INTRODUCTION
T-cell engager (TCE) therapies have emerged as a transformative immunotherapeutic strategy by redirecting endogenous T cells to eliminate malignant cells in a major histocompatibility complex (MHC)-independent manner
1-4. Among these agents, CD3×CD20 bispecific antibodies have shown substantial clinical activity in patients with relapsed or refractory (r/r) B-cell non-Hodgkin lymphoma (B-NHL)
5-8. However, primary nonresponse and acquired relapse remain common and the underlying mechanisms are incompletely defined, limiting the durability of clinical benefit
9,10.
Mechanistic studies have primarily emphasized tumor-intrinsic resistance, including target antigen loss, adaptive PD-L1 upregulation
11-13, and exhaustion programs driven by sustained TCR signaling
12,14,15. Yet these features are not consistently detected in patients who progress on TCE therapy, suggesting that they do not fully account for clinical resistance. Emerging evidence therefore points to additional mechanisms originating from the tumor immune microenvironment that may critically limit TCE efficacy. In particular, whether tumor-infiltrating myeloid populations directly restrain TCE-engaged T cells through contact-dependent interactions remains poorly defined.
Myeloid cells constitute a dominant immune compartment in B-NHL
16 and are increasingly recognized as critical regulators of T-cell function
17,18. Beyond antigen presentation and cytokine production, tumor-infiltrating myeloid cells (TIMCs) can modulate T-cell immunity through metabolic competition, expression of inhibitory ligands, and remodeling of the stromal niche
19,20. Because TCE efficacy requires rapid, durable T-cell activation within a myeloid-rich microenvironment
21,22, suppressive signals from myeloid cells may disproportionately blunt TCE-driven cytotoxicity. However, the specific myeloid-derived pathways that constrain TCE-engaged T cells and thereby promote resistance remain incompletely defined
23.
CD93 is a transmembrane C-type lectin-like receptor expressed across endothelial and myeloid compartments
24-26. While CD93 has been linked to angiogenesis, vascular remodeling, and tumor progression
25,27, its immunomodulatory functions are not fully understood. In silico studies associate CD93 expression with macrophage polarization
28, and CD93 expressed by peritumoral monocytes or intratumoral vasculature has been proposed to impair T-cell function
28,29. However, whether CD93 directly regulates T-cell activity and whether it contributes to resistance in T-cell-redirecting immunotherapies has not been established.
Here, leveraging single-cell multi-omics profiling of B-NHL specimens collected before and after TCE therapy, we identify the enrichment of immunosuppressive TIMCs as a defining feature of therapeutic resistance. Mechanistically, we show that CD93 is induced downstream of BTK signaling in TIMCs and directly binds to CD69 on TCE-activated T cells to suppress their proliferation and activation. Importantly, disrupting the CD93–CD69 interaction restores TCE-driven anti-tumor immunity. Collectively, our findings define a previously unrecognized myeloid−T cell inhibitory axis that constrains TCE efficacy and provides a rational target to overcome resistance.
RESULTS
Immunosuppressive TIMCs were enriched in TCE-treated r/r B-NHL
First, we utilized single-cell RNA sequencing (scRNA-seq) to profile peripheral and tumor-infiltrating immune cells in B-NHL. 51 samples (20 peripheral blood (PB) and 31 tumor tissue samples) from 35 B-NHL patients, including 16 with follicular lymphoma (FL), 18 with diffuse large B-cell lymphoma (DLBCL), and 1 with mantle cell lymphoma (MCL), were collected (Supplementary Table S1). After stringent quality filtering, we identified a total of 459,984 cells, comprising 110,516 CD8+ T cells, 98,442 CD4+ T cells, 20,444 natural killer (NK) cells, and 79,639 TIMCs including 46,374 monocytes/macrophages and 33,265 neutrophils, in addition to 139,450 B cells, 1,905 endothelial cells, 853 platelets/erythrocytes and 8,735 other cells (Fig. 1a–e). To set stricter controls for TCE-resistant patients, we split patients into three groups based on their clinical information: the newly diagnosed group, the untreated group (r/r B-NHL patients not receiving TCE therapy), and the treated group (r/r B-NHL patients receiving TCE therapy). Interestingly, subclusters of immune cells exhibited distinct heterogeneity between PB and tumor, as well as between newly diagnosed and r/r, or between TCE-treated and untreated patients, suggesting dynamic spatial transitions and lineage connections upon therapy resistance (Fig. 1a–o). Of note, naïve CD8T_CCR7, proliferating CD8T_MKI67, and exhausted CD8T_CXCR6 T cells were enriched in TCE-treated tumors, while other more differentiated naïve CD8+ T cells decreased in PB, suggesting that TCE probably engaged naïve CD8+ T cells from PB to lymphoma tissue; however, these T cells either paused in the naïve stage or eventually proceeded to the exhaustion phase in r/r B-NHL patients (Fig. 1f, k). The similar spatial transition was not observed in naïve CD4+ T and NK cells, and Tregs (CD4T_FOXP3) were not likely affected upon TCE therapy (Fig. 1g, h, l, m). More interestingly, we observed marked lineage shifts from granulocytic lineage neutrophils to monocytic lineage monocytes in the PB. In tumor tissues, conventional dendritic cells (cDCs; DC_CLEC9A) and macrophages (Macro_CD163 and Macro_C1QC) were significantly enriched, with the former exhibiting higher antigen-presentation scores and the latter exhibiting higher immunosuppressive scores, in TCE-treated r/r B-NHL patients (Fig. 1i, j, n, o). These observations suggest that TCE treatment promotes the recruitment of bone marrow-derived monocytes into the PB and their subsequent differentiation into macrophages within lymphoma tissues.
We specifically identified two distinct macrophage subclusters characterized by the highest immunosuppressive scores among all myeloid cells: C1QC+ macrophages and CD163+ macrophages (Fig. 1i). Simultaneously, we observed that 7 patients from the untreated group subsequently enrolled in the TCE clinical trial (Fig. 1p–t). To dissect the distinct roles of these populations in therapy resistance, we stratified patients based on their clinical response to TCE treatment. Intriguingly, these two subsets exhibited divergent patterns associated with different phases of treatment failure. In pre-treatment tumor tissues, C1QC+ macrophages were significantly enriched in non-responders compared to responders, suggesting that high baseline infiltration of this subset serves as a key driver of primary (intrinsic) resistance (Fig. 1s). In contrast, analysis of samples from patients who initially responded but subsequently progressed revealed a specific enrichment of CD163+ macrophages in the relapsed tumors compared to baseline (Fig. 1n).
To further confirm the characterization of T and myeloid cell compartments in B-NHL, we collected 48 PB samples (n = 21) and tumor (n = 27) biopsy samples from 35 B-NHL patients, including 18 FL, 14 DLBCL and 3 MCL for high-dimensional flow cytometry analysis (Supplementary Fig. S2 and Table S1). Consistent with scRNA-seq, both the early activated naïve CD8+ T cells (CD3+CD8+CD197+CD45RA+PD-1+GZMB+) and the immunosuppressive myeloid cells (CD11b+CD68+CD163+Arg1+) were significantly increased in TCE-treated r/r B-NHL tissues (Fig. 1u–y). These results further validated the observation of the significant increases in the early activated naïve T cells and immunosuppressive TIMCs in TCE-treated r/r B-NHL patients.
CD163+ myeloid cells are accumulated and adjacent to CD8+ T cells in TCE-resistant B-NHL tumors
Accumulating evidence indicates that the enrichment of immunosuppressive myeloid cells is closely linked to resistance to immunotherapy
30-32. To further validate the enrichment of immunosuppressive TIMCs in TCE-resistant patients, we compared pre- and post-treatment paired tumor samples using multicolour immunofluorescence (mIF) and immunohistochemistry (IHC). Quantitative analysis revealed that the expression levels of target tumor antigens remained stable between baseline and progression samples, suggesting that antigen escape is not the primary driver of resistance in this context (Fig. 2a, b, e; Supplementary Fig. S3). In contrast, we observed a significant enrichment of CD163
+ TIMCs in the progressed tumor tissues compared to paired pre-treatment samples (Fig. 2a−d; Supplementary Fig. S4). Importantly, this consistent increase in CD163
+ TIMCs occurred regardless of whether patients received CD19- or CD20- targeting TCEs (Supplementary Fig. S4), indicating that the expansion of immunosuppressive myeloid populations represents a common, shared tumor microenvironmental response to TCE therapy rather than an antigen-specific phenomenon. Moreover, to provide more subset-specific functional evidence, we performed additional cell-sorting experiments. CD11b
+CD163
+ and CD11b
+CD163
− myeloid cells were isolated from tumor tissue of three FL patients, comprising two newly diagnosed patients and one patient with r/r disease. They were tested in an hTCE co-culture system (Supplementary Fig. S5a), and then co-cultured with anti-CD3/CD28-activated human T cells. As shown, compared to the T cells co-cultured with CD11b
+CD163
– myeloid cells, T cells co-cultured with CD11b
+CD163
+ TIMCs had less potential for cell proliferation (Supplementary Fig. S5b). These results suggest that the enrichment of the CD163
+ TIMC subset in human B-NHL tumors may contribute to T-cell suppression. Furthermore, spatial analysis demonstrated that while T cell infiltration increased post-treatment, these T cells were located in significantly closer proximity to CD163
+ TIMCs in the relapsed tumors (Fig. 2f–i). After TCE treatment, CD163
+ TIMCs and T cells showed reduced spatial separation within the tumor microenvironment. Although this observation does not by itself establish directional recruitment, it suggests increased local proximity between these two cell populations, potentially facilitating immunosuppressive myeloid–T-cell interactions. Based on these observations, we hypothesized that the abnormal accumulation and spatial reorganization of myeloid cells represent a critical mechanism of secondary resistance to TCE therapy in B-NHL.
Myeloid cell depletion synergizes with the anti-tumor efficacy of TCE therapy
To functionally validate the hypothesis that myeloid cells limit TCE efficacy, we established a humanized B-cell lymphoma mouse model. We first constructed a human bispecific TCE (hTCE) comprising an anti-CD3 scFv and an anti-CD19 Fab fragment and confirmed its specific cytotoxicity against B-lymphoma cells in vitro (Fig. 3a). In the humanized in vivo model, hTCE treatment significantly inhibited tumor development in a dose-dependent manner (Fig. 3a); however, consistent with our clinical observations, the residual tumors in hTCE-treated mice exhibited significant infiltration of cytotoxic CD8+ T cells and immunosuppressive TIMCs, whereas neutrophils and DCs were not significantly altered (Fig. 3b–d). To determine whether these enriched TIMCs were responsible for limiting the therapeutic response, we treated the tumor-bearing mice with clodronate liposomes or anti-CSF1R antibodies to deplete the myeloid lineage (Fig. 3e, f, j, k). Strikingly, the depletion of myeloid cells resulted in a potent synergistic antitumor effect when combined with hTCE therapy, significantly surpassing the efficacy of hTCE monotherapy (Fig. 3g, h, l, m). Mechanistic analysis revealed that the reduction of CD11b+F4/80+ TIMCs was accompanied by a marked increase in the infiltration of total T cells, cytotoxic CD8+ T cells, and CD62L–CD44+ effector memory T cells within the tumor microenvironment (Fig. 3i, n). These findings functionally confirm that TIMCs play a pivotal role in dampening the antitumor immune response and that targeting these cells can overcome resistance to TCE therapy.
Inhibition of BTK in TIMCs remodels the immunosuppressive microenvironment and synergistically enhances TCE efficacy
Given the pivotal role of myeloid cells in mediating resistance to TCE therapy and the established function of Bruton’s tyrosine kinase (BTK) in regulating myeloid physiology
33-36, we hypothesized that targeting the BTK axis could reprogram TIMCs to overcome resistance. To systematically validate this hypothesis and assess the therapeutic potential of kinase inhibition in a targeted manner, we performed a high-throughput screen using a 533-compound kinase inhibitor library in a co-culture system comprising human B-NHL Raji cells, IL-4-induced immunosuppressive THP-1 cells, and human primary T cells (Fig. 4a). This screen identified BTK as a top candidate: BTK inhibitors (BTKis) simultaneously enhanced T-cell-mediated cytotoxicity and suppressed the expression of the immunosuppressive marker CD163 on myeloid cells (Fig. 4b–e; Supplementary Table S2). To further exclude the possibility that residual kinase inhibitors directly affected T cells or Raji tumor cells in the co-culture system, we treated human primary T cells or Raji cells with individual kinase inhibitors, respectively. As shown in Supplementary Fig. S6a, no inhibitors, including all BTK inhibitors, exhibited direct cytotoxic effects on Raji cells at 1 μM for 24-h treatment except for isorhamnetin 3-
O-neohesperoside, staurosporine, H3B-6527, ONO-7475, and BGT226. Moreover, none of the kinase inhibitors substantially affected T-cell proliferation under the same conditions (Supplementary Fig. S6b). These data suggested that BTK activity was restricted to myeloid cells in the
in vitro co-culture experiments. Following this, we validated the therapeutic potential of BTK inhibition using
in vivo models. In humanized B-cell lymphoma mouse models, the combination of the BTK inhibitor ibrutinib (IBR) and hTCE yielded a potent synergistic antitumor effect compared to monotherapy in IBR-resistant Raji cell-derived xenograft or IBR-resistant DLBCL patient-derived xenograft (PDX) (Fig. 4f, g, j). Flow cytometry analysis of the tumor tissues revealed that this synergy was accompanied by a marked reduction in hTCE-induced immunosuppressive TIMC infiltration and a concurrent increase in the number of tumor-infiltrating effector CD8
+ T cells and memory T cells (Fig. 4h, i).
To dissect the cellular mechanisms underlying this synergy, we performed scRNA-seq on the tumor microenvironment of humanized PDX mice treated with hTCE, IBR, or their combination (Supplementary Fig. S7a, b). Consistent with our clinical observations in human patients, hTCE treatment in mice induced T cell infiltration and a distinct population of immunosuppressive CD163+ macrophages (Macro_Cd163) (Fig. 4k, l; Supplementary Fig. S7c–l). To further explore the cell origin of these immunosuppressive TIMCs, we performed trajectory analysis on mouse myeloid cells and we identified two major trajectories starting from monocytes (Fig. 4m). In trajectory 1, Mono_Klhl6 cells in state 1 differentiated through Macro_Mrc1 cells in a transitional state before terminally differentiating into state 2-localized immunosuppressive Macro_Cd163 cells; however, in trajectory 2, Mono_Klhl6 in state 1 ultimately differentiated into state 3-resident immunostimulatory Macro_Cd74 cells (Fig. 4m–o). In addition, we observed expansion of immunosuppressive TIMCs in state 2, which could be reduced upon combination treatment with hTCE and IBR (Fig. 4p, q). The addition of IBR effectively blocked this trajectory, preventing the accumulation of Macro_Cd163 cells and shifting the myeloid compartment towards an immunostimulatory phenotype (Fig. 4n–q). Furthermore, we found that the recruitment of these myeloid cells might be mediated by the CCL3/4/5-CCR5 axis between activated T cells and myeloid cells (Supplementary Fig. S8a, b). To confirm this requirement in vivo, we performed a competitive migration assay by co-injecting differentially labelled (Violet/CFSE) CD11b+ bone marrow cells from wild-type (WT) or Ccr5–/– mice at a 1:1 ratio into Raji-bearing humanized mice (Supplementary Fig. S8c). While TCE treatment robustly promoted the infiltration of WT myeloid cells into the tumor microenvironment, the loss of Ccr5 dramatically abrogated this recruitment (Supplementary Fig. S8d). These data demonstrate that BTK inhibition remodels the immune microenvironment by both restricting myeloid recruitment and blocking their polarization into immunosuppressive phenotypes, thereby unleashing the antitumor potential of TCE.
CD93 is a downstream effector of BTK signaling that mediates immunosuppression in TIMCs
To elucidate the molecular effectors downstream of BTK that mediate T-cell suppression, we conducted an integrated comparative analysis of three distinct transcriptomic datasets: genes significantly downregulated in BTKi-treated THP-1 cells; genes upregulated in the specific immunosuppressive Macro_CD163 cell cluster identified in our human scRNA-seq atlas; and genes upregulated in the corresponding TIMC population in our mouse model (Fig. 5a). This cross-species intersection identified CD93 as a robust conserved target. We confirmed that CD93 expression was significantly elevated in TCE-treated patient lymphoma tissues and tightly correlated with BTK activity (Fig. 5b–f).
Consistent with previous findings that protein kinase C (PKC) isoenzymes regulate CD93 expression
37, we also found that the mitogen-activated protein kinase (MAPK) signaling pathway was the most enriched pathway downregulated in IBR-treated or
BTK-knockout (
BTK-KO) THP-1 cells according to KEGG analysis (Supplementary Fig. S9a, b). IBR treatment or knockout of
BTK reduced CD93 expression and the phosphorylation of MAPK and related PKC levels (Supplementary Fig. S9c). Moreover, kinase inhibitors targeting the MAPK and PKC signaling pathways increased the cytotoxicity of T cells elicited by hTCE and decreased the expression of CD163 on THP-1 cells (Supplementary Fig. S9d). We concurrently observed that PKC inhibitors (staurosporine and Go6976) and an MAPK inhibitor (pexmetinib) significantly reduced CD93 expression in THP-1 cells (Supplementary Fig. S9e). To further confirm the regulation of CD93 by BTK in primary TIMCs, CD93
+ TIMCs isolated from the tumor tissues of 9 B-NHL patients were treated with BTK inhibitors, including IBR and ZBR (Supplementary Fig. S10), and significant downregulation of CD93 expression was observed (Fig. 5g). Together, these results suggest that CD93 is regulated by the BTK-MAPK-PKC signaling pathway.
To validate CD93 as the functional downstream effector of BTK, we first utilized CD93-knockout THP-1 cells (Fig. 5h, i). We found that CD93 ablation completely abrogated the synergistic enhancement of hTCE cytotoxicity by BTK inhibition, indicating that CD93 is required for this effect (Fig. 5j, k). Given that BTK is expressed in myeloid lineages but absent in CD93-positive endothelial cells, we generated chimeric mice with hematopoietic-specific CD93 deficiency via bone marrow transplantation to isolate the function of myeloid CD93 in vivo (Fig. 5m). In this model, while the BTKi ZBR robustly synergized with mouse TCE (mTCE) in the WT setting, myeloid CD93 deficiency abolished both the standalone antitumor effect of ZBR and its synergy with mTCE (Fig. 5l–o). Notably, CD93 deficiency alone phenocopied the therapeutic effect of BTK inhibition, and ZBR treatment provided no additional benefit in Cd93–/– mice (Fig. 5o). These findings conclusively demonstrate that CD93 acts as the essential obligate downstream effector of BTK signaling in TIMCs.
CD93 on TIMCs exerts direct immunosuppressive effects independent of known ligands
We first characterized the expression and function of CD93 in the tumor microenvironment. We found that CD93 expression in TIMCs was predominantly localized to immunosuppressive macrophages (Fig. 6a, b). To further validate the function of CD93 in human TIMCs, we isolated both CD93
+ and CD93
– TIMCs from 9 human B-NHL patients and co-cultured them with T cells from the same patient, and found that CD93
+ TIMCs had a significantly greater ability to suppress T-cell proliferation (Fig. 6c, d). Moreover, knockout of
CD93 relieved the suppressive effect of TIMCs, thereby restoring T-cell proliferation and cytotoxic cytokine release (Fig. 6e–g). CD93 expression was negatively correlated with the overall survival (OS) of B-NHL patients in another cohort (Fig. 6h–j). We also observed that CD93 expression was increased in hTCE-treated human B-NHL patients and hPBMC-human DLBCL-PDX model mice compared with untreated patients (Fig. 6k–m), and CD93 expression was positively correlated with BTK, as well as CD163 and PD-L1, in immunosuppressive myeloid cells in a scRNA-seq dataset of 17 DLBCL tumors
38 (Fig. 6n). To explore the potential relevance of this myeloid program beyond TCE therapy, we analyzed a public scRNA-seq dataset of multiple myeloma patients before and after B-cell maturation antigen chimeric antigen receptor (BCMA) CAR-T-cell therapy. Within the myeloid compartment, we identified several monocyte/macrophage and DC subsets (Supplementary Fig. S11a, b). CD163
+ monocyte/macrophage subset showed relatively high CD93 expression and enrichment of immunosuppressive programs (Supplementary Fig. S11c, d). Paired analysis further showed an increased proportion of CD163
+ monocyte/macrophage cells after CAR-T-cell therapy, whereas other monocyte/macrophage subsets did not change significantly (Supplementary Fig. S11e, f). Although CD93 expression was not globally increased after CAR-T-cell therapy, these findings suggest that CD93-associated CD163
+ myeloid cells may also be enriched in the context of CAR-T-cell therapy. Further functional studies will be required to determine whether this population contributes to CAR-T resistance.
CD69 is the functional receptor for CD93 on T cells
To identify this unknown receptor, we performed a genome-wide CRISPR/Cas9 screen using an NFAT-GFP T-cell reporter system, where immobilized CD93-extracellular domain (ECD) inhibited activation-induced GFP expression (Fig. 7a). After analyses of the significantly upregulated sgRNAs via multiple comparisons using the comprehensive approach of EVenn, 67 genes were upregulated in every comparison, and 10 of them (Reps1, Tbc1d20, Ap2m1, Hs6st1, Cd274, Cd69, Nras, Adipor2, Mbtps2, and Nisch) presented more than 1 significantly enriched sgRNA (Fig. 7b). After two rounds of verification by individual gene knockout, we found that knockout of CD69 significantly increased GFP expression (Figs. 7c–e). These results suggest that CD69 on T cells may mediate CD93-induced T-cell suppression. To further confirm whether CD93 interacts with CD69 directly, we performed surface plasmon resonance (SPR) assays using CD93-ECD and CD69-ECD proteins bidirectionally. CD93 directly bound to CD69 with a dissociation constant of 0.20–0.37 μM (Fig. 7f; Supplementary Fig. S12a–c). Moreover, we found that either the C-type lectin-like domain (CTLD) or the 5 epidermal growth factor (EGF)-like domains of CD93 mediated the interaction with CD69 (Fig. 7f; Supplementary Fig. S12d–i). Consistent with these interactions, both CTLD and the 5 EGF-like domains of CD93 inhibited T-cell proliferation (Fig. 7g, h). Furthermore, the CD69-ECD protein rescued CD93-mediated suppression of T-cell proliferation and cytotoxicity (Fig. 7i–k). Because CD69 is classically recognized as an early T-cell activation marker, we next examined whether the suppressive effect of CD93 depended on the activation stage of T cells. We first monitored the expression kinetics of CD69 and PD-1 under continuous stimulation, transient stimulation, and restimulation conditions. CD69 was rapidly induced after initial T-cell activation and subsequently declined after removal of stimulation, whereas restimulation re-induced CD69 expression, defining a temporal window in which CD69 was dynamically expressed on activated T cells (Supplementary Fig. S13a, b). PD-1 displayed a distinct and more sustained expression pattern during T-cell activation and restimulation (Supplementary Fig. S13c). We then compared the inhibitory effect of CD93-ECD when added at different time points. CD93-ECD showed the strongest suppressive activity at the initial activation phase, and its suppressive activity was reduced when CD69 was decreased (Supplementary Fig. S13d). These results indicate that CD93-mediated suppression is CD69-dependent and occurs on the early activation state of T cells.
To further determine whether CD93 engagement actively alters T-cell signaling rather than merely sequestering T cells, we examined early TCR-proximal phosphorylation events after restimulation. Compared with restimulated T cells, CD93-ECD treatment reduced phosphorylation of LCK, ZAP70, and LAT, while its effect on PLCγ1 phosphorylation was less pronounced (Supplementary Fig. S13e). Addition of recombinant CD69 protein partially restored CD93-ECD-mediated suppression of pLCK and pZAP70, supporting the specificity of the CD93–CD69 interaction in regulating proximal TCR signaling. These findings suggest that CD93 engagement of CD69 transmits an inhibitory signal that attenuates early TCR activation events, rather than functioning solely as a passive ligand-binding or steric hindrance mechanism (Supplementary Fig. S13f).
Together, these results demonstrate that CD69 is a functional receptor of CD93 on T cells that mediates T-cell suppression. To further evaluate the potential efficacy of CD93 blockade, we generated CD93-blocking antibodies that strongly interfered with the interactions between CD93 and CD69 (Supplementary Figs. S14a–d). Blockade of CD93 with these antibodies significantly increased T-cell proliferation and the release of cytotoxic cytokines (Granzyme B and Interferon-γ) in vitro (Fig. 7l–n). To further determine whether blockade of CD93 on TIMCs has synergistic effects on TCE immunotherapy, we established a hematopoietic lineage-specific CD93-humanized mouse model by transplanting bone marrow cells from whole-body CD93-humanized mice into lethally irradiated BALB/c mice, followed by inoculation with A20 lymphoma cells (Fig. 7o, p). The results showed that treatment with CD93-blocking antibodies significantly improved the anti-tumor efficacy of mTCE treatment synergistically (Fig. 7q). Moreover, the combination of TCE and anti-CD93 markedly increased the ratio of stem-like (TCF1+_Tem-like_T) or progenitor-exhausted (SLAMF6+_Tex_prog_T) CD8+ T cells to terminally differentiated exhausted CD8+ T cells (Tox+_Tex_Term_T) compared with TCE treatment alone (Supplementary Fig. S15). These findings indicate that CD93 blockade attenuates the accumulation and maintenance of terminally dysfunctional CD8+ T cells, preserving a higher proportion of stem/progenitor-like states within the tumor microenvironment.
More importantly, cell‒cell interactions between CD93-expressing TIMCs and CD69-expressing T cells were observed in B-NHL patients after receiving hTCE therapy (Fig. 7r–t), and the infiltration of immunosuppressive TIMCs was positively correlated with CD8+CD69+ T-cell populations (Fig. 7u). To further evaluate whether this spatial pattern could be observed in an independent TCE model, we analysed tumor tissues from mice treated with a CD19/CD3 TCE. Similar to the human CD20/CD3 TCE-treated samples, CD19/CD3 TCE treatment was associated with reduced spatial separation between CD93+CD11b+ myeloid cells and CD69+CD3+ T cells in the tumor microenvironment (Supplementary Fig. S16). Although these data do not establish directional recruitment, they support the presence of increased local proximity between CD93-expressing myeloid cells and activated CD69+ T cells after TCE treatment across different experimental settings.
Together, our findings demonstrated that CD93 regulated by BTK signaling in TIMCs inhibited T-cell-mediated tumor killing activity via interaction with CD69 and that blockade of the CD93‒CD69 interaction repressed the immune evasion of B-NHL cells to resist TCE therapy.
CD93 on TIMCs supports immunosuppression in solid cancers
To determine the role of CD93 in TIMCs in solid cancers, tumor-infiltrating CD3
+ T cells and CD93
+ or CD93
– TIMCs from human breast cancer, urothelial carcinoma and renal carcinoma patients were co-cultured in a manner similar to that used for B-NHL samples (Supplementary Fig. S17a, d, g). CD93
+ TIMCs showed stronger immunosuppressive ability than CD93
– TIMCs did, and T-cell proliferation was significantly inhibited by CD93
+ TIMCs (Supplementary Fig. S17b, e). Moreover, T-cell suppression by CD93
+ TIMCs was alleviated by treatment with anti-CD93 antibodies (Supplementary Fig. S17g, h). In agreement with the SPP1:CXCL9 indicator of immunosuppression
39, TIMCs with higher CD93 expression also expressed higher SPP1 but lower CXCL9 in human urothelial carcinoma
40(Supplementary Fig. S17j). By analysing gene expression data from the TCGA database, we found that high CD93 expression led to adverse prognosis outcomes in bladder cancer, kidney papillary cell carcinoma and breast cancer patients (Supplementary Fig. S17c, f, i), which implied that the immunosuppressive role of CD93 has strong clinical relevance to the poor survival of these solid cancer patients. Furthermore, to determine whether blockade of CD93 on TIMCs has synergistic effects on TCE immunotherapy in solid cancer, we established a hematopoietic lineage-specific
Cd93-humanized mouse model as previously described, followed by inoculation with mouse CD19-expressing 4T1 cells (Supplementary Fig. S17k, l). We found that blockade of CD93 synergistically improved the antitumor efficacy of mTCE treatment in a mouse model of breast cancer (Supplementary Fig. S17m). Taken together, these data suggest that CD93 also has a potential immunosuppressive role in solid TIMCs to suppress antitumor immunity.
DISCUSSION
T-cell engagement has been successful in recent cancer immunotherapy; however, resistance to T-cell engagement therapy and subsequent relapse remain major clinical challenges. Understanding the mechanism of resistance to T-cell engagement therapy is urgently needed. Here, we profiled the tumor immune microenvironment of human B-NHL patients before or after TCE therapy at the single-cell omics level. Next, we found that immunosuppressive myeloid cells were recruited by engaged T cells via CCLs/CCR5 chemoattraction. Moreover, we revealed that CD93 mediated BTK-MAPK-PKC signaling in these recruited TIMCs to suppress the proliferation and activation of engaged T cells. More importantly, we identified the early activation marker of T cells, CD69, as a functional receptor of CD93 that inhibits T cells. Interference with the interaction between CD93 and CD69 significantly enhanced the antitumor efficacy of TCE in treating B-NHL and solid cancers. Together, these results suggest that there is a novel mechanism of TCE resistance in which engaged T cells are suppressed by TIMCs (Supplementary Fig. S18).
Potential mechanisms of TCE resistance are reportedly associated with tumor antigen loss
41, intrinsic or acquired T-cell dysfunction
15,42-44, and an immunosuppressive tumor microenvironment
10,13,45. In our cohort of TCE-resistant r/r B-NHL patients, we aimed to determine whether other factors were independent of the downregulation or loss of TCE-targeting tumor antigens. Consistent with previous studies focused on resistance to immunotherapy in a variety of malignancies
46-48, we found that many myeloid cells were enriched in the tumor microenvironment after TCE therapy, suggesting that these TIMCs may create an immunosuppressive tumor microenvironment resulting in drug resistance. Myeloid cells in the tumor microenvironment promote tumor proliferation and invasion, regulate intratumoral angiogenesis, and suppress antitumor immunity, which could lead to resistance to immunotherapy in a variety of malignancies
49-53. Increasing evidence indicates that TIMCs are potential therapeutic targets for improving the efficacy of immunotherapy
50. However, the ability of these myeloid-targeting drugs to reverse the immunosuppressive tumor microenvironment has not yet been proven, probably due to the nature of these TIMCs in non-tumor antigen-specific suppression. Interestingly, in our study, unlike neoantigen-specific T cells, TCE-engaged T cells were equipped with bispecific antibodies so that they could bypass pMHC-TCR restriction and kill tumor cells directly. This particular situation may allow TIMCs to be crucial therapeutic targets for unleashing antitumor T-cell suppression.
BTK is an important protein kinase that regulates the activation of the BCR signaling pathway, and it serves as a therapeutic target for B-cell malignancies
54. Many BTKis have been approved for the treatment of B-cell malignancies
55-58. In addition to interfering with BCR signaling by inhibiting BTK and B lymphocyte kinases and impairing the phosphorylation of downstream effectors, BTK also plays an essential role in regulating the function of myeloid cells
59,60. In addition to being a nonselective BTKs, IBR also targets interleukin-2-inducible T-cell kinase (ITK), resulting in the depletion of T helper 2 (Th2) cells and inducing a shift toward Th1 cells or an increase in Th17 cell subsets
61,62. It has been reported that pre-treatment with IBR before T-cell collection for chimeric antigen receptor (CAR)-T-cell therapy can reverse the dysfunction of T cells and increase the generation of CAR-T cells for adoptive immunotherapy
63,64. These results suggest that BTKis may still benefit BTKi-resistant patients when combined with T-cell immunotherapy. In our study, in addition to the relatively low-selective BTKi IBR, which also inhibits ITK, the highly selective BTKi ZBR also had synergistic effects on enhancing TCE antitumor efficacy
in vivo. BTK has been implicated in Toll-like receptor (TLR) signaling pathways that regulate macrophage activation and the production of proinflammatory cytokines
65. The targeting of BTK in MDSCs likely contributes to the inhibition of NFκB signal transduction via mediating MDSC expansion and function
60. However, the underlying molecular mechanisms by which BTK regulates TIMC function have not been elucidated. In this study, we found that BTK is a key regulator that mediates CCR5 signaling to recruit T cells and further promotes the gene expression of immune inhibitory molecules, such as CD93, through the MAPK and PKC signaling pathways. Although our
in vitro THP-1-based system was designed to restrict BTK inhibition to macrophage-like myeloid cells, BTK inhibitors were administered systemically in the
in vivo models. Therefore, we cannot completely exclude potential effects of BTK inhibition on other immune compartments, including lymphoma B cells or T cells. Nevertheless, the use of BTKi-insensitive lymphoma models and the loss of zanubrutinib-mediated synergy in hematopoietic CD93-deficient chimeric mice support a significant contribution of the myeloid BTK-CD93 axis to TCE resistance.
CD93, a transmembrane glycoprotein, is highly expressed on endothelial cells, plays an important role in angiogenesis
24-26 and participates in maintaining the endothelial barrier and reducing tumor metastatic dissemination
27,66,67. It has been reported that the interaction of CD93 with IGFBP7 or MMRN2 contributes to abnormal tumor vasculature
68-71. The anti-CD93 agent has emerged as an effective antiangiogenic strategy
72,73. Abnormal vasculature is a key pathological feature facilitating the generation of the immunosuppressive tumor microenvironment
74, and targeting the tumor vasculature thus represents a major effort to reverse the tumor microenvironment to enhance immunotherapy
75,76. Blockade of CD93 has been reported to suppress tumor growth in mice via normalization of the tumor vasculature, which enhances effector T-cell trafficking into tumors
29,71. However, in our study, we established a myeloid cell-specific CD93-deficient mouse tumor model and found that when reconstituted myeloid cells were deficient in CD93, both the antitumor efficacy of ZBR alone and the synergistic effects of ZBR on the tumor cytotoxicity of TCEs were eliminated. These results suggested that CD93 sufficiently mediated BTK function in TIMCs. Moreover, pan-cancer in silico analysis revealed a strong relationship between CD93 expression and M2-like macrophages in several types of human solid cancers
28. Here, we found that CD93 is a downstream gene of BTK signaling in TIMCs that mediates T-cell suppression. In accordance with previous results showing that PKC can regulate CD93 expression in monocytes
37, CD93 expression was downregulated by MAPK/PKC inhibitors, including staurosporine, Go6976, pexmetinib, and SB203580, in this study. A recent study reported that CD93 in peritumoral monocytes may inhibit T-cell function and infiltration by inducing the production of versican
77. However, we found that CD93 on TIMCs suppressed T cells in a cellular signaling-independent manner, suggesting that CD93 may directly bind to a PD-1-like molecule on T cells.
Interestingly, via CRISPR/Cas9 screening, we identified CD69 as a functional receptor of CD93, which is highly expressed in tumor-infiltrating T cells. CD69 is a type 2 transmembrane glycoprotein that is known as a marker for both T-cell priming and resident memory
78-81. Recent studies have shown that CD69 is a functional molecule rather than a simple activation marker
82 and is considered a negative immune regulator, for CD69 deficiency has been found to lead to exacerbated responses in murine models of asthma, arthritis, colitis, myocarditis, tumors
83, and infection
84. It has been reported that growth and metastasis of murine breast cancer were significantly attenuated in CD69-deficient mice and the inhibition of tumor progression was associated with an increase in tumor-infiltrating T cells and reduced CD8
+ T-cell exhaustion
85. Meanwhile, CD69-deficient tumor-specific CD8
+ T cells had been reported to reduce levels of TOX and enhance antitumor immunity, possibly due to CD69’s contribution to signaling through the TCR, because nuclear NFAT2 was found to be decreased in CD69-deficient mice
86. In addition, anti-CD69 treatment increased the generation of Tim3
+ terminally differentiated CD8
+ T cells and thereby enhanced antitumor immune responses
86, because anti-PD-1 therapy preferentially increased the frequency of Tcf1
+ stem-like CD8
+ T cells
87,88, the combination therapies of anti-CD69 and anti-PD-1 are compatible and showed an effective therapeutic effect even on B16 melanoma
86.
Consistent with this context-dependent role of CD69, our single-cell analysis showed that CD69 expression differed between progenitor-like exhausted and terminally exhausted CD8+ T-cell subsets. Therefore, CD69 should not be interpreted solely as an early activation marker in our system. Rather, CD69 may participate in both early activation-associated regulation and the subsequent differentiation or exhaustion trajectory of engaged T cells. However, based on the current patient and experimental data, we cannot precisely quantify the relative contribution of impaired T-cell activation vs activation followed by dysfunction.
Although Myl9/12, galectin-1, S100A8/A9, and oxidized low-density lipoprotein have been identified as potential CD69 ligands
89-92, no surface protein has been identified as a CD69 receptor. In our study, we identified CD69 as the functional interacting partner of CD93, which could inhibit the activity of TIMCs toward T cells at the early activation stage. Traditional ICB therapy targeting inhibitory receptors has shown promising results in reinvigorating exhausted T cells. For example, PD-1 blockade is one of the main strategies used to prevent T-cell exhaustion
93-96, which could lead tumor-specific T-cell aggregates to persist after ICB
97. However, TCE therapy is designed to target tumor-associated antigens (TAAs) by linking two antibody fragments that recognize distinct epitopes on TAAs and on the T-cell surface, which could mediate T-cell cytotoxicity and activation by bridging all CD3
+ T cells and tumor cells
98. In the context of TCE therapy, not only tumor-specific T cells but also all kinds of T cells can be mobilized. We hypothesized that once T cells are activated, they can be inhibited by TIMCs via the CD93‒CD69 interaction. Interfering with the interaction between CD93 and CD69 with CD93 blocking antibodies unleashes T-cell suppression and enhances TCE antitumor efficacy. It may suggest that CD93–CD69 functions as a myeloid–T-cell priming checkpoint, restraining the transition from initial T-cell activation to productive clonal expansion before classical T-cell exhaustion is established. In summary, our work revealed a novel mechanism of TCE resistance in which engaged T cells are suppressed by TIMCs, highlighting that the BTK-CD93-CD69 axis could be utilized as a new therapeutic target to remodel antitumor immunity.
Despite the novel insights provided by our study regarding the BTK-CD93-CD69 axis in TCE resistance, several limitations should be acknowledged. First, our single-cell omics profiling of B-NHL patients provided high-resolution data, but the clinical cohort size was relatively small. Further validation in larger, multi-center prospective cohorts encompassing various subtypes of B-cell malignancies is necessary to confirm the universal applicability of the CD93-mediated resistance mechanism. Second, we identified CD69 as a functional receptor for CD93; however, the exact intracellular signaling cascade triggered by this interaction within T cells remains partially elusive. Specifically, how CD69–CD93 binding translates into the suppression of T cell proliferation and activation at the molecular level needs to be further investigated. Furthermore, while our myeloid cell-specific CD93-humanized mouse models provided strong evidence for the role of TIMCs, mouse models cannot fully replicate the extreme complexity and spatial heterogeneity of the human tumor microenvironment. The long-term safety and potential systemic immune consequences of therapeutic CD93–CD69 blockade also require extensive toxicological evaluation. In addition, although we demonstrated that interfering with the CD93–CD69 interaction enhances TCE efficacy in solid cancer models, the density and phenotype of myeloid cells in solid tumors differ significantly from those in B-NHL. Future studies should explore whether the recruitment kinetics of these TIMCs via the CCLs-CCR5 axis remain the primary driver of resistance in non-hematological malignancies.
In addition, other limitations of our study should be noted, particularly regarding the PBMC-engrafted NOG mouse model. Although this model is useful for evaluating TCE-induced human T-cell activation and antitumor activity, it does not fully recapitulate the human myeloid compartment. In this system, transferred human PBMCs mainly reconstitute lymphoid cells, whereas myeloid cells remain predominantly host-derived because of limited human myeloid reconstitution
99,100. Therefore, the CD163
+ myeloid cells expanding after TCE treatment in our
in vivo experiments should be interpreted as a murine macrophage-like population rather than human TIMCs. Previous studies have shown that host murine myeloid cells can be recruited or activated by transferred human T cells and may modulate human T-cell infiltration, persistence, and antitumor activity in xenograft models
22,101. Consistently, depletion of host myeloid cells in our PBMC-NOG model enhanced hTCE-induced human T-cell activation. These findings suggest that murine myeloid cells may influence human T-cell responses in this model, but they do not definitively prove direct suppressive signaling between murine CD163
+ myeloid cells and human T cells. Thus, we interpreted these
in vivo results together with human clinical samples, patient-derived single-cell analyses, and human cell-based functional assays. Future studies using models with improved human myeloid reconstitution will be needed to further validate this mechanism in a more fully humanized immune context.
MATERIALS AND METHODS
Patient samples
This study utilized 101 PB samples and tumor biopsy samples from 74 B-NHL patients from Peking University Cancer Hospital, comprising 26 newly diagnosed patients and 25 r/r patients who had not yet received hTCE when biopsy samples were collected (assigned as “untreated”); 23 r/r patients received anti-human CD19×CD3 or anti-human CD20×CD3 bispecific TCEs (hTCEs) (assigned as “treated” for drug-resistant patients). Fifty-one samples (20 PB samples and 31 tumor tissue samples) were collected for scRNA-seq from 35 B-NHL patients across FL (n = 16), DLBCL (n = 18), and MCL (n = 1) subtypes. Forty-eight biopsy samples (21 PB and 27 tumor) from 35 B-NHL patients, including 18 FL, 14 DLBCL and 3 MCL patients, were collected for high-dimensional flow cytometry analysis. Twenty-six biopsy samples (12 PB and 14 tumor), including 16 FL, 6 DLBCL, 1 MCL and 3 marginal zone lymphoma (MZL) samples, were collected for functional assays. Paraffin-embedded samples from 6 pre- and post-TCE paired patients were subjected to IHC and mIF staining (2 paired samples were used for CD3 and CD163 IHC staining, and 4 paired samples were used for CD20 IHC and mIF staining). All the patient samples used in this study are listed in Supplementary Table S1.
Blood samples were obtained from patients in EDTA anticoagulant tubes (BD). Patient surgical tumor samples were stored in tissue storage solution (Miltenyi Biotec, 130-100-008) once separated from patients and then transferred to ice. PBMCs were isolated through Ficoll density gradient centrifugation. CD3+ T cells were negatively selected via a Rosette-Sep kit for T-cell isolation from whole blood (STEMCELL Technologies). Monocyte-derived macrophages (MDMs) were induced from CD14+ cells with 100 ng/mL M-CSF. Clodronate liposomes (Liposoma, CP-005-005) and an anti-CSF1R mAb (BioXCell, clone AFS98) were used to deplete macrophages in vivo. Tumor tissue samples from patients were used for scRNA-seq, IHC/mIF staining, flow cytometry, in vitro culture or generation of a PDX model in mice. For patient samples in in vitro experiments, we stained the tumor sample single-cell suspensions with anti-CD3 (BioLegend, 300306), anti-CD11b (BioLegend, 301322), anti-CD45 (BioLegend, 368514), anti-CD93 (BioLegend, 336108), anti-CD19 (BioLegend, 302212), and 7-AAD viability stains (BioLegend, 420404). We collected CD3+ T cells, which were then labelled with CellTrace Far Red (Thermo Fisher Scientific, C34564), and CD11b+CD93– or CD11b+CD93+ myeloid cells to generate a co-culture system to assess T-cell proliferation after 5 days, or we collected CD11b+CD93+ cells and tested how BTKis, 1 μmoL zanubrutinib (Selleck, S8791) or 1 μmoL IBR (Selleck, S2680) treatment could affect the expression of CD93 after 96 h.
scRNA-Seq analysis
All library preparation and sequencing services were provided by LC-Bio Technology Co., Ltd. (Hangzhou, China). Raw scRNA-seq data derived from human and mouse samples were aligned against species-specific reference genomes via Cell Ranger Software (v7.0.1): human samples, GRCh38-2020-A (refdata-gex-GRCh38-2020-A, 10X Genomics); and mouse samples, GRCh38-and-mm10-2020-A hybrid reference (refdata-gex-GRCh38-and-mm10-2020-A, 10X Genomics). All reference genomes were obtained from the 10X Genomics official website. The generated counts were used for downstream analysis in R (version 4.1.3). We applied the Seurat R package (version 4.3.0) to convert the gene expression matrices into Seurat objects. Further quality control was performed to remove low-quality single cells expressing fewer than 200 genes or with more than 20% mitochondrial gene counts in mouse samples, and for human samples, low-quality single cells expressing fewer than 250 genes or with more than 25% mitochondrial gene counts were removed. Cells from mouse samples with UMI counts over 50,000 and gene expression over 7,500 and human samples with UMI counts over 30,000 and gene expression over 6,000 were filtered, as these cells may reflect potential doublets. In addition, for PDX samples, relatively high criteria were applied to remove human-mouse mixed cells. Specifically, single cells exhibiting higher rates of the human transcriptome (> 90%) or having a mouse transcriptome rate > 90% were defined as human cells or host cells, respectively, all of which were retained. After quality control, 459,984 cells from 51 human samples (35 patients), comprising 110,516 CD8+ T cells, 98,442 CD4+ T cells, 20,444 NK cells, and 79,639 TIMCs, including 46,374 monocytes/macrophages and 33,265 neutrophils, in addition to 139,450 B cells, 1,905 endothelial cells, 853 platelets/erythrocytes and 8,735 other cells were identified. And 71,036 transcriptomes of single cells were obtained from the human PBMC-humanized PDX B-cell lymphoma model, in which 40,649 human T cells, 1,950 human DLBCL tumor cells and 28,437 mouse host cells were identified.
Gene‒cell count matrices from different samples from humans and different samples from mice were merged via Seurat, and two matrices were obtained. Gene expression matrices were log-normalized to correct for library size differences via the normalizeData function of the Seurat package. Then, per-gene z-score scaling for each cell was performed via Seurat’s ScaleData function. Afterward, principal component analysis (PCA) was performed on the top 2,000 genes (in two matrices of humans and mice) identified as highly variable genes (HVGs) via the Seurat function FindVariableFeatures. The top 23 (in the mouse matrix) and top 25 (in the human matrix) principal components (PCs) were further chosen for UMAP dimension reduction with the RunUMAP function from Seurat. The first round of unsupervised clustering (resolution = 0.1 in the mouse dataset) identified three major cell types on the basis of cell type markers of the human transcriptome, including human tumor cells (CD19, MS4A1), human T cells (PTPRC, CD3D, CD3E, and CD3G), and mouse host cells (without human PTPRC expression). In the human dataset, the FindClusters function (resolution = 1) was used for unsupervised clustering, and 7 major cell types were identified, including T cells (CD3D, CD8A, CD4, and FOXP3), B cells (CD79A and MS4A1), natural killer cells (NCAM1), myeloid cells (CD14, CD68, CSF3R, CLEC10A, LILRA4, and TPSAB1), endothelial cells (VWF), platelets/erythrocytes (PPBP), and other cell types.
Next, we performed a second round of clustering to further characterize the subpopulations of major cell types in the human dataset. Owing to the variable numbers and properties of cells within each major cell type, different clustering parameters were used. For the clustering of CD8+ T cells, the top 25 PCs were selected based on 2,000 HVGs (resolution = 0.3). For the clustering of CD4+ T cells, the top 25 PCs were selected on the basis of 2,000 HVGs (resolution = 0.3). For the clustering of natural killers, the top 20 PCs were selected on the basis of 2,000 HVGs (resolution = 0.5). For monocytes or macrophages or dendritic or plasmacytoid dendritic cells, the top 25 PCs were selected based on 2,000 HVGs (resolution = 0.8). For neutrophils or mast cells, the top 25 PCs were selected based on 2,000 HVGs (resolution = 0.8). As a result, we identified 10 CD8+ T clusters, 7 conventional CD4+ T, 1 T regulatory, and 6 NK cell clusters for the lymphoid lineage; 7 monocyte, 5 macrophage, 3 DC, 2 pDC, and 7 neutrophil clusters; and 1 mast cell cluster for the myeloid lineage. To facilitate data visualization, the cells were re-clustered into five embeddings via Seurat, including: (1) CD8+ T cells, (2) CD4+ T cells, (3) NK cells, (4) monocytes and macrophages and dendritic and plasmacytoid dendritic cells, and (5) neutrophils and platelets/erythrocytes and mast cells. Next, we used the FindMarkers or FindAllMarkers function to identify differentially expressed genes (DEGs) with adjusted P < 0.05 via Bonferroni correction.
To identify subclusters of human T cells and mouse host cells accurately in the mouse dataset, we first selected two independent data subsets following the above-described cell type markers from the filtered matrix in both human cells and host cells separately. The second round of dimension reduction and unsupervised clustering was carried out on the two data subsets obtained above, including normalization, finding HVGs, scaling, and calculating the PCA matrix. The number of PCs was selected on the basis of the elbow plot, the use of which was dataset-dependent. For mouse host cells, the second round of clustering parameter PCs and the resolution were set to 25 and 0.8, respectively. We then annotated the mouse host clusters to cell types by known markers of the mouse transcriptome, including pericytes (
Pdgfrb,
Cspg4, and
Trpc6), fibroblasts (
Dcn,
Col1a1,
Col1a2,
Col5a1,
Fgf7, and
Mme), endotheliocytes (
Pecam1,
Vwf,
Emcn,
Nrp1,
Kdr,
Cdh5,
Gpihbp1, and
Plvap), and myeloid cells (
Ptprc,
Cd14,
Cd68,
Adgre1,
Itgax,
Ccr2,
Csf1r,
Ly6c2,
Cd209a, and
Irf7)
102. For human T cells, in the second round of unsupervised clustering, a PCA matrix with 20 PCs and a resolution of 0.8 was summarized via UMAP to present the data. The 15 obtained T-cell cluster subclusters were annotated on the basis of canonical marker genes, including
CD8A,
CD4, and
TRDC, for the lymphoid lineage (CD8
+ T, conventional CD4
+ T, and γδT). The Seurat function FindAllMarkers was used for the differentially expressed genes of subclusters within the human T and mouse host cell types.
The functional marker data were obtained from a previous study
38. The functional scores of the T and myeloid cell types were calculated as the difference between the average expression levels of the functional markers. The proportion of each subcluster in the human samples and PDX model samples was calculated via the R function “prop.table”. The relative proportion of cells in the tumor microenvironment was obtained by multiplying the above proportion by a ratio and multiplying by 100. The Monocle2 R package (version 2.22.0) was used to infer the developmental trajectories of 1 monocyte cell type and 3 macrophage types in the samples. We built a new CellDataSet object from a cluster-annotated Seurat object via the new CellDataSet function. A monocle function, Differential Gene Test, was used to derive differentially expressed genes from the genes obtained via the Variable Features function, and genes with
q < 1 × 10
−5 were used to order the cells in pseudotime.
To investigate cell‒cell interactions among clusters from each module, we analysed the L–R pairs among samples via CellPhoneDB (version 2.1.7). Differentially expressed genes between samples were analysed via the DEseq2 package (version 1.34.0). The topGO (version 2.46.0) was used to enrich different GO pathways for the nontreated and treated or monotherapy and combination treatment pair samples.
RNA-seq analysis
For RNA-seq analysis, Bowtie (version 1.0.1) was used to construct the reference genome, and the resulting clean data were subsequently compared to the reference genome via HISAT2 (version 2.1.0). Differential gene analysis was performed via DEGSeq (version 1.18.0), and an adjusted P value < 0.05 was considered the threshold. We performed Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analysis with Sangebox 3.0 (ivip.sangerbox.com).
For the RNA-seq analysis of co-cultured THP-1 cells and T cells, we labelled T cells with CellTrace Far Red and co-cultured them with EGFP-expressing THP-1 cells at a ratio of 4:1. After 24 h, we separated T cells and THP-1 cells by flow cytometry for RNA-seq analysis.
Public data analysis
We downloaded transcriptomic data and clinical data from the Breast Cancer (BRCA) cohort, Bladder Cancer (BLCA) cohort, and Kidney Papillary Cell Carcinoma (KIRP) cohort in the TCGA database (xenabrowser.net/datapages/). Survival analysis of primary tumor samples was performed via the survival (v 3.5.5) and survminer (v 0.4.9) R packages. We downloaded the single-cell transcriptome sequencing dataset GSE271915 from the Gene Expression Omnibus (GEO) database. It contains scRNA-seq data of bone marrow samples from seven multiple myeloma (MM) patients before and after BCMA chimeric antigen receptor T (CAR-T) cell therapy. Further analyses were performed with the use of R software, version 4.5.1. Firstly, quality control was achieved by removing cells with fewer than 200 or more than 6,000 RNA features, cells with more than 20% mitochondrial genes, and cells with more than 1% hemoglobin. Then, data normalization and PCA dimensionality reduction were performed using the “Seurat” software package (version 5.4.0). The “harmony” method in the “IntegrateLayers” function was used to remove the batch effect, and UMAP cluster analysis was performed. Then myeloid cells (a total of 29,824) were extracted and clustered again to obtain seven types of myeloid cell subsets (including 18,809 CD163_Mono/Mac, 6,944 FCGR3A_Mono/Mac, 1,352 FCGR3B_Mono/Mac, 1,354 MS4A3_Mono/Mac, 1,067 CD1C_DC, 76 CLEC9A_DC and 222 pDC). The “CellDimPlot” function in the “SCP” package was used to draw the beautified UMAP, and the “FindAllMarkers” function was used to determine the top genes and draw the bubble map. The “heatmap” function was used to draw a heat map to compare the gene expression and cell function scores of each subset of cells. The changes in the proportion of various myeloid cells in each patient before and after CAR-T treatment were compared, and the paired t-test was performed.
Cell lines and culture conditions
Human DLBCL cells (HBL-1, TMD8, OCI-Ly10 and DHL-16), MCL cell lines (Mino, Z138 and Jeko-1) and Burkitt lymphoma cells (Raji) were generously provided by Dr. Fu from the University of Nebraska Medical Center, Omaha, NE, USA. The DLBCL cell lines SU-DHL-6 and U2932, the acute myeloid leukemia (AML) cell line THP-1, the A20 murine B lymphoma cell line, the 4T1 murine breast carcinoma cell line, and the 293T cells were purchased from the American Type Culture Collection (Manassas, VA). RPMI-1640 or low-glucose Dulbecco’s modified Eagle’s medium (DMEM) (Thermo Fisher Scientific) supplemented with 10% fetal bovine serum (FBS; Biological Industries), penicillin/streptomycin, glutamine and beta-mercaptoethanol were used for all cell cultures except for OCI-LY10 cells, which were maintained in IMDM (Thermo Fisher Scientific) supplemented with 10% penicillin/streptomycin. For the A20 murine B lymphoma cell line, 50 μM 2-mercaptoethanol (Sigma-Aldrich, M3148) was added to the cultures. For human T cells or PBMCs, 10 ng/mL interleukin-2 (IL-2) (STEMCELL Technologies) was added to the cultures. For mouse T cells, 1% glutamine (Thermo Fisher Scientific, O2956--100), 100 lU/mL, 50 μM 2-mercaptoethanol (Sigma-Aldrich, M3148), and 1% MEM nonessential amino acids (Gibco, 11140-050) were used. The THP-1 cell line is a human monocyte cell line, and we used it as a macrophage tool. The macrophages were induced with 200 ng/mL PMA for 48 h, after which 40 ng/mL IL-4 was added to induce M2-like macrophages. All the cells were maintained in a humidified 5% CO2 incubator at 37 °C and were identified via short tandem repeat DNA fingerprinting analysis (Applied Biosystems).
Virus construction and infection
For retrovirus packaging, plasmid constructs for gene overexpression, including human
CD93, human
IGFBP7,
Cas9, mouse
Cd19 and
OKT-3, and the CRISPR/Cas-9-based mouse membrane protein-targeting gRNA library (constructed from our own library) for gene knockout, were mixed with PCL-ECO (2:1), followed by transfection into 293T cells via Lipofectamine 2000 (Invitrogen). For lentivirus packaging, CRISPR/Cas9-based
CD93-targeting gRNA constructs were mixed with psPAX2 and pMD2G (Addgene) at a ratio of 4:3:1 and transfected into 293T cells via Lipofectamine 2000 (Invitrogen). Virus-containing supernatants were collected 48–72 h post-transfection and used for infection as described previously
103.
Chimeric receptor reporter assay and genome-scale CRISPR-Cas9 knockout screening
We constructed a stable chimeric receptor reporter cell system as described previously
104,105. First, we tested the ability of the CD93-ECD to trigger the inhibition of the activating effect of anti-mouse CD3 or anti-FLAG, which chimerically fuses the intracellular domain of paired immunoglobulin-like receptor β to signal through the adaptor DAP-12 to activate the NFAT promoter, resulting in a decrease in GFP expression. We subsequently constructed an IGFBP7-expressing reporter system. If CD93 or its truncated fusion protein binds the ECD of IGFBP7 and activates the chimeric signaling domain, an increase in GFP expression would be observed. Briefly, for the functional inhibition assay of CD93, CD93-ECD proteins (C405, Novoprotein; 5 μg/mL) or human AB serum (10%, diluted in PBS) and anti-mouse CD3 (BioLegend, 100202; 5 μg/mL) or anti-FLAG antibodies (BioLegend, 637302; 5 μg/mL) were precoated on 96-well plates at 37 °C for 3 h. For the IGFBP7 activation functional assay of CD93, 2.5 µg/mL CD93-ECD or its truncated fusion protein was precoated on a 96-well plate at 37 °C for 3 h. After two washes with PBS, 2 × 10
4 WT or IGFBP7 reporter cells were seeded in each well. After 16 h, the percentage of GFP
+ reporter cells was analysed via flow cytometry.
We utilized the stable chimeric receptor reporter cell system mentioned above, constructed a Cas-9-overexpressing reporter cell system and infected the system with a retrovirus-packed sgRNA library that contained 20,403 sgRNAs targeting 3,400 mouse membrane proteins. After 48 h, 1 µg/mL puromycin was added for 2 days, and the survival rate was controlled at 30% to ensure that one cell was infected with only one virus. Then, 2 × 10
5/mL surviving cells were seeded into 96-well plates precoated with PBS, CD93-ECD (5 µg/mL), anti-CD3 (5 µg/mL) with CD93-ECD (5 µg/mL), or anti-FLAG (5 µg/mL) with CD93-ECD (5 µg/mL). After 16 h, we sorted the GFP
+ cells for next-generation sequencing (NGS) as described previously
106 and analysed the differences between groups via model-based analysis of genome-wide CRISPR-Cas9 knockout (MAGeCK)
107.
Flow cytometry
For tumor samples, the samples were stored in tissue storage solution (Miltenyi Biotec, 130-100-008) once separated from patients or mice and transferred to ice, after which they were turned into single-cell suspensions via gentle MACSTM as soon as possible. The samples were resuspended in 100 μL of viability blocking solution, incubated for 30 min at 4 °C, topped with 1,000 μL of PBS and spun for 5 min at 500× g at 4 °C. Then, we stained surface markers (see the Reagents or Resources in Supplementary Table S3) in saline buffer with Brilliant Stain Buffer Plus (BD, 566385), incubated them for 30 min at 4 °C, topped them with 1,000 μL of PBS and spun them for 5 min at 500× g and 4 °C. The cells were then fixed with 250 µL of fixation buffer (BioLegend, 420801) per tube and incubated in the dark for 20 min at room temperature. Then, we resuspended the fixed cells in diluted Intracellular Staining Permeabilization Wash Buffer (BioLegend, 421002) and centrifuged them at 350× g for 5‒10 min; this process was repeated twice. After that, we suspended the cells in 100 μL of an intracellular antibody mixture with Brilliant Stain Buffer Plus and incubated them for 45 min at room temperature in the dark. We then washed the samples twice in diluted Intracellular Staining Permeabilization Wash Buffer, resuspended them in 250 μL of staining buffer, and analysed them on a spectral cytometer (Cytek® Aurora) or CytoFLEX (Beckman Coulter). BD FACSAria III and BXnationl GSE pro5 in the core facility of state key laboratory of natural and biomimetic drugs were used for cell sorting.
For blood samples, fresh blood was obtained from patients or mice in EDTA anticoagulant tubes (BD), and the following staining protocols were similar to those used for the above tumor samples. For other co-cultured samples, samples were collected as cell suspensions, and the following staining protocols were similar to those used for the above tumor samples. For TCR-proximal phosphorylation, primary human T cells were activated with soluble anti-CD3/anti-CD28 activator in complete RPMI 1640 medium supplemented with 10% FBS and 100 U/mL recombinant human IL-2 for 3 days. The activator was then removed, and cells were rested in fresh activator-free medium for 2 days. Rested T cells were re-stimulated with the same anti-CD3/anti-CD28 activator for 6 h, followed by treatment with recombinant CD93-ECD and/or recombinant CD69 protein at 10 μg/mL for 10 min. Cells were immediately harvested, fixed and permeabilized according to the manufacturer’s instructions, and stained with antibodies against phosphorylated TCR-proximal signaling molecules, including pLCK, pZAP70, pLAT, and pPLCγ1. Phosphorylation levels were quantified by flow cytometry and reported as mean fluorescence intensity (MFI).
Cytotoxicity assessment assay
Human T cells were isolated from the whole blood of healthy adults. Then, T cells were incubated with different target B-cell lymphoma cell lines loaded at different effector-to-target ratios (E:T) or ratios of 5:1, and tumor cell lysis was assessed according to standard procedures. Isolated T cells and tumor cell lines were incubated with increasing concentrations of TCE. The K562 chronic myeloid leukemia (CML) cell line was used as a negative control (CD19 negative). After 8 h of co-culture, the supernatant was collected and subjected to lactate dehydrogenase (LDH) lysis according to the CytoTox 96® Non-Radioactive Cytotoxicity Assay Technical Bulletin using a standard 96-well plate reader. Cytotoxicity (labelled as % target cell lysis in this context) was measured according to the following equation:
To analyse the effects of macrophages on the tumor-killing effect of TCE, macrophages isolated from Raji tumor-bearing mice, T cells, and tumor cells were co-cultured at a ratio of 1:2.5:1 in the presence of TCE. After 24 h of co-culture, the resulting suspensions were collected and analysed via flow cytometry using CD3 and PI staining (CD3–PI+ cells, which are defined as dead tumor cells).
To analyse the synergistic role of TCE and CD93-KO or anti-CD93 antibody, 4 × 105 WT or CD93-KO THP-1 cells were induced to differentiate into macrophages with 200 ng/mL PMA for 48 h. After the supernatant was discarded, IL-4-containing media supplemented with 1 ng/mL TCE were added to precoated plates supplemented with 10 µg/ml anti-CD93 antibody or IgG control with WT or CD93-KO cells. T cells (2 × 105/well) labelled with CellTrace FarRed and 4 × 104/well tumor cells (Raji) labelled with CellTrace Violet (Thermo Fisher Scientific, C34557) in IL-2-containing medium were subsequently plated in precoated 96-well plates.
To analyse the synergistic effects of TCE and BTK inhibitors, 4 × 105 WT or CD93-KO THP-1 cells were induced to differentiate into macrophages with 200 ng/mL PMA for 48 h. After the supernatant was discarded, IL-4-containing media supplemented with 1 ng/ml TCE were added to precoated plates supplemented with increasing concentration of ZBR with WT or CD93-KO cells. T cells (2 × 105/well) labelled with CellTrace FarRed and 4 × 104/well tumor cells (Raji) labelled with CellTrace Violet in IL-2-containing medium were subsequently plated.
Mouse T cells were isolated from the spleens of BALB/c mice. Mouse T cells were then activated with 2 μg/mL anti-mouse CD3 in precoated 96-well tissue culture plates for 3 h. After that, CellTrace Far Red-labelled T cells and the CellTrace Violet-labelled murine B lymphoma cell line or murine breast carcinoma cell line were co-cultured in the presence of increasing concentrations of mouse TCE at an E:T ratio of 5:1, and tumor cell death was measured by flow cytometry after 7-AAD viability staining was performed after 96 h (Far red-violet+7AAD+ defined as dead tumor cells).
To test the impact of CD93 on T-cell cytotoxicity, we constructed an OKT-3-overexpressing reporter cell system that also expressed CD93 or not. We isolated human T cells as mentioned above, labelled them with CellTrace Far Red, and co-cultured with reporter cells that were labelled with CellTrace Violet at an E:T ratio of 5:1. Reporter cell death was measured by flow cytometry after 7-AAD viability staining for 48 h (Far red violet+ 7AAD+ indicates dead reporter cells).
ELISA
A total of 1.0 × 105 THP-1 cells/well were induced to differentiate into macrophages with 200 ng/mL PMA for 48 h. After the supernatant was discarded, IL-4-containing medium was added for another 48 h. Then, 1.0 × 105 T cells/well labelled with CellTrace FarRed in IL-2-containing medium were plated in WT or CD93-KO macrophage-coated 96-well plates in the presence of ImmunoCult™ Human CD3/CD28 T-cell activator (StemCell, 10971), and an anti-CD93 antibody or IgG control was added to the culture medium of WT macrophages. After co-culture for 5 days, the suspensions were collected, and their proliferation was measured via flow cytometry or Granzyme B and IFN-γ secretion via precoated Human Granzyme B (4abio, CHE0127) or an IFN-γ microplate ELISA Kit (4abio, CHE0017).
In vivo murine studies
NOD-SCID-IL2rg-null mice (strain no. 408) were purchased from Charles River. Cd93–/– mice (strain no. 170807) were purchased from Bcgen (Beijing Biocytogen Co., Ltd.). C57BL/6JGpt mice (strain no. N000013), BALB/cJGpt mice (strain no. N000020), Ccr5–/– mice (strain no. T052842) and BALB/c-hCD93 mice (strain no. T054647) were purchased from GemPharmatech (Nanjing, China). Eμ-myc mice (strain no. 002728) were purchased from Jackson Laboratory. Six- to eight-week-old female NOD-SCID-IL2rg-null (NOG) mice from Charles River (Beijing, China) were generated via human PBMC injection into the tail vein. The mice were kept in a specific pathogen-free animal laboratory environment that met the National Institutes of Health Guide for the Care and Use of Laboratory Animals. The mouse experiments were all approved by the Institutional Animal Care and Use Committee of Peking University Health Science Center and Peking University Cancer Hospital & Institute. The PBMCs were derived from healthy individuals. Raji cells (3 × 106) in PBS medium suspended in an equal volume of Matrigel (Corning, 356237) were inoculated subcutaneously into the area under the right flank of each mouse. The PDX model was developed by inoculating a 1 mm3 biopsy specimen tissue from a patient with DLBCL subcutaneously into the right flank of every single NOG mouse. When the tumor size reached 60–80 mm3, 5 × 106 PBMCs were injected into each mouse. After approximately 10 days, when the humanized immune system was rebuilt according to the presence of human CD3+ T cells in the peripheral blood, the mice were randomly divided into different groups (n = 8–10/group). The mice received vehicle control, IBR (20 mg/kg, diluted with 10% DMSO, 40% PEG300, 5% Tween-80, and 45% saline) by gavage once daily, hTCE (10 μg/kg, dissolved in PBS with 0.05% Tween-80) by intraperitoneal injection daily, macrophage depletion by clodronate liposomes twice per week, anti-CD93 antibody by intraperitoneal injection twice a week, and combinations including coadministration of hTCE and IBR, coadministration of hTCE and the macrophage depletion strategy, coadministration of hTCE and the anti-CD93 antibody, and coadministration of hTCE and the CCR5 inhibitor.
CD93-KO chimeric mice were generated by transferring 2 × 106 BM cells from Cd93−/− or WT mice into lethally irradiated WT B6 mice [1100 rads (2 × 550 rads)] via a gamma irradiator with cesium-137. Humanized CD93 chimeric mice were generated by transferring 2 × 106 BM cells from BALB/c-hCD93 mice into lethally irradiated BALB/c mice [1100 rads (2 × 550 rads)] via a gamma irradiator with cesium-137. Chimeric mice were used for tumor experiments 4–6 weeks after reconstitution. For the combination of BTKi and TCE therapy, mice with palpable tumors were treated with vehicle control, zanubrutinib (10 mg/kg, diluted with 10% DMSO, 40% PEG300, 5% Tween-80, and 45% saline) by gavage once daily, or mouse TCE (500 μg/kg, dissolved in PBS with 0.05% Tween-80) by intraperitoneal injection every 3 days. For the combination of anti-CD93 and TCE therapy, mice with palpable tumors were treated with vehicle control, anti-CD93 (5 mg/kg, dissolved in PBS with 0.05% Tween-80) or mTCE (500 μg/kg, dissolved in PBS with 0.05% Tween-80) by intraperitoneal injection every 3 days.
Tumor volume was assessed with a calliper every two days. Tumor tissue samples and spleens were collected and maintained in formalin or ground to obtain a cell suspension for subsequent experiments. The tumor volume was calculated according to the formula V = ab2/2, where a and b denote the long and short diameters of the tumor, respectively.
To assess myeloid cell infiltration, we collected CD11b+ BM cells from Ccr5–/– or WT mice via an APC-conjugated anti-mouse CD11b antibody (BioLegend, 101212) and MojoSort™ mouse anti-APC nanobeads (BioLegend, 480072) and labelled Ccr5–/–CD11b+ cells with CFSE, while WT CD11b+ cells were labelled with CellTrace Violet. After that, we made a 1:1 mixture of the collected cells and intraperitoneally injected them into immunologically reconstituted NOG mice (as mentioned above) bearing palpable Raji tumors. From the day of mouse myeloid cell transplantation, the mice received vehicle control or hTCE (10 μg/kg, dissolved in PBS with 0.05% Tween-80) by intraperitoneal injection daily, and we collected the tumor samples and measured the infiltration of the transplanted myeloid cells by flow cytometry.
Kinase inhibitor library screening
The kinase inhibitor library purchased from Selleck (L1200) was prediluted in RPMI 1640 to obtain a concentration of 10 μM. A total of 2 × 104 THP-1 cells/well were induced to differentiate into macrophages with 200 ng/mL PMA for 48 h. After washing, the THP-1 cells were treated with 1 μM kinase inhibitors or control medium for 24 h in the presence of IL-4. After the supernatant was discarded, 3 × 104 tumor cells (Raji) labelled with CellTrace Violet and 1.5 × 105 T cells labelled with CellTrace CFSE were plated in these precoated 96-well plates in the presence of TCE. After co-culture for 24 h, the cells were collected and stained with PI for assessment of tumor cell death via flow cytometry. Treatment with a single TCE was defined as the control, and relative cell death (%Experimental cell death/% control cell death) was used to evaluate the additional effects of kinase inhibitors on TCE. For further validation of the screened kinase inhibitors, the MFI of CD163 was assessed via flow cytometry. THP-1 cells (2 × 104/well) plated in 96-well plates were induced to differentiate into macrophages with 200 ng/mL PMA for 48 h, after which 40 ng/mL IL-4 and kinase inhibitors were added to the indicated wells for 24 h. The raw CD163 MFI of each well was normalized to that of the IL-4-treated wells. Gene Ontology (GO) term and KEGG pathway enrichment analyses were conducted via the web-based Gene Set Analysis Toolkit (WebGestalt). Effective gene targets (with one or more effective drugs that both upregulated relative cell death and downregulated relative CD163 MFI) were enriched by the overrepresentation analysis method (ORA). All of the gene targets of the tested drugs were included in the reference gene list. Biological process-related GO terms with FDRs < 0.05 were selected.
IHC and mIF staining and analysis
Paraffin-embedded tumor sections were cut into 5-μm thick sections, which were deparaffinized and stained with primary antibodies. After heat-mediated antigen retrieval, the sections were incubated with these primary antibodies at the recommended concentrations overnight at 4 °C. The sections were subsequently incubated with a secondary antibody at 37 °C for 60 min. The slides were then stained with DAB and counterstained with hematoxylin. All the slides were imaged with a microscope (DMI6000B; Leica Microsystems). Staining was quantified using ImageJ software. Briefly, tumor regions of interest (ROIs) were selected while excluding necrotic areas, tissue folds, artifacts, and non-tumor regions. The CD163-positive staining area was measured within each ROI and normalized to the total annotated tissue area. CD163 abundance was therefore expressed as CD163+ area percentage rather than as raw staining area.
For mIF analysis, tissue sections were stained with antibodies against CD3, CD8, CD11b, CD69, CD93, and DAPI, as indicated in each figure. Multichannel fluorescent images were acquired under identical imaging settings within each staining batch. ROIs were defined consistently across pre- and post-treatment samples or across treatment groups. Within each ROI, single-cell segmentation and marker-based phenotyping were performed using HALO software (v3.6, Indica Labs). CD69+ T cells were identified based on CD3 and/or CD8 co-expression with CD69, and CD93+ TIMCs were identified based on CD11b and CD93 co-expression. Cell densities were normalized to the annotated tissue area and reported as cells/mm2.
Spatial proximity analysis was performed within the same annotated ROIs used for cell quantification. CD3+CD8+CD69+ T cells were defined as reference cells, and CD11b+CD93+ TIMCs were defined as target cells. For each reference T cell, the shortest distance to the nearest CD11b+CD93+ TIMC within the same ROI was calculated. Nearest-neighbor distances were measured in micrometers (μm), and analyses were constrained within a fixed 100-μm radius. This ROI-based and cell-level spatial analysis ensured that distance measurements were derived from comparable tissue regions and were not influenced by differences in tissue size or section-level cellularity.
CRISPR/Cas9-based gene knockout in THP-1 cells
THP-1 cells were infected with Cas9-expressing lentivirus (U6-sgRNA-SFFV-Cas9-FLAG-P2A-EGFP) and scramble control sgRNA (sgRNA 5′-CGCTTCCGCGGCCCGTTCAA -3′) or BTK- or CD93-targeting sgRNAs designed with an online tool (crispr.mit.edu): sgbtk-1 (5’-ATCACTTGTGTTGAAACAG-3’), sgbtk-2 (5’-TGTGCACGGTCAAGAGAAAC-3’), sgbtk-3 (5’-AGCAAATTTCAATCATTGAA-3’), sgcd93-1 (5’-GCCCTCAATGTTACTTCCGG-3’), sgcd93-2 (5’-ATGTGTTCGACTGGGGCAGC-3’), and sgcd93-3 (5’-ATCGCAAGCGGAGAGCGAAG-3’). After infection with lentivirus for 5 days, the cells were collected for GFP analysis via flow cytometry. After cell expansion, knockout cells were verified by western blotting or flow cytometry.
AlphaFold3 predictions and multiple sequence alignment
We predicted the structures of the CD93 and IGFBP7 proteins via the AlphaFold3 website (alphafoldserver.com) following the included instructions. We also used the UniProt website (uniprot.org) to obtain the ECD of the CD93 and IGFBP7 sequences.
SPR
Biacore 8K/2383429 (Cytiva, Marlborough, UK) and CM5 chips (Cytiva) were used to analyse the binding of the CD69 ECD (Novoprotein, CU12) or IGFBP7 (CU50 Novoprotein) to the CD93 ECD (Novoprotein, C405) or its functional truncated protein. The CM5 chips were first activated with a mixture of 400 mM N-ethyl-N’-(3-dimethylaminopropyl) carbodiimide (EDC) and 100 mM N-hydroxysuccinimide (NHS) at a flow rate of 10 μL/min. Recombinant proteins (CD93 or its truncated protein, 100 μg/mL) in 10 mM sodium acetate (pH 4.0) were then injected into the sample channel for 420 s at a flow rate of 10 μL/min, achieving an immobilization level of 1,500–2,000 response units (RU). The analyte was diluted to 8–10 concentrations and injected into sample channels at a flow rate of 30 μL/min for an association phase of 120 s, followed by a dissociation phase of 360 s. The cycle was repeated 8–10 times according to the analyte concentrations in ascending order. The affinity between metabolites and proteins was evaluated via Biacore T2000 software (Cytiva).
Biolayer interferometry
Binding interaction analyses, which were performed on Octet RED96 (ForteBio, Pall Corporation), were utilized to assess the competitive ability of CD69 with anti-CD93 or CD69 with IGFBP7. CD93-ECD-coated streptavidin biosensors (100 μg/mL CD93-ECD was loaded for 420 s; streptavidin biosensors, ForteBio, Pall LLC, 18-5019) were washed in PBST (PBS with 0.1% Tween 20) before monitoring the association (300 s) and dissociation (600 s) of CD69, IGFBP7, anti-CD93 or PBST buffer. The CD93-coated sensor was subsequently loaded with CD69 (2 μM), IGFBP7 (2 μM) or anti-CD93 (100 nM) for 750 s to reach a saturated binding state. The CD69-saturated channel was followed by 100 nM anti-CD93 and 2 μM CD69 for 375 s and was incubated with PBST for 225 s. The IGFBP7-saturated channel was followed by 2 μM IGFBP7 and 2 μM CD69. The anti-CD93 saturated channel was then incubated with 2 μM CD69 for 375 s and then with PBST for 225 s. The buffer channel was followed by incubation with CD69 (2 μM) or anti-CD93 (100 nM) for 375 s and incubation with PBST for 225 s.
Reagents and antibodies
hTCE, an anti-human CD19×human CD3 antibody without an Fc fragment, was provided by ITabMed Ltd. (Shanghai, China), which was stored at the original concentration of −80 °C and diluted in PBS prior to use. A library of kinase inhibitors with a 10 mM primary concentration was purchased from Selleckchem. The agents used in vitro or in vivo and the antibodies used for flow cytometry are listed in the reporting summary. The mTCE, anti-mouse CD19×mouse CD3 with LALA-mutant IgG1 and anti-CD93 antibodies (Patent No. PCT/US2021/035542; Clone: 7F3) with a human IgG4 tail were produced in the laboratory. The mTCE and anti-CD93 antibodies were stored at the original concentrations at −80 °C and diluted in PBS prior to use.
Statistical analysis
Data are presented as mean ± SD or mean ± SEM, as indicated in the figure legends. The results were assessed with GraphPad Prism (Version 8.4.0) or IBM SPSS Statistics (Version 22.0; IBM Corp., New York, USA). Student’s t-test was used to assess the significance of differences between two groups. For the analysis of multiple groups, one-way ANOVA was used. Differences with a P value < 0.05 were recognized as statistically significant. Further statistical details for each experiment are presented in the Figure legends.
Ethics approval
This study was conducted in accordance with the Declaration of Helsinki and was approved by the Ethics Committee of Peking University Cancer Hospital & Institute (approval number: 2021KT74, 2023KT87, 2023KT139, 2025KT154 and 2025KT200) and the Peking University Third Hospital (Ethics Approval number LM2024410). Written informed consent was obtained from all participants before enrollment and sample collection. The animal experiments were all approved by the Institutional Animal Care and Use Committee of Peking University Health Science Center (Ethics Approval number DLASBE0853) and Peking University Cancer Hospital & Institute (Ethics Approval number EAEC-2024-22).
DATA AVAILABILITY
The scRNA-sequencing data generated in this study have been deposited in the OMIX repository of the National Genomics Data Center, China National Center for Bioinformation, under accession number OMIX019373 and BioProject accession number PRJCA025327. The data are available under controlled access in accordance with applicable ethical and data-protection requirements.
The Author(s) 2026. Published by Higher Education Press. This is an Open Access article distributed under the terms of the CC BY license (https://creativecommons.org/licenses/by/4.0/).
