SCIENCE IMMUNOLOGY | RESEARCH ARTICLE

CANCER IMMUNOLOGY Copyright © 2021 The Authors, some rights reserved; KIR3DL3-HHLA2 is a human immunosuppressive exclusive licensee American Association pathway and a therapeutic target for the Advancement Yao Wei1†, Xiaoxin Ren1†, Phillip M. Galbo Jr.1,2, Scott Moerdler1,3, Hao Wang1, R. Alejandro Sica1,4, of Science. No claim 5 1 1 1 1 1 to original U.S. Bijan Etemad-Gilbertson , Lei Shi , Liqiang Zhu , Xudong Tang , Qi Lin , Mou Peng , Government Works Fangxia Guan1, Deyou Zheng2,6, Jordan M. Chinai1, Xingxing Zang1,4,7*

The B7 family HERV-H LTR–associating 2 (HHLA2) is an attractive target for cancer immunotherapy because of its coinhibitory function, overexpression in human cancers, and association with poor prognoses. Downloaded from However, the knowledge of the HHLA2 pathway is incomplete. HHLA2 has an established positive receptor trans- membrane and immunoglobulin (Ig) domain containing 2 (TMIGD2) but a poorly characterized negative receptor human killer cell Ig-like receptor, three Ig domains, and long cytoplasmic tail (KIR3DL3). Here, KIR3DL3 and TMIGD2 dim simultaneously bound to different sites of HHLA2. KIR3DL3 was mainly expressed on CD56 NK and terminally + + + + differentiated effector memory CD8 T (CD8 TEMRA) cells. KIR3DL3 CD8 TEMRA acquired an NK-like phenotype + and function. HHLA2 engagement recruited KIR3DL3 to the immunological synapse and coinhibited CD8 T and http://immunology.sciencemag.org/ + NK cell function and killing, inducing immune-evasive HHLA2 tumors. KIR3DL3 recruited SHP-1 and SHP-2 to + attenuate Vav1, ERK1/2, AKT, and NF-B signaling. HHLA2 tumors from human kidney, lung, gallbladder, and + stomach were infiltrated by KIR3DL3 immune cells. KIR3DL3 blockade inhibited tumor growth in multiple hu- manized mouse models. Thus, our findings elucidated the molecular and cellular basis for the inhibitory function of KIR3DL3, demonstrating that the KIR3DL3-HHLA2 pathway is a potential immunotherapeutic target for cancer.

INTRODUCTION with cancer (9, 13–18). However, a few studies also show that higher Cancer immunotherapy has fundamentally shifted the paradigm HHLA2 expression is associated with better survival (19–21). for cancer treatment from targeting the tumor itself to harnessing This paradox could be partly explained by the dual role of HHLA2

the host’s (1). Immune checkpoint blockades have in immune responses. HHLA2 can inhibit (4, 22) or enhance (6) at Albert Einstein College of Medicine on July 9, 2021 achieved durable responses in patients with advanced-stage tumors immune cell function. Transmembrane and immunoglobulin (Ig) that would have otherwise been fatal (2). Several checkpoint inhibi- domain containing 2 (TMIGD2; also called IGPR-1/CD28H) is the tors targeting the cytotoxic T lymphocyte–associated antigen 4 or first characterized receptor for HHLA2 (6, 7) and is primarily ex- programmed cell death receptor 1 (PD-1)/PD ligand 1 (PD-L1) axis pressed on naïve T (TN) and natural killer (NK) cells, rapidly de- have been approved for a broad spectrum of cancers (3). Although creasing after activation (6, 23). TMIGD2 costimulates both T and the percentage of responders to these drugs has steadily increased, NK cell activities (6, 23), which does not explain the inhibitory most patients with cancer still do not benefit from them (3). There- effect of HHLA2. Thus, we reasoned that there was an unidentified fore, it is critical to characterize additional immune checkpoints inhibitory receptor of HHLA2 (5, 7, 24). In 2017, we identified that and develop new therapeutic strategies. killer cell Ig-like receptor, three Ig domains, and long cytoplasmic We previously identified human endogenous retrovirus H long tail (KIR3DL3), an orphan molecule in the killer cell Ig-like recep- terminal repeat–associating protein 2 (HHLA2; also called B7H7/ tor (KIR) family (25, 26), is another receptor for HHLA2 (27). Very B7-H5/B7y) as a functional B7 family member that phylogenetically recently, two groups perform high-throughput interactome screen forms a subfamily with B7x and B7-H3 in the B7 family (4, 5). of Ig superfamily (IgSF) and indicate KIR3DL3 as a binding partner HHLA2 protein is expressed on antigen-presenting cells (APCs) of HHLA2 (28, 29). During the revision of this paper, another group (4, 6) and has a limited expression in normal human organs, whereas also independently report KIR3DL3 as the second receptor for HHLA2 it is highly expressed in many human cancers (7–10). In several and demonstrate that the engagement of KIR3DL3 with HHLA2 tumor types, HHLA2 expression is more frequent than PD-L1, and inhibits the activation of a line and the cytotoxicity of an NK cell coexpression of PD-L1 and HHLA2 is rare; PD-L1–negative tumors line (12). However, numerous key characteristics of the KIR3DL3-­ tend to express HHLA2 (9, 11, 12). In general, elevated HHLA2 is HHLA2 pathway remain unknown, including the full expression pattern associated with more severe pathology and worse prognosis in patients of KIR3DL3, how HHLA2 regulates immune responses through KIR3DL3, the downstream signaling pathway of KIR3DL3, the coexpres- sion of HHLA2 and KIR3DL3 in the tumor microenvironment, 1Department of Microbiology and Immunology, Albert Einstein College of Medicine, and the therapeutic efficacy of KIR3DL3 blockade in human cancers. Bronx, NY 10461, USA. 2Department of Genetics, Albert Einstein College of Here, we demonstrated that KIR3DL3 was mainly expressed Medicine, Bronx, NY 10461, USA. 3Department of Pediatrics, Children’s Hospital, + 4 on terminally differentiated effector memory CD8 T cells and Montefiore Medical Center, Bronx, NY 10461, USA. Department of Medicine, dim + Montefiore Medical Center, Bronx, NY 10461, USA. 5NextPoint Therapeutics CD56 CD16 NK cells. KIR3DL3 mediated coinhibition of primary 6 + + Inc., Cambridge, MA 02142, USA. Departments of Neurology and Neuroscience, CD8 T and NK cells and induced resistance of HHLA2 tumors to 7 Albert Einstein College of Medicine, Bronx, NY 10461, USA. Department of Urology, killing. KIR3DL3 showed immunosuppressive activities by recruit- Montefiore Medical Center, Bronx, NY 10461, USA. *Corresponding author. Email: [email protected] ing Src homology region 2 domain–containing –1 †These authors contributed equally to this work. (SHP-1) and SHP-2 to attenuate downstream Vav guanine nucleotide

Wei et al., Sci. Immunol. 6, eabf9792 (2021) 9 July 2021 1 of 15 SCIENCE IMMUNOLOGY | RESEARCH ARTICLE exchange factor 1 (Vav1), extracellular signal–regulated (ERK1/2), protein kinase B (PKB, also known as AKT), and nuclear factor B (NF-B) signaling in NK cells. Human HHLA2+ tumors were infiltrated by KIR3DL3+ immune cells, and KIR3DL3 blockade promoted antitumor immunity in multiple humanized mouse models. Together, our study provided a detailed functional and mechanistic view of the KIR3DL3-HHLA2 pathway–mediated immunosuppression, supporting this pathway’s therapeutic application for cancer immunotherapy.

RESULTS

KIR3DL3 and TMIGD2 simultaneously bound to nonidentical Downloaded from sites on HHLA2 In addition to the expression of HHLA2 on APCs and tumors (4, 6, 7, 9, 10), we found that its expression could also be induced on activated T and NK cells (fig. S1A). Because HHLA2 is a member of IgSF and has orthologs in humans and monkeys but not in mice or rats, we screened the receptors for HHLA2 through a list of http://immunology.sciencemag.org/ belonging to IgSF, exclusive to primates, and with a predicted trans- membrane domain (fig. S1B). We observed that HHLA2-Ig strongly bound to cells expressing KIR3DL3, but not its closest homologs KIR3DL1 or KIR3DL2 (Fig. 1A and fig. S1C). Inversely, KIR3DL3-Ig bound to an HHLA2+ human lung cancer cell HCC827, but not an HHLA2− cell A427 (Fig. 1B). The binding of KIR3DL3-Ig to HHLA2-expressing cells was completely blocked by anti-HHLA2 monoclonal antibody (mAb) clone B5B5 (Fig. 1C). To further ex- plore whether KIR3DL3 has other binding partners, we used a high-­ throughput screening with cell microarray individually expressing

5484 full-length human membrane or cell surface–tethered at Albert Einstein College of Medicine on July 9, 2021 human secreted proteins and found that HHLA2 was the only spe- cific binding partner for KIR3DL3 (fig. S1, D and E). HHLA2 extracellular region is composed of tandem IgV1-IgC- IgV2 domains (4). Loss of the IgC-IgV2 domains of HHLA2 largely abrogated its binding to KIR3DL3 but did not interfere with its binding to TMIGD2 (Fig. 1D). The extracellular part of KIR3DL3 is composed Fig. 1. Characterization of KIR3DL3 as a receptor for HHLA2. (A) Histograms of tandem D0-D1-D2 domains (fig. S1F). The removal of D1-D2 showing the binding of HHLA2-Ig (open) or hIgG (shaded) to the indicated cell domains from KIR3DL3 did not affect its binding to HHLA2 (Fig. 1E). lines. (B) Top: Histograms showing the HHLA2 expression on the human lung cancer Preincubating HHLA2/3T3 cells with KIR3DL3 did not block TMIGD2 cell line A427 and HCC827 examined by anti-HHLA2 mAb clone B5B5 (open) or mIgG1 binding and vice versa (Fig. 1F). Together, our results demonstrated that (shaded). Bottom: Histograms showing the binding of KIR3DL3-Ig (open) or hIgG KIR3DL3 and TMIGD2 bound to nonidentical sites on HHLA2, with (shaded) to A427 and HCC827. (C) Histograms showing the binding of KIR3DL3-Ig both KIR3DL3 and TMIGD2 simultaneously binding to HHLA2. (open) or hIgG (shaded) to HHLA2/3T3 in the presence of anti-HHLA2 mAbs (clone B5B5 or 566.1) or mIgG1. (D) Left: Cartoons showing the structures of HHLA2-Ig Generation and characterization of KIR3DL3 D0 (top) and HHLA2-IgV1-Ig (bottom). Right: Histograms showing the binding of domain–specific mAbs HHLA2-Ig (open, top) and HHLA2-IgV1-Ig (open, bottom) or hIgG (shaded) to KIR3DL3/3T3 and TMIGD2/3T3 cells. (E) Left: Cartoons showing the structures of To investigate the expression of KIR3DL3, we first generated a panel KIR3DL3-Ig (top) and KIR3DL3-D0-Ig (bottom). Right: Histograms showing the binding of of anti-KIR3DL3 mAbs (fig. S2A). Our lead clone 26E10 bound to KIR3DL3-Ig (open; top) and KIR3DL3-D0-Ig (open; bottom) or hIgG (shaded) to HHLA2/3T3 the KIR3DL3 D0 domain with a high affinity (Kd = 0.17 nM) (fig. S2, cells. (F) Left: The median fluorescence intensity (MFI) of TMIGD2-mIgG binding to B and C) and did not cross-react with other KIRs (fig. S2A). The HHLA2/3T3 cells. HHLA2/3T3 cells were preincubated with KIR3DL3-Ig or hIgG at the specificity of 26E10 was further verified by lack of reactivity with indicated concentrations and then stained by TMIGD2-mIg. Right: The MFI of KIR3DL3- the cell microarray expressing 5484 human proteins (fig. S1D). Ig binding to HHLA2/3T3 cells. HHLA2/3T3 cells were preincubated with TMIGD2- KIR3DL3 is a highly polymorphic , encoding more than mIgG or mIgG at the indicated concentrations and then stained by KIR3DL3-Ig. In 100 alleles (30). The D0 domain, which our lead clone 26E10 binds (A) to (F), data are representative of two to three independent experiments. to, is relatively conserved and contains about 16 amino acid variants (30). We mutated the D0 domain of KIR3DL3*004 and transduced KIR3DL3 protein was expressed on human innate 3T3 to express wild-type (WT) and mutant KIR3DL3*004. We then and adaptive immune cells tested the binding of 26E10 with 11 KIR3DL3 D0 variants and found The mRNA of KIR3DL3 was undetectable in most normal hu- that 9 of them, including KIR3DL3*001 and KIR3DL3*010, which man tissues but high in the blood and spleen (fig. S4A). Thus, we are the most predominant D0 sequence in Europeans and Chinese used 26E10 to examine KIR3DL3 protein expression on different Han, respectively (30), could be detected by 26E10 (fig. S3). immune cells from peripheral blood and found that NK cells, some

Wei et al., Sci. Immunol. 6, eabf9792 (2021) 9 July 2021 2 of 15 SCIENCE IMMUNOLOGY | RESEARCH ARTICLE

 T, CD8+, and CD4+ T cells expressed KIR3DL3 (Fig. 2A and but down-regulated T cell costimulatory [CD28 and inducible T cell co- fig. S4B). stimulator (ICOS)] and coinhibitory (PD-1) receptors in comparison The expression of TMIGD2 and KIR3DL3 was mutually exclu- with their KIR3DL3− counterparts (Fig. 3, G and H). To exclude the sive of CD8+ and CD4+ T cells (Fig. 2B). The major KIR3DL3 potential contamination of NK cells in these CD8+ T cells, we per- + expressing CD8 T cells were terminally differentiated (TEMRA), fol- formed a redirected cytotoxicity assay by using anti-CD16 mAb. + + lowed by an effector memory CD8 T cell (TEM) subset, whereas TN Although anti-CD3 steadily activated these CD8 T cells, we did + and central memory (TCM) CD8 T cells barely expressed KIR3DL3 not observe any activation induced by anti-CD16 (fig. S5J). + + (Fig. 2B and fig. S4C). NK cells were the main immune cells ex- Therefore, our results indicated that KIR3DL3 CD8 TEMRA cells pressing KIR3DL3 (Fig. 2A). KIR3DL3 and TMIGD2 could be co- acquired NK-like phenotypes. expressed or singly expressed on an NK cell (Fig. 2C). KIR3DL3 was KIR3DL3+ CD8+ T cells were able to lyse K562 cells without TCR mainly expressed on a CD56dimCD16+ NK subset, which is more engagement in an effector-to-target (E:T) ratio-dependent manner differentiated and predominant in peripheral blood (31), compared (Fig. 3I). HHLA2 expressing K562 cells were more resistant to Downloaded from with a CD56birght CD16− NK subset, which had higher TMIGD2 KIR3DL3+ CD8+ T cell killing as compared with control K562 cells expression (Fig. 2C and fig. S4D). KIR3DL3+ NK cells expressed acti- (Fig. 3I and fig. S6C). Gene deletion of HHLA2 sensitized HCC827 vating receptors, including CD16, NKG2D, NKp30, NKp46, and to KIR3DL3+ CD8+ T cell lysis (Fig. 3J and fig. S6D). Anti-KIR3DL3 2B4 (Fig. 2D). Most of the NK cells expressed at least one KIR protein, mAb clone 26E10 was capable of completely blocking the binding and KIR3DL3 could be coexpressed with all the other KIRs (Fig. 2D). of KIR3DL3 to HHLA2 (fig. S6E). As expected, 26E10 enhanced the lysis of HCC827 by KIR3DL3+ CD8+ T cells, as well as the degranu- http://immunology.sciencemag.org/ KIR3DL3 hampered both TCR-dependent and lation and cytokine (TNF- and IFN-) production of these T cells + TCR-independent function of human CD8 T cells (Fig. 3K). Together, these results indicated that KIR3DL3+ CD8+ We then asked whether HHLA2 and KIR3DL3 could be recruited to TEMRA cells could be activated by both TCR-dependent and TCR-­ T cell immunological synapses. After coculturing Jurkat cells ex- independent signaling with KIR3DL3 inhibiting T cell function. pressing KIR3DL3 and Raji cells expressing the HHLA2–yellow + fluorescent protein (YFP; or control-YFP) fusion protein in the KIR3DL3 inhibited NK cell function and mediated HHLA2 presence of superantigen staphylococcal enterotoxin E (SEE) (32), tumor resistance against NK cells KIR3DL3 and HHLA2 colocalized in about 70% of the cell conju- We next sought to determine the effect of KIR3DL3 on human NK gates, whereas very little KIR3DL3 clusters were formed in the cell function. We sorted out primary KIR3DL3+ NK cells and per- absence of HHLA2 (Fig. 3A and fig. S5, A to C). We next sought to formed an NK cell–based redirected cytotoxicity assay (fig. S7A). + investigate the function of KIR3DL3 on T cells. Fluorescence-­ CD16-induced lysis of P815 by primary KIR3DL3 NK cells was at Albert Einstein College of Medicine on July 9, 2021 activated cell sorter (FACS)–purified KIR3DL3+ CD8+ T cells were significantly inhibited by the coengagement of KIR3DL3 but was expanded, and the expression of KIR3DL3 was stable during the not affected by coengagement of CD56 (Fig. 4A). Consistently, in vitro culture (fig. S5D). Because the specific antigens of these KIR3DL3 markedly decreased the degranulation and cytokine and CD8+ T cells are unknown, we performed a redirected cytotoxicity chemokine production of NK cells (Fig. 4, B and C). Next, we ex- assay (fig. S5E) (33). After CD3 engagement, the KIR3DL3+ CD8+ plored the effect of the KIR3DL3-HHLA2 engagement on NK-­ T cells displayed cytolytic activity against Fc receptor (FcR)+ target mediated tumor cell lysis. The human NK cell line NK92, which was cells P815, K562, and Raji (Fig. 3B and fig. S5, F to J). In comparison TMIGD2−, was sorted into KIR3DL3+ and KIR3DL3− populations with CD3 alone, coengagement of CD28 and CD3 did not have an and used as effector cells (fig. S7B). We observed that HHLA2-­ additional effect (Fig. 3, B and C), but coengagement of KIR3DL3 expressing tumor cells were more resistant to KIR3DL3+ NK92 and CD3 significantly suppressed the cytolytic function, the de- cell–mediated killing compared with control tumor cells, but this granulation, cytokines [interferon- (IFN-), tumor necrosis factor– effect was lost when KIR3DL3 was absent (fig. S7, C to E). (TNF-), interleukin-13 (IL-13), granulocyte-macrophage colony-­ Similar to T cell immunological synapses, both KIR3DL3 and stimulating factor (GM-CSF), and IL-5], and C-C motif chemokine HHLA2 clustered at the interfaces between the KIR3DL3+ NK92 ligand 1 (CCL1) secretion of CD8+ T cells (Fig. 3, C and D, and fig. S5, cell and HHLA2-expressing tumor cells (fig. S7F). The NK immu- F to I). These results suggested that KIR3DL3 hampered the T cell nological synapses can be lytic or inhibitory (34, 35); thus, we asked receptor (TCR)–­dependent function of CD8+ T cells. whether the KIR3DL3-HHLA2 interaction could mediate inhibitory + + The majority (60 to 97%) of KIR3DL3 CD8 T cells were TEMRA synapse formation. In the absence of HHLA2, we observed F-actin, (Fig. 2B). We performed RNA sequencing (RNA-seq) of sorted but not KIR3DL3, polarization at the interfaces (Fig. 4D), indicat- + − + KIR3DL3 and KIR3DL3 CD8 TEMRA cells (Fig. 3, E and F, and ing that lytic synapses were formed to facilitate tumor killing. By fig. S6, A and B). Gene set enrichment analysis (GSEA) identified contrast, in the presence of both HHLA2 and KIR3DL3, we ob- “–mediated cytotoxicity” as the most significant served HHLA2 and KIR3DL3 colocalization, but no F-actin accu- pathway in the KIR3DL3+ subset and “T cell receptor signal path- mulation, at the interfaces (Fig. 4D), indicating that the inhibitory ways” as the most significant pathway in KIR3DL3− subset (Fig. 3F synapses were formed to prevent cytoskeletal reorganization. and tables S1 and S2). To verify the above results at the protein level, Primary NK cells coexpressed KIR3DL3 and TMIGD2 (Fig. 2C). we performed a high-dimensional flow cytometry analysis and ob- KIR3DL3+ TMIGD2+ NK cells were sorted and expanded (Fig. 4E). served a trend of gradual acquisition of NK cell receptors and loss of The expression of KIR3DL3 was stable, whereas the expression of T cell co-receptors from TN, TCM, TEM, to TEMRA (Fig. 3, G and H). TMIGD2 decreased after primary NK cell activation (Fig. 4E). + + Consistent with our RNA-seq finding, KIR3DL3 CD8 TEMRA HCC827 tumor cells expressed multiple activating ligands and were up-regulated NK-activating (NKG2C, NKp30, NKp46, and CD16) and susceptible to primary NK killing (fig. S7G). Loss of HHLA2 sensi- inhibitory (KIR3DL1, KIR2DL2/3, NKG2A, and KLRG1) receptors, tized HCC827 to primary NK cell killing (Fig. 4F), indicating that

Wei et al., Sci. Immunol. 6, eabf9792 (2021) 9 July 2021 3 of 15 SCIENCE IMMUNOLOGY | RESEARCH ARTICLE Downloaded from http://immunology.sciencemag.org/ at Albert Einstein College of Medicine on July 9, 2021

Fig. 2. KIR3DL3 was expressed on innate and adaptive immune cells. (A) Flow cytometric analysis of KIR3DL3 expression on PBMCs. Left: KIR3DL3 expression on indi- cated subsets from one donor. Right: The frequencies of KIR3DL3+ cells in the indicated subsets (n = 27 for NK, CD4+ T, and CD8+ T cells; n = 7 for  T cells). Data are rep- resented as means ± SD. (B) Left: The expression pattern of KIR3DL3 and TMIGD2 on CD4+ and CD8+ T cells. Middle: The distribution of KIR3DL3+ CD8+ T cells based on the coordinate expression of CD45RA and CCR7. Right: The summary of the distribution of KIR3DL3+ CD8+ T cells (n = 16). Data are represented as means ± SD. (C) Left: The coexpression pattern of KIR3DL3 and TMIGD2 on NK cells. Right: The frequencies of KIR3DL3+ and TMIGD2+ cells in the indicated NK subsets (KIR3DL3, n = 8; TMIGD2, n = 6). P values by a two-tailed paired t test. (D) Human PBMCs were analyzed by spectral flow cytometry. The UMAP (Uniform Manifold Approximation and Projection) was generated on the basis of flow cytometric data gated on NK cells (n = 3). In (B) (left) and (C) (left), data are representative of three independent experiments with different donors.

Wei et al., Sci. Immunol. 6, eabf9792 (2021) 9 July 2021 4 of 15 SCIENCE IMMUNOLOGY | RESEARCH ARTICLE Downloaded from http://immunology.sciencemag.org/ at Albert Einstein College of Medicine on July 9, 2021

+ Fig. 3. KIR3DL3 hampered both TCR-dependent and TCR-independent function of CD8 T cells. (A) KIR3DL3 and HHLA2 colocalized on T cell immune synapses. Left: Representative confocal images of cell conjugates acquired after Jurkat-Raji contact. Scale bars, 10 μm. Right: The bar graph show the frequencies of KIR3DL3 clusters and HHLA2-YFP or control-YFP clusters in all cell conjugates (HHLA2-YFP/Raji, n = 76; control-YFP/Raji, n = 43). Mouse Tim-3-YFP/Raji was used as the control. (B to D) KIR3DL3+ CD8+ T redirected cytotoxicity assay against P815. (B) The lysis of P815 cells (n = 5). (C) The degranulation (CD107a) and cytokine production (IFN- and TNF-) of KIR3DL3+ CD8+ T cells (n = 5). CD28 and KIR3DL1 served as negative and positive control, respectively. (D) Cytokine production in the coculture supernatant of KIR3DL3+ CD8+ T cells with indicated antibody-coated P815. The differentially secreted cytokines are shown in the heatmap (n = 5). Data are represented as means ± SD. (E and F) RNA-seq of KIR3DL3+ and − + + KIR3DL3 CD8 TEMRA (n = 3). (E) The volcano plot of RNA-seq showing down-regulated (blue) and up-regulated (red) differentially expressed genes. (F) GSEA in KIR3DL3 and − + KIR3DL3 subset. Data shown are GSEA plots and heatmaps for the leading-edge genes. (G and H) The T (G) and NK (H) expression on CD8 TN, TCM, TEM, − + KIR3DL3 , and KIR3DL3 TEMRA subsets as accessed by spectral flow cytometry (n = 12). Data are represented as means ± SD. (I) Lysis of HHLA2/K562 or control K562 by KIR3DL3+ CD8+ T cells at indicated E:T ratios. Data are means for duplicate measurements. Results are representative of three independent experiments. (J) Lysis of scrambled control or HHLA2KO HCC827 by KIR3DL3+ CD8+ T cells at E:T = 10:1 (n = 4). (K) The cytotoxicity, degranulation (CD107a), and cytokine production (IFN- and TNF-) of KIR3DL3+ CD8+ T cell against HCC827 in the presence of anti-KIR3DL3 (26E10) or mIgG1 at E:T = 10:1 (lysis) or 2:1 (CD107a, IFN-, and TNF-) (n = 4). P values by one-way ANOVA (B, C, G, and H) and two-tailed paired t test (D, J, and K). *P < 0.05, **P < 0.01, ***P < 0.001, **** P < 0.0001; FDR, false discovery rate; NES, normalized enrichment score.

Wei et al., Sci. Immunol. 6, eabf9792 (2021) 9 July 2021 5 of 15 SCIENCE IMMUNOLOGY | RESEARCH ARTICLE Downloaded from http://immunology.sciencemag.org/ at Albert Einstein College of Medicine on July 9, 2021

+ Fig. 4. KIR3DL3 inhibited primary NK function and mediates HHLA2 tumor immune resistance. (A to C) Primary KIR3DL3+ NK cells redirected cytotoxicity assay against P815. (A) The lysis of P815 cells (n = 4). (B) The degranulation (CD107a) and cytokine production (IFN- and TNF-) of primary KIR3DL3+ NK cells (n = 4). CD56 and KIR3DL1 served as negative and positive control, respectively. (C) Cytokine production in the coculture supernatant of primary KIR3DL3+ NK cells with indicated antibody-­coated P815 (n = 4 to 6). Data are represented as means ± SD. (D) HHLA2-KIR3DL3–mediated inhibitory synapses formation. Left: Cartoons depicting lytic and inhibitory synapses. Middle: Representative confocal images of cell conjugates acquired after KIR3DL3+ NK92 cell–A427 contact followed by staining with anti-KIR3DL3 and phalloidin. Scale bars, 10 μm. Right: The intensity quantification of F-actin, YFP, and KIR3DL3 at the immunological synapses (IS) and the cell surface away from synapses (non-IS) from NK92 cell–control/A427 (n = 20) and NK92-HHLA2/A427 (n = 21) conjugates. Mouse Tim-3-YFP/A427 was used as the control. (E) Primary KIR3DL3+ TMIGD2+ NK cells were sorted and cultured in vitro. Ten days later, KIR3DL3 and TMIGD2 expres- sion was examined by flow cytometry. (F) Lysis of scrambled control or HHLA2KO HCC827 by primary KIR3DL3+ TMIGD2+ NK cells in (E) at indicated E:T ratios. Data are means for duplicate measurements. Data are representative of three independent experiments with three different donors. P values by one-way ANOVA (A and B) or two-tailed paired t test (C and D). ****P < 0.0001; NS, not significant; TRAIL, TNF-related apoptosis-inducing ligand; CXCL1, C-X-C motif chemokine ligand 1; MDC, macrophage-derived chemokine; MIP1-, macrophage inflammatory protein 1 alpha.

Wei et al., Sci. Immunol. 6, eabf9792 (2021) 9 July 2021 6 of 15 SCIENCE IMMUNOLOGY | RESEARCH ARTICLE the inhibitory KIR3DL3-HHLA2 pathway was predominant even was barely expressed in renal cell carcinoma (Fig. 6, A and B). These in the presence of the activating receptor TMIGD2. Together, these results suggested that HHLA2+ tumors were infiltrated by KIR3DL3+ results demonstrated that KIR3DL3 exerted an inhibitory effect on immune cells, and the KIR3DL3-HHLA2 interaction may represent NK cells and mediated HHLA2+ tumor resistance to NK cell killing. an immune checkpoint pathway in these patients with cancer.

KIR3DL3-induced inhibitory signaling in NK cells KIR3DL3 blockade promoted NK cell–based antitumor The intracellular tail of KIR3DL3 contained one immunoreceptor immunity in vitro and in vivo tyrosine-based inhibitory motif (ITIM). We mutated the tyrosine We next explored whether the KIR3DL3-HHLA2 blockade could 381 in this ITIM to phenylalanine (Y381F) and introduced WT enhance NK-mediated lysis of tumors. Treatment with either anti-­ KIR3DL3 or a Y381F mutant into the KIR3DL3− NK92 cell line KIR3DL3 (clone 26E10 or 34B10) or anti-HHLA2 (clone B5B5) (Fig. 5A). After treatment with the tyrosine phosphatase inhibitor blocking mAbs restored the cytotoxicity of the KIR3DL3+ NK92 + pervanadate (36), WT KIR3DL3, but not Y381F mutant, displayed cell line against HHLA2 K562 cells (Fig. 7A). To test this concept Downloaded from tyrosine phosphorylation (Fig. 5B and fig. S8A). Y381F mutation in vivo, we subcutaneously engrafted nonobese diabetic (NOD). did not affect the recruitment of KIR3DL3 by HHLA2 to immuno- Cg-PrkdcscidIl2rgtm1Wjl/SzJ (NSG) mice with HHLA2+ Raji or con- logic synapses (Fig. 5C) but abolished its inhibitory effect on the trol HHLA2− Raji cells and then treated them with KIR3DL3+ NK92 NK92 cell line (Fig. 5D). cell line, together with 26E10 or mIgG1 (Fig. 7B). With the treat- ITIMs recruit the SH2 domain–containing tyrosine phosphatase, ment of the KIR3DL3+ NK92 cell line alone, the HHLA2+ tumors such as SHP-1 or SHP-2, inhibiting signaling induced by activating grew more aggressively than the control tumors (Fig. 7C). Com- http://immunology.sciencemag.org/ receptors (37, 38). Coimmunoprecipitation showed that KIR3DL3 pared with mIgG1, 26E10 significantly suppressed tumor growth in recruited both SHP-1 and SHP-2 in primary NK cells (Fig. 5E and HHLA2+ tumor–bearing mice (Fig. 7C). fig. S8B). We thus investigated KIR3DL3/SHP-1/2–induced down- We next tested the effect of KIR3DL3 blockade on primary NK stream pathways. WT KIR3DL3 or Y381F mutant that transduced cells. 26E10 mAb enhanced the lysis of HCC827 by primary NK92 cells were cocultured with HHLA2+ Raji cells, to initiate sig- KIR3DL3+ NK cells regardless of the presence of TMIGD2 or naling, and then subjected to analysis with human phosphokinase not, but it showed no effect on primary KIR3DL3− NK cells arrays. Compared with WT KIR3DL3, the Y381F mutant displayed (Fig. 7, D and E). Also, 26E10 significantly improved primary a higher phosphorylation level of multiple , including ERK1/2 KIR3DL3+ NK cells’ degranulation as well as IFN- and TNF- pro- and AKT (fig. S8, C and D). Further immunoblot analysis showed duction (Fig. 7F). We then tested the therapeutic efficacy of KIR3DL3 enhanced activation of Vav1, ERK1/2, AKT, and the downstream blockade with primary NK cells in vivo. NSG mice were subcutane- transcription factor NF-B upon Y381F mutation (Fig. 5, F and G, ously engrafted with HCC827 and then treated with expanded pri- at Albert Einstein College of Medicine on July 9, 2021 and fig. S8E). Consistently, upon KIR3DL3 signaling initiation, a mary KIR3DL3+ NK cells intratumorally, as well as 26E10 or mIgG1 substantially reduced activation of Vav1, ERK1/2, AKT, and NF-B (Fig. 7G). 26E10 significantly reduced tumor growth when compared was observed in CD16-stimulated KIR3DL3+ primary NK cells with mIgG1 (Fig. 7H). We next intraperitoneally engrafted the (Fig. 5, H and I, and fig. S8F). These results suggested that the ITIM luciferase-transduced HCC827 (HCC827-luc2) into NSG mice. The in KIR3DL3 was essential for KIR3DL3-mediated NK cell suppres- mice were then treated with primary KIR3DL3+ NK cells and 26E10 sion by recruiting SHP-1/2, which inhibited Vav1, ERK1/2, and or mIgG1 intraperitoneally (Fig. 7I). Recombinant human IL-2 AKT activation and downstream NF-B signaling. (rhIL-2) and rhIL-15 were injected to support primary NK cell survival in vivo. Tumor growth was significantly suppressed in + + KIR3DL3 immune cells infiltrated human HHLA2 tumors 26E10 treatment group compared with control group (Fig. 7, To further explore the KIR3DL3-HHLA2 pathway in human can- J and K). Using a more physiologically relevant human lung can- cers, we analyzed multiple databases through the cer model, mice were inoculated with HCC827-luc2 cells intrave- Omnibus. KIR3DL3 mRNA was up-regulated in several human solid nously and then were reconstituted with primary KIR3DL3+ NK and hematopoietic malignancies as compared with the normal tis- cells and treated with 26E10 or mIgG1, rhIL-2, and rhIL-15 (Fig. 7L). sues (fig. S9A). HHLA2 mRNA was also significantly up-regulated in Tumor growth was significantly inhibited in the mice treated with renal cell carcinoma, whereas TMIGD2 mRNA was down-regulated 26E10 compared with mIgG1 (Fig. 7, M and N). As a control, the in the tumor (fig. S9A), revealing potential immunosuppressive roles therapeutic efficacy of 26E10 was lost when the mice were trans- of the KIR3DL3-HHLA2 pathway in the tumor microenvironment. ferred with primary KIR3DL3− NK cells (fig. S10, A to C). Together, Consistent with the above mRNA data, we observed higher KIR3DL3 our results demonstrated that the KIR3DL3-HHLA2 blockade and lower TMIGD2 expression on both CD4+ and CD8+ T cells of restored the effector function of the NK cells and promoted anti- patients with adult T cell leukemia/lymphoma (ATLL) compared with tumor immunity. healthy donors (fig. S9, B to E). We then used quantitative multiplex immunofluorescence to determine KIR3DL3 and CD45 expression in four tumor tissue micro­ DISCUSSION array (TMA) cohorts, including renal cell carcinoma, lung cancer, HHLA2 plays dual roles in immune response (4, 6, 22, 23), making the gastric cancer, and gallbladder cancer. We observed KIR3DL3+ HHLA2 blockade approach for cancer immunotherapy a challenge. CD45+ cells infiltrated in all four tumors types (Fig. 6, A and B; fig. In this study, we reported that KIR3DL3 mediated HHLA2 coinhibi- S9F; and table S3). Immunohistochemistry (IHC) staining showed tion on primary CD8+ T and NK cells, which provided a cellular and that HHLA2 was strongly and widely expressed in these cancers molecular mechanistic basis for the inhibitory function of HHLA2. (Fig. 6, A and B). By contrast, high PD-L1 expression was limited to a We further showed that the KIR3DL3 blockade, which reinvigo- small portion of the lung, gastric, and gallbladder cancer, whereas it rated immune responses and enhanced antitumor immunity without

Wei et al., Sci. Immunol. 6, eabf9792 (2021) 9 July 2021 7 of 15 SCIENCE IMMUNOLOGY | RESEARCH ARTICLE

Fig. 5. KIR3DL3 ITIM–mediated NK cell suppression through recruiting SHP-1 and SHP-2 to at- tenuate downstream signaling. (A) Tyrosine (Y) in ITIM of KIR3DL3 was mutated to phenylalanine (F). The KIR3DL3− NK92 cell line was then transduced to express a simi- lar level of WT KIR3DL3 or Y381F mutant. (B) Tyrosine phosphoryl­ ation in WT or Y381F KIR3DL3/ NK92 cells cultured in the pres-

ence or absence of pervanadate Downloaded from (VO4). IP, immunoprecipitation; WCL, whole-cell lysates. (C) Images of cell conjugates acquired after indicated NK92 cell line and HHLA2/Raji contact followed by staining with anti-KIR3DL3 and DAPI. Scale bars, 10 μm. (D) Lysis http://immunology.sciencemag.org/ of Raji (top) or K562 (bottom) by WT or Y381F KIR3DL3/NK92 cell line at the indicated E:T ratios. Data are means for duplicate mea- surements. (E) SHP-1 and SHP-2 were coimmunoprecipitated with KIR3DL3 in primary KIR3DL3+ NK cells. (F and G) Expression and phosphorylation of Vav1, ERK1/2, AKT, and NF-B in WT or Y381F KIR3DL3/NK92 cells stimulated by

HHLA2/Raji for the indicated time at Albert Einstein College of Medicine on July 9, 2021 (F). Quantification of immunoblot- ting (G). Data are means ± SD from two independent experiments. (H and I) Expression and phos- phorylation of Vav1, ERK1/2, AKT, and NF-B in primary KIR3DL3+ NK cells after cross-linking with indicated mAbs at the indicated time (H). Quantification of immu- noblotting (I). Data are means ± SD from two independent experi- ments. In (A) to (F) and (H), data are representative of two or three independent experiments. P values by unpaired Student’s t test (G and I). *P < 0.05, **P < 0.01, and ***P < 0.001.

interrupting the costimulatory function of HHLA2, is a potential binding affinity with HLA-Bw4 ligands, and inhibition on NK cells therapeutic strategy for cancers. (39–41). Further studies will be needed to determine whether We demonstrated that our anti-KIR3DL3 clone 26E10 could KIR3DL3 polymorphisms affect its cell surface expression, its bind- recognize most of the KIR3DL3 allelic variants. Although KIR3DL3 ing to HHLA2, its inhibitory signaling capacities, and its association is a highly polymorphic gene, KIR3DL3 D0 domain is relatively con- to cancer risk. served and bound to HHLA2, indicating that targeting KIR3DL3 D0 The expression of TMIGD2 and KIR3DL3 on T cells was mutu- domain is a rational strategy for KIR3DL3-HHLA2 blockade. Differ- ally exclusive, suggesting that they function independently in the + ent KIR3DL1 alleles display different surface expression levels, different stages of T cell differentiation. CD8 TN cells constitutively

Wei et al., Sci. Immunol. 6, eabf9792 (2021) 9 July 2021 8 of 15 SCIENCE IMMUNOLOGY | RESEARCH ARTICLE Downloaded from http://immunology.sciencemag.org/ at Albert Einstein College of Medicine on July 9, 2021

+ + Fig. 6. Human HHLA2 tumors were infiltrated by KIR3DL3 immune cells. (A) Representative images from the indicated cancer type [taken from the patients in (B)] showed coexpression of KIR3DL3 and CD45 detected by immunofluorescence (left), expression of HHLA2, and PD-L1 detected by IHC (right). Scale bars, 50 μm. (B) Quan- titative analysis of KIR3DL3 [quantitative immunofluorescence (QIF) score; red; right y axis], HHLA2 (H score; blue; left y axis) and PD-L1 (H score; green; left y axis) expression on kidney (n = 78), gallbladder (n = 72), lung (n = 54), and gastric (n = 42) tumors from patients with cancer. AU, arbitrary units.

+ express TMIGD2, which transmits a costimulatory signal to T cells CD8 TN cells, suggesting that the acquisition of KIR3DL3 may re- (6). Repetitive stimulation down-regulates TMIGD2 expression; quire long-term differentiation of some antigen-experienced T cells. + + thus, most of the CD8 TEMRA cells lose their TMIGD2 expression. By CD8 TEMRA cells, which accumulate during aging, down-regulate + contrast, KIR3DL3 was mainly expressed on CD8 TEMRA, but not TCR signaling molecules and costimulatory molecules but up-regulate

Wei et al., Sci. Immunol. 6, eabf9792 (2021) 9 July 2021 9 of 15 SCIENCE IMMUNOLOGY | RESEARCH ARTICLE Downloaded from http://immunology.sciencemag.org/ at Albert Einstein College of Medicine on July 9, 2021

Fig. 7. KIR3DL3 blockade promoted NK cell antitumor immunity in vitro and in vivo. (A) The lysis of HHLA2/K562 by KIR3DL3+ NK92 cells in the presence of anti-­ KIR3DL3 mAbs (clones 26E10 and 34B10) or anti-HHLA2 mAbs (clones B5B5 and 566.1) at an E:T = 5:1. Data are means for duplicate measurements. (B and C) Subcutaneous Raji mouse model treated with NK92 cells. (B) Outline of the experiment procedure. (C) The tumor growth in the treatment of 26E10 or mIgG1 (n = 4). Each arrow indicates treatment with NK92 cells and the antibody. Mean tumor volumes ± SD are shown. (D) The indicated primary NK cells cytotoxicity against HCC827 at indicated E:T ratios in the presence of 26E10 or mIgG1. Data are means for duplicate measurements. (E and F) The cytotoxicity (E) (n = 15), degranulation (CD107a), and cytokine production (IFN- and TNF-) (F) (n = 4) of primary KIR3DL3+ NK cells against HCC827 in the presence of anti-KIR3DL3 (26E10) or mIgG1 at an E:T = 10:1. (G and H) Subcutaneous HCC827 mouse model treated with primary KIR3DL3+ NK cells. (G) Outline of the procedure. (H) The tumor growth in treatment of the 26E10 or mIgG1 (n = 4). Each arrow indicates treatment with primary KIR3DL3+ NK cells and the antibody. Mean tumor volumes ± SD are shown. (I and K) Intraperitoneal HCC827 tumor model treated with primary KIR3DL3+ NK cells. (I) Outline of the procedure. (J) Bioluminescence images by ventral imaging of each mouse at indicated days. (K) Tumor growth in the 26E10 or mIgG1 treatment group (n = 6). Each arrow indicates treatment with primary KIR3DL3+ NK cells and the antibody. Data are means ± SD of the total flux from ventral imaging for each mouse. (L to N) HCC827 lung metastasis mouse model treated with primary KIR3DL3+ NK cells. (L) Outline of the experiment procedure. (M) Bioluminescence images by dorsal and ventral imaging of each mouse at indicated days. (N) Tumor growth in the treatment of 26E10 or mIgG1 group (n = 4). Each arrow indicates a treatment with primary KIR3DL3+ NK cells and the antibody. Data are means ± SD of the average of total flux from dorsal and ventral imaging for each mouse. Data are representative of three (A and D) or two (C, H, K, and N) independent experiments. P values by two-way ANOVA (C, H, K, and N), two-tailed paired t test (E), and one-way ANOVA (F).

Wei et al., Sci. Immunol. 6, eabf9792 (2021) 9 July 2021 10 of 15 SCIENCE IMMUNOLOGY | RESEARCH ARTICLE

NK receptors (42). This switch enables them to broaden their capac- Study design ity for immune surveillance by different recognition systems. We This study was designed to elucidate the HHLA2-KIR3DL3 path- + + showed that the KIR3DL3 CD8 TEMRA cells expressed NK recep- way and the therapeutic potential. We used multiple approaches to tors and acquired NK-like function. These KIR3DL3+ CD8+ T cells demonstrate KIR3DL3 as a receptor for HHLA2 and used human were capable of immediately lysing NK-sensitive K562 targets and peripheral blood mononuclear cells (PBMCs) to study the expression less-sensitive solid tumor cells, indicating they may exhibit important pattern, the function, and the downstream signaling of KIR3DL3. immune surveillance function by sensing ligands on stressed and We then used imaging methods to inspect HHLA2 and KIR3DL3 in transformed cells. the immunological synapse and in the tumor microenvironments. The noncompetitive binding of KIR3DL3 and TMIGD2 to HHLA2 We further used humanized NSG mice to test the therapeutic effi- and the coexpression pattern of both receptors on NK cells suggested cacy of KIR3DL3 blockade. Group sizes were typically n = 4 or 5 or that both receptors can function simultaneously. NK cells are regu- as indicated in the figure legends, and experiments were repeated at lated by the balance of signaling from numerous activating and inhib- least twice. Investigators were not blinded to group identity during Downloaded from itory receptors (43). These receptors with opposing effects can share experimentation. the same ligand, such as polio virus receptor (PVR): DNAM1/TIGIT; human leukocyte antigen E (HLA-E): NKG2C/NKG2A (43). The Mice immune response is fine-tuned by these paired receptors. Among BALB/c and NSG mice at 6 to 8 weeks old were purchased from them, the inhibitory receptors usually dominantly bind to the ligand Charles River Laboratory and the Jackson Laboratory, respectively. by a greater affinity to compete with the activating receptor (43). All mice were kept in a specific pathogen–free animal facility. All http://immunology.sciencemag.org/ However, KIR3DL3 and TMIGD2 did not compete for the binding mouse protocols followed the National Institutes of Health (NIH) to HHLA2. Instead, during NK cell activation, TMIGD2 expression guidelines and were approved by the Animal Care and Use Commit- is down-regulated (23), whereas KIR3DL3 expression was stable and tee of Albert Einstein College of Medicine. thus became predominant. Through this spatial and temporal diver- gent expression, the NK immune response maintains . Human primary cells Although the biological function of these negative regulators is Human PBMCs were isolated from the buffy coats of healthy donors to limit immune responses, they can be hijacked by tumor cells to purchased from New York Blood Center, using Ficoll-Hypaque (GE evade immunologic destruction. Ectopic expression of HHLA2 on Healthcare) density gradient separation. Human peripheral blood tumor cells was sufficient to make them resistant to human NK and samples of patients with ATLL were obtained at the time of hospi- CD8+ T cell killing. Meanwhile, gene deletion of HHLA2 from en- talization or clinic visit in Montefiore Medical Center under an Insti- + dogenous HHLA2 tumors sensitized them to the killing of primary tutional Review Board–approved protocol and with written informed at Albert Einstein College of Medicine on July 9, 2021 NK and CD8+ T cells. Tumor cells expressing HLA-C could evade consent from each patient. PBMCs were separated via Ficoll-Hypaque KIR2DL1/2/3+-induced NK killing (44). HLA-E also contributes to Plus (GE Healthcare). protecting tumor cells from killing by NKG2A+ NK cells (45, 46). However, down-regulation or even complete loss of HLA class I Human tumor tissues molecules occurs in human tumors (47). Tumor cells with impaired Formalin-fixed paraffin-embedded (FFPE) commercial human can- expression of HLA I become resistant to CD8+ T cell killing (48), cer TMAs were purchased from US Biomax. There were four dif- whereas these tumors are highly sensitive to NK cell killing because ferent cohorts—including kidney, lung, gastric, and gallbladder of “missing-­self recognition” (49). Therefore, the selective pressure cancer—used in this study. may drive tumors to exploit the KIR3DL3-HHLA2 pathway to evade NK cell surveillance. Human protein cell microarray system The presence of KIR3DL3+ immune cells in HHLA2+ human tu- The cell microarray technology platform (Retrogenix; www.retrogenix. mors and the enhanced antitumor immunity by KIR3DL3 blockade com/the-technology/) was used to identify new binding partners for were other key findings of our study. We showed that KIR3DL3+ KIR3DL3 and to examine the specificity of mAb 26E10. immune cells infiltrated the tumor beds of kidney, lung, gastric, and gallbladder cancers. Its ligand HHLA2 was also widely and highly Human phosphokinase arrays expressed on these tumors. HHLA2 on cancer could suppress the The phosphorylation profiles of kinases downstream of KIR3DL3-­ function of KIR3DL3+ immune cells in the tumor microenvironment, HHLA2 pathway were determined by using a human phosphokinase leading to immune evasion and poor clinical outcomes in these array (R&D, catalog no. ARY003B) following the manufacturer’s patients. Multiple independent studies demonstrate that elevated instructions. HHLA2 in cancers is significantly associated with poor prognosis in various patients with cancer (9, 10, 13–18). Last, we showed that Production and purification of fusion proteins KIR3DL3 blockade effectively suppressed tumor growth in multiple KIR3DL3-Ig, KIR3DL3-D0-Ig, HHLA2-IgV1-Ig, and TMIGD2-mIgG humanized mouse models. Subsequent humanization of these anti-­ proteins were generated in an inducible secreted serum-free Drosophila KIR3DL3 mAbs and testing in clinical settings will provide further expression system as described previously (4). evidence for this new therapeutic application in patients with cancer. Cell binding and blocking assays In the fusion protein cell binding assay, HHLA2-IgG (R&D Systems), MATERIALS AND METHODS HHLA2-IgV1-Ig, KIR3DL3-Ig, KIR3DL3-D0-Ig, or hIgG (5 to 20 g/ml More details on these methods and the cell lines used can be found or indicated concentration) (R&D Systems) were incubated with cor- in the Supplementary Materials. responding NIH 3T3 cells on ice for 45 min, followed by incubation

Wei et al., Sci. Immunol. 6, eabf9792 (2021) 9 July 2021 11 of 15 SCIENCE IMMUNOLOGY | RESEARCH ARTICLE with allophycocyanin (APC)- or phycoerythrin (PE)–conjugated stained with Alexa Fluor Plus 405 Phalloidin (Life Technologies) anti-human IgG Fc antibody (1:100; BioLegend, clone HP6017) on for 1 hour at RT. The slides were then mounted by Gold Antifade ice for 30 min. Then, the cells were acquired on an LSR II flow Mountant without DAPI (Life Technologies). Images were acquired cytometer (BD Biosciences). using a Leica SP8 confocal microscope and processed by ImageJ.

Generation of mAbs against KIR3DL3 and HHLA2 Isolation and culture of human NK and CD8+ T cells Mouse anti-KIR3DL3 mAbs were generated by standard hybridoma KIR3DL3+ CD8+ T cells were purified by FACS and then cultured techniques (4). with allogeneic PBMCs as feeder cells (irradiated at 30 gray, feeder cells:T cells = 10:1) in OpTmizer (Invitrogen) supplemented with Biolayer interferometry 5% human serum from type AB donors (Sigma-Aldrich), 1% The affinities of anti-KIR3DL3 mAbs were assayed by biolayer inter- l-glutamine, penicillin (100 U/ml), streptomycin (100 g/ml), anti-­ ferometry, using an Octet RED96 system (ForteBio, Pall LLC). CD3/CD28 Dynabeads (Gibco), and rhIL-2 (100 ng/ml; BioLegend). Downloaded from Human primary NK cell subpopulations or NK92 cell subpop- Flow cytometry ulations (KIR3DL3+ or KIR3DL3− NK92 cells) were sorted by FACS Human PBMCs were preincubated with FcR blocking reagents on the basis of KIR3LD3 and/or TMIGD2 expression. Primary NK (Miltenyi Biotec) before staining. For surface marker staining, cells cell subpopulations were then expanded by culturing with autolo- were incubated with specific antibodies for 30 to 45 min at 4°C. For gous PBMCs as feeder cells (irradiated at 3000 rads, feeder cells:NK intracellular cytokine and CD107a staining, cells were cultured with cells = 20:1) in OpTmizer (Invitrogen) supplemented with 5% human http://immunology.sciencemag.org/ the Protein Transport Inhibitor Cocktail (eBioscience) and anti-­ AB serum (Sigma-Aldrich), 1% l-glutamine, penicillin (100 U/ml), CD107a (clone H4A3) for 5 hours, followed by staining with the streptomycin (100 g/ml), anti-CD3 OKT3 (10 ng/ml; BioLegend), Live/Dead Ghost Dye UV 450 (Tonbo Biosciences). The cells were rhIL-2 (40 ng/ml; BioLegend), and IL-15 (10 ng/ml; BioLegend). then fixed and permeabilized using the Fixation/Permeabilization Five days later, NK cells were further expanded in the same medium Solution Kit (BD Biosciences) according to the manufacturer’s guide- without anti-CD3 and feeder cells. lines, followed by staining with intracellular antibodies for 30 to 45 min at 4°C. Cytotoxicity assay For the spectral flow cytometry, PBMCs were stained with a fix- Cytotoxicity assays were performed through a flow-based assay. able blue dead cell stain kit (Invitrogen) and then incubated with FcR Briefly, target cells were labeled with PKH26 (Sigma-Aldrich) for blocking reagents (Miltenyi Biotec). The cells were then stained with 2 min at 37°C. Prelabeled target cells were coincubated with effector + anti-KIR3DL3 clone 26E10 on ice for 30 min, followed by goat anti-­ cells (NK92, primary NK, or CD8 T cells) in a 96-well round-bottom at Albert Einstein College of Medicine on July 9, 2021 mouse IgG PE (BioLegend). After blocking with normal mouse se- plate at indicated E:T ratios for 4 to 6 hours at 37°C. 7-aminoactinomycin rum, the cells were then stained by a mixture of surface markers D (7-AAD) was used to differentiate dead cells from live cells. diluted in Brilliant stain buffer (BD Biosciences) on ice for 30 min. Supernatants were collected after 24 hours of coculture for human The cells were then acquired on Aurora (Cytek). cytokine 65-plex assay (Eve Technologies). The standard formula of 100 × PKH26+7-AAD+ cells/PKH26+ cells % was used to calculate Jurkat-Raji and NK92-tumor conjugation assay specific lysis percentages. For KIR3DL3 or HHLA2 blocking, NK For the Jurkat-Raji conjugation assay, Raji cells were prepulsed with cells or target cells were preincubated with anti-KIR3DL3 or anti-­ SEE (100 ng/ml) (Toxin Technology) in RPMI medium for 30 min HHLA2 mAbs (20 g/ml) for 30 min, respectively, before mixing. at 37°C. The cells were then washed twice and resuspended in RPMI medium. The 4 × 105 SEE-loaded Raji and 4 × 105 Jurkat cells were CRISPR-Cas9–mediated knockout of HHLA2 precooled on ice and then mixed in a 96-well round-bottom plate. The scramble control single guide RNA (sgRNA) and HHLA2-­ The plate was centrifuged at 290g for 1 min at 4°C and then incubated targeting sgRNA, generated by GPP sgRNA Designer (https://portals. for 10 min at 37°C. The mixture of cells was resuspended and loaded broadinstitute.org/gpp/public/analysis-tools/sgrna-design), was cloned onto poly-l-lysine–precoated slides at room temperature (RT) for into lentiCRISPR v2 (Addgene, 52961). All these constructs are not 40 min. The cells were fixed with 4% formaldehyde at RT for 15 min predicted to target any known sequences in the human . The and then blocked by 5% normal goat serum at RT for 1 hour, followed lentiviruses were produced as described above. HCC827 were trans- by staining with anti-KIR3DL3 antibodies (a mixture of clones 3B7, duced with viral supernatant and then selected by puromycin (2 g/ml) 8G7, 14F8, 15D2, 26E10, 29H7, 34B10, and 51C3) (20 g/ml) at 4°C for 3 days. Stable knockout of HHLA2 (HHLA2KO) was confirmed overnight and then with goat anti-mIgG (H+L) Alexa Fluor 647 by flow cytometry analysis. (Invitrogen) at RT for 2 hours. The slides were then mounted by the Gold Antifade Mountant with 4′,6-diamidino-2-phenylindole RNA-seq data analysis (DAPI) (Life Technologies). RNA-seq reads were aligned to a (hg38) using Spliced For NK92-tumor conjugation assay, 5 × 105 NK92 and 5 × 105 Transcripts Alignment to a Reference (STAR) (version 2.6.1b) (50). A427 (or Raji) cells were mixed in a 50-ml Falcon tube centrifuged The number of RNA-seq fragments mapped to each gene in the reference at 1500 rpm for 2 min at RT. The tube was then incubated at 37°C gene annotation (downloaded from the UCSC Genome Browser in for 40 min. The cells were then loaded onto poly-l-lysine–precoated November 2019) was then counted using HTSeq (version 0.6.1) (51). slides and then stained as described above. Genes were considered expressed if their average expression counts For F-actin staining, after staining with primary and secondary were greater than or equal to 1 in either of the two sample groups antibodies as described above, the cells were permeabilized by 0.1% and selected for differential expression analysis and principal com- Triton X-100 at RT for 15 min. The permeabilized cells were then ponents analysis by DESeq2 (version 3.11) (52). To perform GSEA,

Wei et al., Sci. Immunol. 6, eabf9792 (2021) 9 July 2021 12 of 15 SCIENCE IMMUNOLOGY | RESEARCH ARTICLE we used the Kyoto Encyclopedia of Genes and database intratumorally (i.t.) and 200 g anti-KIR3DL3 mAb (clone 26E10) and ranked all genes by −log10 (P value) and multiplied by the sign or isotype control (mIgG1) intraperitoneally (i.p.) every 2 days of the log2 (fold change) (53). The ranks were then used for GSEA for four total treatments. For the subcutaneous HCC827 tumor to determine gene sets that were enriched among the genes ex- model, 3 × 106 HCC827 were subcutaneously injected. After 2 weeks, pressed greater in either KIR3DL3+ or KIR3DL3− samples (53, 54). the mice were divided into two groups (n = 4 for each group) and received KIR3DL3+ NK cell (i.t., 1 × 107) and mAb (i.p., 200 g) Coimmunoprecipitation and immunoblotting treatment every 2 days for four times. The tumor volume was cal- NK92 cells or primary NK cells pretreated with 1 mM pervanadate culated as (width2 × length)/2. (New England Biolabs) for 2 min at 37°C, or not treated were lysed For the intraperitoneal tumor mouse model, 4 × 106 HCC827-Luc2 in Pierce immunoprecipitation (IP) lysis buffer supplemented with cells were intraperitoneally injected into NSG mice. KIR3DL3+NK protease and phosphatase inhibitor cocktail (Thermo Fisher Scien- cells were expanded as described before. Four days later, the mice tific). Proteins from whole-cell lysis were further incubated with were imaged and then allocated to 26E10 or mIgG1 group based on Downloaded from anti-KIR3DL3 antibodies and Dynabeads protein G (Thermo Fisher baseline bioluminescence to ensure each group had an equivalent Scientific) for further IP. To analyze phosphorylation status, after tumor burden (n = 6 for each group). Each mouse then received mixing and receptor cross-linking, the cells were lysed in radioimmu- 1 × 107 KIR3DL3+ NK cells together with 1 g of rhIL-2, 1 g of noprecipitation lysis buffer [50 mM tris-HCl (pH 7.5), 0.15 M NaCl, rhIL-15, and 200 g of 26E10 (or mIgG1) intraperitoneally every 1% NP-40, 0.5% sodium deoxycholate, and 0.1% SDS] supplemented other day (five times in total). with protease and phosphatase inhibitor cocktail. Samples were sep- For the intravenous tumor mouse model, 5 × 105 HCC827-Luc2 http://immunology.sciencemag.org/ arated on SDS–polyacrylamide gel electrophoresis gels (GenScript) cells were intravenously (i.v.) injected into NSG mice. Four or five days and transferred onto nitrocellulose membranes (Bio-Rad) for pro- later, the mice were allocated to 26E10 or mIgG1 group based on tein detection. baseline bioluminescence to ensure each group had an equivalent tumor burden (n = 4 to 5 for each group). Each mouse then received KIR3DL3 cross-linking and cell mixing experiments 7 × 106 expanded KIR3DL3+ NK or 4 × 106 expanded KIR3DL3− For mAb-mediated cross-linking of KIR3DL3, KIR3DL3+ NK cells NK cells together with 1 g of rhIL-2, 1 g of rhIL-15, and 200 g of were preincubated with isotype control or anti-KIR3DL3 mAbs 26E10 (or mIgG1) intravenously every other day (three times in total). (clone 26E10) (10 g/ml) in the presence of anti-CD16 (5 g/ml) for Bioluminescence images were acquired by dorsal and/or ventral 30 min on ice. After washing with medium, NK cells were cross-linked imaging of each mouse by the Caliper Life Sciences IVIS spectrum with goat anti-mouse IgG (minimal x-reactivity) (25 g/ml; BioLegend) in vivo imaging system beginning 10 to 18 min after intraperitoneal at 37°C water bath for the indicated times. For cell mixing, prechilled injection of d-luciferin (150 g/g of body weight; Gold Biotechnology). at Albert Einstein College of Medicine on July 9, 2021 NK92 cells and target cells were mixed at a ratio of 2:1 on ice and then The data were analyzed with Living Image 3.0 software. transferred to a 37°C water bath for the indicated time. The cells were moved to ice and then lysed for further immunoblotting analysis. Quantification and statistical analysis Statistical analysis was performed with Prism GraphPad software IHC staining v8.0.2. The error bars in the figures represent SD. Statistical signifi- IHC staining was performed as described before (7). cance was determined with paired or unpaired Student’s t test (two- tailed), one-way or two-way analysis of variance (ANOVA), or Fisher’s Immunofluorescence staining exact test. The statistical test used, the sample size n, and the P values FFPE human cancer TMAs (US Biomax) were deparaffinized, fol- are stated in the corresponding figure legends. N in the figure rep- lowed by antigen retrieval with EDTA unmasking buffer (Cell Signal- resents the biological replicate unless indicated. All in vitro ex- ing Technology) in a steamer for 20 min at a sub-boiling temperature periments with cell lines were done with at least two replicates per (95° to 98°C). Slides were then blocked by 3% hydrogen peroxidase condition and repeated at least twice. In vitro experiments with pri- solution at RT for 10 min and subsequently by 5% normal goat serum mary human cells were performed with at least four independent at RT for 1 hour. A rabbit anti-KIR3DL3 (clone 1136B, R&D) mAb donors. All in vivo experiments were performed with at least four mice was used at a dilution of 1:3500; a mouse anti-CD45 mAb (clone: per group and repeated at least twice. 2B11 + PD7/26, Novus Biologicals) was used at a dilution of 1:75 for overnight incubation. Then, a goat anti-mIgG (H+L) Alexa Fluor SUPPLEMENTARY MATERIALS 647 (Invitrogen) and a goat anti-rabbit IgG (H+L) Alexa Fluor 488 immunology.sciencemag.org/cgi/content/full/6/61/eabf9792/DC1 (Invitrogen) were used both at 1:2000 dilution at RT for 1 hour. The Materials and Methods slides were then mounted by Gold Antifade Mountant with DAPI Figs. S1 to S10 (Life Technologies). Positive and negative controls (FFPE cell blocks) Tables S1 to S4 References (55–57) were included in each staining. View/request a protocol for this paper from Bio-protocol. In vivo mouse models of human cancers and NK cells Six- to 8-week-old NSG mice were used in all tumor models. For the REFERENCES AND NOTES subcutaneous Raji tumor model, HHLA2/Raji cells or control Raji 1. Y. Yang, Cancer immunotherapy: Harnessing the immune system to battle cancer. J. Clin. cells (1 × 106) were implanted into NSG mice subcutaneously (s.c.). Invest. 125, 3335–3337 (2015). 3 2. A. D. Waldman, J. M. Fritz, M. J. Lenardo, A guide to cancer immunotherapy: From T cell Once the tumor size reached around 150 mm , control tumor or Nat. Rev. Immunol. 20 + basic science to clinical practice. , 651–668 (2020). HHLA2 tumor–bearing mice were allocated to two groups (n = 4). 3. A. Haslam, J. Gill, V. Prasad, Estimation of the percentage of US patients with cancer who + 6 Mice were then treated with KIR3DL3 NK92 cells (2 × 10 ) are eligible for immune checkpoint inhibitor drugs. JAMA Netw. Open 3, e200423 (2020).

Wei et al., Sci. Immunol. 6, eabf9792 (2021) 9 July 2021 13 of 15 SCIENCE IMMUNOLOGY | RESEARCH ARTICLE

4. R. Zhao, J. M. Chinai, S. Buhl, L. Scandiuzzi, A. Ray, H. Jeon, K. C. Ohaegbulam, K. Ghosh, 25. A. E. Trundley, S. E. Hiby, C. Chang, A. M. Sharkey, S. Santourlidis, M. Uhrberg, A. Zhao, M. D. Scharff, X. Zang, HHLA2 is a member of the B7 family and inhibits human J. Trowsdale, A. Moffett, Molecular characterization of KIR3DL3. Immunogenetics 57, CD4 and CD8 T-cell function. Proc. Natl. Acad. Sci. U.S.A. 110, 9879–9884 (2013). 904–916 (2006). 5. M. Janakiram, U. A. Shah, W. Liu, A. Zhao, M. P. Schoenberg, X. Zang, The third group 26. V. Beziat, H. G. Hilton, P. J. Norman, J. A. Traherne, Deciphering the killer-cell of the B7-CD28 immune checkpoint family: HHLA2, TMIGD2, B7x, and B7-H3. Immunol. immunoglobulin-like receptor system at super-resolution for natural killer and T-cell Rev. 276, 26–39 (2017). biology. Immunology 150, 248–264 (2017). 6. Y. Zhu, S. Yao, B. P. Iliopoulou, X. Han, M. M. Augustine, H. Xu, R. T. Phennicie, S. J. Flies, 27. X. Zang, New immune checkpoint pathways: HHLA2 and its receptors including TMIGD2, M. Broadwater, W. Ruff, J. M. Taube, L. Zheng, L. Luo, G. Zhu, J. Chen, L. Chen, B7-H5 Cold Spring Harbor Asia Conference on Precision Cancer Biology: From Targeted to Immune costimulates human T cells via CD28H. Nat. Commun. 4, 2043 (2013). Therapies, Suzhou, China, 18 to 22 September 2017. 7. M. Janakiram, J. M. Chinai, S. Fineberg, A. Fiser, C. Montagna, R. Medavarapu, E. Castano, 28. W. M. Wojtowicz, J. Vielmetter, R. A. Fernandes, D. H. Siepe, C. L. Eastman, G. B. Chisholm, H. Jeon, K. C. Ohaegbulam, R. Zhao, A. Zhao, S. C. Almo, J. A. Sparano, X. Zang, Expression, S. Cox, H. Klock, P. W. Anderson, S. M. Rue, J. J. Miller, S. M. Glaser, M. L. Bragstad, J. Vance, clinical significance, and receptor identification of the newest B7 family member HHLA2 A. W. Lam, S. A. Lesley, K. Zinn, K. C. Garcia, A human IgSF cell-surface interactome reveals protein. Clin. Cancer Res. 21, 2359–2366 (2015). a complex network of protein-protein interactions. Cell 182, 1027–1043.e17 (2020).

8. H. Cheng, M. Janakiram, A. Borczuk, J. Lin, W. Qiu, H. Liu, J. M. Chinai, B. Halmos, 29. E. Verschueren, B. Husain, K. Yuen, Y. Sun, S. Paduchuri, Y. Senbabaoglu, I. Lehoux, Downloaded from R. Perez-Soler, X. Zang, HHLA2, a new immune checkpoint member of the B7 family, T. A. Arena, B. Wilson, S. Lianoglou, C. Bakalarski, Y. Franke, P. Chan, A. W. Wong, is widely expressed in human lung cancer and associated with EGFR mutational status. L. C. Gonzalez, S. Mariathasan, S. J. Turley, J. R. Lill, N. Martinez-Martin, The Clin. Cancer Res. 23, 825–832 (2017). immunoglobulin superfamily receptome defines cancer-relevant networks associated 9. C. Y. Jing, Y. P. Fu, Y. Yi, M. X. Zhang, S. S. Zheng, J. L. Huang, W. Gan, X. Xu, J. J. Lin, with clinical outcome. Cell 182, 329–344.e19 (2020). J. Zhang, S. J. Qiu, B. H. Zhang, HHLA2 in intrahepatic cholangiocarcinoma: An immune 30. L. A. Leaton, J. Shortt, K. M. Kichula, S. Tao, N. Nemat-Gorgani, A. J. Mentzer, checkpoint with prognostic significance and wider expression compared with PD-L1. S. J. Oppenheimer, Z. Deng, J. A. Hollenbach, C. R. Gignoux, L. A. Guethlein, P. Parham, J. Immunother. Cancer 7, 77 (2019). M. Carrington, P. J. Norman, Conservation, extensive heterozygosity, and convergence http://immunology.sciencemag.org/ 10. Q. H. Zhou, K. W. Li, X. Chen, H. X. He, S. M. Peng, S. R. Peng, Q. Wang, Z. A. Li, Y. R. Tao, of signaling potential all indicate a critical role for KIR3DL3 in higher primates. W. L. Cai, R. Y. Liu, H. Huang, HHLA2 and PD-L1 co-expression predicts poor prognosis Front. Immunol. 10, 24 (2019). in patients with clear cell renal cell carcinoma. J. Immunother. Cancer 8, e000157 (2020). 31. L. Moretta, Dissecting CD56dim human NK cells. Blood 116, 3689–3691 (2010). 11. H. Cheng, A. Borczuk, M. Janakiram, X. Ren, J. Lin, A. Assal, B. Halmos, R. Perez-Soler, 32. Y. Zhao, C. K. Lee, C.-H. Lin, R. B. Gassen, X. Xu, Z. Huang, C. Xiao, C. Bonorino, L.-F. Lu, X. Zang, Wide expression and significance of alternative immune checkpoint molecules, J. D. Bui, E. Hui, PD-L1:CD80 cis-heterodimer triggers the co-stimulatory receptor CD28 B7x and HHLA2, in PD-L1-negative human lung cancers. Clin. Cancer Res. 24, 1954–1964 while repressing the inhibitory PD-1 and CTLA-4 pathways. Immunity 51, 1059–1073.e9 (2018). (2019). 12. R. S. Bhatt, A. Berjis, J. C. Konge, K. M. Mahoney, A. N. Klee, S. S. Freeman, C. H. Chen, 33. D. Saverino, C. Tenca, D. Zarcone, A. Merlo, A. M. Megiovanni, M. T. Valle, F. Manca, O. A. Jegede, P. J. Catalano, J. C. Pignon, M. Sticco-Ivins, B. Zhu, P. Hua, J. Soden, J. Zhu, C. E. Grossi, E. Ciccone, CTLA-4 (CD152) inhibits the specific lysis mediated by human D. F. McDermott, A. R. Arulanandam, S. Signoretti, G. J. Freeman, KIR3DL3 is an inhibitory cytolytic T lymphocytes in a clonally distributed fashion. J. Immunol. 162, 651–658 receptor for HHLA2 that mediates an alternative immunoinhibitory pathway to PD1. (1999). Cancer Immunol. Res. 9, 156–169 (2021). 34. E. M. Mace, J. S. Orange, Multiple distinct NK-cell synapses. Blood 118, 6475–6476 (2011). 13. D. Chen, W. Chen, Y. Xu, M. Zhu, Y. Xiao, Y. Shen, S. Zhu, C. Cao, X. Xu, Upregulated 35. J. S. Orange, Formation and function of the lytic NK-cell immunological synapse. Nat. Rev.

immune checkpoint HHLA2 in clear cell renal cell carcinoma: A novel prognostic Immunol. 8, 713–725 (2008). at Albert Einstein College of Medicine on July 9, 2021 biomarker and potential therapeutic target. J. Med. Genet. 56, 43–49 (2019). 36. G. Huyer, S. Liu, J. Kelly, J. Moffat, P. Payette, B. Kennedy, G. Tsaprailis, M. J. Gresser, 14. P. Koirala, M. E. Roth, J. Gill, J. M. Chinai, M. R. Ewart, S. Piperdi, D. S. Geller, B. H. Hoang, C. Ramachandran, Mechanism of inhibition of protein-tyrosine by Y. V. Fatakhova, M. Ghorpade, X. Zang, R. Gorlick, HHLA2, a member of the B7 family, is vanadate and pervanadate. J. Biol. Chem. 272, 843–851 (1997). expressed in human osteosarcoma and is associated with metastases and worse survival. 37. M. Daeron, S. Jaeger, L. Du Pasquier, E. Vivier, Immunoreceptor tyrosine-based inhibition Sci. Rep. 6, 31154 (2016). motifs: A quest in the past and future. Immunol. Rev. 224, 11–43 (2008). 15. G. Lin, H. Ye, J. Wang, S. Chen, X. Chen, C. Zhang, Immune checkpoint human 38. W. M. Yokoyama, B. F. M. Plougastel, Immune functions encoded by the natural killer endogenous retrovirus-H long terminal repeat-associating protein 2 is upregulated gene complex. Nat. Rev. Immunol. 3, 304–316 (2003). and independently predicts unfavorable prognosis in bladder urothelial carcinoma. 39. C. M. Gardiner, L. A. Guethlein, H. G. Shilling, M. Pando, W. H. Carr, R. Rajalingam, Nephron 141, 256–264 (2019). C. Vilches, P. Parham, Different NK cell surface phenotypes defined by the DX9 antibody 16. L. Wei, L. Tang, H. Chang, S. Huo, Y. Li, HHLA2 overexpression is a novel biomarker are due to KIR3DL1 gene polymorphism. J. Immunol. 166, 2992–3001 (2001). of malignant status and poor prognosis in gastric cancer. Hum. Cell 33, 116–122 (2020). 40. J. P. Vivian, R. C. Duncan, R. Berry, G. M. O’Connor, H. H. Reid, T. Beddoe, S. Gras, 17. Y. Xiao, H. Li, L.-L. Yang, L. Mao, C.-C. Wu, W.-F. Zhang, Z.-J. Sun, The expression patterns P. M. Saunders, M. A. Olshina, J. M. Widjaja, C. M. Harpur, J. Lin, S. M. Maloveste, D. A. Price, and associated clinical parameters of human endogenous retrovirus-H long terminal B. A. Lafont, D. W. McVicar, C. S. Clements, A. G. Brooks, J. Rossjohn, Killer cell repeat-associating protein 2 and transmembrane and immunoglobulin-like receptor 3DL1-mediated recognition of human leukocyte antigen containing 2 in oral squamous cell carcinoma. Dis. Markers 2019, 5421985 (2019). B. Nature 479, 401–405 (2011). 18. Z. Zhu, W. Dong, Overexpression of HHLA2, a member of the B7 family, is associated with 41. W. H. Carr, M. J. Pando, P. Parham, KIR3DL1 polymorphisms that affect NK cell inhibition worse survival in human colorectal carcinoma. Onco. Targets Ther. 11, 1563–1570 (2018). by HLA-Bw4 ligand. J. Immunol. 175, 5222–5229 (2005). 19. P. P. C. Boor, K. Sideras, K. Biermann, M. Hosein Aziz, I. J. M. Levink, S. Mancham, N. S. Erler, 42. B. I. Pereira, R. P. H. De Maeyer, L. P. Covre, D. Nehar-Belaid, A. Lanna, S. Ward, R. Marches, X. Tang, C. H. van Eijck, M. J. Bruno, D. Sprengers, X. Zang, J. Kwekkeboom, HHLA2 is E. S. Chambers, D. C. O. Gomes, N. E. Riddell, M. K. Maini, V. H. Teixeira, S. M. Janes, expressed in pancreatic and ampullary cancers and increased expression is associated D. W. Gilroy, A. Larbi, N. A. Mabbott, D. Ucar, G. A. Kuchel, S. M. Henson, J. Strid, J. H. Lee, with better post-surgical prognosis. Br. J. Cancer 122, 1211–1218 (2020). J. Banchereau, A. N. Akbar, Sestrins induce natural killer function in senescent-like CD8+ 20. H. Yan, W. Qiu, A. K. Koehne de Gonzalez, J. S. Wei, M. Tu, C. H. Xi, Y. R. Yang, Y. P. Peng, T cells. Nat. Immunol. 21, 684–694 (2020). W. Y. Tsai, H. E. Remotti, Y. Miao, G. H. Su, HHLA2 is a novel immune checkpoint protein 43. F. Levi-Schaffer, O. Mandelboim, Inhibitory and coactivating receptors recognising in pancreatic ductal adenocarcinoma and predicts post-surgical survival. Cancer Lett. the same ligand: Immune homeostasis exploited by pathogens and tumours. 442, 333–340 (2019). Trends Immunol. 39, 112–122 (2018). 21. Z. Zhang, J. Liu, C. Zhang, F. Li, L. Li, D. Wang, D. Chand, F. Guan, X. Zang, Y. Zhang, 44. F. Romagne, P. Andre, P. Spee, S. Zahn, N. Anfossi, L. Gauthier, M. Capanni, L. Ruggeri, Over-expression and prognostic significance of HHLA2, a new immune checkpoint D. M. Benson Jr., B. W. Blaser, M. Della Chiesa, A. Moretta, E. Vivier, M. A. Caligiuri, molecule, in human clear cell renal cell carcinoma. Front. Cell Dev. Biol. 8, 280 (2020). A. Velardi, N. Wagtmann, Preclinical characterization of 1-7F9, a novel human anti-KIR 22. S. A. Rieder, J. Wang, N. White, A. Qadri, C. Menard, G. Stephens, J. L. Karnell, C. E. Rudd, receptor therapeutic antibody that augments natural killer-mediated killing of tumor R. Kolbeck, B7-H7 (HHLA2) inhibits T-cell activation and proliferation in the presence cells. Blood 114, 2667–2677 (2009). of TCR and CD28 signaling. Cell. Mol. Immunol. 18, 1503–1511 (2021). 45. E. Lo Monaco, E. Tremante, C. Cerboni, E. Melucci, L. Sibilio, A. Zingoni, M. R. Nicotra, 23. X. Zhuang, E. O. Long, CD28 homolog is a strong activator of natural killer cells for lysis P. G. Natali, P. Giacomini, Human leukocyte antigen E contributes to protect tumor cells of B7H7+ tumor cells. Cancer Immunol. Res. 7, 939–951 (2019). from lysis by natural killer cells. Neoplasia 13, 822–830 (2011). 24. M. Janakiram, J. M. Chinai, A. Zhao, J. A. Sparano, X. Zang, HHLA2 and TMIGD2: New 46. T. Kamiya, S. V. Seow, D. Wong, M. Robinson, D. Campana, Blocking expression immunotherapeutic targets of the B7 and CD28 families. Onco. Targets. Ther. 4, e1026534 of inhibitory receptor NKG2A overcomes tumor resistance to NK cells. J. Clin. Invest. 129, (2015). 2094–2106 (2019).

Wei et al., Sci. Immunol. 6, eabf9792 (2021) 9 July 2021 14 of 15 SCIENCE IMMUNOLOGY | RESEARCH ARTICLE

47. P. Leone, E.-C. Shin, F. Perosa, A. Vacca, F. Dammacco, V. Racanelli, MHC class I antigen phenotype defines non-small cell lung carcinomas sensitive to immune checkpoint processing and presenting machinery: Organization, function, and defects in tumor cells. blockers. Nat. Commun. 9, 3196 (2018). J. Natl. Cancer Inst. 105, 1172–1187 (2013). 48. M. Sade-Feldman, Y. J. Jiao, J. H. Chen, M. S. Rooney, M. Barzily-Rokni, J. P. Eliane, Acknowledgments: We thank H. Huang for the IVIS Spectrum Imaging System, H. Guzik for S. L. Bjorgaard, M. R. Hammond, H. Vitzthum, S. M. Blackmon, D. T. Frederick, the 3DHISTECH PANNORAMIC 250 Flash II slide scanner, and J. Zhang for the Cytek Aurora. M. Hazar-Rethinam, B. A. Nadres, E. E. Van Seventer, S. A. Shukla, K. Yizhak, J. P. Ray, Funding: The Flow Cytometry Core Facility and Analytical Imaging Facility are supported by D. Rosebrock, D. Livitz, V. Adalsteinsson, G. Getz, L. M. Duncan, B. Li, R. B. Corcoran, National Institutes of Health (NIH) P30CA013330 (Einstein Cancer Center) and P30DK020541 D. P. Lawrence, A. Stemmer-Rachamimov, G. M. Boland, D. A. Landau, K. T. Flaherty, (Einstein-Mount Sinai Diabetes Research Center). The 3DHISTECH PANNORAMIC 250 Flash II R. J. Sullivan, N. Hacohen, Resistance to checkpoint blockade therapy through slide scanner is supported by SIG 1S10OD019961-01. P.M.G.J. is supported by NIH inactivation of antigen presentation. Nat. Commun. 8, 1136 (2017). 5TL1TR002557. This work was supported by NIH R01CA175495 and R01DK100525, 49. K. Karre, H. G. Ljunggren, G. Piontek, R. Kiessling, Selective rejection of H–2-deficient lymphoma Department of Defense BC190403, and Cancer Research Institute awarded to X.Z. Author variants suggests alternative immune defence strategy. Nature 319, 675–678 (1986). contributions: Y.W., X.R., S.M., and X.Z. conceived the studies and designed the experiments. 50. A. Dobin, C. A. Davis, F. Schlesinger, J. Drenkow, C. Zaleski, S. Jha, P. Batut, M. Chaisson, Y.W., X.R., and S.M. performed experiments with support from B.E.-G. and J.M.C. P.M.G.J. and T. R. Gingeras, STAR: Ultrafast universal RNA-seq aligner. Bioinformatics 29, 15–21 (2013). D.Z. performed the bioinformatic and statistical analysis. S.M., H.W., R.A.S., L.S., L.Z., X.T., Q.L., Downloaded from 51. S. Anders, P. T. Pyl, W. Huber, HTSeq—A Python framework to work with high- M.P., and F.G. provided necessary resources. Y.W. and X.R. wrote the manuscript. All authors throughput sequencing data. Bioinformatics 31, 166–169 (2015). reviewed the manuscript. Competing interests: Y.W., X.R., S.M., and X.Z. are inventors on a 52. M. I. Love, W. Huber, S. Anders, Moderated estimation of fold change and dispersion pending patent (KIR3DL3 is an inhibitory receptor of the immune system and uses thereof); for RNA-seq data with DESeq2. Genome Biol. 15, 550 (2014). B.E.-G. is an employee and X.Z. is a cofounder of NextPoint Therapeutics Inc. The other authors 53. M. Kanehisa, M. Furumichi, M. Tanabe, Y. Sato, K. Morishima, KEGG: New perspectives declare that they have no competing interests. Data and materials availability: The RNA-seq on genomes, pathways, diseases, and drugs. Nucleic Acids Res. 45, D353–D361 (2017). data have been uploaded to the Gene Expression Omnibus database with accession number 54. A. Subramanian, P. Tamayo, V. K. Mootha, S. Mukherjee, B. L. Ebert, M. A. Gillette, GSE175550. All data needed to evaluate the conclusions in the paper are present in the paper http://immunology.sciencemag.org/ A. Paulovich, S. L. Pomeroy, T. R. Golub, E. S. Lander, J. P. Mesirov, Gene set enrichment or the Supplementary Materials. analysis: A knowledge-based approach for interpreting genome-wide expression profiles. Proc. Natl. Acad. Sci. U.S.A. 102, 15545–15550 (2005). Submitted 2 December 2020 55. S. A. Stewart, D. M. Dykxhoorn, D. Palliser, H. Mizuno, E. Y. Yu, D. S. An, D. M. Sabatini, Resubmitted 22 April 2021 I. S. Chen, W. C. Hahn, P. A. Sharp, R. A. Weinberg, C. D. Novina, Lentivirus-delivered stable Accepted 28 May 2021 gene silencing by RNAi in primary cells. RNA 9, 493–501 (2003). Published 9 July 2021 56. B. E. Welm, G. J. Dijkgraaf, A. S. Bledau, A. L. Welm, Z. Werb, Lentiviral transduction 10.1126/sciimmunol.abf9792 of mammary stem cells for analysis of gene function during development and cancer. Cell Stem Cell 2, 90–102 (2008). Citation: Y. Wei, X. Ren, P. M. Galbo Jr., S. Moerdler, H. Wang, R. A. Sica, B. Etemad-Gilbertson, 57. S. N. Gettinger, J. Choi, N. Mani, M. F. Sanmamed, I. Datar, R. Sowell, V. Y. Du, E. Kaftan, L. Shi, L. Zhu, X. Tang, Q. Lin, M. Peng, F. Guan, D. Zheng, J. M. Chinai, X. Zang, KIR3DL3-HHLA2 is S. Goldberg, W. Dong, D. Zelterman, K. Politi, P. Kavathas, S. Kaech, X. Yu, H. Zhao, a human immunosuppressive pathway and a therapeutic target. Sci. Immunol. 6, eabf9792 J. Schlessinger, R. Lifton, D. L. Rimm, L. Chen, R. S. Herbst, K. A. Schalper, A dormant TIL (2021). at Albert Einstein College of Medicine on July 9, 2021

Wei et al., Sci. Immunol. 6, eabf9792 (2021) 9 July 2021 15 of 15 KIR3DL3-HHLA2 is a human immunosuppressive pathway and a therapeutic target Yao Wei, Xiaoxin Ren, Phillip M. Galbo, Jr., Scott Moerdler, Hao Wang, R. Alejandro Sica, Bijan Etemad-Gilbertson, Lei Shi, Liqiang Zhu, Xudong Tang, Qi Lin, Mou Peng, Fangxia Guan, Deyou Zheng, Jordan M. Chinai and Xingxing Zang Downloaded from Sci. Immunol. 6, eabf9792. DOI: 10.1126/sciimmunol.abf9792

KIR3DL3-HHLA2 interactions protect tumors HHLA2 (a B7 family member) has both immune inhibitory and activating abilities and is expressed in many human

cancers. These differing functions rely on the receptor to which it binds, yet many of these receptors are uncharacterized. http://immunology.sciencemag.org/ Here, Wei et al. found that KIR3DL3 bound to HHLA2 and was expressed on effector memory CD8+ T cells and CD56 dim NK cells. Interactions of KIR3DL3 with tumoral HHLA2 inhibited the function and killing capacity of both CD8+ T cells and NK cells. KIR3DL3 + immune cells infiltrated various types of HHLA2+ tumors from patients with cancer, and blockade of KIR3DL3 inhibited the growth of tumors in various mouse models. Thus, KIR3DL3-HHLA2 inhibits immune-mediated clearance of tumor cells and presents a possible immunotherapeutic target for cancer.

ARTICLE TOOLS http://immunology.sciencemag.org/content/6/61/eabf9792

SUPPLEMENTARY

http://immunology.sciencemag.org/content/suppl/2021/07/02/6.61.eabf9792.DC1 at Albert Einstein College of Medicine on July 9, 2021 MATERIALS

REFERENCES This article cites 56 articles, 18 of which you can access for free http://immunology.sciencemag.org/content/6/61/eabf9792#BIBL

PERMISSIONS http://www.sciencemag.org/help/reprints-and-permissions

Use of this article is subject to the Terms of Service

Science Immunology (ISSN 2470-9468) is published by the American Association for the Advancement of Science, 1200 New York Avenue NW, Washington, DC 20005. The title Science Immunology is a registered trademark of AAAS. Copyright © 2021 The Authors, some rights reserved; exclusive licensee American Association for the Advancement of Science. No claim to original U.S. Government Works

Supplementary Materials for

KIR3DL3-HHLA2 is a human immunosuppressive pathway and a therapeutic target

Yao Wei et al.

Corresponding author: Xingxing Zang, [email protected]

Sci. Immunol. 6, eabf9792 (2021) DOI: 10.1126/sciimmunol.abf9792

The PDF file includes:

Figs. S1 to S10 Tables S1 and S2 References (55–57)

Other Supplementary Material for this manuscript includes the following:

Tables S3 and S4 Supplemental materials and methods

Cell lines

Human cell lines in this study include: Phoenix-ampho, retrovirus producer lines (ATCC,

CRL-3213); HEK 293T cell, lentivirus producer lines (a gift from Dr. Wenjun Guo); K562, human chronic myelogenous leukemia (ATCC, CCL-243); Jurkat, a human T lymphoblastic leukemia cell line (ATCC, TIB-152); Raji, human B-cell lymphoma (ATCC, CCL-86); A427, human lung adenocarcinoma (a gift from Dr. Haiying Cheng); HCC827, human lung adenocarcinoma (ATCC, CRL-2868). These cells were cultured in either DMEM or RPMI1640

(Gibco) medium supplemented with 10% FBS, 100 U/mL penicillin, and 100 μg/mL streptomycin. Human NK-cell leukemia cell line NK92 (ATCC, CRL-2407) were cultured in

MEM α medium (Gibco) supplemented with 12.5% horse serum and 12.5% FBS, 0.2 mM inositol, 0.1 mM 2-mercaptoethanol, 0.02 mM folic acid, 200 U/ml recombinant human IL-2.

Mouse cell lines used in this study were mouse fibroblast line NIH-3T3 (ATCC, CRL-1658), mouse mast cell line P815 (ATCC, TIB-64) and mouse myeloma cell line NSO (a gift from Dr.

Matthew D. Scharff). Cells were cultured in DMEM medium supplemented with 10% FBS, 100

U/mL penicillin, and 100 μg/mL streptomycin. All cell lines were cultured at 37°C in a humidified atmosphere containing 5% CO2.

The human protein cell microarray system

KIR3DL3-Ig protein (20 μg/ml) was primarily screened for binding to fixed human

HEK293 cells individually expressing 5484 full-length human plasma membrane proteins and secreted proteins. This screen revealed 11 primary hits, including HHLA2. Then each primary hit was re-expressed and probed with KIR3DL3-Ig protein or control protein to determine repeatable and specific interaction for KIR3DL3-Ig protein. The similar screen was performed with anti-KIR3DL3 clone 26E10.

Human phospho-kinase arrays

KIR3DL3 WT or Y381F mutant expressing NK92 cells were rested in serum-free medium at 37 °C for 2 hr to reduce the phosphorylation background. Pre-cooled HHLA2/Raji cells (4×106) and rested NK92 cells (4×106) were then mixed and incubated at 37 °C water bath for 5 min. Cells were immediately transferred to ice to stop reaction and then lysed with cell lysis buffer, followed by analyzing the relative levels of protein phosphorylation according to the manufacturing instructions.

Production and purification of fusion proteins

The coding region of the whole or indicated part of the extracellular domain without signal peptide of KIR3DL3, HHLA2, or TMIGD2 was fused to a human or mouse IgG1 Fc tag in a pMT/BiP vector. Each construct was co-transfected with a blasticidin resistant plasmid into

Drosophila Schneider 2 (S2) cells by the calcium phosphate transfection kit (Invitrogen). The stably transfected S2 cell lines were cultured and expanded in Schneider's Drosophila Medium

(Gibco) supplemented with 10% FBS, 100 U/mL penicillin, 100 μg/mL streptomycin, and 25

µg/ml of blasticidin (Gold Biotechnology). The S2 cells were induced to secrete fusion proteins in Express Five serum-free medium (Life Technologies) in the presence of 0.75 mM CuSO4.

Proteins were purified using Protein G resin (GenScript) columns. The purity and identity of fusion proteins were confirmed by SDS-PAGE.

Virus production and generation of stable cell lines

Molecules expressed in NIH 3T3 cells, K562, Raji, and A427 cells were introduced by retrovirus transduction. Retrovirus was produced in Phoenix-ampho cells transfected with pCMV-VSV-G (55) and MSCV-YFP containing the gene of interest using jetPRIME reagents

(Polyplus Transfection).

Molecules expressed in Jurkat, NK92, HCC827 cells, KIR3DL3*004 wild type and mutants expressed in 3T3 cells were introduced by lentiviral transduction. Lentivirus was produced in HEK293T cells transfected with pCMVR8.74, pCMV-VSV-G, and a lentiviral backbone vector (56) containing the gene of interest using jetPRIME reagents (Polyplus

Transfection).

Virus-containing supernatant was harvested 48-72 h post-transfection and filtered through a 0.45 μm filter. Cells were spin infected at 2000×g for 120 min at 37°C in the presence of 5 μg/ml of polybrene (Merck Millipore) and 1-2 ml virus supernatant. Transduced cells were sorted using a BD FACSAria Fusion Cell Sorter (BD Biosciences).

Cell binding and blocking assays

In the anti-HHLA2 mAbs blocking assay, HHLA2/3T3 cells were preincubated with 5 or

10 μg/ml anti-HHLA2 mAb or mIgG1 on ice for 30 min. After wash, the cells were then incubated with 0.15 μg/ml KIR3DL3-Ig or hIgG on ice for 45 min. In the anti-KIR3DL3 mAb blocking assay, 0.15 μg/ml KIR3DL3-Ig was pre-incubated with anti-KIR3DL3 mAbs at the indicated concentration in room temperature (RT) for 30 min. Then HHLA2/3T3 cells were incubated with the protein-mAb mixtures on ice for 45 min. HHLA2/3T3 cells were then incubated with APC anti-human IgG Fc antibody on ice for 30 min and acquired on LSR II (BD

Biosciences).

In the competition binding assay, HHLA2/3T3 cells pre-coated with indicated concentrations of KIR3DL3-Ig or hIgG were incubated with 10 μg/ml TMIGD2-mIg on ice for

45 min, followed by APC anti-mouse IgG (1:200, eBioscience) on ice for 30 min. In the reverse direction, HHLA2/3T3 cells pre-coated with indicated concentrations of TMIGD2-mIg or mIgG were incubated with 10 μg/ml KIR3DL3-Ig on ice for 45 min, followed by APC anti-human IgG

Fc on ice for 30 min. Then the cells were acquired on LSR II (BD Biosciences).

Generation of monoclonal antibodies against KIR3DL3 and HHLA2

Splenocytes from KIR3DL3-Ig or KIR3DL3-D0-Ig immunized BALB/c mice were fused with NSO myeloma cells. Fourteen independent clones that recognized KIR3DL3, but not other irrelevant protein presented on 3T3 cell surface, were selected by high throughput flow cytometry. Hybridoma cells were cultured in CELLine 350 Bioreactor Flask (DWK Life

Sciences). The cell compartment media was DMEM (Hyclone) supplemented with 10% ultra- low IgG FBS (Thermo Fisher), 10% NCTC-109 (Thermo Fisher), 1% nonessential amino acids,

100 U/mL penicillin and 100 μg/mL streptomycin. The medium compartment media was

DMEM (Hyclone) supplemented with 100 U/mL penicillin and 100 μg/mL streptomycin.

Antibodies were purified from hybridoma supernatant by Protein G resin (GenScript) columns.

The purity and identity of antibodies were confirmed by SDS-PAGE. Clone 26E10 was conjugated with PE by SiteClick™ R-PE Antibody Labeling Kit (Invitrogen) for the following analysis. Similarly, mAbs B5B5 and A3H11 against the IgV1 domain of HHLA2 were generated.

Biolayer interferometry (BLI)

Anti-human Fc capture biosensors (ForteBio, Pall LLC) were pre-loaded with KIR3DL3-

Ig and then dipped into a solution containing mAb at serial dilutions. Data were analyzed using

Forte Pall (Port Washington, NY) software 9.0. The Global data fitting to a 1:1 binding model was used to estimate values for the Kon (association rate constant), Koff (dissociation rate constant), and KD (equilibrium ).

ELISA

KIR3DL3-Ig and KIR3DL3-D0-Ig or hIgG (5 μg/ml) were immobilized on a 96-well flat-bottom plate in PBS buffer at 4°C overnight. The immobilized proteins were then incubated with 5 μg/ml mAbs in a solution of PBS containing 0.5% BSA at 37°C for 1 hr. Plates were then incubated with alkaline phosphatase-conjugated anti-mouse IgG (SouthernBiotech) followed by detection with 1 mg/ml pNPP substrate (SouthernBiotech). The absorbance at 405 nm was measured by a plate reader.

Flow cytometry

The following fluorophore-conjugated antibodies were used for staining: CD45 (clone

HI30), CD3 (clone HIT3a), CD4 (clone OKT4), CD7 (clone M-T701), CD8 (clone HIT8a),

CD14 (clone 63D3), CD16 (clone 3G8), CD19 (clone HIB19), CD56 (clone 5.1H11), TCR γδ

(clone B1), CCR7 (clone G043H7), CD45RA (clone HI100), ULBP-1 (clone 170818), ULBP-

2/5/6 (clone 165903), ULBP-3 (clone 166510), MICA/B (clone 6D4), CD48 (clone BJ40),

CD155 (clone SKII.4), CD112 (clone TX31), ICAM-1 (clone HCD54), ICAM-2 (clone CBR-

IC2/2), IFN-γ (clone B27), TNF-α (clone mab11).

The following fluorophore-conjugated antibodies were used for the spectral flow cytometry: KLRG1 (clone SA231A2), NKP46 (clone 9E2), KIR3DL2 (clone 539304), NKG2D

(clone 1D11), CD28 (clone 28.2), CD56 (clone 5.1H11), KIR2DL1/S1/S3/S5 (clone HP-MA4),

NKG2C (clone 134591), KIR2DL2/3 (clone DX27), PD-1 (clone ebioJ105), CD26 (clone L272),

CD4 (clone OKT-4), KIR3DL1 (clone DX9), ICOS (clone C398.4A), CD45RA (clone HI100),

2B4 (clone C1.7), NKG2A (clone 131411), NKP44 (clone p44-8), CCR7 (clone G043H7),

NKP30 (clone p30-15), CD16 (clone 3G8), CD8 (clone RPA-T8), CD3 (clone UCHT1). For co-expression of KIR3DL3 and TMIGD2 staining, PBMCs were incubated with anti-

TMIGD2 (R&D, clone 953743) on ice for 30 min, followed by staining with APC anti-mouse

IgG on ice for 30 min. After blocking with normal mouse serum, the cells were then stained by surface markers and anti-KIR3DL3 conjugated with PE. The cells were acquired on LSR II (BD

Biosciences).

All the data were analyzed with FlowJo (BD Biosciences) software. DownSample and

FlowSOM plugins were used for generating the UMAP. All samples were gated to exclude doublets and dead cells using forward-scatter area versus height and viability dye, respectively.

Gates were set with FMO controls.

Cytotoxicity assay

For redirected cytotoxicity assays, PKH26-labeled P815 (or K562, Raji) cells were pre- incubated with 2 μg/ml of anti-KIR3DL3 mAbs, mIgG1, anti-CD28 (clone CD28.2), anti-CD56

(clone 5.1H11) or anti-KIR3DL1 (clone DX9) in the presence of anti-CD3 (0.05-0.5 μg/ml,

CD8+ T cell assay) or anti-CD16 (0.5 μg/ml, NK cell assay) for 15 min at RT. Effector CD8+ T cells or NK cells were added at the indicated E:T ratio. After 4-6 hr at 37°C, the lysis of target cells was analyzed by flow cytometry as described above. Supernatants were collected after 24 hr co-culture for Human cytokine 65-Plex Assay (Eve Technologies).

Plasmid construction and site-directed mutagenesis

The plasmids coding membrane IgSF genes were purchased from Einstein Molecular

Cytogenetics Core, Harvard plasmid or R&D systems. All genes were cloned into the MSCV-

YFP vector, as previously described (4). The pHIV-KIR3DL3-Zsgreen construct was generated by using PCR to amplify the coding sequence of KIR3DL3 without stop codon and then cloned into NotI and HpaI site of the pHIV-Zsgreen (Addgene). Mutagenesis of KIR3DL3 Y381 and KIR3DL3*004 D0 domain was performed using the Q5® Site-Directed Mutagenesis Kit (NEB). The mutants were constructed using the following primers:

Y381F Forward: GGAGGTGACATTCGCACAGTTGA

Y381F Reverse: TGAGGGTCTTGTTCATCAG

H52R (KIR3DL3*001) Forward: GTCGCTCTCGTCTTGGGTTTAACG

H52R (KIR3DL3*001) Reverse: ACTGAAGAGTCACATGTTGTCCTT

H52R and N56K (KIR3DL3*006) Forward: TGGGTTTAAAGAATTCAGTCTGTCC

H52R and N56K (KIR3DL3*006) Reverse: AGACGAGAGCGACACTGAAGAGTC

H52R and R77W (KIR3DL3*010) Forward: CAGAATATTCTGGAACAGCTTTC

H52R and R77W (KIR3DL3*010) Reverse: TTGTAGAGCTCAGGGACAG

M66K (KIR3DL3*033) Forward: AAGACGGGAAGCCTGTCCCTGAGC

M66K (KIR3DL3*033) Reverse: CTTTGGACAGACTGAATTCGTTAAA

H52R and S107P (KIR3DL3*039) Forward: CTGGGTGGCCGGCACCCAGCAAC

H52R and S107P (KIR3DL3*039) Reverse: TGGGGGAGTGTGGGTGTGAACTGC

H52R and P100A (KIR3DL3*043) Forward: AGTTCACACGCACACTCCCCCACT

H52R and P100A (KIR3DL3*043) Reverse: GCAACATCTGTAGGTCCCTGCATG

H52R and V38M (KIR3DL3*047) Forward: CCGGCACTATGGTGTCTGAAGGAC

H52R and V38M (KIR3DL3*047) Reverse: GCCAGGCAGAGAGGAAGGGCTTGT

H52R and R77Q (KIR3DL3*068) Forward: AGAATATTCCAGAACAGCTTTCTCATG

H52R and R77Q (KIR3DL3*068) Reverse: GTTGTAGAGCTCAGGGAC H52R and G36S (KIR3DL3*070) Forward: TGCCTGGCCCAGCACTGTGGT

H52R and G36S (KIR3DL3*070) Reverse: GAGAGGAAGGGCTTGTCCTGAC

H52R and P103L (KIR3DL3*091) Forward: ACACTCCCTCACTGGGTGGTCG

H52R and P103L (KIR3DL3*091) Reverse: GGGTGTGAACTGCAACATCTGTAG

CRISPR/Cas9-mediated knockout of HHLA2

The sequences of sgRNAs are: scramble control sgRNA: 5’-GCACTACCAGAGCTAACTCA-3’

HHLA2 targeting sgRNA#1: 5’-ACTTGGTCCAGAATGAAAAG-3’

HHLA2 targeting sgRNA#2: 5’-ATCCTCGTCCAATTATCACG-3’

RNA isolation

CD8+ T cells were enriched from fresh PBMCs by CD8+ microbeads (Miltenyi Biotec).

+ - 10,000 to 20,000 KIR3DL3 and KIR3DL3 CD8+ TEMRA, defined as

CD3+CD8+CD45RA+CCR7- were sorted by the BD FACSAria directly into RNA lysis buffer.

RNA was then extracted with the RNeasy Micro Kit (Qiagen) and stored at -80°C. Library preparation (by SMARTer Stranded Total RNA Pico Kit) and sequencing (at 40M paired-end reads per sample) was performed by Admera Health.

Co-immunoprecipitation and immunoblotting

The following antibodies were used: anti-phospho-tyrosine 4G10 (Sigma-Aldrich,

1:1,000), anti-SHP-1 (CST, 1:500), anti-SHP-2 (CST, 1:500), anti-Vav1 (CST, 1:2000), anti- phospho-Vav1 Tyr160 (Invitrogen, 1:2,000), anti-ERK1/2 (CST, 1:2,000), anti-phospho-ERK1/2

Thr202/Tyr204 (Biolegend, 1:1000), anti-AKT (CST, 1:1,000), anti-phospho-AKT Ser473 (CST,

1:1000), anti-phospho-NF-κB p65 Ser536 (CST, 1:1,000), anti-β-actin (Santa Cruz, 1:2,000), horseradish peroxidase (HRP)-conjugated goat anti-mouse (Jackson ImmunoResearch, 1:1,0000), rabbit anti-goat (Jackson ImmunoResearch, 1:1,0000), goat anti-rabbit (CST, 1:2,000) secondary antibodies and enhanced chemiluminescent substrate (ECL, Bio-Rad).

Immunohistochemistry staining

FFPE human cancer TMAs (USBiomax) were deparaffinized, followed by antigen retrieval with EDTA unmasking buffer (CST) in a steamer for 20 min at a sub-boiling temperature (95-98°C). Slides were then blocked by 3% hydrogen peroxidase solution at RT for

10 min and subsequently by 5% normal goat serum at RT for 1 hr. A rabbit anti-PD-L1 (clone

E1L3N, CST) mAb was used at a dilution of 1:200; a mouse anti-HHLA2 mAb (clone 566.1) was used at a concentration of 10 μg/ml for overnight incubation. The anti-HHLA2 mAb 566.1 has been validated by previously published studies (7). The slides were then incubated with boost detection reagent (HRP, CST) at RT for 1 hr, followed by SignalStain DAB (CST) and hematoxylin nuclear counterstaining. Positive and negative controls (FFPE cell blocks) were included in each staining.

The slides were scanned by a 3DHistech P250 high-capacity slide scanner (Budapest,

Hungary). Staining was analyzed with the DensitoQuant module of 3DHistech Quant Center by a trained researcher. Briefly, the brown signal's pixel intensity was scaled and displayed in three levels: weak, moderate, and strong by color thresholding. The module automatically calculated histoscores (H-score) using each pixel intensity level and the total area of pixels (8).

Immunofluorescence staining

The same cohorts as the IHC staining were used. The slides were scanned by a 3DHistech

P250 high-capacity slide scanner (Budapest, Hungary) by three channels with filter settings for

DAPI, FITC, and Cy5. Staining was analyzed with the HistoQuant module of 3DHistech Quant

Center by a trained researcher. Briefly, two clusters were generated by color thresholding of the red pixel (CD45) and green pixel (KIR3DL3), respectively. The colocalization cluster was then activated to indicate colocalized objects. The QIF score of KIR3DL3 in the CD45+ cell compartment was calculated by dividing the pixel intensities of colocalization by the area of

CD45 positivity in the core (57). A CD4 T CD8 T NK cells

Resting Activated Resting Activated Resting Activated

x

a

M

f

o

%

Anti-HHLA2 B C

a b c

A H

H

- -

-

C C

C

S S

S

S F S

FSC-A FSC-A SSC-A

e d

A

-

C

S S

FITC DAPI D E CTLA4-Ig KIR3DL3-Ig FCGR2A IGHG3 IGHG3 CXCL12 HHLA2 CXCL12

IGF1 IGF1 26E10

CD86

F Signal peptide MSLMVVSMACVGFFLLEGPWPHVGG QDKPFLSAWPGTVVSEGQHV D0 TLQCRSRLGFNEFSLSKEDGMPVPELYNRIFRNSFLMGPVTPAHAGT

YRCCSSHPHSPTGWSAPSNPVVIMVTGVHRKPSLLAHPGPLVKSGE D1 TVILQCWSDVRFERFLLHREGITEDPLRLVGQLHDAGSQVNYSMGPM

TPALAGTYRCFGSVTHLPYELSAPSDPLDIVVV GLYGKPSLSAQPGP D2 TVQAGENVTLSCSSRSLFDIYHLSREAEAGELRLTAVLRVNGTFQANF

PLGPVTHGGNYRCFGSFRALPHA WSDPSDPLPVSVTGNSRHLHVL I Transmembrane Cytoplasmic tail GTSVVIIPFAILLFFLLHRWCANKKNAVVMDQEPAGNRTVNREDSDEQ ITIM DPQEVTYAQLNHCVFTQRKITRPSQRPKTPPTDTSV

Fig. S1 Screening the binding partners for HHLA2 and KIR3DL3. (Related to Fig. 1)

Fig. S1 (A) HHLA2 expression was induced on activated T and NK cells. T cells were stimulated with human T-activator CD3/CD28 Dynabeads. NK cells were activated by rhIL-2/rhIL-15, irradiated autologous PBMCs, and anti-CD3 mAb. HHLA2 expression was examined by anti-HHLA2 mAb clone B5B5 (open) or mIgG1 (shaded) by flow cytometry. Data are representative of three independent experiments.

(B) Schematic diagram of screening IgSF members to identify the receptor(s) for HHLA2.

(C) Gating strategy for transduced 3T3 cells. (a) Gate for total 3T3 cells based on the FCS-A and

SSC-A; (b-c) Doublets were excluded based on plotted in FSC-H/FSC-A and SSC-H/SSC-A. (d)

From the single cell gate, dead cells were excluded based on DAPI staining; (e) YFP positive cells were gated.

(D-E) Screening other binding partners for KIR3DL3 and cross-reactivity of 26E10 by high- throughput cell microarray. (D) Schematic diagram of screening. (E) The binding of CTLA-4-Ig

(left) and KIR3DL3-Ig (right) to the target-expressing cells was shown.

(F) The protein sequence of KIR3DL3. The extracellular domain of KIR3DL3 was composed of tandem D0-D1-D2 domains. The cytoplasmic tail of KIR3DL3 contained an immunoreceptor tyrosine-based inhibition motif (ITIM). Domains were predicted and annotated based on

UniProtKB (Q8N743).

A 26E10 34B10 3B7 8G7 14F8 15D2 29H7 37A7 51C3

KIR3DL1

KIR3DL2

KIR3DL3

KIR2DL1

KIR2DL2

KIR2DL3

KIR2DL4

KIR2DL5

Anti-KIR3DL3

B KIR3DL3-D0-Ig C KIR3DL3-Ig Clone KD (nM) hIgG 0.8 26E10 34B10 3B7 0.51± 0.007 1.0 1.0

m Kd=0.17 nM Kd=1.51 nM

n 8G7 0.25± 0.004

5 0.6 0.8 167 nM 0.8

0

4 g

42 nM 14F8 0.69± 0.009

g

n e 0.6 i 0.6 167 nM

n 21 nM

c d

0.4 i

n d

n 84 nM

i a n 0.4

i 0.4 15D2 0.56± 0.009 B

b 42 nM

r B

o 0.2 21 nM s 0.2 0.2

b 29H7 1.13± 0.016 A 0.0 0.0 0.0 200 400 600 800 200 400 600 800 37A7 8.50± 0.14 2266E1E100 334B104B10 Time(S) Time(S) Time (s) 51C3 0.19± 0.005

Fig. S2 Fig. S2 Specificity and affinity of anti-KIR3DL3 monoclonal antibodies. (Related to Fig. 2) (A) The specificity of anti-KIR3DL3 mAbs tested by incubating KIR3DL1/3T3, KIR3DL2/3T3,

KIR3DL3/3T3, KIR2DL1/3T3, KIR2DL2/3T3, KIR2DL3/3T3, KIR2DL4/3T3, and

KIR2DL5/3T3 with indicated anti-KIR3DL3 antibody clones (open) or mIgG (shaded).

(B) The binding of 26E10 and 34B10 to immobilized KIR3DL3-Ig, KIR3DL3-D0-Ig or hIgG detected by ELISA.

(C) The affinity of anti-KIR3DL3 mAbs. Left: Kinetic binding curves for 26E10 and 34B10 as detected by the Octet Red96 BLI instrument. Right: Data were acquired from kinetic binding curves from indicated clones.

In A-C, data are representative of three independent experiments.

26E10 34B10 3B7 8G7 14F8 15D2 29H7 37A7 51C3

KIR3DL3*004 WT

KIR3DL3*001 D0

KIR3DL3*006 D0

KIR3DL3*010 D0

KIR3DL3*033 D0

KIR3DL3*039 D0

KIR3DL3*043 D0

KIR3DL3*047 D0

KIR3DL3*068 D0

KIR3DL3*070 D0

KIR3DL3*091 D0

Anti-KIR3DL3 Fig. S3 Fig. S3 Anti-KIR3DL3 monoclonal antibodies recognized different KIR3DL3 variants.

(Related to Fig. 2) The binding of anti-KIR3DL3 mAbs (open) and mIgG (shaded) with the 3T3 cells expressing wild type and D0 domain mutant KIR3DL3*004. Data are representative of two independent experiments.

A GeneGen exepexpressioressionn foforrKIR3DL KIR33D(ENSG0L3 (E0000242NSG019.1)0000242019.1) 0.0.5050

0.40

) 0.40

1

+

) M

1 0.30

0+ .30

P

M

P

T

(

T

0

(

1

g

o

0 l 0.20

1 0.20

g

o L 0.0.1010

0.0.0000 Whole VagVBloodUUterusThyroidTTestisSStomachSSpleenSmallS IntestSSkin - SunPProstate ExposedPituitaryP Pancreas (LowerOvaryO leg) Nerve Muscle- Tibi Minor - Skeletal SaliLung Liver KidneyHeart H- Cortex -Fallop Lef EsophagusE ColonC - -ColonGastroesophageal Transverse -CervixC SigmoidC Cells- Ec -CJunct CellsEB - BreastCulturedB Brain- Mammary -Blad BAmyg TissueArteryA AdrenalA- AortaAdiposeA Gl - Subcutaneous W ina T P N M M L L K F C B d r a t h e t p m k r i v i e ian Tubes o e e e r l er t d d e o l i o t a e i u v i a o r V e r a e h g y s e a inne - Terminalu Ileumn a ua n vary Glande d a t Ventricleo l l tocervixl -transformedll lymphocytesa d r andip o i r r t m s i r r sl n ll p o o v l- - a i dala r e n u o i e ll t t c y v o g r n r o n i E C s fi n d y n o le a s s a n a a r e c r e t p h - n x t brob e s id In t r e - l a T - B u l a b c e y a T e S y ia S V l asts r l e h te i -S a n g ra i tu G T lo s s b l u g -t r o t ia k iv T s n m ion r e la is d in l e a u s o a d n s e le r b v n d u t y e e id s fi e a G rs fo b l e r ro la m b n e l d d a s ly ts m p h o c y te s B

a b c

A

-

H

H

H

-

-

-

C

C

C

C

S

S

S

S

S

S

F S

FSC-A FSC-A SSC-A DAPI

d e f

δ

/

6

γ

8

5

D

R

D

C

C

C T

CD3 CD3 CD4 g h

TCM TN

6

7

5

R

D

C

C C

TEM TEMRA

CD3 CD45RA

C D CD56bright CD16- NK CD56dim CD16+ NK

CD8 TN CD8 TCM CD8 TEM CD8 TEMRA

A

A

- -

0.1% 0.1% 0.3% 5.5% C

C

S S

S 35.1%

S 5.7%

Anti-KIR3DL3 Anti-KIR3DL3 Fig. S4 Fig. S4 KIR3DL3 was expressed on innate and adaptive immune cells. (Related to Fig. 2)

(A) KIR3DL3 mRNA levels in normal human tissues from the Genotype-Tissue Expression

(GTEx) database.

(B) Gating strategy for immune subsets from PBMCs. (a) Gate for lymphocytes based on the

FCS-A and SSC-A from PBMCs; (b) Doublets were excluded based on plotted in FSC-H/FSC-A and SSC-H/SSC-A. (c) From the single cell gate, dead cells were excluded based on DAPI staining; (d) CD3+ γδ TCR+ cells were defined as γδ T cells; From CD3/CD56 plot (e), CD3+ T cells were then divided into CD4+ and CD8+ subsets (f); (g) NK cells were further divided into

bight dim + + CD56 and CD56 subsets; (h) From CD8 CD3 gate, CD8+ T cells then divided into TN,

TCM, TEM and TEMRA based on the expression of CD45RA and CCR7.

(C) KIR3DL3 expression on CD8+ TN, TCM, TEM, and TEMRA cells from one donor.

(D) KIR3DL3 expression on CD56bright CD16- and CD56dim CD16+ NK cells from one donor. A B C KIR3DL3/Jurkat Control Raji HHLA2/Raji % of Max % of Max

Anti-KIR3DL3 Anti-HHLA2

E D Redirected cytotoxicity assay A - anti-CD3/CD28 feeder cells SSC IL-2

Anti-KIR3DL3 Anti-KIR3DL3 F a b c d e H A A H A - - - - - SSC FSC SSC SSC SSC

FSC-A FSC-A SSC-A PE 7-AAD G H Target cell: Raji Target cell: K562 P=0.0027 P=0.0316 100 40 40 40 6 P=0.0047 P=0.0408 P=0.0179 80 30 30 30 T (%) T (%) cells (%) +

+ 4 +

60

20 20 CD8 20 CD8 CD8 + + + α 40 γ 2 10 10 10 Specific Lysis (%) Specific Lysis IFN- Specific Lysis (%) Specific Lysis 20 TNF-

0 0 CD107a 0 0 0 CD3 + + CD3 + + CD3 + + CD3 + + CD3 + + KIR3DL3 - + KIR3DL3 - + KIR3DL3 - + KIR3DL3 - + KIR3DL3 - +

I a b c J 60 P<0.0001 A H H - - - P=0.9998 40 FSC SSC SSC

20 Specific Lysis (%) Specific Lysis FSC-A FSC-A SSC-A f e d 0 CD3 - - + CD16 - + - A - A A - - SSC SSC SSC

IFN-ɣ Live/dead PE g h A A - - SSC SSC

TNF-ɑ CD107a Fig. S5 KIR3DL3 exerted an inhibitory effect on CD8+ T cells. (Related to Fig. 3)

(A) KIR3DL3/Jurkat cells were generated and examined for KIR3DL3 expression by 26E10

(open) or mIgG1 (shaded).

(B) Control Raji and HHLA2/Raji were generated and examined for HHLA2 expression by anti-

HHLA2 mAb B5B5 (open) or mIgG1 (shaded).

(C) The cartoons depict KIR3DL3+ Jurkat forming conjugates with HHLA2-YFP+ Raji (top) or

Control-YFP+ Raji (bottom) pre-pulsed with superantigen SEE.

(D) KIR3DL3+ CD8+ T cells were sorted and expanded in vitro. The expression of KIR3DL3 was detected by 26E10 (open) or mIgG1 (shaded).

(E) The cartoon illustrates the principle of the redirected cytotoxicity assay.

(F) Gating strategy for the cytotoxicity assay. (a) Gate for total cells based on the FCS-A and

SSC-A; (b-c) Doublets were excluded based on plotted in FSC-H/FSC-A and SSC-H/SSC-A. (d)

PKH-26 labeled target cells were gated. (d) Dead cells were gated.

(G-H) KIR3DL3+ CD8+ T redirected cytotoxicity assay against K562 and Raji. (G) Lysis of

Raji (left) or K562 (right) cells at an E:T= 2:1. (H) The degranulation (CD107a) and cytokine production (IFN-γ and TNF-α) of KIR3DL3+ CD8+ T cells targeting Raji cells.

(I) Gating strategy for the degranulation and intracellular staining of CD8+ T cells. (a) Gate for total cells based on the FCS-A and SSC-A; (b-c) Doublets were excluded based on plotted in

FSC-H/FSC-A and SSC-H/SSC-A. (d) PKH-26 negative CD8+ T cells were gated. (e) Dead cells were excluded. (f-h) IFN-γ+, TNF-α+ and CD107a+ CD8+ T cells were gated.

(J) KIR3DL3+ CD8+ T redirected cytotoxicity assay against P815 in the absence or presence of anti-CD3 or anti-CD16 (n=4). Data are represented as mean ± SD. In A, B and D, data are representative of three independent experiments. P values by a two-tailed paired t-test (G-H) and a one-way ANOVA (J).

A B Transcription Factor Signaling Molecules 2 2 SIPA1L2 2 NR4A2 SIPA1L2PA1L2 1 1 1 SPRY2 0 CD8 T FOS SPRY2Y2

EMRA A 0 0

- LAT2 -1 KIR3DL3- KIR3DL3+ IKZF2 -1 LAT2T2 --12 NCALD

SSC -2 -2

CCR7 BAZ2B NCALD TRIO

- - - TRIO NOG D2+ D1 D2 D3 D1+ D3+ NOG CD45RA KIR3DL3 RGS1 RGS1GS1 Surface Proteins RNF157 2 RNF157 TIMD4IMD4 SESN3 1 SESN33 C LRRN3 LRRN3 0 HPCAL4 HPCAL4PCAL4 CD248 -1 PTPRK Control K562 HHLA2/K562 PTPRKPRK DPP4PP4 -2 HK2 NRCAMNRCAM CERK CERKERK DSC1SC1 SH3RF3 SH3RF33RF3 CNR2 PDE7A PDE7AE7A SMIM10SMIM10M10 DDHRS3HRS3

% of Max HRS3 METTL7ATL7A - - - CD63 D2+ Anti-HHLA2 D1 D2 D3 D1+ D3+ FCRL3FCRL3 - - - D1 D2 D1+ D2+ D3+ D3

D E KO Scrambled ctrl HHLA2 7000 HCC827 HCC827 mIgG1 mIgG1566.1 6000 566.1B5B5 I Ig binding Ig 26E10 - F MFI M 5000 34B10 MFI of MFI B5B5 4000 % of Max 500 26E10 KIR3DL3 250 0 34B10 0 1.25 2.5 Anti-HHLA2 Concentration (μg/ml)

Fig. S6 KIR3DL3+ CD8+ T cells displayed NK-like phenotype and function. (Related to Fig.

3)

+ - + (A-B) RNA sequencing of KIR3DL3 and KIR3DL3 CD8+ TEMRA (n=3). (A) KIR3DL3 and

- KIR3DL3 CD8+ TEMRA for were sorted from fresh PBMCs. (B) The heatmaps of DEGs between

+ - KIR3DL3 and KIR3DL3 CD8+ TEMRA.

(C) Control K562 and HHLA2/K562 were generated and examined for HHLA2 expression by

B5B5 (open) or mIgG1 (shaded).

(D) Scrambled control and HHLA2KO HCC827 were generated and examined for HHLA2 expression by B5B5 (open) or mIgG1 (shaded). (E) Anti-KIR3DL3 mAbs blocked the binding of KIR3DL3 to HHLA2. The binding of

KIR3DL3-Ig to HHLA2/3T3 cells in the presence of anti-HHLA2 mAbs (Clone B5B5 and

566.1), anti-KIR3DL3 mAbs (Clone 26E10 and 34B10) or mIgG1 at indicated concentrations were detected by flow cytometry.

In C-E, data are representative of three independent experiments.

A B Enriched primary + NK cells KIR3DL3 NK KIR3DL3- NK92 KIR3DL3+ NK92

anti-CD3 98.9%

A 60.1% 2.30% 0.58% -

feeder cells A 0.45%

-

C C

S IL-2; IL-15

S

S S

Anti-KIR3DL3 Anti-KIR3DL3 Anti-TMIGD2 Anti-KIR3DL3 Anti-KIR3DL3

D C KIR3DL3+ NK92 KIR3DL3- NK92

Control A427 HHLA2/A427 100 100

) %

( 80 80

s

i s

x 60 60

y

a

L

M

c

i 40 f

f 40

i

o

c

e Control Raji %

p 20 20

S HHLA2/Raji 0 0 Anti-HHLA2 E:T 1:10 1:5 1:2 1:1 1:10 1:5 1:2 1:1 E G

30 80 ULBP-1 ULBP-2/5/6 ULBP-3

) %

( 60 s

i 20

s

x

y a

L 40

M

c

i

f f

i 10

c

o

e Control K562 20

p Control A427 % S HHLA2/K562 HHLA2/A427 0 0 E:T 1:1 2:1 5:1 10:1 2:1 5:1 10:1 20:1 MIC-A/B CD48 CD155

F DAPI YFP KIR3DL3 Merge

x

a M

HHLA2/ f

A427 o

%

CD112 ICAM-1 ICAM-2 Control

A427

x

a

M

f

o

HHLA2 Control % Target cell P value /A427 A427 No. of conjugates imaged 73 46 N/A No. of KIR3DL3 clustering (%) 65 (89%) 9 (19%) 0.001 No. of YFP clustering (%) 65 (89%) 3 (6%) 0.001 No. of both KIR3DL3 and YFP 62 (85%) 1 (2%) 0.001 clustering (%) No. of neither KIR3DL3 nor YFP 5 (7%) 34 (74%) 0.001 clustering (%)

Fig. S7 Fig. S7 KIR3DL3 colocalized with HHLA2 at the NK92 cell immunological synapses and suppressed NK92 cell function. (Related to Fig. 4)

(A) Primary KIR3DL3+ NK cells were sorted and cultured in vitro. KIR3DL3 expression was examined by 26E10 (open) or mIgG1 (shaded). (B) Flow cytometry analysis of TMIGD2 and KIR3DL3 expression on NK92 cell line (left).

KIR3DL3 expression on sorted KIR3DL3+ and KIR3DL3- NK92 cells was examined by 26E10

(right).

(C) Control A427 and HHLA2/A427 were generated and examined for HHLA2 expression by

B5B5 (open) or mIgG1 (shaded).

(D-E) Lysis of HHLA2- versus HHLA2+ tumor cells by NK92 cells. (D) Control Raji or

HHLA2/Raji cells were co-incubated with KIR3DL3+ or KIR3DL3- NK92 cells at the indicated

E:T ratios. (E) Control K562 or HHLA2/K562 cells (left) and control A472 or HHLA2/A427 cells (right) were co-incubated with KIR3DL3+ NK92 cells. Data are mean for duplicate measurements.

(F) NK92 cell-A427 conjugate assay. Top: Representative confocal images of cell conjugates after KIR3DL3+ NK92 cell-A427 contact followed by staining with anti-KIR3DL3 and DAPI.

YFP was shown in yellow. Scale bars, 10 휇m. Bottom: The table shown is the frequencies of

KIR3DL3 and/or HHLA2-YFP (or control-YFP) clustering at intercellular conjugates. Mouse

Tim-3-YFP/A427 was used as the control. P values by Fisher’s exact test.

(G) The expression of activating ligands on HCC827. HCC827 was stained by the indicated markers (open) and isotype control (shaded).

In A-E, G, data are representative of two-three independent experiments.

A B C 1 2 WT Y381F mIgG1 + - + - α-KIR3DL3 - + - + 3 VO4 - + - + (kD) (kD) VO4 - - + + 4 WT

3 pTyr 55

L KIR3DL3 D

55

:

3

P

R

I

I

P

I K

- WT Y381F α SHP-1 70 2 VO4 - + - + 1 3 70 4 KIR3DL3 Y381F

70 SHP-2

WT Y381F KIR3DL3 55 VO4 - + - + KIR3DL3 WT

70 D KIR3DL3 Y381F L C KIR3DL3 KIR3DL3 WT W 0.4

70 KIR3DL3 Y381F e

L SHP-1 e

c 1

C

c

n

n e

W 0.3

r e r

e 2

f

e f

e 3

e

R

R

70 SHP-2 o

t 0.2 o

4

t

e

e

v

i

v

i

t

t

a

a

l l

e 0.1

e

R R β-actin 40 0.0 ) 7 ) ) ,8 )9 )3 in 41 S 37 nin 0Y ) S9/ 74 en / 7 /1 S t Y25 12 ( tea /8 8 (S (S4 -ac 21 / 3 Β 0 T /Y1 (S2/β //32 -C 2, 5 α 21 β 4 8 /3β t/ (T0 1 α- k1 2 K At //2T T -3S k 12 0 G A 2K SK T G ER ( /2 1 K R E

E WT Y381F F 26E10 - - - - + + + + anti-CD16 + + + + + + + + Min. 0 2 5 10 0 2 5 10 (kD) GαM 0 2 5 10 0 2 5 10 (kD)

pVav1 90 pVav1 90

100 Vav1 Vav1 100

pERK1/2 pERK1/2 40 40

ERK1/2 ERK1/2 40 40

pAKT 60 pAKT 60

AKT AKT 60 60

pNF-κB 60 pNF-κB 60

β-actin 40 β-actin 40

Fig. S8 Human phospho-kinase array screen for KIR3DL3 downstream signalingFi gpathway. S8 s. (Related to Fig. 5) (A) The entire blots for Figure 5B.

(B) The entire blots for Figure 5E.

(C-D) A human phospho-kinase array of WT or mutant KIR3DL3 NK92 cells stimulated with

HHLA2/Raji. Kinase spots showing significantly different densities between WT and Y381F are indicated (C). Relative quantification of the phosphorylation level of indicated kinases (D). Data are mean ± SD. Data are representative of three independent experiments.

(E) The entire blots for Figure 5F.

(F) The entire blots for Figure 5H.

A B + + Renal cell Merkel cell Breast cancer Adult T cell NK cells CD8 T C HD CD4 CD7

carcinoma carcinoma GSE7904 Leukemia/Lymphoma ) ATLL CD4+CD7+ ) 25

80 %

% ( ) + - GSE53757 GSE19069 ( GSE39612

P=0.0285 ATLL CD4 CD7

)

%

s

T (

l 20

P=0.2607 l

+

% s

P=0.0164 e

] P=0.0076

P<0.0001 l

P=0.0005 (

8

l c

60

A 15 8 7 8 12 Normal e

D 15 P=0.1018

T

K

c

)

N

C

8

N

K R

Tumor %

D

+

+ ( 10 N 10

40

C

m 3

3 10

+

6 6 6 +

+

L

L 3

3 P=0.0058

3 3

L 5

L

D

D L

8 L

D 3

KIR3DL3 3

D

D D

3 20 5

3 R 3 R

4 5 4 3

R I

I 0 I

mRNA R

R

I R

6 K

K

I

K

I

K K -5 K [ 0 0 2 2 4 2 4 HD ATLL

g HHDD ATATLLL HD ATLL o L 2 -5 0 3 0 )

3 ) 4 + + -

%

)

%

(

(

) % ] + CD7

20 T s

P<0.0001 5 P=0.1177 P=0.4248 8 P=0.7786 ( + CD7 + CD7 l

15 3 P=0.1169

l %

A P=0.1876 6 4 + s 4

( P=0.0129 e

l D

l D

N

) 8

c C

e C

K 2 R

D L

c T

% L

0 2 L (

N HD CD4 L

C

m 8

K 4

6 + AT AT D

10 +

+ 2

N P=0.0121

2

2

C

2 +

HHLA2 A

1

A

2

+

A A L -20

4 L

1 2

A L L 2

mRNA H

A

L

H

H

H L

H 0

H H

[ 5 4

H

H

H 2

-40 H H g 0 o 0 -1 L HD ATLL -60 HHDD ATATLLL HD ATLL 0 3 2 -2

+ + - P<0.0001 ) P=0.0019 ] P=0.1255 P=0.1286 P=0.0052 P=0.0031 ) 100 P<0.0001

100 %

A 9 9 12 CD7

) + )

10 (

% + CD7 + CD7

( N

4 4

%

% T

( P=0.0058

s D

A 100 D

l

( 80 R

l C C 80 +

s N

8 l

e L

) l

m L 8 10 K P=0.0042

c L L

R e

60 HD CD4

2

D

N % 8 c 80 8 T AT

AT

(

m 60

D C

TMIGD2 +

6 8

K

+

2

+

2

G D

N 40

2

I 8 2 mRNA D

C 60

+

D

D

2 40

M

+ D

4 G

G

I

2 G

D 20 T

7 7 I

I

G

[

D

I

G 2

M 40 I

M 6

M

G g

T 20 I

2 M M

T 0

T

o

T

T

M L

T 20 0 6 6 4 0 -20 HD ATLL HHDD ATATLLL HD ATLL 0 + + -

+ CD7 + CD7 + CD7 4 4 D D C C L L D HD CD4 L L AT AT a b c Live CD45+ cells

51.8

H H

5 -

PBMCs - Single cells

H

4

- C

62.4 C 90.5

D C

Single cells S

S

C

F S

S 98.2 F

SSC-H SSC-A FSC-A Live/Dead Blue g f d e

CD3+CD4+ CD3+ CD3- )

CD4+ 9 ATLLs 1 Normal CD4 T 45.3

89.5 9.90 D C

CD8 T +

4

3

4 1

4 21.7 D

D CD56+ NK

D

D

C C

C 38.5

C CD3- (

5.74 e

g CD3+ a

e 66.4

n

i L CD7 CD8 CD3 CD56 E F + Age/Sex Diagnosis KIR3DL3 NK92 KIR3DL3- NK92 KIR3DL3/DAPI 25/F ATLL KIR3DL3/DAPI

82/F ATLL

73/F ATLL

56/F ATLL 69/M ATLL

75/F ATLL 43/F ATLL Fig. S9 Fig. S9 KIR3DL3 was upregulated in solid and hematopoietic tumors. (Related to Fig. 6)

(A) The mRNA expression of KIR3DL3, HHLA2, and TMIGD2 in human tumors versus corresponding normal tissues by analyzing Gene Expression Omnibus (GEO) databases. (renal cell carcinoma, n = 72 versus 72; Merkel cell carcinoma, n = 27 versus 64; breast cancer, n = 43 versus 7; adult T cell leukemia/lymphoma, n = 12 versus 10 samples of human tumor versus normal tissues). Data are mean ± SD.

(B-C) KIR3DL3, HHLA2, and TMIGD2 expression on peripheral blood from healthy donors

(HD) versus adult T cell leukemia/lymphoma (ATLL) patients by flow cytometry. Data shown are the frequencies of the KIR3DL3+, HHLA2+, or TMIGD2+ cells on NK and CD8+ T cells (B), normal (CD4+CD7+) and atypical (CD4+CD7-) CD4 T cells (C). HD: n=7 (HHLA2) or 10

(KIR3DL3 and TMIGD2), ATLL: n=7. Data are mean ± SD.

(D) Gating strategy for ATLLs and normal immune cell subsets in PBMCs from ATLL patients.

(a) The major cell population was gated based on the FCS-A and SSC-A; (b) Doublets were excluded based on plotted in SSC-H/SSC-A and FSC-H/FSC-A. (c) Live CD45+ cells were gated from single cells based on CD45 and Live/Dead blue staining; (d) Lineage CD14+ and CD19+ cells were excluded from live CD45+ cells; (e) CD3- CD56+ cells were defined as NK cells; (f)

CD3+ T cells were then divided into CD4+ and CD8+ subsets; (g) ATLL tumor cells (CD4+CD7-) and normal CD4 T cells (CD4+CD7+) were gated based on CD4 and CD7 staining.

(E) The age and sex of ATLL patients.

(F) Validation of anti-KIR3DL3 antibody clone 1136B. Immunofluorescence images of positive staining on KIR3DL3+ NK92 cell line and negative staining on KIR3DL3- NK92 cell line.

Images are representative of three independent experiments.

P values by a two-tailed unpaired student t-test (A, B) and a one-way ANOVA (C).

A C 3×108 HCC827-luc2 Primary KIR3DL3- NK (i.v.) mIgG1 inoculation (i.v.) mIgG1 or 26E10 (i.v.) 26E10

x 2×108

u

l

f

l

a

t o T 1×108 5 d 22 d Analysis of tumor growth 0 0 5 10 15 20 25 Days after NK transfer

mIgG1 26E10

B Ventral Dorsal Ventral Dorsal Luminescence

D1

2.0×107

D5

1.0×107

D10

Radiance (p/sec/cm2/sr) D15 Color Scale Min=1.2×105 Max=3.0×107

Fig. S10 Fig. S10 The therapeutic efficacy of KIR3DL3 blockade was dependent on KIR3DL3 expressed by NK cells. (Related to Fig. 7)

HCC827 lung metastasis mouse model treated with primary KIR3DL3- NK cells. (A) Outline of the experiment procedure. (B) Bioluminescence images by dorsal and ventral imaging of each mouse at indicated days. (C) Tumor growth in the treatment of 26E10 or mIgG1 group (n=5).

Each arrow indicates treatment with primary KIR3DL3- NK cells and the antibody. Data are mean ± SD of the average of total flux from dorsal and ventral imaging for each mouse. i.v. intravenously.

Table S1. Gene list for NK cell-mediated cytotoxicity pathway in KEGG database

+ Probe Core enrichment in KIR3DL3 CD8+ TEMRA KIR3DL1 Yes TYROBP Yes KLRC2 Yes KLRC3 Yes KIR2DL4 Yes SH2D1B Yes KIR2DS4 Yes KIR2DL3 Yes NCR1 Yes KIR2DL1 Yes PLCG2 Yes SYK Yes NCR3 Yes KIR3DL2 Yes ICAM1 Yes ITGB2 No PRKCG No SH3BP2 No VAV3 No RAC3 No CD244 No PIK3R1 No KLRD1 No RAC1 No PIK3CA No TNFRSF10A No NFATC1 No NFATC3 No SHC2 No PIK3CG No SH2D1A No CHP1 No HLA-A No BID No HRAS No RAC2 No HLA-C No MICA No MICB No ULBP3 No PAK1 No MAPK1 No PIK3CB No PRF1 No FCGR3A No HCST No GZMB No SOS2 No PPP3R1 No HLA-B No NRAS No PRKCA No SHC1 No ITGAL No FCER1G No FAS No SHC4 No MAP2K2 No PTPN11 No PPP3CC No FCGR3B No PPP3CB No KLRC1 No TNF No GRB2 No PRKCB No IFNG No KLRK1 No VAV1 No HLA-E No LCK No KRAS No MAP2K1 No CD48 No PTK2B No PIK3R5 No PTPN6 No PIK3R3 No SOS1 No NFAT5 No VAV2 No LCP2 No BRAF No FASLG No PPP3R2 No NFATC2 No SHC3 No PPP3CA No CASP3 No ARAF No TNFSF10 No CD247 No PIK3CD No HLA-G No RAF1 No FYN No MAPK3 No IFNAR1 No IFNGR2 No ICAM2 No IFNAR2 No IFNGR1 No PLCG1 No TNFRSF10B No ZAP70 No TNFRSF10D No LAT No

Table S2. Gene list for T cell receptor signaling pathway in KEGG database

- Probe Core enrichment in KIR3DL3 CD8+ TEMRA AKT3 No MAPK12 No NFKBIE No VAV3 No PIK3R1 No PIK3CA No NFATC1 No NFATC3 No PIK3CG No TEC No CHP1 No MAPK11 No HRAS No MAP3K8 No CDC42 No RHOA No PAK1 No CBLB No CDK4 No PAK2 No MAPK1 No PIK3CB No SOS2 No PPP3R1 No NRAS No MAPK13 No PDPK1 No RELA No PAK4 No CD3D No CARD11 No NFKBIB No MAP2K2 No PPP3CC No AKT1 No NFKB1 No NCK1 No MAP2K7 No CD3G No CD3E No PTPRC No PPP3CB No TNF No GRB2 No CD8A No IFNG No VAV1 No MAP3K7 No PAK6 No LCK No KRAS No GSK3B No MAP2K1 No PIK3R5 No PTPN6 No PIK3R3 No SOS1 No CHUK No NFAT5 No VAV2 No PRKCQ No LCP2 No PPP3R2 No RASGRP1 No NFATC2 No MALT1 No IL5 No MAPK14 No CBL No PPP3CA No CD4 No CD247 No BCL10 No DLG1 No PIK3CD No AKT2 No NFKBIA No JUN No IKBKB No NCK2 No RAF1 No MAP3K14 No IKBKG No FYN No MAPK3 No MAPK9 No CD8B Yes CD40LG Yes PLCG1 Yes ZAP70 Yes PDCD1 Yes ITK Yes ICOS Yes GRAP2 Yes CTLA4 Yes LAT Yes FOS Yes CD28 Yes