Antigen-specific CD8+ T cell feedback activates NLRP3 inflammasome in antigen-presenting cells through perforin

The connection between innate and adaptive immunity is best exemplified by antigen presentation. Although antigen-presenting cells (APCs) are required for antigen receptor-mediated T-cell activation, how T-cells feedback to APCs to sustain an antigen-specific immune response is not completely clear. Here we show that CD8+ T-cell (also called cytotoxic T lymphocytes, CTL) feedback activates the NLRP3 inflammasome in APCs in an antigen-dependent manner to promote IL-1β maturation. Perforin from antigen-specific CTLs is required for NLRP3 inflammasome activation in APCs. Furthermore, such activation of NLRP3 inflammasome contributes to the induction of antigen-specific antitumour immunity and pathogenesis of graft-versus-host diseases. Our study reveals a positive feedback loop between antigen-specific CTLs and APC to amplify adaptive immunity.

N ACHT, LRR and PYD domains-containing protein 3 (NLRP3) is the most studied member of the Nod-like receptor (NLR) family. NLRP3 is activated primarily in innate immune cells such as dendritic cells and macrophages, and by a variety of stimuli, including pathogens and danger signals such as monosodium urate (MSU) and ATP [1][2][3][4][5] . Upon stimulation, NLRP3 recruits the adaptor Apoptosis-associated Speck-like protein containing a CARD (ASC) through PYD-PYD domain association, and ASC further recruits caspase-1 through CARD-CARD domain interaction, forming the signalling complex known as the inflammasome. Activated caspase-1 then cleaves pro-IL-1b to form mature IL-1b with pro-inflammatory functions 3,4 . In addition to caspase-1, bacterial infections also activate caspase-11 for the 'non-canonical' NLRP3 inflammasome pathway 6,7 . Dysregulation of NLRP3 inflammasome activation is associated with a variety of inflammatory disorders, such as cryopyrin-associated periodic syndromes and diabetes [8][9][10][11] . However, the functions of NLRP3 inflammasome in the pathogenesis of tumours and graft-versus-host disease (GVHD) are less defined [12][13][14] and it is unclear whether the NLRP3 inflammasome has a function in antigen-specific antitumour immunity.
Antigen-presenting cells (APCs) bridge innate and adaptive immunity. Antigens are processed and presented in APCs through MHC class II or MHC class I to activate naïve CD4 þ or CD8 þ T cells, respectively 15 . b2 microglobulin (b2M) is a subunit of MHC class I and has been shown to be required for antigen-specific CD8 þ T cells (also called cytotoxic T lymphocytes, CTLs) differentiation, activation and proliferation 16 . Antigen-activated CTLs have critical functions in host defense against tumours and pathogens, as well as in the pathogenesis of GVHD 17 . The cytolytic killing of target cells by CTLs requires perforin-mediated release of granzymes, mainly granzyme B, from cytotoxic granules [18][19][20] . Fas-FasL signalling also contributes to CTLmediated effects 21 .
Although innate immunity instructs adaptive immunity for antigen-specific immune responses, adaptive immunity has also been shown to suppress innate immunity to modulate abnormal inflammatory responses during viral infection in an antigen-independent manner 22 . T regulatory (Treg) cells are well-defined suppressors of both adaptive and innate effector cells and function via the secretion of suppressive cytokines or by cell-cell contact 23 . One study reported that anti-CD3-activated T cells dampen innate immune responses through suppressing the NLRP3 inflammasome in macrophages in an antigenindependent manner 24 . However, it is not completely clear how innate immunity-driven adaptive immunity feedback promotes innate immunity to amplify antigen-specific immune responses. Here, we show that CTLs activate the NLRP3 inflammasome in APCs which amplifies antigen-specific CTL-mediated effector functions.

Results
Inflammasome assembly induced by antigen-specific CTLs. ASC is a key adaptor of several inflammasomes such as NLRP3 and AIM2, and its activation is reflected by ASC speck assembly or oligomerization 4 . We utilized ASC speck assembly as a readout to search for potential new ASC inflammasome activators and found that OT1 CTLs induced ASC speck assembly in bone marrow-derived dendritic cells (BMDCs) pulsed with OVA peptide during co-culture, similar to MSU treatment (Fig. 1a,b). We also found that OT1 CTLs induced ASC oligomerization (Fig. 1c). Consistent with the ASC activation, we observed that OT1 CTLs activated Caspase-1 and consequently induced IL-1b maturation and secretion (Fig. 1c,d). However, protein levels of IL-6 and pro-IL-b were not induced (Supplementary Fig. 1a;   Fig. 1c), indicating that inflammasome activation is responsible for IL-1b secretion. Kinetic experiments showed that OT1 CTLs induced IL-1b secretion as early as 1 h after the incubation (Fig. 1e). OT1 CTLs also induced IL-1b secretion in OVA-pulsed bone marrow-derived macrophages (BMDM) or peritoneal macrophages (PMs) ( Fig. 1f; Supplementary Fig. 1b,c). CTLs are the primary killer cells in the mixed-lymphocyte reaction (MLR) assay in an antigen-specific manner 25 . Similar to OT1 CTLs, CTLs from the MLR induced ASC speck assembly in BMDCs (Fig. 1g,h). The CTLs also induced ASC oligomerization, caspase-1 activation and consequently IL-1b maturation and secretion (Fig. 1i,j;  Supplementary Fig. 1d). We then compared the ability of CTLs and the other remaining cells in the MLR to induce IL-1b production in BMDCs and found that CTLs were the predominant cells for IL-1b secretion and target cell killing ( Fig. 1k; Supplementary Fig. 1e). Alloantigen-specific CTLs can be determined by loss of CFSE staining in MLRs 26 . Therefore, we isolated CFSE high CTLs from MLR as non-responding cells and CFSE low CTLs from MLR as alloantigen-specific CTLs, and then co-cultured them with allogeneic DC cells and we found that alloantigen-specific CTLs (CFSE low CTLs) induced IL-1b secretion ( Fig. 1l; Supplementary Fig. 1f). Together, our data suggest that antigen-specific CTLs induce the ASC-mediated inflammasome assembly for IL-1b maturation and secretion in APCs.
These results indicate that MHC I-mediated antigen presentation by APCs activates CTLs, and the activated antigen-specific CTLs feedback activate NLRP3 inflammasome for IL-1b secretion in APCs.
Perforin from CTLs modulates IL-1b secretion by APC. The danger signal ATP that activates NLRP3 inflammasome is secreted from dead cells 14 . To determine whether any potential soluble factors secreted by cells are important for antigen-specific CTL-induced IL-1b secretion, we collected cell culture medium (supernatant) of the mixed OT1 CTLs and BMDCs and used the supernatant to treat BMDCs. We found that the supernatant did not further induce IL-1b secretion in BMDCs (Fig. 4a), indicating that cell-cell contact is likely required for OT1 CTL-induced IL-1b secretion. We further utilized transwell experiments to confirm that cell-cell contact is required for OT1 CTLs to induce IL-1b secretion from OVA-pulsed BMDCs (Fig. 4b) or BMDMs ( Supplementary Fig. 3a). Fas-FasL signalling is reported to induce non-canonical IL-1b maturation 27 . We found that OT1 or MLR CTL-induced IL-1b secretion was not significantly reduced in Fas-deficient (Fas lpr ) BMDCs ( Fig. 4c; Supplementary Fig. 3b), indicating that Fas signalling is not important for CTL-mediated IL-1b maturation. To determine whether cell apoptosis contributes to the observed IL-1b secretion, we compared the kinetics of cell apoptosis with that of IL-1b secretion and found that both target cell apoptosis ( Supplementary Fig. 3c) and IL-1b secretion ( Fig. 1e) occurred at 1 h while cell death occurred at 2 h ( Supplementary Fig. 3c). We then knocked down Casp8 and RIP3, the key signalling molecules in cell apoptosis and necroptosis, respectively, and found that the two molecules were not required for OT1 CTL-induced IL-1b secretion in BMDMs ( Supplementary Fig. 3d,e). Similarly, IL-1b secretion was not affected by a combination of apoptosis and necroptosis inhibitors ( Supplementary Fig. 3f,g) or by RIP3 deficiency plus an apoptosis inhibitor ( Supplementary Fig. 3h,i). These data suggest that target cell apoptosis and necroptosis are not required for antigen-specific CTL-induced IL-1b secretion in APCs. The cytolytic killing of target cells by CTLs requires perforinmediated release of granzymes from cytotoxic granules 28 . We found the expression of perforin and granzymes was induced in CTLs after antigen-mediated activation, and the expression of granzyme B was much higher than that of other granzymes ( Supplementary Fig. 4a). We next checked perforin and granzyme B levels in naive, effector memory, and central memory CTLs. We found effector memory CTLs comprise the majority of the population of CTLs (Fig. 4d) and had the highest expression of perforin and granzyme B in the MLR (Fig. 4e). Similarly, the effector memory population of OT1 CTLs had the highest expression of perforin and granzyme B, and their expression levels were highly induced in OT1 CTLs by OVA peptide stimulation ( Supplementary Fig. 4b). Notably, we observed that the effector memory CTLs had the strongest ability to induce IL-1b secretion in APCs of the populations of the MLR and OT1 CTLs ( . We then utilized a perforin inhibitor as reported previously 29,30 and found that the inhibitor suppressed IL-1b secretion ( Fig. 4g; Supplementary  Fig. 4d), suggesting that perforin is likely important for CTL-induced IL-1b production. To confirm this, we utilized Perforin-deficient mice and found that perforin deficiency in CTLs almost blocked IL-1b secretion and cell death in BMDCs ( Fig. 4h; Supplementary Fig. 4e). We then determined the potential contribution of granzyme B and found that deficiency of granzyme B in CTLs also reduced IL-1b secretion, while the granzyme B-mediated effect was much weaker than that of perforin (Fig. 4h). Furthermore, double deficiency of granzyme A and B in CTLs had an effect on IL-1b secretion similar to single granzyme B deficiency (Fig. 4h), indicating that granzyme A is not important for MLR CTL-induced IL-1b production. Consistent with the very weak expression of perforin and granzymes in BMDCs ( Supplementary Fig. 4a), deficiency of perforin, granzyme B or granzymes A and B in BMDCs did not affect the specific CTL-mediated induction of IL-1b and cell death ( Supplementary Fig. 4f). However, it is reported that Pasteurella multocida toxin induces granzyme A expression in macrophage and the exocytosed granzyme A enters target cells and mediates IL-1b maturation independently of caspase-1 and without inducing cytotoxicity 31 . The differential use of granzyme A in IL-1b maturation is probably due to the different experimental systems.    WT Nlrp3 Anti-ASC Anti-CD8α Anti-NLRP3 DAPI Figure 2 | The essential role of NLRP3 for CTL-induced IL-1b secretion. (a) IL-1b release from LPS-primed and ova-pulsed wild-type or Aim2 À / À BMDCs co-incubated with OT1-CTLs at the ratios indicated for 4 h. pAT (poly(dA:dT)) was used as a positive control. (b) The levels of IL-1b from LPS-primed and ova-pulsed wild-type or Nlrp3 À / À BMDCs co-incubated with OT1-CTLs at the ratios indicated. MSU and ATP were used as positive controls. (c) Confocal microscopy analysis of ASC specks assembled of LPS-primed and ova-pulsed wild-type (WT) or Nlrp3 À / À BMDCs incubated with activated OT1-CTLs (DC þ CTL). Far right, magnification of the area in red colour outlined at left showing the typical NLRP3 inflammasome structure. Scale bar, 5 mm.
There are two models for perforin functions. The first model for perforin and granzymes mediated killing is that granzymes enter target cells via plasma membrane pores formed by perforin 18,32 . The second suggests that perforin forms transient pores in the plasma membrane to induce the endocytosis of perforin and granzymes into the target cells and then forms pores in endosomes to trigger cytosolic release 18,33,34 . We next determined whether perforin can induce IL-1b secretion in APCs. Perforin needs to be dissolved in calcium containing ARTICLE perforin buffer when added to medium to treat target cells, and calcium is important for the function of perforin [34][35][36] . We found that IL-1b secretion was already highly induced in the supernatant of BMDCs that were treated with only perforin buffer (Fig. 4j), likely due to the high extracellular calcium concentration in the buffer because calcium is reported to be essential for NLRP3 activation 37 and the calcium inhibitor or NLRP3 deficiency blocked the buffer-induced IL-1b secretion (Fig. 4j). Under this condition, perforin still statistically increased IL-1b secretion although the increase was not dramatic (Fig. 4j), and the increase of IL-1b secretion by perforin was blocked by the calcium inhibitor or NLRP3 deficiency (Fig. 4j). We also utilized the profectP1 protein delivery system and found that perforin delivered to the BMDCs markedly induced IL-1b secretion and NLRP3 was required for perforin-induced IL-1b secretion under this delivery system (Fig. 4k). We then explored how perforin affects NLRP3 inflammasome activation for IL-1b maturation and secretion. As delivery of perforin to target cells by antigen-specific CTLs has been reported to induce calcium influx into the cytosol 38 , and calcium has been shown to be required for NLRP3 inflammasome activation 37 , we determined the effect of calcium on antigen-specific CTL-induced IL-1b production in APCs. The endoplasmic reticulum (ER) is the major source of intracellular calcium influx under stress conditions 37 . We checked expression levels of ER Ca 2 þ channels, including IP3Rs and RyRs, in BMDCs and found that IP3Rs were highly expressed ( Supplementary Fig. 4g). We further observed that either inhibition of calcium influx from ER by the inhibitor 2-APB for IP3Rs or depletion of intracellular calcium by BAPTA-AM blocked CTL-induced IL-1b secretion in BMDCs in an antigendependent manner ( Fig. 4l; Supplementary Fig. 4h). In addition to calcium influx, potassium efflux is also required for NLRP3 inflammasome activation by multiple triggers 39 . We found that extracellular potassium chloride also suppressed OT1 CTL-induced IL-1b secretion ( Fig. 4m; Supplementary Fig. 4i).
As perforin delivered to target cells forms membrane pores 18,34 , these data suggest that the pore-forming ability of perforin likely induces potassium efflux and calcium influx to activate NLRP3 inflammasome.
CTLs play an important role in killing tumour cells through tumour antigen-specific immunity 40 . To understand whether CTLs induce IL-1b production during antigen-specific antitumour immunity, we established an in vivo model in which mice were injected subcutaneously (s.c.) with the tumour cell line EL4 (a T-cell-derived tumour cell line) or OVA containing EL4 (EG7) for tumour formation, and then injected intravenously (i.v.) with activated OT1 CTLs to kill EG7 target cells (as depicted in Supplementary Fig. 5a). We found that OT1 CTLs quickly induced IL-1b secretion and IFN-g production in the EG7 tumours but not in EL4 tumours ( Fig. 5a; Supplementary  Fig. 5b). Furthermore, OT1 CTL-induced IL-1b and IFN-g production was blocked in b2M-deficient mice ( Fig. 5b; Supplementary Fig. 5c), indicating that antigen presentation is required for IL-1b secretion in vivo. To determine whether the NLRP3 inflammasome activated by tumour antigen-specific CTLs contributes to their mediated antitumour immunity, we utilized a similar strategy (as depicted in Supplementary Fig. 5d) to analyse the ability of OT1-mediated antigen-specific killing of target tumour cells. As expected, OT1 rejected EG7 tumours but not EL4 tumours ( Supplementary Fig. 5e,f), and IL-1b secretion was increased during OT1-mediated antigen-specific rejection of EG7 tumours (Supplementary Fig. 5g). Importantly, deficiency of NLRP3 or Caspase-1 significantly blocked the OT1 CTL-mediated rejection of EG7 tumours (Fig. 5c,d). Consistently, we found that IL-1b secretion was blocked in Nlrp3-, Caspase-1and b2M-deficient mice (Fig. 5e), indicating that antigen-specific activation of NLRP3 inflammasome is required for IL-1b maturation in the CTL-mediated tumour rejection model. IL-1b has been reported to affect CTLs priming and activation 41 and tumour antigen-activated CTLs are crucial for tumour clearance. We observed that IFN-g producing CTLs were not detected by deficiency of b2M and were partially blocked by deficiency of NLRP3 or Caspase-1 in tumour model (Fig. 5f). DC cells from WT, Nlrp3-, Caspase-1-deficient mice had a comparable MHC I level while b2M deficiency reduced MHC I expression ( Supplementary Fig. 5h), indicating that the reduction of IFN-g producing CTLs of NLRP3 or Caspase-1 deficiency is likely due to IL-1b decrease but not due to the differences in antigen load while the reduction of IFN-g producing CTLs of b2M deficiency is due to differences in antigen load. We previously showed that perforin was essential for CTL-induced IL-1b secretion in vitro (Fig. 4h). Therefore, Perforin-deficient OT1 CTLs were injected to estimate the contribution of perforin in the tumour model. We found deficiency of perforin significantly blocked the OT1 CTL-mediated rejection of EG7 tumours (Fig. 5g,h) and also reduced IL-1b secretion (Fig. 5i). We have showed above that effector memory CTLs were the majority population to induce IL-1b secretion in APCs in vitro ( Supplementary  Fig. 4b,c). Effector memory CTLs have been reported to migrate to peripheral tissues and perform immediate effector functions, while naïve and central memory CTLs usually home to secondary lymphoid organs and have little or no effector functions 42 . Therefore, we analysed CTL populations in the tumours and found that effector memory CTLs were the majority of the population and had the highest expression of perforin and granzyme B (Supplementary Fig. 5i,j), suggesting that effector memory CTLs are critical for IL-1b secretion in the tumour model. We then determined whether antigen-specific CTL-induced IL-1b secretion contributes to antitumour immunity by utilizing Il-1r1-deficient mice. Indeed, the receptor deficiency resulted in significant blockade of OT1 CTL-mediated antigen-specific antitumour immunity (Fig. 5j,k). Consistently, antigen-specific CTL activation was reduced in the receptor-deficient mice (Supplementary Fig. 5k). To identify the cellular source of IL-1b secretion, we first examined the expression levels of NLRP3 inflammasome components in APCs (BMDCs and BMDMs) and EG7 tumour cells and found that NLRP3 and pro-IL-1b were not detected in EG7 cells, even after LPS stimulation ( Supplementary Fig. 5l) CTL-induced IL-1b secretion in BMDCs but not in EG7 cells in vitro (Fig. 5l). We then utilized CD45.1 background mice to prepare the tumour model by injecting EG7 cells s.c. (CD45.2 background) and then i.v. injecting OT1 CTLs (CD45.2 background) to analyse IL-1b-producing cells in vivo. Consistent with the in vitro data, IL-1b-producing cells were CD45.2 À (Fig. 5m,n; Supplementary Fig. 5m), indicating that EG7 tumour cells and OT1 CTLs do not produce IL-1b. Further analyses showed that both macrophage and dendritic cells produce IL-1b (Fig. 5o,p; Supplementary Fig. 5m). These data indicate that OVA antigen in EG7 tumour cells is presented by APCs to activate OT1 CTLs, and the OT1 CTLs then activate NLRP3 inflammasome for IL-1b secretion in APCs via a feedback mechanism.
CTL-activated NLRP3 inflammasome in GVHD. We showed above that the inflammasome was activated in the MLRs (Fig. 1g-k). To determine whether NLRP3 is responsible for the inflammasome mediated IL-1b maturation in MLRs, we utilized Nlrp3-deficient mice and found that NLRP3 deficiency in BMDCs markedly blocked IL-1b secretion but did not affect cell death in the MLR of BALB/c anti-C57/BL6 or C3H anti-C57/BL6 (Fig. 6a,b; Supplementary Fig. 6a,b). As MLR CTLs play a critical role in the pathogenesis of GVHD 43 , we explored whether the NLRP3 inflammasome activation contributes to GVHD pathology. We established a CTL-mediated GVHD model by i.v. injection of C3H.SW background bone marrow (BM) with or without CTLs into irradiated C57BL/6 background recipient mice. Similar to the in vitro results in the MLR (Fig. 6b), IL-1b secretion was induced in the GVHD model (Fig. 6c), while deficiency of NLRP3, Caspase-1 or b2M reduced IL-1b induction (Fig. 6c), indicating that the MLR CTLs activate NLRP3 inflammasome in APCs for IL-1b maturation and secretion in vivo. Furthermore, the GVHD-associated body weight loss was significantly blocked in the Nlrp3-or Caspase-1-deficient mice (Fig. 6d). Deficiency of NLRP3 or Caspase-1 led to reduced GVHD mortality, similar to the effect of b2M deficiency (Fig. 6e). Consistent with the phenotypes of body weight change and mouse survival rate, histological examination also showed that deficiency of NLRP3, caspase-1 or b2M clearly suppressed lymphocyte infiltration and tissue damage in the liver and small intestine ( Supplementary Fig. 6c,d). We showed above that perforin was essential for CTL-induced IL-1b secretion in vitro (Fig. 4h). To determine whether the observation happens in vivo, we transferred BM with wild-type or Perforin-deficient CTLs to bm1 mice to induce GVHD. We found deficiency of perforin in CTLs resulted in reduced IL-1b induction (Fig. 6f). Perforin deficiency also led to reduction of GVHD-associated body weight loss and mortality ( Supplementary Fig. 6e,f), consistent with previous reports 44 . Similar to the observation in the tumour model ( Supplementary Fig. 5i,j), effector memory CTLs were the major population and had the highest expression of perforin and granzyme B in the target tissues of the GVHD model ( Supplementary Fig. 6g,h), indicating their roles in GVHD pathogenesis. We next determined the potential role of MLR-CTL-induced IL-1b secretion in APCs in GVHD. We found that antibody blockade of IL-1b significantly reduced the CTL-mediated GVHD pathogenesis, including body weight change, survival (Fig. 6g,h) and pathology ( Supplementary  Fig. 6i,j). Furthermore, IL-1R1 deficiency significantly blocked pathogenesis in the GVHD model (Fig. 6i,j; Supplementary  Fig. 6k,l). Consistently, CTL activation was reduced in the receptor-deficient mice in the GVHD model ( Supplementary  Fig. 6m). However, we found that IL-1b did not promote CTL activation and proliferation in vitro under the conditions with or without TCR activation (Supplementary Fig. 7). IL-1b has been reported to affect DC maturation and migration in a CCR7-dependent manner to prime CTLs 41 . We observed that the percentage of CCR7 þ DC was reduced in the GVHD model of the receptor-deficient mice ( Supplementary  Fig. 6n). These results indicate that IL-1b does not directly target CTLs but instead activates other cell types, such as DCs, to prime CTLs.
The increased IL-1b in b2M-deficient spleen (Fig. 7d) is probably due to increased viral titre that has been reported 45,46 . We also used the LM bacteria or LCMV virus to infect wild-type or Perforin-deficient mice. Similar to the observations in the b2M-deficient mice, we found that IL-1b levels were not reduced in the spleen of Perforin-deficient mice compared to wild-type mice in the infection models (Fig. 7e,f). We then checked into the potential effect of memory CTLs on IL-1b secretion in vivo by immunizing mice with OVA or gp33 peptides for one month and then challenging the mice with LM-OVA or LCMV infections respectively. There existed OVA antigenspecific CTLs in the OVA peptide immunized mice because the antigen-specific CTLs induced IL-1b secretion in BMDCs pulsed with OVA peptide in vitro ( Supplementary Fig. 8c). However, OVA pre-immunization did not affect IL-1b level after LM-OVA challenge in vivo (Fig. 7g). Similarly, while the antigen-specific CTLs from gp33 peptide immunized mice induced IL-1b secretion in BMDCs in vitro ( Supplementary  Fig. 8d), gp33 pre-immunization did not affect IL-1b level after LCMV infection in vivo (Fig. 7h). These data suggest that antigen-specific CTL-induced IL-1b in APCs is not important for host defense against infections of virus and intracellular bacteria.

Discussion
Here, we found that antigen-specific CTLs feedback promoted NLRP3 inflammasome activation in APCs for IL-1b maturation. By utilizing the MLR-CTL and OT1-CTL systems and b2M genetic-deficient mice, we demonstrated that CTL-mediated NLRP3 inflammasome activation required antigen presentation both in vitro and in vivo. As IL-1b is reported to play an important role in activating APCs to promote priming, proliferation and activation of CD4 þ T cells and CTLs 41,47 , our study suggests that NLRP3 inflammasome activation by CTLs forms a positive feedback amplification loop to amplify antigen-specific adaptive immune responses. While the amplification loop is critical for antitumour immunity and GVHD pathogenesis, it is not important for host defense against either primary or secondary infections of virus and intracellular bacteria. The reason is likely because pathogens activate pathogen recognition receptors to induce very strong inflammasome activation for IL-1b secretion in APCs which overrides the effect of antigen-specific CTLs on IL-1b secretion 48 . Therefore, the feedback amplification loop is specifically needed for antigen-specific antitumour immunity or GVHD pathogenesis where innate immunity is probably not directly activated by tumour antigens or allo-antigens. NLRP3 inflammasome can be activated by a variety of stimuli including pathogens and danger signals such as ATP and MSU 3,4,5,[49][50][51][52] . We demonstrated that perforin in CTLs activated NLRP3 inflammasome in APCs in an antigen-specific manner. In contrast to the reported danger signals that are secreted from cells, we found that the CTL-mediated NLRP3 inflammasome activation requires cell-cell contact, consistent with perforin-mediated cell killing in a cell-cell contactdependent manner 53 . We also provided evidence that apoptosis and necroptosis were not required for perforin-induced IL-1b secretion. NLRP3 has been reported to be activated by several common mechanisms, including calcium influx 37 and potassium efflux 39 . Perforin can form membrane pores in target cells and induce calcium influx 54 . We found that both calcium influx and potassium efflux were important for CTL-induced IL-1b secretion. Therefore, the perforin pore-forming ability likely induces calcium influx and potassium efflux to activate NLRP3 inflammasome.
Antigen-specific CTLs have important functions in host antitumour immunity and in the pathogenesis of GVHD. We found that the feedback activation of NLRP3 inflammasome in APCs by CTLs contributed to their killing effects. We established an antigen-specific antitumour mouse model through OT1-mediated killing of EG7 tumour cells containing OVA. The OT1-mediated killing of target tumour cells in vivo induced IL-1b production through antigen presentation and NLRP3 inflammasome, indicating that the OVA tumour antigen was presented by APCs to activate OT1 T cells, and the CTLs then activated NLRP3 inflammasome in APCs. Importantly, we found that CTL-mediated antitumour immunity required NLRP3 inflammasome. Although one recent study reported that mice with a genetic deficiency of NLRP3 inflammasome had increased colon cancer 12 , the mechanisms of the enhanced cancer pathology are not known 55 . To our knowledge, our study provides the first evidence that NLRP3 inflammasome is required for antigen-specific immunity against tumours. Similarly, we showed that IL-1b was required for antigenspecific antitumour immunity, consistent with the reported antitumour function of IL-1b in several studies 14,56 . However, many studies reported the function of IL-1b in promoting tumourigenesis likely through its pro-inflammatory effects as chronic inflammation promotes cancer 57 . These data suggest that IL-1b has dual functions in tumour development, that is, antitumour immunity through tumour antigen-specific CTL-mediated effects and pro-tumour effects in a chronic inflammatory environment in an antigen-independent manner. Thus, targeting IL-1b for cancer therapy should consider its dual effects by differential strategies such as blocking the chronic inflammation-driven pro-tumour effect and strengthening antigen-specific antitumour immunity in a tumour-antigen vaccine, CAR-T therapy and immune checkpoint (CTLA-4 and PD-1) blockade. Similar to tumour antigen-specific CTL-mediated antitumour immunity, we found that MLR antigen-specific CTLs activated NLRP3 inflammasome for IL-1b production in APCs, and the CTL-mediated GVHD induced IL-1b production through NLRP3 inflammasome. While NLRP3 and IL-1b has been reported to be required for T-cell-(CD4 þ and CD8 þ ) mediated GVHD through pathological Th17 induction 13,58 , our study showed that NLRP3 inflammasome is also required for CTL-mediated GVHD through promoting CTL functions. We have data to show that deficiency of perforin in CTLs resulted in significantly decrease in IL-1b level in GVHD target tissues as well as GVHD pathology, indicating that perforin is an important NLRP3 activator in GVHD model in vivo. Similarly in the antigen-specific antitumour model, we showed that deficiency of perforin in OT1 CTLs resulted in significantly reduced IL-1b level and consequently increased tumour size and weight. However, perforin can also mediate cell killing and thus may release factors like ATP to activate inflammasome secondarily or indirectly in vivo 59 . Together, our data indicate that antigen presentation by APCs activates CTLs, and the antigen-specific CTL feedback then activates NLRP3 inflammasome in APCs to form a positive feedback loop to amplify the CTL-mediated effects in antitumour immunity and GVHD pathogenesis.
Innate immunity instructs antigen-specific adaptive immunity. Here, we discovered an amplification loop of adaptive immunity to innate immunity through antigen-specific delivery of perforin to APCs to boost immune responses via NLRP3 inflammasome activation. Consistent with the perforin requirement for amplifying effective immunity, we observed that perforin expression was highly induced in effector memory CTLs (T EM ) compared to naïve and central memory CTLs, and the T EM cells are the majority of the population of CTLs in models of both antitumour immunity and GVHD. Coincident with the perforin level, antigen-specific T EM cells induced much higher IL-1b levels in APCs than the other CTL populations. Our study indicates that innate immunity needs to be promoted during the initial or effector stage to enhance the antigen-specific antitumour immunity and GVHD responses. However, innate immunity might be deleterious during the resolution stage and may need to be suppressed for central memory formation. One potential mechanism is the suppression of adaptive immunity on innate immunity 22 . Treg cells have been reported to suppress both innate and adaptive immune responses either through cell-cell contact or through secreting suppressive cytokines, such as IL-10 and TGFb, in an antigen-independent manner 23 . One study showed that anti-CD3-activated CD4 þ and CTLs suppressed NLRP3 inflammasome activation in macrophages in vitro in an antigen-independent manner 24 . We also confirmed these results that anti-CD3-activated CTLs inhibited NLRP3 inflammasome. In contrast, we unexpectedly found that antigenspecific CTLs promoted NLRP3 inflammasome in APCs. The reason for the differential antigen-dependent versus antigenindependent effects of CTLs is that antigen presentation is required for the cell-cell contact-dependent delivery of perforin from CTLs to APCs. Consistently, we found that perforin was required only for antigen-specific CTL-mediated promoting effects but was not important for antigen-independent CTLmediated suppressive effect.
In summary, we demonstrate that antigen-specific CTLs activate NLRP3 inflammasome in APCs, and this is specifically required for the CTL-mediated effects in antitumour immunity and GVHD. We discover that perforin from antigen-specific CTLs is essential for the CTL-mediated innate NLRP3 inflammasome activation. Our study reveals a positive feedback loop of adaptive immunity to promote innate immunity to amplify antigen-specific adaptive immune responses in situations where innate immunity is not strongly activated.

Methods
lysates were centrifuged at 6,000 r.p.m. for 10 min at 4°C. After the pellets were washed with 1 ml PBS and chemically cross-linked with 2 mM of DSS (Pierce) for 30 min at room temperature with rotation, the cross-linked pellets were centrifuged at 6,000 r.p.m. for 15 min and dissolved directly in SDS sample buffer before being subjected to immunoblotting. :100) and anti-CD11c (N418, 1:100) were from eBioscience. Anti-CD103 (2E7, 1:100) was from Biolegend. Annexin V apoptosis Detection kit (556547) was from BD Pharmingen. For intracellular cytokine staining, Cytofix/Cytoperm plus kit (BD, cat. 55508) was used. For Foxp3 staining, Foxp3 staining buffer set (eBioscience, cat. 00-5523-00) was used. Multiple-colour flow cytometric analysis was performed using FACSAria (BD Biosciences; Franklin Lakes, NJ, USA). For FACS sorting, cells stained with FACS antibodies were sorted on BD FACSAria device. FlowJo software was used for data acquiring and analysis.
CTL assay. CTL killing assay was measured using a non-radioactive method based on release of lactate dehydrogenase (LDH) from target cells. BMDCs were primed with LPS for 3 h, and pulsed with the ova peptides (50 nM) for 1 h at 37°C. After three times washing, 5 Â 10 4 BMDCs in 100 ml were added to 100 ml of various numbers of effector cells that had been plated in 96-round well plates to obtain target: effector cell (CTL cells) ratios of 1:2.5, 1:5 or 1:10. After 4 h of incubation, the supernatant was collected to measure killing efficiency and cytokine analysis. The medium or the target cells alone were used as the low-level control (spontaneous LDH release). For the high-level control (maximum LDH release), lysate buffer was added to the target cells. The mixed cells were incubated for 4 h and assayed for LDH release using a cytotoxicity detection kit (LDH). The percentage of cell-mediated cytotoxicity was determined by the following equation: cytotoxicity (%) ¼ [[(effector target cell mix À effector cell control) À low-level control]/(high-level control À low-level control)] Â 100.
Immunofluorescence. Cells were cultured on glass in complete medium. For staining ASC speck, LPS-primed and ova-pulsed BMDCs on coverslips were washed twice with PBS, incubated with CTLs that were pre-labelled with Alexa Fluor 647-CD8a (eBioscience) and fixed for 15 min at room temperature with 4% formaldehyde in PBS, and then were washed three times with PBS. After permeabilization with 0.1% Triton X-100 in PBS and blocking with 5% BSA in PBS for 1 h, cells were incubated with anti-NLRP3 (Cryo-2; AdipoGen) and anti-ASC (sc-22514-R; Santa Cruz Biotechnology, Inc.) primary antibodies (in 5% BSA) for 1 h at room temperature. Cells were washed three times with PBS and then were incubated for 1 h at room temperature with the appropriate fluorescenceconjugated secondary antibody (Invitrogen). The samples were washed three times in PBS, and then incubated with DAPI for 5 min. Samples were observed with a Zeiss LSM510.
Mouse peritoneal neutrophil recruitment model. Mice were pre-injected with 10 mg ova peptide for 1.5 h, and then neutrophil influx was induced by injection i.p. of 2 Â 10 7 CTL cells in 0.3 ml sterile PBS. After 6 h, mice were killed and peritoneal cavities were washed with 10 ml of PBS. The lavage fluids were analysed for neutrophils recruitment by FACS using the neutrophil marker CD11b-FITC (M1/70, 1:100) and Ly6G-PE (RB6-8C5, 1:100).
Tumour graft model. Mice received s.c. injections of 1 Â 10 6 tumour cells in 100 ml of PBS in the right flank. Tumour size was measured at the indicated time points with a vernier caliper, and tumour size was estimated with the following formula 62,63 : width Â width Â length Â 0.5. Raw data and tumour size calculation result are included as Supplementary Data 1. The mice were euthanized by CO 2 asphyxiation when they had tumour size over the maximum size 4,000 mm 3 or whenever they were observed in severe sick. Five to seven days after inoculation of the tumour cells, the mice were adoptively transferred i.v. with 1.5 Â 10 7 OT1 CTLs that were isolated and purified by MACS anti-CD8a microbeads. Tumours were then isolated from the treated mice at the indicated times.
GVHD induction model. Allogeneic lethally irradiated mice underwent allogeneic BM transplantation as previously described 64 . Briefly, recipient mice were irradiated with 9.5 Gy administered in three fractions of 137Cs source. C3H.SW CD8 þ T cells (2 Â 10 6 ) plus C3H.SW TCD BM (5 Â 10 6 ) or C3H.SW TCD BM (5 Â 10 6 ) alone were i.v. given into the irradiated recipient mice. The bm1 mice possess a mutant class I allele that differs from B6 mice. Bm1 mice were irradiated with 5 Gy total body irradiation (TBI) and injected with 2 Â 10 6 CD8 þ T cells together with 5 Â 10 6 TCD BM cells from donors. T cells were isolated from the donor spleen with the MACS cell separation system (Miltenyi Biotec, Germany).
Day of BM or BM plus CD8 þ T-cell injection was assigned as day 0. Recipient mice were monitored for survival and body weight two to three times a week. The levels of IL-1b in tissue homogenates were analysed by ELISA on day 7. Slides of small intestine and liver samples collected on day 35 were stained with H&E. The tissue was examined using a semiquantitative scoring system known to be associated with GVHD 65 . Seven parameters were scored for small intestine (villous blunting, crypt regeneration, crypt epithelial cell apoptosis, crypt loss, luminal sloughing of cellular debris, lamina propria inflammatory cell infiltrate and mucosal ulceration) and ten parameters for liver (portal tract expansion by an inflammatory cell infiltrate, lymphocytic infiltrate of bile ducts, bile duct epithelial cell apoptosis, bile duct epithelial cell sloughing, vascular endothelialitis, parenchymal apoptosis, parenchymal microabscesses, parenchymal mitotic figures, hepatocellular cholestasis and hepatocellular steatosis). The scoring system for each parameter was as follows: 0 indicates normal; 0.5, focal and rare; 1, focal and mild; 2, diffuse and mild; 3, diffuse and moderate; and 4, diffuse and severe.
L. monocytogenes and LCMV infection. L. monocytogenes strain 10403s (LM) or was prepared by shaking the bacteria overnight at 37°C in brain heart infusion broth with 50 mg ml À 1 streptomycin. For infections, LM were grown to mid-logarithmic phase, then resuspended in PBS and quantified by visible spectrometry readings at 600 nm. LCMV Armstrong strain was provided by Dr Lilin Ye. LM (2 Â 10 4 by i.v.) or LCMV (5 Â 10 5 by i.p.) infected mice for 5-7 day, then the levels of IL-1b in tissue homogenates were analysed by ELISA.