MADD regulates natural killer cell degranulation through Rab27a activation Free

Natural killer (NK) cells are innate cytotoxic lymphocytes capable of killing stressed, virally infected and cancerous cells without prior sensitization. Unlike CD8⁺ T cells, NK cells do not undergo germline receptor rearrangement and instead rely on a balance of activating and inhibitory receptors to identify potential targets (Ham et al., 2022; Paul and Lal, 2017). With sufficient activation, NK cells will undergo a tightly regulated stepwise process culminating with the release of specialized pre-formed secretory lysosomes, named lytic granules (LGs) (Orange, 2008). This process begins with adhesion receptor ligation, which induces clustering of LGs at the microtubule-organizing center (MTOC) (Mentlik et al., 2010). Upon sufficient activating receptor stimulation, the clustered LGs will polarize with the MTOC to the cytolytic synapse where they offload from the microtubules onto F-actin and dock near the plasma membrane (Andzelm et al., 2007; Mentlik et al., 2010; Sanborn et al., 2009).

Terminal transport and fusion of LGs at the plasma membrane is mediated by the small GTPase Rab27a, which cycles from an active GTP-bound state to an inactive GDP-bound state through GTP hydrolysis. Loss of Rab27a results in a rare autosomal recessive disease characterized by hypopigmentation and variable immunodeficiency called Griscelli syndrome type-II (GS-II) (Menasche et al., 2000). T cells and NK cells isolated from individuals with GS-II, or Rab27a-deficient mice, polarize LGs to the synapse when stimulated but are unable to fuse their LGs with the membrane and degranulate (Haddad et al., 2001; Menasche et al., 2000; Stinchcombe et al., 2001). Unsurprisingly, mutations in key Rab27a effector proteins can result in similar deficiencies. One such protein, Munc13-4 (also known as UNC13D), binds to active GTP-bound Rab27a and anchors LGs to the membrane, priming them for release (Feldmann et al., 2003; Fukuda, 2013). Additionally, Munc13-4 initiates activation of the SNARE complex, which mediates the fusion of the lytic granule membrane with the plasma membrane of the cell (Boswell et al., 2012). Loss of Munc13-4 expression or function results in a disease similar to GS-II called familial hemophagocytic lymphohistiocytosis type 3 (FHL3) (Crozat et al., 2007; Feldmann et al., 2003; Yamamoto et al., 2004), which is characterized by defective LG exocytosis and severe systemic inflammation.

Although Rab GTPases require guanine nucleotide exchange factors (GEFs) for activation, a majority of the 63 human Rabs lack a known GEF (Yoshimura et al., 2010). Mitogen-activated protein kinase activating death domain protein or MADD, also called Rab3GEP, is a DENN domain-containing protein that was first identified as a GEF for isoforms of Rab3 in rat neurons (Wada et al., 1997), but subsequent screening for a Rab27a GEF in melanocytes identified MADD as a key regulator of Rab27a activation in melanosome transport (Figueiredo et al., 2008). MADD has also been identified as a crucial GEF for the Rab27a-dependent isoproterenol-induced release of amylase from rat parotid acinar cells and the release of Weibel–Palade bodies from endothelial cells (Imai et al., 2013; Kat et al., 2021). Biallelic mutations or variations in MADD results in a pleiotropic disorder including failure to thrive, mild-to-severe developmental delay, endocrine and exocrine dysfunction, and hematological anomalies (Abu-Libdeh et al., 2021; Schneeberger et al., 2020). However, the ability of MADD to function as a Rab27a GEF to regulate LG degranulation in lymphocytes has not been addressed.

In this study, we used a combination of molecular, cellular and genetic approaches to investigate the role of MADD in cell-mediated killing by NK cells and CD8⁺ T cells. Herein, we find that loss of MADD expression by CRISPR Cas9 KO reduced GTP-bound Rab27a in both stimulated and unstimulated NK cells leading to a significant reduction in cytotoxicity and degranulation. Moreover, we show that MADD colocalizes with Rab27a on lytic granules and is enriched at the cytolytic synapse in NK cells. Altogether, our data highlight the important contribution of MADD to Rab27a activation and the regulation of cytolytic immune cell killing.

Given that loss of Rab27a in cytotoxic lymphocytes severely ablates cytotoxicity, we sought to examine whether MADD-deficient lymphocytes also display cytotoxic deficiencies. To test this, we generated MADD and Rab27a knockouts (denoted crRab27 and crMADD, respectively) in human CD8⁺ T cells (Fig. 1A) and used these cells in reverse antibody-dependent cell cytotoxicity (rADCC) assays against the p815 mastocytoma cell line coated with anti-CD3 and anti-CD28 antibodies. Importantly, loss of MADD decreased cytotoxicity to the same level seen with Rab27a knockout over all the effector cell-to-target cell (E:T) ratios tested (Fig. 1B), thus highlighting the importance of MADD in CD8⁺ T cell killing. Similarly, MADD- and Rab27a-knockout primary human NK cells were generated and assessed for cytotoxicity against various target cell lines (Fig. 1C). Interestingly, targets such as 721.221 and p815 cells coated with either anti-CD16 or anti-NKG2D and anti-2B4 antibodies (NKG2D is also known as KLRK1, and 2B4 is also known as CD244), showed decreased, but higher than expected levels of cytotoxicity when either MADD or Rab27a were knocked out (Fig. 1D; Fig. S1A,B). We also tested cytotoxicity against several pancreatic ductal adenocarcinoma cell lines that are more resistant to NK cell-mediated lysis, such as Panc1 and 03.27 (Fig. 1E,F), and L3.6 and 4535 (Fig. S1C,D). The use of these cell lines revealed a more obvious defect in cytotoxicity, similar to that seen with Rab27a deficiency. Interestingly, regardless of the target or antibody stimulation used, deletion of either MADD or Rab27a resulted in a similar level of decreased killing, suggesting that MADD is likely the sole activator of Rab27a involved in the regulation of cell-mediated killing in these cytotoxic lymphocytes.

To rule out the possibility that the observed defect in cytotoxicity might be attributed to earlier stages of killing, we first assessed the ability of NK cells to bind target cells using a standard conjugate assay. Significantly, we saw no difference in the ability of crNC (CRISPR negative control) and crMADD NK cells to form conjugates with target cells (Fig. 2A; Fig. S2A). Importantly, no changes in activating receptors, integrins or perforin expression were seen by flow cytometry (Fig. S2B,C). Given that it has been demonstrated that loss of MADD impacts the trafficking of melanosomes in melanocytes, we investigated whether MADD also impacted the clustering of lytic granules to the MTOC and the polarization of the MTOC to the cytolytic synapse. To test this, NK–Panc1 conjugates were stained with anti-perforin antibodies to label LGs and anti-γ-tubulin antibodies to label the MTOC, with the distance of the MTOC to the cytotoxic synapse and the mean LG distance to the MTOC being measured. Neither crMADD or crRab27a knockout significantly impacted the ability of LGs to cluster at the MTOC or polarize to the synapse (Fig. 2B,C).

Given that the processes leading to LG fusion were not impaired by the deletion of MADD, we also assessed the ability of NK cells to degranulate. NK cells were incubated with Panc1 cells for 3 h in the presence of anti-CD107a antibody. Upon degranulation, CD107a is exposed to the surface of the cell where it can be bound by the antibody. In crMADD and crRab27a NK cells, both the percentage of cells positive for cell surface exposed CD107a and the mean fluorescence intensity of CD107a signal was decreased compared to that seen for crNC NK cells (Fig. 2D). Altogether, these data indicate that MADD is a crucial regulator of NK cell-mediated LG plasma membrane fusion through the activation of Rab27a.

To investigate the ability of MADD to activate Rab27a in NK cells, we generated a GST fusion protein containing the SLP homology domain (SHD) from Slac2b because it has been previously used to assess active GTP-bound Rab27a (Fig. 3A) (Kat et al., 2021). To verify the specificity of the GST–SLAC2b-SHD fusion protein, HEK293T cells were transfected with GFP–Rab27a, dominant-negative GFP–Rab27a.T23N and constitutively active GFP–Rab27a.Q78L. As expected, the GST–SLAC2b-SHD fusion protein was able to selectively pulldown GFP–Rab27a and GFP–Rab27a.Q78L but not GFP–Rab27a.T23N (Fig. 3B). Furthermore, a higher proportion of GFP–Rab27a.Q78L was enriched in the pulldown compared to GFP–Rab27a, demonstrating the ability of the GST–SLAC2b-SHD to enrich active GTP-loaded Rab27a (Fig. 3C).

To test whether MADD was able to activate endogenous Rab27a in NK cells, crNC and MADD-knockout (crMADD) NK cells were stimulated with anti-NKG2D and anti-2B4 antibodies, and the resulting lysates were assessed for active Rab27a with the GST–SLAC2b-SHD fusion protein. At rest, crMADD NK cells had reduced levels of active GTP-bound Rab27a compared to those in crNC NK cells (Fig. 3D,E). Importantly, total Rab27a protein levels were unaffected by MADD knockout. Upon stimulation with anti-NKG2D and anti-2B4 crosslinking, the amount of active GTP-bound Rab27a in crNC cells increased slightly, but insignificantly, above basal levels (Fig. 3D,E). However, in stimulated crMADD NK cells the amount of GTP-bound Rab27a was not increased and remained similar to that in unstimulated crMADD cells and below that of unstimulated crNC NK cells (Fig. 3D,E). Although, some donors had small, insignificant, increases in GTP-bound Rab27a in crMADD NK cells upon stimulation (Fig. 3D,E), this might be attributed to incomplete MADD knockout or an intrinsic GTP loading capability of Rab27a. Additionally, the fold change reduction in active Rab27a seen with loss of MADD matched what has previously been seen in human umbilical vein endothelial cells (HUVECs) (Kat et al., 2021).

Given that MADD has been reported to activate the MAPK kinase pathway downstream of TNFR1 (also known as TNFRSF1A) (Schievella et al., 1997), and to eliminate the possibility that the decreased GTP-bound Rab27a seen with loss of MADD could be the result of decreased NK cell activation, we examined levels of phospho-ERK (pERK1/2; ERK1 and ERK2 are also known as MAPK3 and MAPK1, respectively) and phospho-tyrosine (Fig. S3A,B). Importantly, the knockout crMADD NK cells displayed no significant defects in MAPK activation downstream of NKG2D and 2B4 (Fig. S3A,B). Additionally, no differences were seen in total phospho-tyrosine levels between crNC and crMADD NK cells (Fig. S3A). Taken together, our data implicate MADD in the regulation of Rab27a-GTP loading in NK cells both basally and downstream of NKG2D and 2B4.

To investigate the cytosolic structures where MADD localizes in NK cells, we examined the colocalization and overlap of MADD with various endosomal and LG markers by confocal microscopy (Fig. 4A,B; Fig. S4A). Antibody specificity for immunostaining was demonstrated by confocal microscopy in MADD and Rab27a knockout NK cells (Fig. S4B). MADD showed association with markers of the rapid recycling endosome (Rab4 proteins), early endosomes (EEA1), lysosomes (LAMP1), recycling endosomes (RAB11FIP1) and perforin (LGs). However, the highest correlation for colocalization was with Rab27a, which is also enriched on LGs in NK cells (Fig. 4B).

To further explore the association of MADD with Rab27a, individual NK cells (Fig. 4C, top) and NK cells in conjugate with the pancreatic cancer cell line, Panc1, were stained with MADD and Rab27a antibodies and imaged by confocal microscopy (Fig. 4C, bottom). In both instances, MADD and Rab27a were largely vesicular in nature demonstrating punctate staining within the cytoplasm, with a majority of the signal appearing proximal to each other, suggestive of a potential interaction (Fig. 4C). Indeed, in NK cells in conjugate with Panc1 target cells, both Rab27a and MADD were enriched together at the cytolytic synapse (Fig. 4C,D). To further substantiate the interaction of MADD and Rab27a, high resolution Airyscan microscopy was used to assess NK cells spread on fibronectin-coated coverslips. In some instances, MADD and Rab27a appeared to be colocalized along filamentous structures within the cell (Fig. 4C), consistent with reports of MADD interacting with kinesin motor proteins to regulate the transport of synaptic vesicles in neurons along microtubules (Hummel and Hoogenraad, 2021; Niwa et al., 2008).

Recent reports have indicated that Rab27a is highly enriched on multicore LGs, compared to single-core LGs, which can be identified by pre-incubating NK cells with wheat germ agglutinin (WGA) and staining for perforin or granzyme B (Balint et al., 2020; Chang et al., 2022). To examine whether MADD also localized with Rab27a on multicore LGs, we performed the proximity ligation assay (PLA) on WGA-labeled NK cells in Panc1 conjugates (Fig. S5A–C). Interestingly, some but not all WGA-positive structures contained PLA spots. Whereas MADD and Rab27a had high positive correlation, their correlation on WGA-positive structures was only weakly positive (Fig. S5A–C).

Because MADD has been shown to regulate kinesin-dependent transport, and MADD and Rab27a localized on filamentous structures in NK cells, we investigated whether MADD was associated with the microtubule network. In NK–Panc1 conjugates co-stained for MADD and α-tubulin, MADD-positive puncta were found in coincident spots along microtubules (Fig. 5A). Disruption of the microtubule network by nocodazole treatment did not result in the loss of MADD puncta, but did slightly reduce the association of MADD with perforin, suggesting that the microtubule network is only minimally involved in the recruitment of MADD to vesicular structures (Fig. 5B,C).

Previous studies in melanocytes and endothelial cells have shown that the depletion of MADD resulted in Rab27a-containing vesicles clustering in the perinuclear region of the cell (Figueiredo et al., 2008; Kat et al., 2021). We therefore investigated whether the depletion of MADD in NK cells would affect the localization of Rab27a with perforin-containing LGs. Interestingly, MADD depletion did not affect the interaction of Rab27a and perforin in resting or activated NK cells (Fig. 6A,B). Furthermore, Rab27a and perforin colocalized together at the cytotoxic synapse in MADD-depleted cells, in a similar way to what is seen in control NK cells (Fig. 6B).

In this study, we describe a role for MADD in NK cell and CD8⁺ T cell degranulation through the activation of Rab27a (Fig. 7). We demonstrate that MADD knockout in cytotoxic lymphocytes phenocopies the defect in cytotoxicity previously seen with Rab27a deficiency, supporting the hypothesis that MADD is the crucial GEF involved in Rab27a activation in lymphocytes. Furthermore, we found that MADD is required for the activation of Rab27a, as loss of MADD significantly decreased the proportion of GTP-bound Rab27a in both unstimulated and stimulated NK cells. We also found that MADD colocalizes with Rab27a and polarizes to the cytotoxic synapse with Rab27a. Similar to previous observations, MADD appeared to align along the microtubule network, although treatment with nocadazole only partially displaced MADD from perforin.

Individuals with GS-II display severely reduced levels of cytotoxicity in NK and CD8⁺ T cells (Klein et al., 1994; Menasche et al., 2000). Our data is in line with this previous observation, as loss of MADD reduced cytotoxicity to the same level as loss of Rab27a; however, this was only apparent when using CD8⁺ T cell rADCC assays or NK cells with pancreatic ductal adenocarcinoma cell lines. Surprisingly, MADD-deficient NK cells displayed reduced but higher than expected cytotoxicity against 721.221 target cells and in CD16 and NKG2D+2B4 rADCC assays compared to crNC NK cells. In one individual with GS-II with a novel homozygous nonsense Rab27a Q118X mutation, it was found that NK cells display diminished cytotoxicity in NKp30 (also known as NCR3) rADCC assays and against 721.221 (Gazit et al., 2007). Surprisingly, this individual also had intact cytotoxicity in CD16 rADCC assays despite the lack of Rab27a (Gazit et al., 2007). However, high levels of residual cytotoxicity and degranulation in the absence of Rab27a remains controversial (Wood et al., 2009) and might have been enhanced by ex vivo expansion of the NK cells prior to Rab27a knockout. It would be of interest to test whether high doses of IL-2 or ex vivo expansion of NK cells isolated from individuals with GS-II can also enhance Rab27a-independent degranulation.

We also demonstrated that MADD-deficient NK cells had severely reduced active GTP-bound Rab27a at rest and upon stimulation. Although low levels of GTP-bound Rab27a were seen in MADD bulk knockout NK cells, this could be attributed to the small proportion of MADD-sufficient cells remaining in the bulk culture. Indeed, the decrease in GTP-bound active Rab27a was comparable to that seen in HUVECs, suggesting that MADD is the GEF responsible for Rab27a activation in NK cells (Kat et al., 2021). Further supporting this, MADD-deficient NK cells displayed reduced cytotoxicity to the same degree as Rab27a-deficient NK cells, suggesting that no alternative Rab27a GEFs are present in NK cells that could overcome the loss of MADD.

In HEK293T cells, it has been demonstrated that overexpression of full-length MADD and DENN domain-deficient MADD are able to weakly co-precipitate with GTP-bound Rab27a upon GST–Slac2b pulldown (Kat et al., 2021). Although the authors interpreted this to mean that the DENN domain was not essential to bind GTP-bound Rab27a, it could also imply that MADD remains associated with other Rab27a effectors independently of its ability to bind Rab27a. Furthermore, the ability of MADD to activate Rab27a requires the full-length protein and not solely the DENN domain, suggesting that MADD might associate with other proteins to bind Rab27a (Sanza et al., 2019). Interestingly, the terminal transport of LGs to the plasma membrane in T cells has been shown to be dependent on a kinesin-1, SLP3, and Rab27a complex (Kurowska et al., 2012). Owing to the ability of MADD to bind to the stalk domain of kinesins (KIF1β and KIF1A) through its death domain in neurons (Hummel and Hoogenraad, 2021; Niwa et al., 2008), it is reasonable to hypothesize that MADD might be part of a similar complex in lymphocytes. Interestingly, we saw no defects in MTOC polarization to the synapse or perforin accumulation at the membrane in the absence of MADD, suggesting that Rab27a-LG movement to the cytolytic synapse is MADD independent. Additionally, although MADD can associate with microtubules through kinesins, our data suggest that the presence of MADD on LGs only partially requires intact microtubules. We would therefore speculate that MADD requires the association of other proteins, such as kinesin-1, to activate Rab27a but does not require microtubules for recruitment to LGs.

Our data is consistent with the previous observations that MADD is required for Rab27a-mediated degranulation in other cell types. We have demonstrated that MADD and Rab27a colocalize with each other on LGs in the resting state and during cell-mediated killing when the LGs polarize to the synapse. This is in line with the observation that individuals with GS-II have intact LG polarization but impaired LG docking and degranulation (Feldmann et al., 2003). Upon MADD knockout in NK cells, no changes in total Rab27a protein or localization to LGs were seen in our study. Although this differs from the decrease in total Rab27a seen in MADD-deficient HUVECs, it is less clear in MADD-deficient melanocytes (Figueiredo et al., 2008; Kat et al., 2021; Sanza et al., 2019). This could be due to the reciprocal dependence of Rab27a and Munc13-4 for recruitment to LGs in resting NK cells (Wood et al., 2009). Our use of ex vivo expanded NK cells in this study might have also influenced the recruitment of Rab27a and MADD to LGs, as their interaction appeared to be constitutive even in unstimulated cells spread on fibronectin.

During the preparation of this manuscript, a short report on two individuals with MADD deficiency was published (Schutze et al., 2023). This study examined the cytotoxic ability of NK cells isolated from these individuals; however, whether this was dependent on the GEF activity of MADD was not addressed. Interestingly, hypopigmentation was not seen in these individuals despite the presence of immunodeficiency. It has previously been demonstrated that specific mutations in Rab27a can affect binding to Munc13-4 but not melanophilin, thus causing GS-II without hypopigmentation (Cetica et al., 2015; Ohishi et al., 2020). However, given that MADD has been shown to be the Rab27a GEF in melanocytes and loss of MADD is able to induce perinuclear clustering of melanosomes, it remains unclear why loss of active Rab27a would not have the same hypopigmentation defect (Figueiredo et al., 2008). It is therefore possible that another putative Rab27a GEF, such as DENND10, or an intrinsic GDP/GTP exchange by Rab27a, might also be required to activate Rab27a in melanocytes (Zhang et al., 2019). However, given that knockout of MADD phenocopied the cytotoxicity defect seen with Rab27a deficiency, MADD appears to be the only Rab27a GEF required for cytotoxicity.

A remaining question not addressed in this study is what factors induce MADD to activate Rab27a in resting and stimulated NK cells. In the absence of stimulation, Rab27a has been shown to regulate LG movement within the cytoplasm, driving membrane-directed trafficking of secretory vesicles (Liu et al., 2010). Our data suggests that MADD associates with and activates Rab27a in unstimulated NK cells. In non-secretory cells, such as polarized epithelial cells, Rab27a constitutively shuttles proteins and lipids to the apical or basolateral membrane to maintain polarity (Fukuda, 2013; Galvez-Santisteban et al., 2012; Yasuda et al., 2012). Similarly, Rab27a is constitutively active and maintained in the GTP-bound form in unstimulated platelets (Kondo et al., 2006). Therefore, MADD might constitutively activate Rab27a in unstimulated NK cells, and maintain a higher amount of GTP-bound Rab27a after expansion. This large pool of constitutively active GTP-bound Rab27a might therefore prime LGs for release immediately after their arrival at the cytolytic synapse, not requiring further activation by MADD. Upon stimulation with anti-NKG2D and anti-2B4 antibodies, we saw a slight, but insignificant, increase in GTP-bound active Rab27a in some donors suggesting that signaling pathways downstream of activating receptors in NK cells can directly increase the activity of MADD. It is possible that this interaction is regulated by post-translational modification of MADD; however, further research is required to assess the requirements for this interaction.

In conclusion, we have demonstrated that loss of MADD from NK cells causes a defect in degranulation due to deficient Rab27a activation. Given that early stages in LG trafficking were unaffected by loss of MADD or Rab27a, MADD is likely required to activate Rab27a for the terminal fusion events of LGs at the cytolytic synapse. Activation of MADD, therefore, is a further regulatory element required for degranulation. Future studies aimed at identifying the impact of mutations of MADD on cytotoxic lymphocyte killing and Rab27a activation, defining MADD-interacting proteins and uncovering how MADD is activated downstream of NK cell activating receptors will be of significant interest to the field.

Primary human CD8+ T cells and NK cells were isolated from anonymized healthy donor apheresis cones obtained from the Mayo Clinic Blood Bank (Rochester, MN, USA) as previously described (Phatarpekar et al., 2020; Wilton et al., 2019). Briefly, human peripheral blood mononuclear cells (PBMCs) were isolated by Ficoll-Hypaque (GE Healthcare, Chicago, IL, USA) density gradient centrifugation. The resulting PBMCs were then mixed with red blood cells and incubated with antibodies from either Rosette Separation CD8+ T cell or NK cell enrichment kits (StemCell Technologies, Vancouver, Canada), for 20 min at room temperature as per the manufacturer's instructions. Ficoll-Hypaque density gradient centrifugation was then repeated, and the resulting CD8+ T cells and NK cells were analyzed for purity by flow cytometry using anti-CD3, anti-CD8 and anti-CD56 antibodies (BioLegend, San Diego, CA) and with Ghost Dye™ (Tonbo, San Diego, CA).

CD8+ T cells were expanded for 12 days with ImmunoCult™ Human CD3/CD28T Cell Activator (StemCell Technologies) as per the manufacturer's instructions. Primary human NK cells were expanded using the modified mbIL-21 K562 line, CSTX002, which was kindly supplied by Dean Lee (Ohio State University, Columbus, OH, USA), and accomplished as described in Somanchi et al. (2011). Briefly, mbIL-21 K562 cells were irradiated at 100 Gy and mixed with NK cells at a E:T ratio of 1:2 for 7 days with fresh medium and IL-2 supplied on days 3, 5 and 7. CD8+ T cells, NK cells, 721.221 (a gift from Dr Peter Parham, Stanford University, USA) and p815 cells (ATCC, TIB-64) were expanded and maintained in RPMI 1640 supplemented with 50 U/ml recombinant human IL-2, 10% FBS, and 1% each of penicillin-streptomycin, sodium pyruvate, MEM nonessential amino acids and L-glutamine from Corning. HEK293T and Panc1, 03.27, L3.6, and 4535 cells were maintained in DMEM (Gibco, 12100-038), RPMI (Gibco, 31800-022), MEM (Gibco, 61100-053) and DMEM-F12 (Gibco, 12500-039), respectively, and supplemented with 5% FBS, 5% BCS, 1× of a 100× stock penicillin-streptomycin (Corning, 30-002-Cl) and 2 mM L-glutamine (Corning, 25-005-Cl). HEK293T, Panc1 and L3.6 cells were obtained from the ATCC. 03.27 and 4535 cells were generated from patient-derived xenografts (Ding et al., 2019). All cell lines were routinely tested for mycoplasma. All antibodies used are listed in Table S1.

Guide RNA complexes were prepared by heating equimolar amounts of crRNA and tracrRNA at 95°C for 5 min as previously described (Wen et al., 2022). The resulting complexes were allowed to cool to room temperature and incubated with NLS-Cas9 (QB3 Macrolab, Berkeley, CA, USA) for 10 min. For each target, three crRNAs were pooled so that 450 pmols of crRNA-tracrRNA complex were delivered with 180 pmol Cas9. Cells were nucleofected by Lonza 4D nucleofector program CM-137 in a nucleofection buffer consisting of 5 mM KCl, 15 mM MgCl₂, 15 mM HEPES and 105 mM Na₂HPO₄ and 50 mM mannitol, all supplied by Sigma (St Louis, MO, USA). Nucleofections were performed on day 5 of expansion and all knockouts were verified by immunoblotting. crRNAs were designed using Broad's CRISPick tool (Doench et al., 2016; Sanson et al., 2018). crRNA and tracrRNA were ordered from Integrated DNA Technologies (Coralville, IA, USA).

GFP–Rab27a bacteria stab cultures were purchased from Addgene (Watertown, WT 89237, USA) and isolated with Promega (Madison, WI, USA) Wizard mini prep columns. Rab27a mutagenesis was performed using a Quikchange site-directed mutagenesis kit (Agilent, Santa Clara, CA, USA). PEI transfection was performed to express the constructs in HEK293T cells. Similarly, SLAC2b-SHD fusion protein was cloned from NK cell cDNA and introduced into the pGEX plasmid backbone (lab stock originally from GE Healthcare; now available from the ATCC, no. 77103). GST fusion protein was purified as previously published (Ham et al., 2013). All crRNA guides and DNA oligonucleotides are included in Table S2.

Chromium release assays were performed as previously described (Phatarpekar et al., 2020). In brief, target cells were labeled with ⁵¹Cr (Perkin-Elmer, Waltham, MA) and co-cultured with dilutions of NK cells for 1 or 3 h where indicated. The resulting ⁵¹Cr release was measured using Luma-96 plates (PerkinElmer) on a Top Count NXT Microplate Scintillation and Luminescence Counter. CD8⁺ T cell and NK rADCC cell assays were performed using the P815 cell line and antibodies against CD3 and CD28 (BioLegend, San Diego, CA) and NKG2D (R&D Systems, Minneapolis, MN), 2B4 (C1.7) and CD16 (3G8) as previously described (Ham et al., 2013). Specific lysis was calculated by determining the spontaneous counts per minute (cpm) (in the cell culture medium) and maximum cpm (0.5% Triton X-100) release using the following formula: percentage specific lysis=(sample cpm−spontaneous cpm)/(maximal cpm−spontaneous cpm)×100.

Primary human NK cells and 721.221 cells were labeled with Cell Tracker Deep Red (Thermo Fisher Scientific, C34565) and CMAC (Invitrogen, C2110), respectively, for 30 min in serum free medium as per manufacturer's instructions. Cells were then washed and incubated on ice for 15 min in serum-free medium before resuspension at a 1:1 ratio. The cells were then centrifuged at 200 rpm for 5 min at 4°C without brake and incubated at in a 37°C water bath for the indicated amount of time. Conjugate formation was terminated by vortexing and fixation with 4% paraformaldehyde in cold PBS. Conjugate formation was assessed using a FACSCanto II flow cytometer (BD, Franklin Lakes, NJ) and analyzed by quantifying the percentage of target-bound NK cells (CMAC+ CellTracker Deep Red+) from the gated population of total number of NK cells (Cell Tracker Deep Red+) using Flowjo software (BD).

Primary NK cells were resuspended in 100 µl serum-free media with 20 μl FITC-conjugated antibody against CD107a (cat. no. 555800, BD) and mixed with Panc1 target cells at a 1:1 ratio. The cells were incubated at 37°C for 30 min, washed twice with PBS and fixed with 4% formaldehyde in PBS. Degranulation was assessed by flow cytometry and analyzed by quantifying the percentage of CD107a+ cells from the gated population of total number of NK cells using FlowJo software (BD).

In experiments involving NKG2D and 2B4 cross-linking, 15×10⁶ NK cells were incubated on ice with 1 µg of each antibody for 3 min, washed, then resuspended with 10 µl goat anti-mouse-IgG antibody (MP Biomedicals, Santa Ana, CA, USA) in 100 µl of RPMI. Crosslinked NK cells were then incubated for the indicated amount of time. Stimulation was terminated by addition of 1 ml ice-cold PBS and centrifugation (370 g for 5 min), prior to lysis with 50 mM Tris-HCl pH 7.5, 10 mM MgCl₂, 0.5 M NaCl and 2% Igepal, 1 mM sodium orthovanadate, 1 mM PMSF, 10 µg/ml leupeptin and 5 µg/ml aprotinin. The resulting lysate were then clarified by centrifugation at 14,000 rpm, and quantified by performing a Bradford assay, before rotating with GST-coated beads (Thermo Fisher Scientific, 16101) for 2 h. The beads were then washed, boiled at 95°C for 10 min, and separated by SDS-PAGE. CD8⁺ T cell and NK cell lysates were similarly prepared in a lysis buffer containing 10 mM Tris-HCl pH 7.4, 50 mM NaCl, 5 mM EDTA, 50 mM NAF, 30 mM Na₄P₂O₇ and 1% Triton X-100.

For immunoblotting, lysates were separated by SDS-PAGE prior to transfer to PVDF membranes. The resulting membranes were blocked with 4% BSA in TBS (10 mM Tris-HCl, 15 mM NaCl; pH 7.4) for 30 min, rocked overnight at 4°C with the indicated primary antibody (Table S1) in 2% BSA in TBST (TBS plus 0.05% Tween 20). The membranes were then washed three times in TBST, dectected with the secondary antibodies [Table S1, secondary antibodies were used at 1:10,000 and were obtained from Jackson ImmunoResearch; goat anti-mouse light chain specific (115-035-174), mouse anti-rabbit light chain specific (211-032-171), goat anti-mouse Fc specific (115-035-008), goat anti-rabbit Fc specific (111-035-046)] for 1 h, then washed four times with TBST prior to the addition of SuperSignal West Pico Plus Chemiluminescent substrate (Thermo Fisher Scientific, Waltham, MA, USA) and development. Blots are cropped for clarity. When relevant, the relative protein band density was quantified using the FIJI image processing package. Where indicated, the relative band density was normalized to that for α-tubulin and β-actin loading controls and displayed as a fold change relative to the crNC value. For GST pulldown, the active Rab27a level was normalized to the Rab27a level from the input sample and displayed as a fold change relative to the input GFP–Rab27a or that in unstimulated crNC. For ERK phosphorylation, pERK1/2 levels were quantified and normalized to total ERK1/2 levels relative to the 0-min crNC sample.

NK cells were plated onto poly-L-lysine (PLL; Millipore, P2636) or fibronectin-coated coverslips (Millipore, F4759) for 30 min at 37°C. For images with Panc1–NK cell conjugates, Panc1 cells were seeded onto PLL-coated coverslips in 12-well plates and allowed to adhere overnight. After 12 h, NK cells were added to each well and the plates were spun at 400 rpm for 5 min before 30 min of incubation at 37°C. For experiments involving wheat germ agglutinin staining, NK cells were incubated with 2 µg of WGA-488 (Millipore, W11261) for 90 min at 37°C prior to use. For experiments using nocodazole (Selleckchem, S2775), NK cells were treated with DMSO or 1 μg/ml for 2 h, washed in 1× PBS and plated onto coverslips for microscopy. Cells were fixed with molecular grade 4% paraformaldehyde (Electron Microscopy Sciences, Hatfield, PA) in PBS (18 min), permeabilized for 3 min with 0.15% Surfact-Amps (Thermo Fisher Scientific) in PBS and blocked with FBS-based blocking reagent. Coverslips were then stained with secondary antibodies conjugated to AF488, AF568 or AF647 (Invitrogen). F-actin was stained using phalloidin conjugated to AF488 or Rhodamine–phalloidin (Invitrogen). Coverslips were mounted onto glass slides using SlowFade (Invitrogen) and examined using a C-Acromat 63×/1.2W lens on an LSM-800 laser scanning confocal microscope with Airyscan (Carl Zeiss). Images were captured using the ZEN blue software package (Carl Zeiss) and analyzed using FIJI software and the JaCoP plugin (including to determine Pearson's and Manders' coefficients). Unless otherwise indicated, representative images are from single plane snaps, with brightness adjusted for clarity.

The distance of the MTOC to the synapse and the distance of the individual granules to the MTOC were quantified using the FIJI length function as previously described (Wilton et al., 2019). To quantify the average distance of the granules to the MTOC, the coordinates of the MTOC and each granule were collected, with distance from the MTOC calculated for each granule and averaged for each cell. Representative images were adjusted to increase clarity of the location of the MTOC and granules.

In situ interaction between MADD and Rab27a was assessed by PLA kit Duolink with PLA probe anti-rabbit minus and PLA probe anti-mouse plus (Sigma-Aldrich), following the manufacturer's protocol. In brief, cells were prepared and stained with primary antibodies as above, before incubation with the plus and minus probes, ligation and amplification. The resulting coverslips were then stained with Rhodamine–phalloidin prior to mounting. Z-stack images were captured using an LSM-800 laser scanning confocal microscope as indicated above. PLA spots within each cell were counted by generating maximum intensity projection of Z-stack images, which were then thresholded, and counted using the analyze particle tool in Fiji software.

Figures were created in BioRender.