As we have seen so far, each LG component has its own cytolytic activity. This can potentially cause self-destruction of NK cells during synthesis and maintenance of LGs. NK cells have several protection layers to ensure safe storage and trafficking of cytolytic contents until degranulation. First, the acidic environment inside of LGs prevents the activity of the cytolytic proteins. In this low pH environment, perforin and granzymes also interact with chondroitin sulfate proteoglycan known as serglycin ( Figure 1 ). The association with serglycin keep both cytolytic proteins in an inactive state until secretion (95–98). In addition, several perforin-specific protection mechanisms have been identified (99). As previously mentioned, perforin is N-linked glycosylated at the C-terminus in the ER ( Figure 1 ). This prevents perforin oligomerization and pore-forming activity during its transit to LGs, regardless of calcium concentration and pH (51). Upon arrival at the LGs, the mature perforin without the inhibitory C-terminus is still kept inactive due to very limited availability of calcium (100). In addition, interaction of calreticulin with perforin in the ER and LGs was also suggested to contribute to the inhibition of perforin activity (101).

Our current understanding of the biogenesis of the LGs is mainly focused on the biosynthesis and sorting of cytolytic proteins into the LGs but not on the LG organelle itself. Are LGs derived from pre-existing lysosomes or are they generated independently from lysosomes? In addition, components of the lysosomes and LGs are often mediated by the same trafficking and sorting machineries. Therefore, it remains unclear how cells containing LGs distinguish cargoes between the two organelles. In this regard, it is interesting to note that proteins like AP3 and CHS1/LYST involved in LG biogenesis and/or LG protein sorting are ubiquitously expressed. Therefore, it would be interesting to elucidate how the mutations in these proteins only impact cells with secretory lysosomes. It was also recently shown that LG size and contents are associated with the efficiency of NK cell cytotoxicity (102). Future studies aimed at elucidating the mechanisms by which NK cells regulate the amount of cytolytic contents and the size of LGs will also be of interest. Finally, we have very limited understanding of the heterogeneity of the LGs. Thus, it will be interesting to examine potential differences among the LGs inside a single NK cell and define not only how these distinct LGs mature but also the signaling mechanisms regulating their exocytosis.

The release of cytolytic granules is accomplished through a heavily regulated stepwise process beginning with the convergence of cytolytic granules to the MTOC (10). This process occurs rapidly and is initiated through the engagement of adhesion receptors, such as the leukocyte function associated antigen-1 (LFA-1), in combination with other activating NK cell receptors. The function of convergence is both to prepare the LGs for directed secretion to a target cell and to effectively concentrate LGs for enhanced delivery (115, 116). This minimizes off target effects of LG secretion and ensures sufficient delivery of the cytolytic contents. Interestingly, LG convergence occurs in both activating and inhibitory NK CS and is independent of PI3K, MEK, and PLCγ activation, although these signals are required for maturation of the NK CS and degranulation (117). LG convergence also occurs prior to microtubule or F-actin reorganization as Taxol, cytochalasin D, and latrunculin A inhibited MTOC polarization to the synapse but not LG convergence (118, 119). This suggests that LG convergence is an early event downstream of adhesion and prior to large cytoskeletal reorganization events. Interestingly, high dose IL-2 can induce LG convergence independent from adhesion (117). This was found to be dependent on Src kinase activity, which is induced by IL-2 through a non-canonical, JAK3-independent, pathway and is also downstream of LFA-1 activation ( Figure 3 ) (117, 120). LG convergence, therefore, rapidly occurs downstream of activation but prior to a commitment to cytotoxicity.

The rapid accumulation of LGs at the MTOC is dependent on dynein/dynactin mediated minus-end-directed movement along the microtubule network ( Figure 3 ) (118). Although the dynein/dynactin complex is constitutively localized with LGs in NK cells, dynein-mediated LG movement requires additional adaptor proteins (118, 121). For example, HkRP3, which is localized at LGs and interacts with the dynein/dynactin complex, was found to regulate dynein complex-mediated LG convergence (122). Interestingly, Grb2 interacts with Src and the P150^Glued subunit of dynactin which could link Src activation to dynactin signaling (118). This alternative pathway of Src kinase-dependent LG convergence may help explain how high-dose IL-2 can rescue the phenotype of WASP deficiency through the activation of the WASP family member WAVE2 (123–125). Another mechanism that might regulate dynein function is its potential interaction with, and recruitment by, Rab7a and Rab Interacting Lysosomal Protein (RILP) ( Figure 3 ) (122, 126). Rab7a was identified in the lysosome fraction of the NK cell line YTS (127) and, with RILP, recruits dynein/dynactin complexes to lysosomes (128, 129). Furthermore, overexpression of RILP in CTLs causes clustering of LGs and prevents plus-end-directed movement, suggesting an important role in LG minus-end trafficking (130). However, the precise mechanisms regulating dynein-directed NK cell movement, and the role of Rab7a, have yet to be fully elucidated.

LG convergence is a prerequisite for the polarization of LGs to the NK CS (131–133). This is accomplished through the polarization of the MTOC and converged LGs to the maturing CS through mechanisms that include F-actin reorganization and continued signaling through clustered receptors ( Figure 3 ) (118, 133). Although there are differences in the rate of LG convergence and MTOC polarization between CTLs and NK cells, the mechanisms that control these processes are thought to be similar (134). Indeed, many studies investigating synapse formation and microtubule dynamics performed in the CD4⁺ Jurkat T cell line may be extrapolated, with care, to CD8⁺ T cells and NK cells, despite the lack of cytolytic ability in Jurkat cells. In CTLs, two mechanisms for MTOC polarization have been proposed. The first mechanism is a dynein-dependent cortical sliding mechanism where dynein, anchored to the cell cortex, pulls on microtubules to bring the MTOC toward the synapse (115, 135, 136). This method is supported by the role of adhesion and degranulation promoting adaptor protein (ADAP) in microtubule anchoring and the previous observations of synaptic microtubule anchoring in MTOC movement (135). The second proposed method of MTOC polarization is a capture-shrinkage mechanism where anchored dynein pulls on microtubules which depolymerize, effectively pulling the MTOC to the synapse (135, 137). This mechanism is also plausible as Taxol, which stabilizes microtubules thus preventing depolymerization, and ciliobrevin, which inhibits dynein activity, abrogated MTOC polarization in Jurkat T cells when used together, whereas use of Taxol alone only slowed polarization (135). Additionally, it was recently demonstrated that the kinesin-4 family member KIF21B, regulates microtubule organization and growth by inducing microtubule pausing and depolymerization (138). Knockout of KIF21B in Jurkat T cells resulted in decreased synaptic MTOC polarization attributed to overgrown microtubules at the synapse. MTOC polarization and microtubule organization was rescued by low dose vinblastine, which induces microtubule depolymerization (138). Although, Hooikaas et al. do not believe KIF21B directly participates in dynein-driven capture-shrinkage, they do not exclude the indirect impact excessively elongated microtubules may have on this process. While capture-shrinkage may be the predominant model when the MTOC and CS are diametrically opposed, when modeled with cortical sliding, the two mechanisms appear to work in synergy suggesting that a mechanistic combination may be more appropriate and applicable to a wider variety of interactions (139).

Furthermore, it was observed that MTOC polarization in CTLs appeared to occur through a two-step mechanism where LGs rapidly polarized to the synapse before slowing down to complete their journey (135). This was proposed to occur through the initial localization and function of dynein at the central region of the CS (central SMAC) followed by dynein activity at the pSMAC (140). The specific mechanisms regulating NK cell MTOC polarization remain unclear and warrant further investigation. In CTLs, it has been suggested that the strength of TCR signaling may regulate the specific mechanisms of MTOC polarization (131, 141). How this translates to NK cell signaling is unknown, especially as it relates to strength of signaling emanating from NK activating receptors.

Several cytoskeletal regulatory proteins are known to be critical for NK cell MTOC polarization including the small GTPase CDC42. CDC42 and WASP localize to the MTOC after LG convergence and are required for polarization (10, 142). This is mediated by CDC42 Interacting Protein (CIP4) which couples both the actin and microtubule networks through binding tubulin, CDC42, and WASP. In activated NK cells, CIP4 localizes with the MTOC to the NK CS and could function to anchor the MTOC through WASP or CDC42 activation (142). Furthermore, CIP4 is not required for F-actin accumulation at the synapse suggesting its primary role is on the LGs (142). ADAP, another cytoskeletal regulatory protein could also help apply force to the MTOC, as it is required for insertion of the microtubule plus-end into the ring-like F-actin network at the peripheral region of the CS, known as the peripheral supramolecular activation cluster (pSMAC) (10, 136, 143). Although ADAP is required for CTL degranulation, its role in NK cells is less clear with some reporting that it may be dispensable for NK cell killing (144).

After MTOC polarization toward the CS, the clustered LGs need to navigate the dense F-actin network at the cell cortex to dock and fuse with the NK cell membrane (145, 146). Although LGs can undergo kinesin-1 plus-ended microtubule movement (147), the proximity of the polarized MTOC and LGs to the synapse is likely sufficient to offload the LGs onto the F-actin network in CTLs (132). Indeed, LG trafficking at the CS has been shown to be independent of plus-ended microtubule movement, as overexpression of RILP kept LGs clustered at the MTOC, therefore preventing plus-ended trafficking, without any impact on lysis (132). However, there are some reports of kinesin-1 regulating plus-ended movement in CTLs (148). Interestingly, it was recently demonstrated that Arl8b regulates MTOC polarization in NK cells through its interaction with the kinesin-1 heavy chain KIF5B and SifA and kinesin-interacting protein (SKIP) (149). Silencing of KIF5B, SKIP or Arl8b led to defective MTOC polarization, suggesting that in NK cells, kinesin could regulate LG trafficking at a much earlier cytolytic stage than in CTLs (149). However, the role of kinesin in NK cell degranulation and the specific mechanisms regulating lytic granule transfer to the F-actin network are still unclear and require further investigation. Lastly, in CTLs it was shown that HDAC6, which deacetylates α-tubulin at Lys40 and interacts with kinesin-1 light chain, is required for proper lytic granule migration to the CS (150). Indeed, although CTLs taken from HDAC6-deficient mice showed a decreased MTOC to target cell distance, lytic granules appeared much more diffuse, suggesting a role for HDAC6 in LG trafficking at the CS (150).

Clearances in the cortical F-actin at the cSMAC have been identified in CTLs (132) and NK cells (151), suggesting a role for an actin motor protein to mediate the final stretch of LG trafficking to the membrane. Indeed, the movement of LGs on F-actin has been shown to be dependent on the non-muscle actin motor myosin IIA ( Figure 3 ). Myosin IIA is a hexamer consisting of two heavy chains, two regulatory light chains, and two essential light chains (152, 153) and is constitutively associated with LGs as single molecules rather than a filament (154). This association with LGs could be mediated through direct recognition of phosphatidylserine, binding of Rab27a, or through binding of the WASP/WIP complex (153). Inhibition or depletion of the myosin IIA heavy chain, MYH9, prevents degranulation but does not impair conjugate formation, LG convergence, synaptic actin reorganization, or MTOC polarization (153, 155, 156). The importance of myosin IIA in NK cell function can be fully appreciated in a group of diseases, now referred to MYH9-related disease (MYH9-RD), caused by mutations in the heavy chain MYH9 ( Table 1 ) (157, 158). Patients with a truncation in MYH9 had ablated cytotoxicity with intact conjugate formation, MTOC convergence, and MTOC polarization (153). Interestingly, the truncation affected both a region of MYH9 important for cargo binding and removed a constitutively phosphorylated serine (MYH9 S1943) required for MYH9 recruitment to LGs (154). Disruption of this key residue resulted in LGs that were present at the synapse but unable to penetrate the F-actin network to reach the membrane (154).

The interaction of myosin IIA with LGs is also dependent on the chaperone protein UNC-45A. UNC-45A colocalizes with LGs in both resting and activated NK cells and polarizes with the LGs to the NK CS upon target engagement ( Figure 3 ) (159). Like MYH9 deficiency, depletion of UNC-45A did not impair conjugate formation, LG convergence, or MTOC polarization but is critical for degranulation (159). Depletion of UNC-45A reduces myosin IIa binding to F-actin without impacting myosin IIA expression or stability (159). In addition to regulating myosin IIA, UNC-45A could have an additional independent role in regulating LG priming, docking and fusion, however, this has not been fully elucidated in NK cells (160).

After transport to the synapse, cytolytic granules dock at the membrane and are prepared for release. This is mediated through the small GTPase Rab27a which was first identified to play a critical role in degranulation through the study of Griscelli syndrome (GS) patients ( Figure 3 ) (161). GS is a rare autosomal recessive disease characterized by partial albinism due to defective melanosome transport. Although originally described as being caused by mutations in myosin Va (GS type 1), it was discovered that mutations in Rab27a (GS type 2) is the predominant disease etiology and is responsible for all GS cases with hemophagocytic lymphohistiocytosis ( Table 1 ) (161, 162). Interestingly, loss of Rab27a but not myosin Va resulted in defective CTL and NK cell degranulation and cytotoxicity (161). This degranulation defect was recapitulated in the mouse model of GS type 2, the ashen mouse, where it was observed that LG convergence and MTOC polarization was intact (163, 164), however, LG membrane docking was not observed by electron microscopy (165). Additionally, in the absence of stimulation, Rab27a regulates microtubule and actin-dependent LG movement at the cell cortex (166). Rab27a therefore regulates LG movement in unstimulated NK cells and is required for the final stages of NK cell-mediated cytotoxicity. Unsurprisingly, Rab27a is also a key secretory protein required in the degranulation of melanocytes, neutrophils, and pancreatic beta cells, suggesting a similar method of action in the terminal stages of LG export (167–169).

The crucial role of Rab27a in NK cell degranulation is mediated through effector proteins which bind to active GTP-bound Rab27 (170). So far eleven effector proteins have been identified in humans and mice and can be categorized into three distinct groups based on domain composition. The first group is comprised of rabphilin and the synaptotagmin-like proteins (Slp): Slp1, Slp2-a, Slp3-a, Slp4-a, and Slp5. The proteins within this group contain an N-terminal Slp homology domain (SHD), which mediates binding to the switch II region of GTP-Rab27a, and two tandem C-terminal C2 domains (C2A and C2B), which bind phospholipids (170). Slp1 and Slp2-a are expressed in NK cells and CTLs and are thought to mediate LG docking and tethering via their C2 domains. However, their role in cytotoxicity is still unclear as CTLs from Slp1- or Slp2-a-deficient mice had intact degranulation (171). Furthermore, expression of the SHD from Slp2-a, which has a dominant negative effect, only resulted in a partial reduction in cytotoxicity suggesting that there might be other important proteins in complex with Rab27a.

The next group of GTP-Rab27a effectors is characterized by the presence of the N-terminal SHD and the distinct absence of C-terminal C2 domains (170). This group is comprised of Noc2 and the Slp homolog lacking C2 domain (Slac2) proteins: Slac2-a, Slac2-b, and Slac2-c. However, no role for Slac2 proteins has been identified in NK cells. The last group contains the protein Munc13-4 which has an N-terminal C2 domain followed by a Rab binding domain (RBD), a MUN domain, and a second C2 domain (170). Interestingly, Munc13-4 lacks a C1 domain, which mediates DAG binding, found in other Munc13 family members and the mechanism of Rab27 binding by the Munc13-4 RBD is uncharacterized despite the importance of Munc13-4 in LG exocytosis. In addition to the Rab27a interactors described above, proteins involved in Rab27 prenylation, as well as several proteins known to bind GDP-bound Rab27 including CORONIN-3, RabGDI, and MAP kinase activating death domain (MADD) have remained largely uncharacterized as it pertains to their roles in NK cell-mediated killing (170, 172).

The tethering of cytolytic granules at the synapse refers to the process by which LGs make initial interaction with the membrane and are prepared for fusion. This is Rab27a-dependent and is thought to be mediated by Slp1, Slp2, and Munc13-4 ( Figure 3 ) (165, 171). Munc13-4 mutations were first identified as the cause of familial hemophagocytic lymphohistiocytosis type 3 (FHL3), where it was observed that mutation of Munc13-4 resulted in impaired CTL degranulation with intact conjugate formation and MTOC polarization ( Table 1 ) (165). Upon stimulation, Munc13-4 localizes to the CS in CTLs and strongly colocalizes with LGs (162, 165). Interaction with Rab27a is critical to Munc13-4 function as wild type but not RBD-deficient or RBD mutant Munc13-4 was able to restore degranulation in FHL3 CTLs (173). Interestingly, recruitment or retention of Rab27a and Munc13-4 on LGs is co-dependent as Rab27a is not recruited or retained on granules in FHL3 patients and Munc13-4 recruitment or retention on LGs is impaired in GS type 2 patients (162, 173, 174). In contrast to Rab27a deficiency, however, Munc13-4 is not required for LG docking at the membrane, suggesting its main function is through the tethering of LGs (165). Indeed, both the C2A and C2B domains are required for cytotoxicity as they mediate binding to both lipids and soluble N-ethylmaleimide-sensitive factor activating protein receptor (SNAREs) (175). Munc13-4 may have a similar function to synaptotagmin in neuronal degranulation, as C2 domain binding is calcium-dependent, and thus Munc13-4 might be the calcium sensor that triggers synchronous granule release (176, 177). Interestingly, nicotine acid adenine dinucleotide phosphate (NAADP), a Ca²⁺-mobilizing second messenger is involved in exocytosis through their regulation of the two-pore channel (TPC) 1 and TPC2 which are present on lytic granules (178, 179). It was shown in CTLs that NAADP activates TPCs in a manner that is independent of global Ca²⁺release stimulated by IP₃ suggesting that a local release of Ca²⁺ facilitates lytic granule vesicle fusion (178) ( Figure 3 ). The molecular targets of this local calcium release could be molecules involved in vesicle fusion such as Munc13-4.

Interestingly, low levels of degranulation were observed in GS type 2 deficiency, suggesting Munc13-4 might also partner with other Rab proteins. Indeed Munc13-4 binds to Rab11 and Rab15 regulating recycling endosomes and Weibel-Palade body exocytosis (170, 180, 181). In CTLs, it has been demonstrated that Munc13-4 mediates the fusion of Rab11 positive recycling endosomes with Rab27a positive late endosomes which are then transported to the synapse for release (174, 181). However, this does not seem to occur in NK cells. Despite low levels of perforin in recycling endosomes, Rab11 and other recycling endosome markers are not tightly associated with NK cell synapses. This is further confirmed as Munc13-4 deficiency does not impact release of IFNγ and TNFβ, which is mediated through a Rab11 positive recycling endosome pathway (182), and recycling endosome inactivation does not greatly impact preformed LG release (174).

LG priming is mediated by the assembly and recruitment of SNARE complex proteins that facilitate the fusion of the LG with the NK plasma membrane at the CS. This is dependent on Munc13-4, which promotes SNARE complex formation through its interaction with the SNARE protein syntaxin 11 ( Figure 3 ) (183, 184). Munc13-4 carries out this function by opening the conformation of syntaxin 11 through the removal of its chaperone protein, syntaxin binding protein 2 (STXBP2), which is required for syntaxin 11 stability and subcellular localization (173, 185, 186). Interestingly, overexpression of syntaxin 11 in NK cells and activated CTLs increased their cytolytic potential (187, 188). Like Munc13-4 mutations, mutations in syntaxin 11 cause FHL4 which presents as a defect in NK cell degranulation with intact synapse formation and LG polarization ( Table 1 ; Figure 3 ) (189). This was also recapitulated in the mouse model of FHL4 (184). STXBP2 mutations detrimental to its interaction with syntaxin 11 cause FLH5, which has the same phenotype as FLH4 ( Table 1 ) (190, 191). Interestingly defective degranulation in FLH4 and FLH5 can partly be rescued by IL-2 treatment. This is thought to occur through alternate pairing as IL-2 induces expression of syntaxin-3, which replaces syntaxin 11 in FHL4, and STXBP1, which replaces STXBP2 in FLH5 (192). Additionally, syntaxin 7 was shown to be required for CTL degranulation but has not been investigated in NK cells (193). Although the function of alternative pairing was investigated in FLH5 patients, the roles of other syntaxins, such as syntaxin-1 and syntaxin-7, warrant further investigation (194).

Lytic granule priming and SNARE complex assembly was recently shown to be supported by septin filaments (195). Septins are GTP-binding proteins which can be organized into 4 subgroups (represented by septin 1, septin 3, septin 6, and septin 7), and can assemble into hetero-hexamers and hetero-octamers, which can assemble to form complex structures (196). Septin 7 is the only member of its subgroup and is indispensable for septin complex formation. Interestingly, depletion of septin 7 decreases septin 1 and septin 2 expression and impairs NK cell degranulation without impacting conjugate formation or lytic granule accumulation at the CS (195). Despite being concentrated to the cell cortex away from the CS, septin 7 puncta were observed in apposition to lytic granules and were found in the crude lysosomal fraction of NK cells (195). Mass spectrometry analysis of septin 2 crude lysosomal fraction immunoprecipitates revealed associations with lytic granule regulatory proteins including syntaxin 11 and STXBP2, which were confirmed by proximity ligation assay and immunoprecipitation (195). Additionally, depletion of septin 7 or septin stabilization with forchlorfenuron, decreased association of STXBP2 with syntaxin 11. This suggests that septin filaments stabilize SNARE complex assembly and are critical for facilitating syntaxin11 and STXBP2 interaction and ultimately, LG membrane fusion.

The final stage of LG degranulation is the fusion of the LGs with the plasma membrane, which is mediated by formation of trans-SNARE complexes. Due to the number of SNARE protein combinations required to make a complete fusion complex, they have not been fully defined in NK cells (197). In human CTLs, membrane fusion is promoted by STXBP2 which forms a trans-SNARE complex with STX11, SNAP23, and VAMP8 (198). In mice, however, this is mediated by VAMP2 and VAMP8 which colocalize with LGs. Loss of VAMP2 and VAMP8 in mice resulted in impaired degranulation and cytotoxicity (199–201), suggesting other VAMPs cannot compensate for the loss of VAMP2 and VAMP8. Indeed, in NK cells, VAMP4 and VAMP7 are required for NK cell cytotoxicity, while VAMP1, VAMP3 and VAMP8 have limited colocalization with LGs and are therefore likely dispensable for SNARE complex assembly (197, 202). In addition to STX11, STX6 may also play a role in NK cell SNARE complex formation as they interact with VAMP7 and VAMP4, and other syntaxins, like STX4, have been shown to regulate mast cell degranulation ( Figure 3 ) (203–206). After SNARE complex formation, a fusion pore between the plasma membrane and NK cells are formed through which degranulation occurs. Interestingly, the size and fusion status of the pore determines the amount of granule content release, which in turn may regulate LG membrane recycling (197). In neuronal cells this has been called the “kiss and run” pathway, although the mechanisms regulating this process in NK cells are unclear, especially in the context of cell-to-cell interactions and warrant further investigation.