymphocytes such as cytotoxic T lymphocytes (CTL) and natural killer (NK) cells can infiltrate the CNS under pathological conditions [74, 104]. These infiltrating lymphocytes release effector molecules, which can amplify and prolong neuroinflammation, ultimately accelerating neurodegeneration within the CNS.

Among these effector molecules, granzymes, a family of serine proteases traditionally studied in the context of immune defense against viral infection and cancer [20], are gaining attention for their roles in the CNS. Humans express five granzymes: Granzyme A (GzmA), Granzyme B (GzmB), Granzyme H (GzmH), Granzyme K (GzmK), and Granzyme M (GzmM) [20]. While GzmB has been most extensively studied, evidence is emerging that other family members, including GzmA, GzmH, and GzmK, may also contribute to neurodegenerative processes through distinct mechanisms. These proteases can act intracellularly to induce cell death or, in many cases, function extracellularly to cleave ECM components and generate bioactive fragments that contribute to inflammatory signaling [11, 20, 48, 90]. In contrast, GzmM has not been implicated in CNS pathology, and to date, no studies have examined its role in neurodegeneration; therefore, it is not discussed further in this review.

To provide a unified framework for interpreting granzyme activity across both chronic neurodegenerative diseases and acute CNS injuries, we categorize granzyme actions into four mechanistic domains:

1. Intracellular cytotoxicity including perforin- or receptor-mediated granzyme entry and caspase-dependent or caspase-independent cell death, and direct intracellular proteolysis of neuronal proteins

2. Receptor-mediated signaling such as protease-activated receptor (PAR) activation and downstream inflammatory cascades, including PAR-1-dependent perturbations of neuronal proteostasis (e.g., tau hyperphosphorylation)

3. Extracellular proteolysis encompassing extracellular matrix (ECM) cleavage, blood–CNS barrier disruption, cleavage of extracellular or released proteins, and release of bioactive fragments

4. Immune amplification involving cytokine induction, complement activation, and modulation of microglial and infiltrating immune responses.

Building on this framework, we synthesize current knowledge on the roles of GzmA, GzmB, GzmH and GzmK in CNS pathology, highlight mechanistic insights into their contribution to neurodegeneration and neuroinflammation, and evaluate the therapeutic potential of targeting granzyme activity. By integrating evidence across multiple disease contexts, we aim to clarify the emerging roles of granzymes in CNS pathology and identify critical gaps for future investigation.

Granzyme B in neurodegeneration: from classical cytotoxicity to extracellular signaling

Granzyme B-mediated neuronal death: intracellular and extracellular mechanisms

Among the granzyme family, GzmB has the most robust experimental support in CNS pathology, with convergent evidence from human tissue, animal models, and mechanistic studies. Its canonical cytotoxic intracellular function is well documented across diverse disease contexts [126]. It is a potent cytotoxic protease that cleaves after aspartic acid residues [5]. During target cell killing, CTLs and NK cells form an immunological synapse with the target cell and release perforin, which creates pores in the plasma membrane [16]. Through these pores, GzmB enters the cytoplasm and initiates intrinsic apoptosis by cleaving pro-caspase-3 and Bid, triggering mitochondrial cytochrome c release, apoptosome formation, and the activation of caspase-9 and downstream executioner caspases, and ultimately leading to DNA fragmentation [117]. However, accumulating evidence indicates that GzmB can also trigger neuronal death extracellularly in a perforin-independent manner.

Early work by Wang et al. demonstrated that supernatants from activated CD4⁺ and CD8⁺ T cells contain GzmB and induce neurotoxicity in human fetal neurons in vitro [128]. This effect was independent of perforin, as the addition of perforin did not significantly enhance GzmB-mediated neurotoxicity [128]. Furthermore, pre-treatment with mannose-6-phosphate (M6P) failed to block GzmB-mediated neurotoxicity, indicating that GzmB does not enter neurons via M6P receptors (M6PR) [128]. In contrast, pertussis toxin, an inhibitor for G_iα/G_O-coupled receptors, prevented GzmB-mediated neurotoxicity, supporting that GzmB can act extracellularly to induce neuronal death via G protein activation [128]. On the contrary, Haile et al. reported that GzmB can enter human cortical fetal neurons via M6PR and induce neuronal death [39]. They showed that purified granule-derived GzmB resulted in neurotoxicity to a similar extent as activated T cells, whereas activated T cells from GzmB KO mice failed to induce neurotoxicity [39]. They further demonstrated that internalized GzmB is taken up into early endosomes, diffuses out into the soma, activates caspases, and cleaves α-tubulin, and that M6P pre-treatment reduced GzmB-mediated neuronal death [39]. Together, these studies indicate that GzmB can induce neuronal death through either extracellular signaling or receptor-mediated internalization, with the dominant mechanism likely varying by neuronal subtype, developmental stage, or experimental conditions.

In a follow-up study, Wang et al. showed that GzmB-mediated neurotoxicity involves activation of protease-activated receptor (PAR)-1, using the PAR-1 inhibitor SCH 79797 [129]. PAR-1 activation by GzmB led to increased neuronal expression of the voltage-gated potassium channel Kv1.3 in vitro, a finding that was also observed in cerebral cortical lesions from patients with MS [129]. The functional involvement of Kv1.3 in neuronal death was further validated in vivo. Stereotaxic injection of GzmB into the hippocampus of Sprague Dawley rats followed by intraperitoneal administration of Clofazimine, a Kv1.3 inhibitor, for 10 days promoted neuroprotection of immature hippocampal neurons against GzmB-induced neuronal death [129]. In a separate study, they also demonstrated that blocking Kv1.3 can also protect hippocampal neurogenesis against GzmB in vivo [130]. Lee et al. further established the neurotoxic role of GzmB in MS in a subsequent study [61]. When cultured neurons were treated with cerebrospinal fluid (CSF) samples from patients with MS, neuronal death was significantly increased [61]. This effect could be blocked by the cell-permeable GzmB inhibitor Z-Ala-Ala-Asp-Chloromethylketone and partially prevented by the interleukin (IL)-1 receptor antagonist AF 12198 [61]. Chronic exposure of cultured neurons to GzmB for seven days revealed that neurotoxicity was first detectable at 24 h and plateaued by Day 4 [61]. Interestingly, this plateau coincided with reduced PAR-1 expression and increased interleukin (IL)-1 receptor expression [61]. Based on this observation, the authors treated cultured neurons with both GzmB and IL-1β, demonstrating that IL-1β upregulates PAR-1 expression and abolishes the plateau of GzmB-mediated neurotoxicity [61]. Collectively, these findings indicate that GzmB-mediated neuronal death is not solely protease-driven but is amplified by pro-inflammatory cytokines, underscoring a synergistic interaction between extracellular GzmB and neuroinflammation. Figure 1 summarizes both intracellular and extracellular GzmB-mediated neuronal death mechanisms.

Fig. 1.

Granzymes contribute to neurodegeneration and neuroinflammation in the central nervous system (CNS). Peripheral lymphocytes, such as cytotoxic T lymphocytes (CTLs), natural killer (NK) cells or NK-like T (NKLT) cells, can cross the inflamed or non-inflamed blood–brain or blood-spinal cord barrier. Once within the CNS, these immune cells release distinct granzymes that drive neurodegenerative or neuroinflammatory processes. Granzyme B (GzmB) induces neuronal death through intracellular entry via perforin or mannose-6-phosphate receptors (M6PRs) or extracellularly activates protease-activated receptor-1 (PAR-1) to trigger neuronal death. Extracellular GzmB may also cleave extracellular matrix (ECM) proteins, such as thrombospondin-1 (TSP-1) and decorin, generating fragments that can activate infiltrating macrophages and T cells. Granzyme A (GzmA) cleaves tau to generate aggregation-prone fragments and promotes retraction of neuronal and astrocytic processes. Granzyme K (GzmK) stimulates neurons to produce hyperphosphorylated tau via PAR-1 signaling, activates resident microglia and infiltrating macrophages, and may initiate the complement cascade by directly cleaving C2 and C4. Granzyme H (GzmH) promotes neurite retraction through its proteolytic activity. Together, these pathways illustrate how diverse granzyme activities converge to promote neurodegeneration and neuroinflammation in the CNS (created with Biorender.com)

Granzyme B-mediated ECM cleavage as a potential driver of neuroinflammation

Although direct evidence that GzmB promotes inflammation in the CNS is lacking, GzmB is well known to generate a pro-inflammatory milieu by ECM cleavage [4, 48]. Matricellular proteins expressed in the CNS, particularly thrombospondin-1 (TSP-1) and decorin, are established substrates of GzmB [9, 75, 76]. Proteolytic cleavage of these ECM proteins can release bioactive fragments that directly activate immune cells. For instance, the type 1 repeat domain of TSP-1 interacts with CD36 [113] on macrophages [79], while its carboxyl terminal binds to CD47 [31] on T cells [123]. Likewise, decorin fragments can bind to toll-like receptor (TLR) 2 and 4 on macrophages and induce production of pro-inflammatory molecules, such as IL-12 and tumor necrosis factor-α (TNF-α) [71]. In the CNS, both TSP-1 and decorin are expressed by neurons [13, 105] and astrocytes [19, 99], making them reasonable candidate substrates for GzmB released by infiltrating lymphocytes. Once fragmented, these ECM components may amplify neuroinflammation by activating resident microglia or infiltrating immune cells, as illustrated in Fig. 1. GzmB can also release growth factors that are sequestered to the ECM, such as transforming growth factor β1 (TGF-β1) [9] and vascular endothelial growth factor (VEGF) [42]. The release of ECM-sequestered TGF-β1 is particularly relevant in the CNS, as TGF-β signaling exerts context-dependent effects: it promotes neuronal survival and dampens inflammation [85], but also contributes to astrogliosis [98] and fibrosis [80] in the CNS. In contrast, VEGF release may exacerbate neurovascular dysfunction as VEGF increases vascular permeability [42] and can promote blood–brain barrier (BBB) permeability [112].

Granzyme B in multiple sclerosis: pathogenic and regulatory T cell programs

GzmB has been implicated across several neurodegenerative diseases, with elevated levels reported in the serum or CSF in patients with MS [61], amyotrophic lateral sclerosis (ALS) [51] and AD [78]. Lymphocytes from individuals with ALS [56, 57] and AD [56] also exhibit increased GzmB expression and release, suggesting heightened cytotoxic potential in these conditions. Despite these associations, mechanistic evidence linking extracellular GzmB to neuronal death in chronic neurodegenerative diseases remains limited. Among CNS disorders, MS provides the clearest insight, with past work demonstrating that GzmB released by infiltrating cytotoxic T cells can induce neurodegeneration and exacerbate disease severity in experimental autoimmune encephalomyelitis (EAE).

MS is a chronic autoimmune disease characterized by infiltration of autoreactive lymphocytes into the CNS, where they contribute to axonal damage and neurodegeneration [93]. EAE, the most widely used model of MS, recapitulates T cell-driven neuroinflammation and provides a tractable system for testing mechanisms of neurodegeneration [23]. Using this model, Raveney et al. established that CD4⁺ T cell-specific nuclear receptor subfamily 4 group A member 2 (NR4A2) conditional knockout (cKO) suppresses acute EAE but still permits the late-onset form of the disease [87], indicating that additional pathogenic mechanisms sustain chronic progression. The authors showed that this late-onset, chronic phase is driven by a CNS-infiltrating cytotoxic CD4⁺ T cell population expressing the T-box transcription factor Eomesodermin (Eomes) and GzmB [87]. Systemic siRNA knockdown of Eomes markedly ameliorated late-onset EAE in NR4A2 cKO mice, reducing clinical scores and cumulative disease burden [87]. Building on prior evidence that GzmB can induce neuronal death via PAR-1 signaling [129], the authors directly tested this pathway in NR4A2 cKO mice using two complementary interventions: systemic siRNA knockdown of GzmB, and systemic pharmacological inhibition of PAR-1 using the antagonist FR171113. Both treatments suppressed the late stage of EAE in NR4A2 cKO mice, supporting a model in which Eomes⁺ CD4⁺ T cells promote neurodegeneration in MS through an extracellular GzmB–PAR-1 signaling axis.

More recently, Shi et al. extended the link between GzmB⁺ CD8⁺ T cells and MS progression to human peripheral immune signatures [106]. Using single-cell RNA sequencing, flow cytometry, and T cell receptor sequencing, they found that patients with secondary progressive MS (SPMS) exhibited expansion of terminally differentiated effector memory CD8⁺ T cells re-expressing CD45RA (CD8⁺ TEMRA cells) that expressed high levels of GzmB [106]. This GzmB⁺ CD8⁺ TEMRA population showed evidence of clonal expansion and was regulated in part by the transcription factor T-bet [106]. Importantly, the proportion of GzmB⁺ CD8⁺ T cells, particularly GzmB⁺ CD8⁺ TEMRA cells, positively correlated with disability measures including Expanded Disability Status Scale and showed strong diagnostic performance in distinguishing SPMS from relapsing-remitting MS [106]. These findings support the idea that GzmB⁺ CD8⁺ T cell programs are associated with progressive MS and may provide peripheral immune signatures of disease progression.

However, GzmB expression in T cells should not be interpreted as uniformly pathogenic. Human regulatory T cells (Tregs) can express GzmB and use perforin/GzmB-dependent mechanisms to suppress immune responses, including elimination of activated effector T cells or antigen-presenting cells [36]. In autoimmune diseases such as MS, such suppressive activity may counterbalance pathogenic cytotoxic T cell responses. However, GzmB-mediated immune regulation is context-dependent as extracellular GzmB has also been reported to promote resistance to Treg-mediated suppression in CD4⁺ T cells from patients with MS [7]. Thus, while GzmB⁺ cytotoxic T cell programs are associated with MS progression, the net effect of GzmB in autoimmune neuroinflammation likely depends on the producing T cell subset, target cell, and extracellular versus intracellular mode of action.

Extracellular granzyme B in neovascular age-related macular degeneration: vascular pathology, fibrosis, and retinal neurodegeneration

Extracellular GzmB-driven mechanisms have also been implicated in nAMD, a retinal neurodegenerative disease characterized by progressive loss of photoreceptors. Although the retina and brain differ in anatomical organization, cellular composition, immune environment, and disease-specific mechanisms, the retina, including the retinal pigment epithelium (RPE), is developmentally and anatomically an extension of the CNS [91, 137]; therefore, while nAMD does not directly model brain disorders such as AD or MS, it provides a tractable CNS-derived tissue context for studying extracellular granzyme activity in blood–CNS barrier disruption, vascular pathology, fibrotic remodeling, neuronal death, and glial responses. Over the years, our group has demonstrated that extracellular GzmB contributes to two major pathological processes of nAMD: choroidal neovascularization (CNV) and subretinal fibrosis (Fig. 2). Our earlier work showed that extracellular GzmB accumulates in the outer retina with age and is significantly elevated in human AMD eyes, with retinal pigment epithelium (RPE) and choroidal mast cells serving as the major cellular sources [70]. More recently, we found that mast cells accumulate in the aged human and mouse choroid, further supporting their role as a key contributor to age-dependent increases in extracellular GzmB in the outer retina [122]. Based on the hypothesis that this age-dependent accumulation of extracellular GzmB may disrupt the outer blood–retina barrier and initiate nAMD pathology, we applied exogenous GzmB to adult retinal pigment epithelial cell line-19 (ARPE-19) and demonstrated cleavage of RPE ECM proteins (e.g. fibronectin, collagen IV and laminin) as well as tight junction components (e.g. zona occludens-1, occludin and junctional adhesion molecule A) [70]. In a follow-up study, ex vivo choroidal sprouting assays showed that exogenous GzmB cleaves the anti-angiogenic ECM proteins TSP-1 and decorin, resulting in robust vascular sprouting [75, 76, 122]. Consistent with these findings, GzmB deficient mice exhibit significantly smaller CNV lesions in the laser-induced mouse model of CNV compared to age-matched wild-type controls [75].

Fig. 2.

Granzyme B (GzmB) contributes to the pathogenesis of neovascular age-related macular degeneration (nAMD). Choroidal mast cells accumulate with age and serve as a major source of extracellular GzmB in the outer retina. Age-dependent increases in extracellular GzmB lead to cleavage of extracellular matrix (ECM) proteins and tight junction components that maintain outer retinal integrity, thereby initiating early pathogenic events in nAMD. GzmB promotes both choroidal neovascularization (CNV) and subretinal fibrosis by degrading anti-angiogenic and anti-fibrotic ECM proteins (e.g. thrombospondin-1 and decorin), contributing to the angiofibrotic switch. These processes lead to photoreceptor degeneration and irreversible vision loss, and GzmB may further exacerbate photoreceptor vulnerability through direct intracellular or extracellular neurotoxic mechanisms. Retinal macroglia (astrocytes and Müller cells) in the retina undergo reactive gliosis in response to CNV and subretinal fibrosis (created with Biorender.com)

As nAMD progresses, CNV undergoes an angiofibrotic switch that culminates in the development of subretinal fibrosis, a late-stage pathological phenomenon characterized by excessive ECM deposition and fibrotic remodeling [94, 114]. Because GzmB drives both angiogenesis and fibrosis in peripheral, non-nervous tissues [6, 102, 103], subsequent work investigated whether extracellular GzmB also contributes to subretinal fibrosis. In a two-stage laser-induced mouse model of subretinal fibrosis, GzmB deficiency significantly reduces fibrotic lesion size in aged mice, suggesting that age-dependent accumulation of extracellular GzmB in the outer retina may contribute to fibrotic remodeling in nAMD [136]. GzmB deficient mice also exhibited increased levels of the anti-fibrotic proteoglycan decorin within fibrotic lesions, consistent with a model in which GzmB-mediated cleavage of decorin contributes to fibrosis [136]. In addition, GzmB deficiency resulted in increased macrophage-to-myofibroblast transition (MMT) and reduced mast cell accumulation within fibrotic lesions, indicating that GzmB may disrupt normal wound healing processes and sustain chronic inflammation [136]. Notably, GzmB deficiency also improved photoreceptor cell survival above subretinal fibrotic lesions without significantly altering activation of retinal macroglia (astrocytes and Müller cells) [136]. Although enhanced photoreceptor cell survival may be due to the reduced size of fibrotic lesions, it is also possible that loss of GzmB eliminates GzmB-mediated neurotoxicity, given its ability to induce neuronal death through both intracellular and extracellular mechanisms [136]. The observation that activation of retinal macroglia remains unchanged despite improved photoreceptor cell survival raises the possibility that extracellular GzmB may influence glial phenotypes, potentially shifting them towards neurotoxic A1-like states, through direct or indirect mechanisms [136]. Future studies are required to determine how extracellular GzmB influences glial responses in the CNS. Overall, findings in the retina demonstrate that extracellular GzmB can contribute to neuroinflammation and neurodegeneration through ECM cleavage. Figure 2 summarizes these findings, highlighting how extracellular GzmB drives age-dependent outer retina barrier disruption, CNV and subretinal fibrosis in nAMD.

GzmB in acute CNS injury: intracellular cytotoxic mechanisms

In contrast to the progressive, extracellular GzmB-mediated mechanisms observed in nAMD, acute CNS injury models have revealed a distinct role for intracellular GzmB in neuronal death. The Babu group used both spinal cord injury (SCI) [14] and transient middle cerebral artery occlusion (tMCAO) [15] models to investigate how infiltrating CD8⁺ CTLs contribute to CNS neurodegeneration. In rat SCI, thoracic dorsal hemisection induced upregulation of the CTL-attracting chemokine C–X–C motif chemokine ligand (CXCL) 10, increased GzmB levels, and cleavage of poly-ADP ribose polymerase (PARP), including a 64 kDa fragment consistent with GzmB-mediated cleavage [14]. Immunohistochemistry further showed co-localization of GzmB with dying neurons, supporting a model in which infiltrating CTLs release GzmB that enters neurons and contributes to cell death [14]. Similar findings were reported in rat tMCAO, where ischemic injury induced CD3⁺ and CD8⁺ CTL infiltration, increased GzmB levels, appearance of the 64 kDa PARP fragment, and intracellular GzmB localization within neurons in the ischemic penumbra [15]. Together, these studies suggest that GzmB contributes to early neurodegeneration following acute CNS injury by acting intracellularly within neurons.

Granzyme A in neurodegeneration: extracellular signaling and tau pathology

Non-apoptotic mechanisms of GzmA

GzmA is predominantly produced by cytotoxic lymphocytes, including CTLs, NK cells, Natural Killer-like T (NKLT) cells, and γδ T cells [107]. In contrast to GzmB, GzmA cleaves after basic amino acid residues (arginine or lysine), exhibiting trypsin-like activity [145]. Although GzmA was associated with caspase-independent cell death, it has been demonstrated that native human and mouse GzmA lack cytotoxic activity even when delivered intracellularly with perforin, failing to induce mitochondrial depolarization, membrane permeabilization, or apoptosis under conditions that readily support GzmB-mediated cell death [72, 81]. Instead, GzmA functions primarily as a pro-inflammatory protease, including caspase 1-dependent secretion of IL-1β, IL-6, and tumor necrosis factor (TNF)-α from monocytes and macrophages, and similar cytokine responses are triggered by GzmA-secreting NK cells and CTLs [72].

Despite this predominant inflammatory role, several intracellular pathways have been proposed to explain how GzmA might contribute to cell stress or injury under specific conditions. After perforin-mediated entry, GzmA can localize to mitochondria and disrupt components of the electron transport chain, leading to impaired respiration and increased production of ROS [64, 68]. These oxidative signals have been linked to activation of the SET complex and downstream DNA damage pathways [64] although the physiological relevance of this cascade remains unclear given the lack of cytotoxicity observed with native GzmA [72, 81].

GzmA has also been reported to induce a slower, nonapoptotic form of cell death with distinctive morphological changes (“athetosis”), which becomes prominent in the absence of GzmB [111]. This phenomenon depends on an intact actin cytoskeleton and may reflect a context-restricted stress response rather than a primary cytotoxic mechanism [111]. While this process has not been demonstrated in neurons, its reliance on cytoskeletal integrity raises the possibility that GzmA could influence neuronal structure in inflammatory environments.

Finally, GzmA can cleave Gasdermin B (GSDMB) to trigger pyroptosis in cells that express this protein [147]. However, because neurons and glia lack GSDMB expression [18, 52], this pathway is unlikely to contribute to neurodegeneration and neuroinflammation in the CNS.

GzmA-mediated retraction of neuronal and astrocytic processes

Compared to GzmB, the evidence for GzmA in CNS pathology is more limited and is derived almost entirely from in vitro studies. To date, no in vivo studies have demonstrated a causal role for GzmA in neurodegeneration, and the mechanisms described below should therefore be interpreted as emerging rather than established. In 1994, Suidan et al. demonstrated that granule extracts from activated CTLs trigger immediate and complete neurite retraction in NB2a neuroblastoma cells, and purification experiments identified GzmA as the active component [110]. Purified GzmA induced rapid neurite swelling followed by full retraction within minutes, accompanied by vigorous cell body movement and membrane blebbing [110]. GzmA also reversed the stellate morphology of primary rat astrocytes, collapsing their processes into an epithelial-like shape [110]. These effects were dependent on GzmA’s proteolytic activity, as GzmA-mediated neurite retraction was abolished by a trypsin-like protease inhibitor Ph-HNCONHCiTEtOIC (also known as 7-(phenylureido)-4-chloro-3-(2-isothioureidoethoxy)-isocoumarin or IGA) but not by a potent thrombin inhibitor protease nexin-1 [110].

Mechanistically, GzmA acts as a thrombin-like protease [110]. Suidan et al. showed that GzmA cleaves a synthetic peptide corresponding to the N-terminal extracellular domain of thrombin receptor (or PAR-1) at the canonical Arg41-Ser42 site, generating the same tethered-ligand sequence produced by thrombin [110]. Function-blocking antibodies against PAR-1 completely prevented GzmA-mediated neurite retraction, confirming that PAR-1 activation is required [110].

Together, these findings position CTL-derived GzmA as an extracellular modulator of neuronal and astrocytic architecture rather than a classical cytotoxic effector. Given the close structural and functional coupling between neurons and astrocytes in maintaining synaptic transmission and plasticity [138], GzmA-mediated sublethal retraction of neuronal and astrocytic processes could destabilize local neuronal circuits and increase their susceptibility to subsequent degenerative cascades. Importantly, CD4⁺ and CD8⁺ T cells can infiltrate the CNS early in neuroinflammatory conditions, including MS [25, 125] and AD [77, 92], raising the possibility that GzmA released during the early T cell infiltration may prime neurons for later degeneration. Although speculative, GzmA-induced collapse of astrocytic processes could impair astrocytic functions or bias astrocytes toward pro-inflammatory A1-like states; however, further study is required to determine whether such pathological phenomena occur in vitro and in vivo. Figure 1 highlights these GzmA‑mediated effects on neuronal and astrocytic processes within the broader context of granzyme‑driven CNS pathology.

GzmA-mediated tau pathology

In addition to its ability to influence neuronal and astrocytic processes, GzmA also directly targets cytoskeletal proteins. This function is particularly relevant to AD, which is characterized by the accumulation of extracellular amyloid β plaques and intracellular neurofibrillary tangles composed of aggregated tau [142]. Although amyloid pathology is thought to initiate AD, tau pathology correlates more closely with synaptic dysfunction, neuronal loss, and clinical severity [47]. Under physiological conditions, tau stabilizes microtubules; however, in AD, post-translational modifications and proteolytic processing convert tau into aggregation-prone species that drive neurodegeneration and propagate pathology across brain regions [134].

Recently, Quinn et al. identified GzmA as a tau-cleaving protease with direct implications for tau pathogenesis in AD [86]. The authors identified three GzmA cleavage sites on full-length tau: R194–S195, R209–S210, and K240–S241 [86]. They also demonstrated that both pharmacological inhibition (FUT-175) and mutation of the critical amino acid residues at the predicted GzmA cleavage sites in tau abolish GzmA-mediated tau proteolysis [86]. Notably, several of these cleavage sites correspond to tau fragments detected in human tauopathy brain tissue, supporting in vivo relevance [86]. Functionally, GzmA-generated C-terminal tau fragments exhibited enhanced phosphorylation and a markedly increased propensity to aggregate compared with full-length tau [86]. Generation of these fragments promoted the accumulation of sarkosyl-insoluble tau species [86], a defining feature of pathological tau [88]. The most aggregation-prone GzmA-generated fragment, tau195–441, demonstrated efficient intercellular transfer and seeded aggregation in recipient cells, consistent with prion-like mechanisms thought to underlie the spread of tau pathology in AD [86].

These findings establish extracellular GzmA activity as a mechanistic link between neuroinflammation and tauopathy. By directly cleaving tau into aggregation-competent, propagation-prone fragments, GzmA, as illustrated in Fig. 1, provides a pathway through which CTL activity could accelerate tau pathology without overt cytotoxic effects. Notably, GzmA⁺ CD8⁺ T cells are present in the hippocampus of individuals with Alzheimer’s disease, which suggests that GzmA⁺ CTLs may contribute to pronounced tau pathology and neuronal vulnerability within this region [34]. Whether tau cleavage occurs intracellularly following perforin-mediated GzmA delivery or extracellularly remains unresolved. Nevertheless, GzmA-mediated tau proteolysis represents an emerging mechanism by which immune effector proteases may drive neurodegeneration, highlighting extracellular GzmA and its downstream tau fragments as potential therapeutic targets in AD and related tauopathies.

Granzyme K in neurodegeneration: context-dependent signaling and complement-driven inflammation

Non-cytotoxic, pro-inflammatory mechanisms of GzmK

GzmK is expressed by cytotoxic lymphocytes that typically express GzmB and GzmA and non-cytotoxic populations, including macrophages and mast cells [11, 89, 121]. GzmK is closely linked to GzmA on chromosome 5, and these two tryptase-like proteases share overlapping substrate preferences, including cleavage of SET complex components and Ape1, reflecting their common preference for cleaving after basic residues (lysine or arginine) [11, 37, 118]. Although historically viewed as functionally redundant, emerging evidence indicates that GzmK possesses distinct biochemical activities and unique pathophysiological roles in multiple diseases [11, 118]. Studies using human and rat GzmK demonstrate that GzmK can exert measurable cytotoxicity and cleave several GzmA-associated substrates, leading to non-apoptotic forms of cell stress and DNA damage [11]. However, these cytotoxic effects are substantially weaker in mouse GzmK, highlighting unresolved species differences and ongoing debate regarding the true cytotoxic potential of GzmK [11, 53]. Mouse GzmK studies have shown non-cytotoxic roles of GzmK targeting intracellular and extracellular substrates [11]. Non-cytotoxic intracellular substrates include importin α1 or β [146], β-tubulin [12] and heterogeneous nuclear ribonucleoprotein K (hnRNP K) [12]. Although the functional consequences of GzmK-mediated cleavage of these proteins in neurons remain unclear, these findings raise the broader possibility that GzmK may influence cellular processes relevant to neuronal resilience under inflammatory conditions.

In addition to these non-cytotoxic intracellular effects, GzmK also exerts several well-characterized extracellular activities that are increasingly recognized as its dominant mode of action. GzmK was first shown to cleave and activate PAR-1 on human lung fibroblasts, leading to robust secretion of pro-inflammatory cytokines, including IL-6, IL-8 and monocyte chemoattractant protein-1 (MCP-1), and promoting fibroblast proliferation through activation of mitogen-activated protein kinase (MAPK) pathway [24]. Through the same PAR-1-dependent mechanism, GzmK can also elicit pro-inflammatory signaling in endothelial cells without inducing proliferation or cell death, and can also drive keratinocyte proliferation [89, 101]. GzmK also directly stimulates macrophages to release IL-23 in vitro, indicating that it amplifies inflammation by acting on immune cells as well [89]. The pro-inflammatory role of extracellular GzmK has been further validated across multiple inflammatory skin conditions [89, 119, 121]. GzmK was elevated in human burn tissue and secreted by classically activated macrophages, and GzmK deficient mice exhibited improved wound closure, matrix organization, and tensile strength, accompanied by reduced early expression of IL-6, intercellular adhesion molecule 1 (ICAM-1), vascular cell adhesion molecule 1 (VCAM-1), and MCP-1 [121]. GzmK⁺ cells, including dermal mast cells, were enriched in human atopic dermatitis lesions, and GzmK deficient mice displayed reduced scaling, erosions, erythema, microvascular injury, and fibrosis [119]. GzmK was also elevated in human psoriasis lesions, and GzmK deficiency reduced disease severity in a mouse model of psoriasis, with diminished skin plaque formation, epidermal thickening, and inflammatory infiltrate [89]. Together, these studies demonstrate that extracellular GzmK amplifies inflammation and promotes pathological tissue remodeling, raising the possibility that infiltrating GzmK⁺ lymphocytes in the CNS may similarly contribute to neuroinflammation and neurodegeneration through PAR-1-dependent and pro-inflammatory mechanisms.

A context-dependent role for GzmK in the CNS

Compared to GzmB, the evidence for GzmK in CNS pathology is still emerging and varies substantially across disease models. Current findings derive from a mixture of human tissue studies and mechanistic work in MS, SCI, and AD models, and should therefore be interpreted as context-dependent rather than uniformly pathogenic. In particular, the net effect of GzmK appears to depend on disease stage, pathological context, cellular target, and experimental readout, including whether the model primarily reflects amyloid-associated pathology, tauopathy, autoimmune neuroinflammation, or traumatic CNS injury.

Emerging evidence links GzmK⁺ lymphocytes to neuroinflammatory and neurodegenerative changes in the CNS. In MS, Hoeks et al. identified a transcriptionally distinct population of GzmK⁺ CD4⁺ T cells, primarily Runx3⁺ Eomes⁺ Th17.1 cells, that displayed high C–C motif chemokine receptor (CCR) 5 expression and were uniquely capable of crossing a non-inflamed BBB in vitro [46]. These cells were enriched in the CSF of treatment-naïve MS patients and were clearly separable from the GzmB⁺ perforin⁺ CD4⁺ CD28⁻ cytotoxic subset, which was not preferentially enriched among cells migrating across the non-inflamed BBB in vitro [46]. This distinction underscores that GzmK marks a non-cytotoxic, CNS-infiltrating T cell subset that is functionally separate from the GzmB⁺ perforin⁺ cytotoxic population and may participate in early pathological events in MS.

In the context of CNS injury, Kong et al. demonstrated that aging is accompanied by the emergence of CD8⁺ NKLT cells that express high levels of natural killer cell granule protein 7 (NKG7) and GzmK and accumulate in the spinal cord even before injury [60]. These cells are recruited through a C-X-C motif chemokine receptor 6 (CXCR6)–CXCL16 axis and position themselves near MHC I-presenting myeloid cells, such as macrophages and microglia, at the lesion site [60]. Within this injury environment, NKG7⁺ CD8⁺ NKLT cells exacerbate tissue injury by degranulating GzmK in proximity to myeloid cells and disrupt their wound healing functions, contributing to impaired repair in the aged spinal cord [60]. Genetic deletion of NKG7 or antibody-mediated depletion of CD8⁺ T cells restores myeloid repair capacity and improves axonal integrity, indicating that these GzmK⁺ NKLT cells play an active pathogenic role in SCI [60]. Together, these findings position GzmK⁺ T cells as recurring modulators of CNS vulnerability, as illustrated in Fig. 1, across both autoimmune and injury-driven contexts.

In AD-related models, however, GzmK⁺ CD8⁺ T cells have been linked to both pathogenic and protective effects depending on the model system, pathological substrate, and cellular target. Terrabuio et al. recently identified GzmK as a direct immune mediator of neurodegeneration, using 3xTg-AD mice and human AD brain tissue [115]. In both 3xTg-AD mouse and human AD brain tissues, activated CD103⁻ CD8⁺ tissue-resident T cells accumulated in the parenchyma, selectively expressed high levels of GzmK but not GzmB or perforin, and localized in close proximity to hippocampal neurons [115]. GzmK-mediated cleavage of PAR-1 on cultured neurons triggered intracellular calcium dysregulation and activated MAPK and nuclear factor kappa-light-chain-enhancer of activated B cells (NF-κB) signaling, ultimately driving tau hyperphosphorylation at pS199 and pS396 without altering total tau levels [115]. Pharmacological inhibition of PAR-1 prevented GzmK-induced intracellular calcium release and tau phosphorylation in vitro, establishing a mechanistic link between GzmK signaling and tau pathology [115]. Depletion of circulating CD8⁺ T cells reduced brain infiltration of CD103⁻ GzmK⁺ cells and improved long-term hippocampus-dependent spatial learning and memory performance in the Morris water maze, while also lowering amyloid β and tau phosphorylation in 3xTg-AD mice. These findings together support a pathogenic role for GzmK⁺ CD8⁺ T cells in AD. As depicted in Fig. 1, GzmA and GzmK may contribute to tauopathy in AD through distinct yet complementary mechanisms.

In contrast, Mason et al. reported a protective function for GzmK⁺ CD8⁺ T cells in the P301S tauopathy model [69], which lacks amyloid β pathology [1]. Unlike the 3xTg-AD study, where GzmK⁺ CD8⁺ T cells were positioned near neurons and implicated in neuronal PAR-1 signaling, GzmK⁺ CD8⁺ T cells in the P301S model deposited GzmK onto microglia, as shown in Fig. 1, and were regulated by immune checkpoint receptors, such as T cell immunoreceptor with Ig and ITIM domains (TIGIT) and programmed cell death protein 1 (PD-1) [69]. Notably, they lacked classical effector molecules including interferon-gamma (IFNγ), TNF, GzmA, and GzmB [69]. Genetic deletion of CD8⁺ T cells in the tauopathy model accelerated phosphorylated tau spread and exacerbated short-term hippocampus-dependent working memory performance in the Y-maze, suggesting that GzmK⁺ CD8⁺ T cells may restrain disease progression by modulating microglial responses [69]. GzmK⁺ CD8⁺ T cells were also observed in human AD and chronic traumatic encephalopathy (CTE) lesions, indicating relevance across neurodegenerative conditions [69].

Together, these divergent findings indicate that GzmK is not uniformly pathogenic in the CNS. Instead, its effects appear to be shaped by disease stage, amyloid versus tau pathology, CD8⁺ T cell phenotype, cellular target, and the readout used to define disease modification. In 3xTg-AD mice, GzmK⁺ CD8⁺ T cells are associated with neuronal PAR-1 signaling, tau phosphorylation, amyloid burden, and long-term cognitive impairment, whereas in P301S tauopathy, GzmK⁺ CD8⁺ T cells appear to modulate microglia and restrain tau spread and short-term cognitive decline. Importantly, neither study provides evidence that GzmK directly induces neuronal death in the manner of GzmB. Rather, GzmK appears to operate through non-cytotoxic pathways, including neuronal PAR-1 mediated signaling and microglial modulation. GzmK may also engage PAR-2 in the CNS, as it has been shown to induce activation of dorsal root ganglion neurons in a PAR-2-dependent manner and pruritus [44]. Clarifying how GzmK⁺ lymphocytes influence neurons and glia across disease context will be essential for defining their net contribution to neurodegeneration.

GzmK as a potential amplifier of neuroinflammation via complement activation

Beyond its effects on neurons and microglia, GzmK has also been proposed as an activator of the entire complement cascade, providing an additional mechanism through which infiltrating T cells may amplify neuroinflammation in the CNS. Donado et al. demonstrated that GzmK released from tissue CD8⁺ T cells binds heparan sulfate on cell surfaces and directly cleaves complement components C2 and C4, enabling assembly of membrane-bound C3/C5 convertases [28]. This initiates the full complement cascade, generating C3a and C5a, promoting C3b opsonization, and facilitating membrane attack complex formation [28]. In rheumatoid arthritis synovium and mouse models of arthritis and psoriasiform dermatitis, GzmK co-localized with complement activation markers, and GzmK-deficient mice exhibited reduced inflammation and diminished C3d and C4d deposition [28]. Given that complement dysregulation is increasingly implicated in AD, MS, and other neurodegenerative conditions [73], GzmK-mediated complement activation represents a plausible pathway through which T cell infiltration could further escalate neuroinflammation within the CNS, as illustrated in Fig. 1.

Granzyme H in neurodegeneration: primate-specific mechanisms and nerve injury

GzmH: a mechanistically divergent member of the granzyme family

GzmH is a primate-specific granzyme without a direct ortholog in rodents [82, 84]. This has limited in vivo mechanistic studies and contributed to the relatively small literature base. GzmH is located adjacent to GzmB on chromosome 14, and the two genes are thought to share regulatory elements [100]. However, unlike GzmB, GzmH cleaves after aromatic hydrophobic residues (mainly phenylalanine and tyrosine), reflecting its chymotrypsin-like specificity [141]. Despite this genomic proximity, their expression patterns diverge across cytotoxic lymphocyte subsets. Human NK cells and cytotoxic CD4⁺ and CD8⁺ T cells can co-express GzmB and GzmH [100, 149]. In contrast, NKLT cells do not co-express GzmB and GzmH [100], underscoring that GzmH is not part of the canonical cytotoxic program in all T cell lineages.

Sharing 71% homology with GzmB [30], GzmH is considered an alternative cytotoxic effector protease. However, its apoptotic mechanisms appear more heterogeneous across studies. One study found that GzmH-mediated apoptosis occurred without caspase activation, cytochrome c release, or cleavage of Bid, suggesting a noncanonical death pathway [30]. In contrast, another study demonstrated that GzmH can trigger these same apoptotic events in a manner similar to GzmB [49]. A third study showed that GzmH induces apoptosis independently of Bid processing, instead cleaving DNA fragmentation factor 45 (DFF45) directly and promoting DNA fragmentation [29]. These discrepancies likely reflect differences in recombinant GzmH sources, variations in GzmH concentrations, and the use of distinct target cell lines. Despite these inconsistencies, the collective evidence indicates that GzmH can initiate apoptosis through mechanisms that are partially overlapping with, yet distinct from, those of GzmB.

Interestingly, mast cells were identified as a novel immune source of GzmH in humans. Rönnberg et al. demonstrated that mast cells isolated from the bone marrow of a patient with mast cell leukemia and cord blood-derived mast cells express both GzmB and GzmH but not other granzymes [95]. Notably, unstimulated mast cells express high levels of GzmH mRNA, whereas activation leads to downregulation of GzmH and upregulation of GzmB. This suggests that these two proteases are reciprocally regulated in mast cells [95]. Because mast cells typically do not express perforin, mast cell-derived granzymes do not induce perforin-dependent apoptosis [95]. Instead, mast cells may trigger apoptosis through perforin-independent mechanisms involving either M6PR or PAR-1. Alternatively, given that GzmH is highly expressed by unstimulated mast cells, it is possible that GzmH may function as an extracellular effector protease involved in ECM remodeling under resting or homeostatic conditions. Further study is required to clarify the function of mast cell-derived GzmH, particularly in the context of neurodegenerative diseases as mast cells are increasingly recognized as CNS-border-associated immune cells that could contribute to neuroinflammation and neurodegeneration [54]. In support of this idea, choroidal mast cells have been shown to contribute to CNV and subretinal fibrosis in nAMD [26, 75, 122, 136].

GzmH-mediated proteolytic nerve injury

Compared to other granzymes, GzmH remains the least characterized in the CNS. To date, its involvement in neurodegeneration is supported by a single study by Yu et al. in aging-related atherosclerotic cerebral small vessel disease (aCSVD), an uncommon cerebrovascular condition [139]. Noting an increased proportion of CD56^DIM CD16^BRIGHT NK cells in patients with aCSVD, the authors performed proteomic analysis on this particular subset of NK cells and found that GzmH was markedly elevated in patient-derived cells compared with controls [139]. They did not report differential expression of GzmB [139], consistent with the fact that CD56^DIM CD16^BRIGHT NK cells constitutively express high levels of GzmB [27]. CSF proteomics further revealed significant increases in neuronal pentraxin-2 and neurofilament medium polypeptide, both indicative of axonal and synaptic injury [139]. Considering the correlation between increased NK cells and nerve damage in patients aCSVD, the authors co-cultured primary human neurons with CD56^DIM CD16^BRIGHT NK cells derived from individuals with severe aCSVD to explore the role of NK cells in nerve damage [139]. In this culture system, CD56^DIM CD16^BRIGHT NK cells localized to axon hillocks and reduced the number and length of axons, dendrites, and spines [139]. Phosphorylation of death receptor was not significantly changed in the co-culture system, indicating that the observed injury was not driven by classical apoptosis mechanisms and may instead reflect extracellular, perforin-independent GzmH activity [139]. Treatment with the serine protease inhibitor 3,4-dichloroisocoumarin (3,4-DCI) attenuated these effects, leading the authors to propose that GzmH released from CD56^DIM CD16^BRIGHT NK cells mediates this form of nerve damage [139]. However, because 3,4-DCI is a general serine protease inhibitor [41, 62, 132] and CD56^DIM CD16^BRIGHT NK cells do express GzmB [27], this interpretation should be considered with caution. Moreover, the mechanism by which GzmH induces nerve injury remains unresolved. The authors did not identify neuronal substrates of GzmH, and it is unknown whether GzmH can directly cleave axonal, dendritic, or synaptic proteins or instead acts through indirect pathways. The primate-restricted nature of GzmH further complicates mechanistic analysis, underscoring the need for studies in human tissues and non-human primate models to identify its physiological substrates and determine how GzmH-mediated proteolysis contributes to neuroinflammation and neurodegeneration within the CNS.

Endogenous and pharmacological inhibition of granzymes

Endogenous GzmB inhibitor serpinA3N: a double-edged sword

In mice, serpinA3N is a secreted serpin that forms an irreversible complex with GzmB, cathepsin G, and leukocyte elastase, thereby neutralizing their extracellular activities [148]. In contrast, the human ortholog SERPINA3 (α1-antichymotrypsin) does not inhibit GzmB and instead targets a narrower set of chymotrypsin-like serine proteases, including cathepsin G, chymotrypsin, and human mast cell chymase [148]. This highlights the absence of an intrinsic mechanism for restraining extracellular GzmB activity in humans. By comparison, SERPINB9 (also known as proteinase inhibitor 9) in humans provides potent intracellular protection against GzmB, but this safeguard does not operate outside the cell [50]. The rodent ortholog serpinB9 (also known as serine protease inhibitor 6) performs the same intracellular protective function in mice and rats [109]. Together, these species differences underscore a therapeutic opportunity to inhibit extracellular GzmB in human disease contexts where its unopposed activity contributes to chronic inflammation, ECM cleavage, and tissue degeneration.

Across multiple mouse models of CNS injury, serpinA3N, which is not present in humans, has been shown to promote neuroprotection and improve functional outcomes. In ischemic stroke, serpinA3N is rapidly induced in both neurons and astrocytes during the acute phase (6 h to 3 days post-stroke), and both overexpression and loss-of-function studies support a protective role in limiting infarct size, motor dysfunction, neuroinflammation, oxidative stress, apoptosis and BBB breakdown [63, 144]. These effects have been linked to several mechanisms, including inhibition of cytotoxic lymphocyte-derived GzmB and neutrophil elastase, clusterin-dependent Akt-mTOR (mechanistic target of rapamycin) pro-survival signaling and suppression of ferroptosis through regulation of mitochondrial complex I activity [63, 66, 144]. Similar protective roles have been reported in hippocampal stab injury and SCI, where serpinA3N induction is associated with reduced neuronal apoptosis, improved functional outcomes and regulation of injury-associated protease activity [17, 131]. In immune-driven CNS pathology, recombinant serpinA3N also attenuates extracellular GzmB-driven fetal neuroinflammatory abnormalities in a maternal immune-activation model [8] and reduces neuroinflammation, axonal injury and clinical severity in EAE [38].

However, emerging evidence indicates that serpinA3N can also promote neuroinflammation and neurodegeneration, particularly when expressed by reactive astrocytes. SerpinA3N was first identified as a pan marker of astrogliosis by Dr. Ben Barres’s group, with robust induction in reactive astrocytes following stroke, lipopolysaccharide-induced neuroinflammation, and aging [21, 140]. Subsequent studies showed that astrocyte-derived serpinA3N is not merely a marker of reactivity but can actively drive pathological signaling. In human induced pluripotent stem cell (iPSC)-derived BBB co-cultures, TNF-induced astrocytic SERPINA3 promoted vascular inflammation and BBB dysfunction [59]. In kainic acid-induced temporal lobe epilepsy, astrocyte-derived serpinA3N amplified hippocampal inflammation, increased astrocyte and microglial reactivity, worsened seizure severity, and activated NF-κB signaling [65]. In aging-related neurodegeneration, serpinA3N was enriched in senescent astrocytes, including in APP/PS1 mice, and contributed to a senescence-associated secretory phenotype that impaired neurite outgrowth and increased tau phosphorylation in neuro-2A cells [40]. These findings indicate that chronic or astrocyte-derived serpinA3N can actively drive neuroinflammatory and neurodegenerative processes.

Although serpinA3N can neutralize GzmB and provide acute neuroprotection in mice, its therapeutic potential is limited by its rodent-restricted nature, likely immunogenicity in humans and pleiotropic, context-dependent functions. Its beneficial effects may be most relevant during acute injury or early inflammatory responses, whereas chronic or excessive serpinA3N expression may become maladaptive. These long-term liabilities underscore the need for more selective, human-compatible GzmB inhibitors capable of achieving extracellular blockade without the species-specific and context-dependent constraints of serpinA3N.

Exogenous GzmB inhibitor VTI-1002: a selective GzmB inhibitor with therapeutic potential

VTI-1002 is a potent and specific small molecule GzmB inhibitor that effectively suppresses the cleavage of GzmB substrates and alleviates GzmB-associated pathology in vitro, ex vivo, and in vivo. Shen et al. first demonstrated its therapeutic potential in a mouse model of dermal scarring, where topical VTI-1002 gel was applied daily for 30 days [103]. VTI-1002 prevented decorin cleavage and reduced scarring, while improving collagen organization and tensile strength [103]. Turner et al. subsequently demonstrated that pre-incubation of GzmB with VTI-1002 effectively prevented cleavage of E-cadherin and filaggrin in vitro [120]. In a mouse model of atopic dermatitis, daily topical VTI-1002 treatment reduced pathological loss of E-cadherin in the epidermis and filaggrin in the stratum corneum, resulting in markedly improved wound severity [120]. Consistent with these findings, Hiroyasu et al. also reported that VTI-1002 prevents GzmB-mediated cleavage of collagen XVII in vitro and significantly ameliorates disease severity in mouse models of epidermal blistering [45].

In line with these skin studies, our lab has also demonstrated the efficacy of VTI-1002 in suppressing pathological processes relevant to nAMD. Using the mast cell activator 48/80 and the mast cell inhibitor ketotifen fumarate, we showed that choroidal mast cells contribute to vascular sprouting in mouse outer retina tissues ex vivo [75]. Based on the observation that GzmB deficiency significantly reduced CNV development in vivo, we hypothesized that extracellular GzmB drives the angiogenic response ex vivo [75]. Consistent with this, ex vivo inhibition of GzmB with VTI-1002 significantly reduced mast cell-mediated vascular sprouting in mouse outer retina tissues [75]. In a separate study, we further demonstrated that VTI-1002 rescues the GzmB-mediated delay in ARPE-19 wound healing in vitro [136]. Together, these findings suggest that VTI-1002 may be an effective strategy for mitigating GzmB-associated pathological processes within the outer retina and may hold therapeutic potential for nAMD and other neurodegenerative diseases involving extracellular GzmB.

Pharmacological inhibition of GzmA, GzmK and GzmH: broad and non-selective

Although there are no selective inhibitors for GzmA, GzmK, or GzmH, several broad-spectrum serine protease inhibitors have been used experimentally to suppress their activity. As previously mentioned, Ph-HNCONHCiTEtOIC and FUT-175 have been applied to inhibit GzmA activity [3, 86, 110], and 3,4-DCI has been used to block GzmH activity [139]. However, these compounds are non-specific serine protease inhibitors [3, 41, 62, 132] and thus cannot be used to attribute biological effects to inhibition of individual granzymes in vivo. Endogenous regulation of these granzymes is also limited and highly species-specific. In humans, extracellular GzmA is inhibited by antithrombin III, an anticoagulant serpin that regulates fibrin formation by inhibiting thrombin and factors IXa, Xa, and XIa [96, 108]. In contrast, rodents lack a comparable extracellular inhibitor for GzmA. In mice, the species-specific intracellular serpin, serpinB6B, protects cytotoxic lymphocytes from intracellular GzmA, but this safeguard does not extend to the extracellular space [55]. Nonetheless, recombinant mouse serpinB6B has been applied experimentally to inhibit extracellular mouse GzmA activities in vitro [97] and in vivo [33], although this does not reflect its physiological role. No endogenous inhibitor for GzmH has been identified in any species. GzmK is uniquely regulated extracellularly in humans by inter-α inhibitor 1 protein (IαIp), a circulating plasma serine protease inhibitor [11]. IαIp has been used experimentally to block GzmK-mediated pro-inflammatory signaling in endothelial cells in vitro [101] and GzmK-mediated pruritus in mice [44]. Taken together, these gaps highlight a significant unmet need: the development of selective, extracellular inhibitors capable of specifically targeting GzmA, GzmK, and GzmH activity in physiological and pathological settings.

Translational challenges for granzyme inhibition in the CNS

Despite the therapeutic promise of extracellular granzyme inhibition, several translational challenges must be addressed before this strategy can be broadly applied to CNS diseases. First, target validation must be established in a disease- and stage-specific manner, as individual granzymes may exert pathogenic, protective, or compensatory functions depending on the cellular source, tissue compartment, and pathological context. Second, effective delivery remains a major barrier, particularly for brain disorders where therapeutic access is limited by the blood–brain barrier [133], whereas local delivery to CNS-derived tissues such as the retina may provide a more tractable route for targeting extracellular granzyme activity. Third, therapeutic timing is likely to be critical as granzyme activity may have distinct consequences during acute injury, chronic inflammation, tissue remodeling, and late-stage neurodegeneration. Finally, inhibition strategies must distinguish pathological extracellular granzyme activity from canonical intracellular granzyme-mediated cytotoxicity as systemic blockade could impair anti-viral and anti-tumor immune defense. Thus, extracellularly restricted or locally delivered granzyme inhibitors may offer the most feasible path forward, particularly for conditions in which extracellular proteolysis, blood–CNS barrier disruption, and tissue remodeling are dominant pathogenic mechanisms.

Future directions

Although substantial progress has been made in understanding how granzymes contribute to neurodegeneration and neuroinflammation, many fundamental questions remain unanswered. A major priority for future research is to define the precise cellular and spatial context in which each granzyme operates within the CNS. While infiltrating CD4⁺ T cells, CD8⁺ T cells, NK cells, and tissue resident mast cells have all been implicated as potential sources, the timing, anatomical localization, and activation states that govern granzyme release remain poorly resolved. Single-cell and spatial transcriptomic approaches, combined with in situ zymography [35, 135] or activity-based probes for mapping granzyme activity, will enable high-resolution mapping of granzyme expression and activity, substrate engagement, and cellular targets, revealing how extracellular and intracellular granzyme activity affects neurons, glial cells, and the CNS ECM across disease stages.

Another critical priority is to dissect the mechanisms of extracellular granzyme signaling. While granzymes can act independently of perforin, the downstream pathways they activate in neurons and glia remain incompletely understood. PAR-1 activation may be a shared mechanism for GzmB and GzmK, but it is unlikely to be the only receptor involved. Identifying additional G-protein coupled receptors, coreceptors, or ECM-derived ligands that mediate granzyme signaling could reveal new routes through which immune activity shapes neuronal vulnerability. Likewise, the full repertoire of extracellular substrates for each granzyme in the CNS has not been systematically characterized. Proteomic substrate-trapping [10] and degradomics will be crucial for uncovering novel pathways linking granzyme activity to synaptic dysfunction, cytoskeletal destabilization, complement activation, and inflammatory amplification.

Species differences represent another major barrier to translation. Rodents possess endogenous inhibitors such as serpinA3N and serpinB6B that have no human functional equivalents, while humans rely on distinct extracellular regulators such as antithrombin III and IαIp. Because these regulatory systems differ fundamentally between species, granzyme activity, substrate engagement, and downstream signaling in mice may not accurately reflect human biology. This underscores the need for human-relevant experimental systems. Future work should incorporate human iPSC-derived neurons, astrocytes, microglia, and organoids, as well as ex vivo human tissues, to validate granzyme-driven mechanisms identified in animal models or immortalized cell lines.

Emerging insights from our recent work on subretinal fibrosis in nAMD [136] highlight the need to understand how granzymes shape glial phenotypes and neuroimmune interactions. Accumulating evidence suggests that extracellular GzmB and GzmK may influence astrocyte and microglia/macrophage activation states [60, 69, 136], potentially biasing them toward neurotoxic phenotypes or altering their output of cytokines and neurotrophic factors. Whether granzymes directly reprogram glia or influence glial responses to injury remains largely unexplored. Given the central role of glial dysfunction in aging and neurodegenerative diseases, elucidating how granzymes released by infiltrating peripheral immune cells act on glia could reveal new mechanisms driving CNS vulnerability and degeneration.

Despite strong associations between granzymes and neurodegenerative diseases, causal evidence in major neurodegenerative diseases remains limited. For GzmB, in vivo studies using GzmB deficient mice and serpinA3N provide evidence that extracellular GzmB activity can contribute to long-term neuroinflammatory and neurodegenerative processes [38, 75, 87, 136], but mechanistic insights for GzmA, GzmK, and GzmH still derive largely from in vitro systems or acute CNS injury models. Determining whether these granzymes act as early drivers, amplifiers, or downstream byproducts of neurodegeneration will require a combination of systemic and cell-specific knockout models, stage-specific granzyme overexpression or suppression, and longitudinal studies that track granzyme activity across disease stages. These approaches will be essential for establishing whether targeting granzymes can meaningfully alter disease trajectories.

Finally, the development of selective extracellular granzyme inhibitors represents a major therapeutic opportunity. While VTI-1002 has demonstrated the feasibility of targeting GzmB, no selective inhibitors exist for GzmA, GzmH, GzmK, or GzmM. The lack of selective extracellular inhibitors for GzmA, GzmK, and GzmH in humans further underscores the need for pharmacological strategies capable of precisely modulating their activity. Structure-guided drug design and improved biochemical characterization of granzyme-substrate interactions will be critical for generating inhibitors with the specificity required for therapeutic use. Given the role of GzmA and GzmK in tau cleavage and PAR-1 activation, and GzmK-dependent complement cascade activation, selective extracellular inhibitors could represent a new class of neuroprotective agents with broad applicability across neurodegenerative diseases.

Together, these directions help define the emerging landscape of granzyme research in the CNS. Integrating mechanistic studies, human-relevant models, and therapeutic development across the brain, spinal cord, and retina will clarify whether granzyme-targeted interventions can alter neurodegenerative disease progression.

Abbreviations

3,4-DCI

3,4-Dichloroisocoumarin

aCSVD

Aging-related atherosclerotic cerebral small vessel disease

AD

Alzheimer’s disease

ALS

Amyotrophic lateral sclerosis

AMD

Age-related macular degeneration

Ape1

Apurinic/apyrimidinic endonuclease 1

ARPE 19

Adult retinal pigment epithelial cell line-19

BBB

Blood–brain barrier

BDNF

Brain-derived neurotrophic factor

CCR5

C–C motif chemokine receptor 5

CD3

Cluster of differentiation 3

CD4

Cluster of differentiation 4

CD8

Cluster of differentiation 8

CD16

Cluster of differentiation 16

CD28

Cluster of differentiation 28

CD49a

Cluster of differentiation 49a

CD56

Cluster of differentiation 56

CD103

Cluster of differentiation 103

CNS

Central nervous system

CNV

Choroidal neovascularization

CSF

Cerebrospinal fluid

CTL

Cytotoxic T lymphocyte

CXCR6

C–X–C motif chemokine receptor 6

CXCL1

C–X–C motif chemokine ligand 1

CXCL 10

C–X–C motif chemokine ligand 10

CXCL-16

C–X–C motif chemokine ligand 16

DFF45

DNA fragmentation factor 45

ECM

Extracellular matrix

Eomes

T-box transcription factor Eomesodermin

GSDMB

Gasdermin B

GzmA

Granzyme A

GzmB

Granzyme B

GzmH

Granzyme H

GzmK

Granzyme K

hnRNP K

Heterogeneous nuclear ribonucleoprotein K

IαIp

Inter-α inhibitor 1 protein

ICAM-1

Intercellular adhesion molecule 1

IFNγ

Interferon-gamma γ

IL-1β

Interleukin 1β

IL-2

Interleukin 2

IL-6

Interleukin 6

IL-8

Interleukin 8

IL-23

Interleukin 23

iPSC

Induced pluripotent stem cell

ISG

Interferon stimulated gene

M6P

Mannose-6-phosphate

M6PR

Mannose-6-phosphate receptor

MCP-1

Monocyte chemoattractant protein-1

MHC I

Major histocompatibility complex I

MMP2

Metalloproteinase-2

MMT

Macrophage-to-myofibroblast transition

MS

Multiple sclerosis

mTOR

Mechanistic target of rapamycin

nAMD

Neovascular age-related macular degeneration

NDUFS3

NADH:ubiquinone oxidoreductase iron-sulfur protein 3

NF-κB

Nuclear factor kappa-light-chain-enhancer of activated B cells

NGF

Nerve growth factor

NK cell

Natural Killer cell

NKG7

Natural killer cell granule protein 7

NKLT

Natural Killer-like T

NM23 H1

Non-metastatic clone 23, isoform H1

PAR-1

Protease-activated receptor 1

PARP

Poly-ADP ribose polymerase

PD-1

Programmed cell death protein 1

Treg

Regulatory T cell

RPE

Retinal pigment epithelium

ROS

Reactive oxygen species

SCI

Spinal cord injury

SPMS

Secondary progressive MS

STAT3

Signal Transducer and Activator of Transcription 3

CD8⁺ TEMRA cells

Terminally differentiated effector memory CD8⁺ T cells re-expressing CD45RA

tMCAO

Transient middle cerebral artery occlusion

TLR

Toll-like receptor

TNF-α

Tumor necrosis factor-α

TIGIT

T cell immunoreceptor with Ig and ITIM domains

TSP-1

Thrombospondin-1

VCAM-1

Vascular cell adhesion molecule 1

Author contributions

HSY led the conceptualization of the manuscript, wrote the majority of the manuscript, and generated the figures. CZ and PY assisted with literature curation and contributed to initial drafting of sections on Granzyme A and Granzyme K, respectively. AH assisted with figure preparation and critically reviewed the manuscript. JAM and DJG supervised the work and provided critical revisions of the manuscript. All authors reviewed and approved the final version of the manuscript.

Funding

This work was funded by Canadian Institutes of Health Research (CIHR) (JAM and DJG), Natural Sciences and Engineering Research Council of Canada (NSERC) (JAM), Fighting Blindness Canada (HSY and JAM), and Vancouver Coastal Health Research Institute (VCHRI) (JAM). DJG was also funded by an International Collaboration on Repairs Discoveries (ICORD) Seed Grant. HSY was funded by CIHR Fellowship.

Declarations

Conflict of interest

The authors declare no competing interests.

Ethical approval

Not applicable.

Consent to participate

Not applicable.

Consent for publication

Not applicable.