LGs were initially characterised as killing entities of 300-1100 nm consisting of an electron-dense core often surrounded by 30-70 nm vesicles enclosed by the delimiting outer membrane (5). LG fractionation allowed for the identification of their cytotoxic contents, which consist of a battery of serine proteases with different substrate specificity, the granzymes (Gzm) (13), and Prf, a protein with structural and functional homology to bacterial pore-forming toxins (14), packed together on a scaffold of the proteoglycan serglycin (Srgn) (15) in the dense core. Additionally, LGs contain the processed, 9 kDa isoform of granulysin, a saposin-like membrane-disrupting protein (16). Several pieces of evidence witness to the lysosomal origin of LGs, including the low pH and the presence of lysosomal hydrolases (e.g. cathepsin D) and of lysosomal membrane glycoproteins (e.g. LAMP-1) (5, 17). The multivesicular cortex of LGs is also enriched in the cation-independent mannose-6-phosphate receptor (CI-MPR), which transports the acid hydrolases to lysosomes (18). The multivesicular structure of LGs is generated through the invagination of the membrane of early endosomes (EE) (19), the sorting hub for proteins endocytosed at the cell surface, which accounts for the identification of plasma membrane-associated proteins such as T cell receptors (TCRs) and integrins in the multivesicular cortex (17) and highlights the endolysosomal origin of these organelles.

The pathways that regulate LG biogenesis downstream of expression of their components have been in part elucidated ( Figure 1 ). Granzymes, of which the best characterized is GzmB, enter the endoplasmic reticulum (ER) as inactive precursors that are sorted at the trans-Golgi network (TGN) by the CI-MPR (20) following N-glycosylation and acquisition of a mannose-6-phosphate moiety, and bud off in endocytic carriers in a process involving the endosomal sorting complex required for transport (ESCRT) machinery (21). Gzms are first transported to EEs, wherefrom they are sorted into MVBs and directed to late endosomes (LE) (19). At the low local pH they complete their proteolytic maturation but are maintained inactive both by the low pH and through their sequestration by Srgn in the dense core (13). By contrast, the pathway that regulates Prf transport to maturing LGs is as yet poorly characterized. A molecular determinant within its C-terminus and N-linked glycosylation (22), as well as LAMP-1 interaction with the AP-1 sorting complex (23), are required for its efficient export from the ER and LG localisation. Multiple levels of control ensure that Prf is kept inactive until release upon CTL activation. To polymerise and form pores on biological membranes Prf requires Ca²⁺ binding, which is prevented by the low pH of LGs. Additionally, Prf is kept in a monomeric form by Srgn binding (24). As opposed to the lytic components of the dense core, integral LG membrane proteins such as LAMP-1 or LAMP-2 are sorted at the TGN through canonical tyrosine-based or di-leucine-based motifs that are recognized by AP-1 (25).

TCR engagement on CTLs triggers centrosome polarisation towards the contact with the target cell and its close apposition to the synaptic membrane in a process tightly regulated by the actin and tubulin cytoskeletons, which sets the stage for the dynein- and AP-3-dependent transport of LGs to the immunological synapse (IS) center (26, 27), as well as by TCR signal strength (28). Interestingly, rapid LG secretion on target cell encounter can occur in the absence of centrosome polarisation, which is instead essential for the establishment of stable, stimulatory synapses (29), suggesting an alternative mechanism for LG mobilisation to the target cell contact.

Gzm- and Prf-enriched LGs acquire the ability to dock to the plasma membrane and deliver their cytotoxic cargo into the synaptic cleft through two sequential maturation steps that have been extensively characterized following the identification of gene mutations responsible for primary immunodeficiency disorders associated with defective CTL function (30). LG docking to the synaptic membrane requires the acquisition of two key trafficking regulators: the Rab GTPase Rab27 and its effectors synaptotagmin-like proteins SLP1/2, and the adaptor Munc13-4 (31–37). According to one model, this involves fusion of maturing LGs with an intermediate exocytic vesicle generated through the Munc13-4-dependent fusion of recycling endosomes (RE), which carry Munc13-4, with LEs, which carry Rab27a (37). The resulting docking-competent LGs exploit the phospholipid-binding ability of SLP1/2 to interact with the synaptic membrane. Docked LGs become fusogenic through a priming step, which also depends on Munc13-4 (37). This involves the formation of a complex between the vesicle-soluble NSF attachment protein receptor (v-SNARE) VAMP7 at the LG membrane and the target-soluble NSF attachment protein receptor (t-SNARE) syntaxin 11 and its partners SNAP23 and Munc18-2 at the plasma membrane (38–40). More recent data indicate that, rather than fusing before delivery to the IS, REs carrying Rab27 and Munc13-4 and LGs polarise to the IS independently and undergo sequential fusion (41). Accordingly, syntaxin 11 has also been shown to be transported to the IS by REs and released with the assistance of the v-SNARE VAMP8, thereby marking the location for LG docking (42).

Once released into the synaptic cleft, Prf and Gzms cooperate to promote the apoptotic demise of the CTL target. Although the key role for Prf in the delivery of Gzms to the cognate target has been well established, different mechanisms have been proposed, all involving the pore-forming activity of Prf, the polymerisation of which is enabled by its dissociation from Srgn at the higher pH of the synaptic cleft and Ca²⁺ binding: i) delivery of Gzms to the cytosol of the target cell through Prf pores, either directly at the synaptic membrane (43) or following their co-internalisation in endosomes (44, 45); or ii) internalisation of Gzms as a membrane repair response triggered by Prf-induced membrane damage (46). In the cytosol Gzms induce target cell apoptosis by activating caspase-dependent and caspase-independent pathways (13). Granulysin also contributes to the cytotoxic activity of LGs by interacting with the target cell membrane through its positive charges and inducing the influx of Ca²⁺, which leads to mitochondrial damage and caspase-3 activation (16).

In addition to their LG-dependent cytotoxic activity, CTLs exploit the Fas pathway to kill target cells (47). CTLs express FasL, a type II transmembrane protein that is upregulated at the cell surface in response to target cell recognition. Similar to Gzms and Prf, pre-formed FasL is stored in LROs that had been proposed to correspond to LGs, based on its co-localisation with LG markers and the observation that FasL delivery to the cell surface was dependent on degranulation (48). More recent findings indicate that FasL, while indeed stored in LROs, is associated with granules that appear distinct from canonical, dense-core LGs (49). Proteomic analyses identified two subpopulations of cytotoxic granules of different size, of which the larger (300-700 nm) is enriched in FasL and lysosome/MVB markers, and the smaller (<300 nm) in Gzms and Prf (50). Additionally, release of pre-stored FasL was found to have a lower TCR signal threshold than LG release, to be microtubule- and extracellular Ca²⁺-independent, and under certain stimulation conditions to occur in the absence of degranulation (49, 51, 52). The association of intracellular FasL with two populations of LGs, as documented in NK cells (53), may account for these discrepancies.

FasL is targeted to MVB, where it localises together with APO2L/TRAIL both at the limiting membrane and in CD63⁺ intraluminal vesicles (ILV) that are released into the synaptic cleft as EVs (8, 9, 48, 54). Sorting of FasL into MVBs is regulated by phosphorylation by Src kinases that interact with its proline-rich domain and by mono-ubiquitylation (10, 55) which allows for its routing to the ESCRT pathway (21). In this pathway proteins destined for LEs and lysosomes are sorted into ILVs at EEs through recruitment of ESCRT-0, which binds phosphatidylinositol 3-phosphate (PI3P) on the endosomal membrane and mono-ubiquitylated proteins, followed by sequential binding of ESCRT-I and ESCRT-II to nucleate ESCRT-III filaments. These drive the process of membrane deformation that is essential for invagination and pinching off of ILVs (21).

Upon target cell recognition, MVBs polarise towards the centrosome and undergo fusion with the synaptic membrane in a process that in T cells is regulated by diacylglycerol kinase α and the tetraspanin MAL (56, 57). Depending on its localisation within the MVB, FasL is either redistributed to the plasma membrane or released in association with EVs. Using supported lipid bilayers (SLB) to allow for tight control of target membrane composition, Balint and colleagues showed that FasL⁺ puncta were present in the synaptic cleft only when Fas was included in the SLB (11). While this indicates that FasL is associated to EVs, how their delivery is linked to the availability of Fas at the target membrane is not clear. A possible mechanism is suggested by the requirement of CD40 in the SLB for the synaptic release of CD40L⁺ vesicles from helper T cells, which is triggered by co-clustering of CD40L-CD40 with TCR-peptide major histocompatibility complex (pMHC) (58). Whether plasma membrane- or EV-associated, FasL can engage Fas on the target cell, triggering the assembly of a signaling complex that leads to activation of the apoptotic cascade (47).

The recent discovery of SMAPs (11) has unveiled a third, unconventional mechanism of CTL-mediated cytotoxicity, which is shared by NK cells (59). Balint and colleagues used SLBs functionalised with anti-CD3 mAbs and ICAM-1 to activate CTLs and observed that, after CTL removal, glycoprotein complexes -the SMAPs- were left behind. A proteomic analysis of SMAPs revealed an unexpected composition featuring a lack of membrane proteins and an enrichment in canonical LG effectors (Prf, Gzms, Srgn), as well as glycoproteins, of which thrombospondin-1 (TSP-1) and galectin-1 were prominent components. Super-resolution imaging and structural analyses showed that SMAPs are highly stable ∽120 nm particles, with the lytic effectors concentrated in a core surrounded by a glycoprotein shell. Consistent with their lytic cargo, purified SMAPs have the ability to kill cells autonomously (11).

Within CTLs SMAPs are stored in multicore granules (11). A recent report by the Rettig lab has shed light on the identity of this LG population. Using a mouse knock-in for fluorescently tagged Synaptobrevin2 (Syb2), the murine homologue of VAMP7 that marks fusion-competent LGs, Chang and colleagues (12) purified and characterized mature LGs. They identified two distinct Syb2⁺ populations that were confirmed to be LGs based on co-fractionation and co-localization with GzmB: a homogeneous population of smaller granules with a single dense core (SCG), and a heterogenous population of larger granules with multiple dense cores (MCG). Proteomic analysis revealed a remarkably different composition of SCGs and MCGs, with SCGs enriched in lysosomal proteins and MCGs in endosomal trafficking regulators. TSP-1 was found to selectively associate with MCGs and to be released in particles with a similar core-shell structure as the human CTL-derived SMAPs. SCGs and MCGs release involved distinct fusion events, suggesting that the two classes of LG mature and undergo exocytosis through independent pathways (12).