SHP2 inhibitors (SHP2i) alone and in various combinations are being tested in multiple tumors with overactivation of the RAS/ERK pathway. SHP2 plays critical roles in normal cell signaling; hence, SHP2is could influence the tumor microenvironment. We found that SHP2i treatment depleted alveolar and M2-like macrophages, induced tumor-intrinsic CCL5/CXCL10 secretion, and promoted B and T lymphocyte infiltration in Kras- and Egfr-mutant non–small cell lung cancer (NSCLC). However, treatment also increased intratumor granulocytic myeloid-derived suppressor cells (gMDSC) via tumor-intrinsic, NFκB-dependent production of CXCR2 ligands. Other RAS/ERK pathway inhibitors also induced CXCR2 ligands and gMDSC influx in mice, and CXCR2 ligands were induced in tumors from patients on KRASG12C inhibitor trials. Combined SHP2 (SHP099)/CXCR1/2 (SX682) inhibition depleted a specific cluster of S100a8/9hi gMDSCs, generated Klrg1+ CD8+ effector T cells with a strong cytotoxic phenotype but expressing the checkpoint receptor NKG2A, and enhanced survival in Kras- and Egfr-mutant models. Our results argue for testing RAS/ERK pathway/CXCR1/2/NKG2A inhibitor combinations in patients with NSCLC.
Our study shows that inhibiting the SHP2/RAS/ERK pathway triggers NFκB-dependent upregulation of CXCR2 ligands and recruitment of S100A8hi gMDSCs, which suppress T cells. Combining SHP2/CXCR2 inhibitors blocks gMDSC immigration, resulting in enhanced Th1 polarization, induced CD8+KLRG1+ effector T cells with high cytotoxic activity, and improved survival in multiple NSCLC models.
SHP2, encoded by PTPN11, is required for activation of RAS upstream of the RAS guanine nucleotide exchange proteins SOS1/2. Consequently, SHP2 inhibitors (SHP2i) can block downstream signaling by overactive receptor tyrosine kinases (RTK) and “cycling” RAS mutants (e.g., KRASG12C), which retain significant intrinsic RAS-GTPase activity and thus rely on continued SOS1/2 activity (1). In addition to its potential tumor cell–autonomous actions, SHP2 plays critical roles in normal RTK, cytokine, integrin, and immune checkpoint receptor signaling (2). “Driver” mutations (e.g., amplified or mutant RTKs, mutant KRAS) significantly—and differentially—also have tumor cell–intrinsic and –extrinsic effects and evoke distinct cellular and humoral responses in different tissues (3). Consequently, SHP2is likely have important, potentially driver-specific, effects on the tumor microenvironment (TME), including potentially complex effects on antitumor immunity (2, 4–6).
Most preclinical studies of SHP2is have used cell-derived xenografts (CDX) or patient-derived xenografts (PDX) established in immune-deficient mice or syngeneic tumor models implanted in the subcutaneous space. The former models lack adaptive immune responses; the latter rarely harbor the mutational spectrum of the cognate human disease and fail to reproduce key aspects of tissue-specific immunity (e.g., resident macrophages, T cells, etc.). As the response to targeted therapies in patients almost certainly reflects the composite of direct antitumor actions and effects on the TME, CDXs, PDXs, and subcutaneous syngeneic models could provide incomplete or even misleading information about SHP2i action. For example, we found that SHP2i, alone or in combination with KRASG12C inhibitor (G12Ci), increased intratumor T cells in KRASG12C-driven non–small cell lung cancer (NSCLC) and pancreatic ductal adenocarcinoma (PDAC; ref. 4). However, the degree of T-cell function in these models is unknown. Moreover, combination with anti–PD-1 treatment resulted in only minimal improvement in efficacy (4), urging for a more efficacious, rational combination strategy that enhances immune-modulatory effects of SHP2is.
Clinical trials of SHP2is alone or in various combinations are ongoing for multiple disease indications. Systematic characterization of the immune-modulatory effects of SHP2is in genetically defined, orthotopic or autochthonous, immune-competent tumor models that better reflect human cancers might provide important insights into how best to combine these agents. To this end, we characterized the tumor cell–autonomous and nonautonomous effects of SHP2 inhibition in genetically engineered mouse models (GEMM) of Kras- and Egfr-mutant NSCLC and used this information to identify and evaluate a novel, rational combination of SHP2 and CXCR1/2 inhibitors.
We and others previously demonstrated the efficacy of SHP2i in various KRAS-mutant malignancies, including KRAS-mutant NSCLC (4, 7–11). To systematically explore rational new SHP2i/immune-oncology (I/O) combinations, we first performed a detailed analysis of the effects of the tool compound SHP099 on orthotopic KrasG12D;Trp53−/− (KP) NSCLC allografts. KP cells were injected intravenously, and lung tumor formation was monitored by MRI. Once tumors had reached 100 mm3 (∼4 weeks), SHP099 (75 mg/kg/day) treatment was initiated. As expected from previous results (4, 7), SHP099 had significant single-agent efficacy (Fig. 1A; Supplementary Fig. S1A). Immune profiling and IHC staining revealed a significant increase in T lymphocytes in tumor nodules from SHP099-treated mice, compared with vehicle controls, without any change in CD8/CD4 ratio (Fig. 1B and C; Supplementary Fig. S1B). We also observed a marked reduction in alveolar macrophages (CD11b− CD11c+ Siglec-F+) and M2-like (CD206+/PD-L1+) bone marrow (BM)–derived macrophages (CD11b+ CD11c− Ly6C− Ly6G− F4/80+), with a significant increase in M1/M2 macrophage ratio, as well as an increase in tumor-infiltrating B lymphocytes (CD19+ cells; Fig. 1D; Supplementary Fig. S1B and S1C). There was no significant change in the proportion of the M1-like BM-derived macrophages, but the total number of BM-derived macrophages decreased. Notably, in a previous study, the effects of SHP2i on subcutaneous syngeneic models revealed a similar effect on M2 macrophages and M1/M2 ratio (5). Alveolar and M2-like BM-derived macrophages suppress T-cell function (12–14), so the observed decrease in these cells could help explain the increased T-cell infiltration and antitumor effects of SHP099. The effects of B lymphocytes in various tumor models are complex (15–18), although recently they were found to exert important antitumor effects, including in NSCLC (19, 20). Unfortunately, although monocytic myeloid-derived suppressor cells (mMDSC; CD11b+ Ly6C+ Ly6G−) were unaffected, SHP099 treatment led to a significant increase in infiltrating granulocytic myeloid-derived suppressor cells (gMDSC; CD11b+ Ly6G+; Fig. 1D), consistent with our earlier observations (4). These cells can potently inhibit T-cell function (21) and thus could limit the antitumor effects of SHP099.
We next explored the functional effects of these immune cell populations on KP allograft growth in the absence or presence of SHP099. Depletion experiments indicated significant antitumor actions of B cells on KP allografts, as well as secondary effects on specific T-cell subsets (Supplementary Fig. S1D and S1E). Mice depleted for CD4 or CD8 T cells also had significantly larger tumors compared with control IgG–injected mice. SHP099 treatment still suppressed KP tumor size, but its effects were reduced in mice lacking either T-cell population (Fig. 1E and F). Although SHP099 treatment significantly decreased KP allograft growth, it prolonged median overall survival by only 1 week. Thus, although the infiltrating CD4 and CD8 T cells have demonstrable antitumor effects, they clearly cannot orchestrate a substantial and sustained antitumor response. More detailed characterization of these cells revealed that they did not show a phenotype consistent with “exhaustion” (PD-1+ TIM3+), but a significant proportion were naïve (CD44− CD62L+) and a minority exhibited effector phenotype (CD44+ CD62L−; Fig. 1G). Most importantly, only rare infiltrating CD8+ T cells expressed granzyme B (Gzmb), a functional marker of cytotoxic T cells (CTL), and a small fraction of CD4+ T cells expressed Tbet (Tbx21), the defining marker for Th1 cells (Fig. 1G). These results suggest that although SHP099 treatment promotes T-cell immigration into KP tumors, these cells are largely nonfunctional and have only minimal antitumor activity.
We hypothesized that gMDSCs, the largest immune cell population in KP tumors before treatment, whose abundance increases even further following SHP099 administration (Fig. 1D), were responsible for the observed impairment in T-cell function. As an initial test of this hypothesis, we depleted gMDSCs (and other granulocytic populations) by injecting rat anti-Ly6G along with anti-rat antibody (22) and monitored the effects of SHP099 (Fig. 1H). This approach resulted in the expected depletion of gMDSCs along with an increase in intratumor T cells, which increased further in SHP099-treated mice (Fig. 1I; Supplementary Fig. S1F). Strikingly, CD8 T-cell activation (as marked by increased CD44+/CD62L− and concomitantly decreased CD44−CD62L+ cells) also increased, and there was a marked increase in Gzmb-expressing CD8 cells (Fig. 1J). Depletion of gMDSCs also resulted in a basal (without SHP099) increase in activated CD4 T cells and enhanced Th1 differentiation in vehicle- and SHP099-treated mice. These results confirm that intratumor gMDSCs exert immune-suppressive effects on tumor-associated CD8+ and CD4+ T cells. Consistent with its influence on the tumor-infiltrating T cells, depletion of gMDSCs also dampened tumor growth (Supplementary Fig. S1G).
Antibody-based depletion of granulocytes is not accomplished easily (nor desirable, given the attendant risk of infection) in patients. To search for more clinically applicable strategies for preventing SHP2i-induced gMDSC infiltration, we examined immune modulators produced by SHP099-treated KP tumor cells. Transcriptomic analysis identified Cxcl1 and Cxcl5, whose protein products (CXCL1 and CXCL5) signal via CXCR2, a major chemotaxis receptor for gMDSCs (23), as the most upregulated chemokines following SHP099 treatment (Fig. 2A; Supplementary Table S1). Cxcl1/5 was not induced upon SHP099 treatment of KP cells expressing PTPN11TM/QL, which encodes an SHP099-resistant mutant, as confirmed by pERK immunoblotting (Fig. 2B and C). These results show that Cxcl1/5 induction is a consequence of SHP2 inhibition, rather than an off-target effect of SHP099. Allografts established with SHP099-resistant KP cells still exhibited significant increases in T- and B-cell infiltration, along with depletion of alveolar and BM-derived macrophages, following SHP099 treatment. By contrast, gMDSCs did not increase in the same tumors (Fig. 2D). This result indicates that KP-produced CXCL1/5 is essential for SHP099-evoked gMDSC immigration, emphasizes how SHP2i action reflects complex mix of tumor cell–autonomous and –nonautonomous effects, and comports with a previous report that CXCR2 ligands play important roles in gMDSC recruitment in KRAS-mutant colorectal cancer (24).
Promoter-enrichment analysis of the KP cell RNA-sequencing (RNA-seq) data suggested activation of genes with RELA-binding sites (P = 0.012), which include CXCL1/5 (Fig. 2E) and CXCL6 (see below). As RELA is an NFκB family member, we evaluated the effects of SHP2i on the transcriptional activity of a GFP reporter containing four NFκB sites that was introduced stably into KP cells. Consistent with a functional role for NFκB, SHP099 treatment significantly increased reporter expression as shown by flow cytometry for GFP (Fig. 2F). Treatment of KP cells with the MEK inhibitor trametinib also evoked increased NFκB reporter activity (Fig. 2F). These results suggested that NFκB-mediated Cxcl1/5 induction might be a general response to RAS/ERK pathway inhibition in KP cells. Immunoblotting showed that, after a significant delay, SHP2 inhibition (and most likely RAS/ERK pathway inhibition) induced NFκB pathway activation (Fig. 2G) and Cxcl1/5 expression (Fig. 2H) upstream of I-KB (Fig. 2G). Combining SHP099 or trametinib with the IKK inhibitor ML120B abolished transcriptional upregulation of Cxcl1/5, confirming the NFκB dependence of Cxcl1/5 induction upon SHP2/MEK inhibition (Fig. 2H). To test whether these effects of SHP2/MEK inhibition on Cxcl1/5 regulation translated to altered levels of the cognate proteins, we performed multiplexed, Luminex analysis on supernatants from KP cells treated with SHP099 or trametinib (Fig. 2I; Supplementary Table S2). In addition to confirming elevated CXCL1 and CXCL5 secretion upon SHP099 or trametinib treatment, we also observed increased levels of CCL5 (RANTES) and CXCL10 (IP-10) upon SHP2/MEK inhibition. CCL5 and CXCL10 are key chemokines that recruit and activate T cells and thereby play important roles in antitumor immunity (25–32). Moreover, CCL5 has effects on the myeloid cell compartment (33, 34). Consequently, their upregulation could contribute to the enhanced T-cell infiltration upon SHP099 treatment, as well as the depletion of alveolar and M2-like macrophages. However, T-cell infiltration/macrophage depletion occur even in tumors that are SHP099 resistant (Fig. 2D), arguing for additional sources of these agonists. CXCL1/5/10 and CCL5 levels also were increased in KP tumor lysates from mice treated with SHP099 (Fig. 2J), and we observed significant increases in nuclear p65 and overall CXCL1 expression in the same cells upon SHP099 treatment in wild-type (WT), but not SHP099-resistant, KP tumors (Fig. 2K). In concert, these data confirm activation of the NFκB pathway and increased levels of CXCR2 ligands in KP allografts upon SHP2 inhibition.
We then explored the potential relevance of these observations to human KRAS-mutant NSCLC. CXCL1/5 are members of the GRO family of chemokines, whose genes reside in a common chromosomal region (4q13.3 in humans) and are often coregulated. Treatment of a panel of human KRASG12C-mutant cell lines with SHP2i-, MEKi-, or the G12Ci MRTX849 led to upregulation of multiple GRO family genes, including CXCL1 and CXCL6 (n.b., CXCL6 is the homolog of mouse Cxcl5). Induction of most of these genes was blocked by ML120B treatment (Supplementary Fig. S2A). Luminex analysis confirmed increased secretion of CXCL1, CCL5, and CXCL10 in these cell lines upon SHP2/KRAS/MEK inhibition, and again, induction was abrogated by ML120B treatment (Supplementary Fig. S2B; Supplementary Table S2), indicating NFκB dependence. These results predict that increased immigration of gMDSCs (and perhaps other GRO-dependent immune-modulatory cells) might limit the efficacy of SHP2is, MEKis, or G12Cis in human NSCLC as well. Consistent with this hypothesis, we observed increased gMDSC infiltration in tumors from mice treated with trametinib or MRTX849 (Supplementary Fig. S3A).
SX682 is a potent inhibitor of CXCR2 (24, 35) currently in multiple clinical trials (NCT03161431, NCT04477343, NCT04574583, and NCT04245397). Given the results above, we tested the effects of combined SHP2 and CXCR1/2 inhibition. SHP099/SX682 significantly reduced (although did not eliminate) gMDSC infiltration compared with SHP099 alone (Fig. 3A; Supplementary Fig. S4A). Combination therapy also led to a further increase (over single-agent SHP099) in intratumoral CD4 and CD8 T cells, which exhibited a stronger effector phenotype and enhanced proliferation. Tumor-associated CD8+ T cells in combination-treated groups expressed higher levels of Tbet and Gzmb, consistent with a stronger cytolytic phenotype, whereas Th1 polarization (increased percentage of Tbet+ and decreased percentage of GATA3+ cells) was enhanced in the CD4+ population. Injection of anti-CXCL1 and anti-CXCL5 neutralizing antibodies had similar effects on gMDSCs and T cells (Supplementary Fig. S4B–S4D), providing further evidence that the adverse effects of SHP2i on promoting gMDSC infiltration are driven mainly by CXCL1/5 activation of CXCR2.
Treatment with single-agent SHP099 or SX682 for two weeks, at which time vehicle-treated mice start to die, significantly reduced, but did not eliminate, KP allograft growth (Fig. 3B). By contrast, the SHP099/SX682 combination completely suppressed tumor growth at this time point. More importantly, combination treatment significantly prolonged the survival of KP tumor–bearing mice (median, 38 days) as compared with single-agent SHP099 (median, 27 days) or SX682 (median, 21.5 days), and more than doubled overall survival, compared with vehicle-treated (median, 18 days) mice (Fig. 3C). There was no apparent toxicity following long-term (more than 5 weeks) SHP099/SX682 treatment (Supplementary Fig. S4E and S4F).
Previous work indicated that CD11b and Ly6G expression (e.g., CD11b+ Ly6G+ cells) alone does not reliably distinguish gMDSCs from normal neutrophils (36, 37). Therefore, we used single-cell RNA-seq analysis (scRNA-seq) to systematically explore the potential heterogeneity of the gMDSC population following SHP099/SX682 treatment. Data were analyzed using K-nearest-neighbor–based Network graph drawing Layout (KNetL) maps, a graph drawing–based dimensionality reduction algorithm that shows better distinctions in the complex cell communities as compared with t-distributed stochastic neighbor embedding (t-SNE) or uniform manifold approximation and projection (UMAP) analysis (Fig. 3D; Supplementary Fig. S5A–S5C). As expected, Cxcl1 and Cxcl5 were largely restricted to tumor cells, although there was some expression in cancer-associated fibroblasts (CAF; Fig. 3E). Consistent with a previous survey of human NSCLC and KP allografts (38), we identified six (N1–N6) S100a8+ granulocytic cell clusters in KP tumors (Fig. 3F and G; Supplementary Table S3). Most of these (N1–5) increased after SHP099 treatment, comporting with our flow cytometry results. Interestingly, however (and consistent with the incomplete reduction in the gMDSC population by flow cytometric analysis, Fig. 3A), coadministration of SX682 only blocked infiltration of cells in cluster N4 (Fig. 3H). Notably, N4 cells express significantly higher levels of S100a8/9, compared with the other granulocytic cell populations (Fig. 3G), a phenotype that distinguishes gMDSCs from normal neutrophils (36, 37, 39, 40). In addition, N4 cells preferentially express Cybb (which encodes NOX2) and Lcn2 (Fig. 3G), raising the possibility that they might suppress T cells by generating superoxide and/or inducing apoptosis in T cells (21, 41). N2 and N3 cells are also induced upon SHP099 treatment but were unaffected by coadministration of SX682 (Fig. 3H). These cells preferentially express genes encoding cytokines and chemokines that promote T-cell recruitment and/or T-cell activation/differentiation, including TNFα, IL1α, CXCL10, CCL3, and CCL4 (42, 43). Consequently, these findings suggest a proinflammatory role for these granulocytic cells.
Although flow cytometric analysis revealed induction of cytotoxic markers in the CD8+ T cells and more Th1 differentiation in combination-treated mice (Fig. 3A), scRNA-seq also revealed substantial phenotypic heterogeneity in the T-cell population. We identified nine clusters (T1–T9) of Cd3e+ T cells in KP tumors (Fig. 3I and J; Supplementary Table S4). SHP099/SX682 treatment led to a specific increase in cells in the Cd8+ clusters T2 and T5 (Fig. 3J and K). Cluster T2 comprises Klrg1+Cx3cr1+ effector T cells (Teff) that preferentially express cytotoxic genes including Gzmb and Gzmk (Fig. 3J). KLRG1 marks highly cytotoxic and proliferative CD8+ Teffs in other settings (44–47). Interestingly, these cells also expressed Klrc1, which encodes the immune checkpoint receptor NKG2A, but not Pdcd1, Ctla4, or Lag3 (Fig. 3J). NKG2A blockade was recently suggested to enhance antitumor immunity in subcutaneous syngeneic models (see Discussion; ref. 48). Cells in cluster T5 preferentially express genes associated with proliferation (e.g., Ccna2, Aurka, and Tk1) but not with effector function (Fig. 3J). In concert, these findings suggest that CXCR2 inhibition specifically blocks infiltration of S100a8/9hi gMDSCs induced by SHP2i, which in turn leads to the generation of Klrg1+ CD8+ Teffs with high cytotoxic and proliferative capability.
Activating mutations in EGFR are another major cause of NSCLC, and SHP2 plays a critical role in EGFR signaling. Given the effects of SHP099 on the KP TME, we asked whether SHP099/SX682 might also have utility in Egfr-mutant NSCLC GEMMs. We first confirmed that, consistent with a previous report (49), SHP099 possesses cell-intrinsic ability to suppress MEK/ERK activity and proliferation in EGFR-mutant NSCLC cell lines (Fig. 4A and B). SHP2i, MEKi, or EGFRi induced NFκB-dependent expression of several CXCR2 ligands in these lines (Fig. 4C; Supplementary Fig. S6A). EGFR, SHP2, or MEK inhibition also induced enhanced secretion of CXCL1, CCL5, and CXCL10 in these lines, and induction can be abrogated with ML120B treatment (Supplementary Fig. S6B; Supplementary Table S2), comporting with their effects on KRAS-mutant NSCLC cells. We then tested SHP099 and SX682 alone or in combination in an osimertinib-resistant Egfr-mutant GEMM (EgfrT790M,L858R,C797S, hereafter EGFR-TLC). The EGFR driver in this model has the classic L858R mutation, in addition to the gatekeeper mutation (T790M) that renders it resistant to first-generation EGFR inhibitors and the C797S mutation that eliminates covalent binding of osimertinib (Fig. 4D).
Osimertinib-resistant EGFR-mutant NSCLC constitutes a major unmet medical need, as there is currently no effective treatment for these tumors. Compared with KP tumors (Fig. 1B and D), EGFR-TLC tumors contained more alveolar macrophages and fewer T cells and gMDSCs (Fig. 4E). Consistent with its effects on KP allografts, SHP099 treatment depleted alveolar macrophages and increased T-cell infiltration in tumor-bearing EGFR-TLC mice, but also increased intratumor gMDSCs (Fig. 4E; Supplementary Fig. S6C). Single-agent SHP099 also failed to evoke T-cell activation, CD8+ CTLs, or Th1 cells in this EGFR-mutant model (Fig. 4F). By contrast, SHP099/SX682 significantly suppressed gMDSC infiltration and promoted more T-cell infiltration, greater activation of CD8+ T cells, accompanied by increased proliferation and acquisition of cytolytic phenotype, and enhanced Th1 polarization of CD4+ T cells (Fig. 4F). Single-agent SHP099 led to a 50% reduction in tumor size by 4 weeks, but SHP099/SX682 significantly enhanced treatment efficacy, leading to >80% tumor shrinkage (Fig. 4G and H).
We also investigated the effects of “up-front” administration of the triple combination of SHP099, SX682, and osimertinib in an osimertinib-sensitive GEMM, EgfrT790M,L858R (EGFR-TL; Fig. 4I). Again, SHP099 alone or in combination strongly depleted alveolar macrophages, while increasing intratumor gMDSCs and T cells. The SHP099/SX682/osimertinib combination resulted in lower levels of gMDSCs and evoked the largest increase in tumor-associated T cells (Fig. 4J; Supplementary Fig. S6D). Single-agent SHP099 treatment had largely cytostatic effects in this model, whereas osimertinib-containing combinations led to complete responses, which were durable for at least 8 weeks (Fig. 4K). Upon drug withdrawal, however, there were marked differences in the rate of tumor recurrence, with single-agent osimertinib-treated mice recurring fastest, followed by the osimertinib/SHP099 group, and finally mice treated with the three-drug combination (Fig. 4K).
Finally, we asked whether RAS/ERK pathway inhibition results in induction of CXCR2 ligand genes and gMDSC recruitment in patients with NSCLC. Remarkably, CXCL1 and CXCL6 mRNA levels were increased after MRTX849 treatment in matched biopsy samples from 4 of 5 patients with KRASG12C-mutant NSCLC (Fig. 4L). Moreover, weaker induction of CXCL1/CXCL6 expression was associated with more beneficial RECIST response and longer response duration (Fig. 4L and M). All of the MRTX849-treated patients also showed a substantially stronger neutrophil transcriptional signature (Fig. 4N). Finally, we performed scRNA-seq on several (unmatched) biopsy samples from G12Ci-naïve and G12Ci-treated patients with NSCLC (Fig. 4O). As predicted by our preclinical studies, tumor cells from G12Ci-treated patients had higher levels of CXCL1 and CXCL6 than those from G12Ci-naïve patients (Fig. 4P). We also observed significantly higher proportions of gMDSCs in G12Ci-treated, compared with G12Ci-naïve or non-KRAS–mutant, tumors (Fig. 4Q).
SHP2is have antitumor effects in KRAS-mutant and EGFR-mutant NSCLC GEMMs, and initial reports demonstrate efficacy of a clinical-grade SHP2i on patients with cycling KRAS-mutant NSCLC (4–10, 49). The extent to which these therapeutic effects reflect direct actions on cancer cells versus cells in the TME has not been studied extensively in orthotopic or autochthonous models. Also, potentially adverse consequences of SHP2 inhibition, particularly on tumor-associated immune cells, which might limit the efficacy of SHP2is alone or in combination, have not been defined. We find that in addition to direct antitumor effects on the highly aggressive KP allograft model, the tool SHP2i SHP099 has several beneficial effects on the immune TME, lowering the level of tumor-promoting alveolar macrophages and M2 BM-derived macrophages, while increasing tumor-associated B and T lymphocytes. SHP2i-evoked T cells have significant antitumor effects as revealed by depletion experiments. Yet these effects are limited, because SHP099 also induces immigration of gMDSCs, which suppress T-cell activation, proliferation, and cytolytic differentiation. Furthermore, analysis of pre- and posttreatment biopsies shows that similar events occur in patients with NSCLC on RASG12C inhibitor trials.
Others have reported that SHP2is evoke meaningful antitumor T-cell responses (5, 6). These studies analyzed subcutaneous syngeneic tumors, which lack tissue-specific immunity and display mutational spectra that do not reflect the cognate human malignancies. We previously showed that the same syngeneic cancer cells evoke different TMEs and have quantitatively different drug responses when established at subcutaneous versus orthotopic sites (4), in accord with other reports (50). Our group previously demonstrated that SHP2i, alone or in combination with G12Ci, increased intratumor T cells in KRASG12C-driven NSCLC and PDAC. Consistent with the effects observed here, most of those T cells failed to express Gzmb (4). Our depletion studies, flow cytometry, and scRNA-seq analysis reveal that SHP2i alone, despite evoking significant T-cell infiltration in multiple genetically defined, orthotopic, and autochthonous NSCLC models, is unable to generate CD8+ Teffs or enable a strong, highly effective antitumor T-cell response.
Transcriptional profiling, reporter assays, inhibitor treatment, and neutralizing antibody studies indicate that gMDSC immigration is a consequence of NFκB-dependent CXCL1/5 production by KP tumor cells. Increased NFκB-driven transcription appears to result from decreased ERK activation downstream of SHP099, as similar effects are observed upon MEKi and G12Ci treatment. Furthermore, analogous induction of CXCR2 ligands occurs upon SHP2i, G12Ci, or MEKi treatment of human KRAS-mutant NSCLC lines. Most importantly, induction of CXCR2 ligands and evidence of enhanced gMDSC infiltration is seen in biopsy samples from patients treated with two different G12Cis. We found that EGFRi, SHP2i, or MEKi treatment of EGFR-mutant NSCLC lines also induced such chemokines. Taken together, these data suggest that CXCR2 ligand induction (and possibly gMDSC immigration) might be a general response to EGFRi/SHP2i/KRASi/MEKi treatment. We note that two human NSCLC cell lines did not show induction of CXCR2 ligands (H23 and H1650). These cells have PTEN alterations, raising the possibility that our observations might not be applicable to PTEN/PI3KCA-mutant NSCLC.
In addition to CXCR2 ligands, we also saw NFκB-dependent upregulation of CCL5 and CXCL10 upon EGFR/SHP2/KRAS/MEK inhibition. Inhibition of EGFR/SHP2/KRAS/MEK has been shown to enhance T-cell infiltration in several studies (4–6, 51–54), but little is known about the underlying mechanism. Our findings suggest that treatment-induced CCL5 and CXCL10 might promote such T-cell recruitment, although as discussed above, experiments with SHP099-resistant mutants indicate that such effects are unlikely to solely reflect tumor cell–autonomous actions of SHP2is.
Earlier studies assigned tumor-infiltrating granulocytes to “antitumor” N1 or “protumor” N2 subsets (42). N1 granulocytes secrete proinflammatory cytokines (e.g., TNFα, CXCL10, and IL12) that facilitate antitumor T-cell responses; N2 subsets suppress or kill T cells via production of reactive oxygen species, arginase, and/or nitric oxide (21, 42). An scRNA-seq analysis of a large panel of human and mouse lung tumors revealed more extensive phenotypic heterogeneity (38), but the functional importance of these subpopulations remained largely unknown. Recent reports suggest that gMDSCs can be distinguished from normal neutrophils by virtue of high S100a8/9 expression (36, 37, 39, 40). Consistent with these findings, we identified six unique subsets of tumor-infiltrating granulocytes, with transcriptomic profiles suggesting proinflammatory (N2: Tnf, Il1a, Cxcl10; N3: Ccl3, Ccl4, Tnf) and immunosuppressive (N4: S100a8/9, Cybb, Lcn2) roles, respectively. SHP099 treatment evokes increased infiltration of nearly all granulocytic subpopulations into KP tumors, including those with proinflammatory and immunosuppressive function. Intriguingly, coadministration of SX682 specifically blocks infiltration of S100a8/9hi immunosuppressive (N4) granulocytes. NOX2 is an NADPH oxidase, which catalyzes superoxide generation, whereas LCN2 has iron-chelating function and induces T-cell apoptosis (41). S100A8/9 secreted by gMDSCs can also reactivate dormant KP tumor cells (39). Dormancy is increasingly recognized as a key mechanism by which tumor cells, including KRAS-mutant NSCLC (55), escape the effects of targeted therapy and even conventional chemotherapy (56–58). Hence, blocking N4/S100A8/9hi gMDSC infiltration by CXCR2 inhibition could have multiple beneficial effects in therapeutic combinations. Because CXCR1/2 inhibition specifically affects immunosuppressive gMDSCs without impairing proinflammatory granulocyte infiltration, and SX682 and other CXCR1/2 inhibitors already are in clinical trials (NCT03177187 and NCT03473925), our findings could have immediate clinical implications.
Consistent with the above analysis, gMDSC-mediated suppression by SX682 treatment resulted in increased numbers of intratumor CD8 T cells. These cells are more activated and exhibit Klrg1+ Teff phenotype marked by high expression of cytolytic genes. Intratumor CD4 T-cell number, activation, and Th1 polarization also increased. Notably, KLRG1 marks a subset of highly cytotoxic, proliferative, and often short-lived CD8+ Teffs induced in the setting of certain infections by the combination of strong T-cell receptor and inflammatory signals (44–47). For example, Klrg−/− mice have more total and activated CD4+ T cells and survive longer after infection with Mycobacterium tuberculosis (59), suggesting that KLRG1 might act as an immune checkpoint receptor. However, its role in cancer remains largely unknown.
Intriguingly, we found that these CD8+Klrg1+ Teffs preferentially express the checkpoint Klrc1 (NKG2A) but not more conventional immune checkpoints such as PD-1, CTLA4, and LAG3. Such findings may explain why KRAS-mutant NSCLC responds incompletely to PD-1/PD-L1 blockade (60, 61). NKG2A is best known as a killer inhibitory receptor on NK cells (62, 63). More recent work showed that it is also expressed in early CD8+ Teffs, and NKG2A blockade significantly enhances antitumor protective immunity induced by cancer vaccines (48). Anti-NKG2A mAb (monalizumab) treatment was also found to enhance CD8+ T-cell function in a phase II clinical trial (64), and a phase III trial of the effects of NKG2A blockade is ongoing (NCT04590963).
Notably, combined SHP099/SX682 inhibition more than doubled the survival of mice bearing extremely aggressive KP allografts and substantially reduced tumor burden in osimertinib-sensitive and -resistant NSCLC GEMMs. The tumor (neo)antigen(s) responsible for the antitumor T-cell responses in the Kras- or Egfr-mutant models are undefined and future efforts should be directed toward their identification. Although our results support the testing of SHP2i/CXCR1/2i combinations in patients, analysis of the remaining T-cell populations suggests that combining SHP2i (and/or other RAS/ERK pathway inhibitors) with CXCR2 and NKG2A blockade might be even more efficacious. As drugs targeting each are currently being tested in various clinical trials, expeditious testing of these concepts should be possible. Although our focus here was on KRAS- and EGFR-mutant NSCLC, our results suggest that induction of CXCR2 ligands and consequently increased granulocytic infiltration into tumors could also be a common, unavoidable “side effect” of RAS/ERK pathway inhibition in other NSCLC subsets and tumor types.
Cell Culture and Reagents
Mouse KP cells and the H358, H1373, H2122, H23, CALU1, H1975, H1650, and HCC827 cell lines were from stocks in the Wong laboratory. The PC9 cell line was purchased from Sigma-Aldrich. Cells were cultured in RPMI media supplemented with 10% FBS and 1% penicillin–streptomycin in a 37°C incubator with 5% CO2. Cells were tested routinely (every 3 months) for Mycoplasma contamination by PCR (65), and genotyped by short tandem repeat analysis at IDEXX Bioresearch. SHP099 was synthesized by Wuxi AppTec. SX682 was obtained from Syntrix Pharmaceuticals. Trametinib was purchased from Selleckchem. The NFκB–GFP-Reporter was purchased from System Biosciences. The PTPN11T253M/Q257L (TM/QL) expression construct was reported previously (66).
Viruses were produced by cotransfecting HEK293T cells with lentiviral constructs and packaging vectors (pVSV-G+dR8.91) in DMEM supplemented with 10% FBS. Transfection media were replaced by fresh media after 12 hours. Virus-containing supernatants were collected 60 hours later, passed through 0.45-μm filters, and then used to infect various cultured cells in the presence of 8 μg/mL polybrene (Sigma).
All animal studies were reviewed and approved by the Institutional Animal Care and Use Committee at New York University Grossman School of Medicine. For the orthotopic allograft lung cancer model, six-week-old male B6 WT mice were purchased from The Jackson Laboratories. KP cells (106) in 200 μL PBS were injected into the tail vein of each mouse. MRI was used to monitor tumor formation and progression in orthotopic models. Mice were treated with vehicle or SHP099 (75 mg/kg) daily (qd) by oral gavage. For CXCR1/2 inhibitor studies, mice were dosed with SX-682 (100 mg/kg qd), either alone or in combination with SHP099 (75 mg/kg qd).
The EGFR-mutant NSCLC GEMMs harbor a conditional activating mutation of human of EGFRL858R/T790M with/without C797S (TLC GEMM or TL GEMM, respectively) at the collagen I locus (67). Cre-recombinase was induced by intranasal inhalation of 5 × 107 plaque-forming units of adeno-Cre (University of Iowa adenoviral core), and tumors (adenocarcinomas) typically appeared 16 weeks after induction. For drug treatment studies, age-matched littermates (6- to 8-week-old) were induced, and tumor burden was monitored by MRI. Once tumor size reached 300–400 mm3 (∼20 weeks after adenoviral inoculation), mice were randomized to experimental groups. No gender differences were observed in tumor growth or drug response. Mice were evaluated by MRI every other week to quantify lung tumor burden before and after drug treatment. TLC tumor–bearing mice were treated with vehicle, SHP099 (75 mg/kg qd), SX-682 (100 mg/kg qd), or both drugs. For TL mice, in addition to these 4 arms, osimertinib (5 mg/kg qd) was introduced in combination with SHP099 or SHP099+SX-682.
To specifically deplete T cells, 400 μg of rat IgG2b (clone LTF-2; Bio X Cell), anti-mouse CD4 antibody (clone GK1.5; Bio X Cell), or anti-mouse CD8 antibody (clone 2.43; Bio X Cell) were injected into the mouse peritoneum (i.p.) one day before commencement of vehicle/SHP099 treatment. Subsequently, antibodies (200 μg) were administered twice a week throughout the experiments. For specific depletion of gMDSCs, 25 μg of rat IgG2a (clone 2A3; Bio X Cell) or anti-mouse Ly6G antibody (clone 1A8; Bio X Cell) were administered daily in the first week, 2 days before commencing vehicle/SHP099 treatment. The dose was increased to 50 μg starting from the second week; in addition, mice received 50 μg of anti-rat secondary antibody (clone MAR18.5; Bio X Cell) every other day. For in vivo neutralization of CXCL1 and CXCL5, 80 μg of anti-CXCL1 (clone 20326; Leinco) and anti-CXCL5 (clone 61905; Leinco) antibodies were administered i.p. every 5 days. For depletion of B cells, 250 μg of anti-CD20 antibody (clone SA271G2; BioLegend) was administered two days before commencing vehicle/SHP099.
Animals were anesthetized with isoflurane, and MRI of the lung field was performed using the BioSpec USR70/30 horizontal bore system (Bruker) to scan 24 consecutive sections. Overall, tumor volumes within the whole lung were quantified using 3-D slicer software to reconstruct MRI volumetric measurements as described (68). Acquisition of the MRI signal was adapted according to cardiac and respiratory cycles to minimize motion effects during imaging.
Tumor nodules were resected from lungs with visible tumors. Nodules were then cut into small pieces and digested with collagenase/hyaluronidase (STEMCELL Technologies) and DNase I (STEMCELL Technologies) in Advance DMEM/F12 media (Gibco) for 45 minutes in 37°C. Cell suspensions were then filtered through 70-μm cell strainers (Fisher) and washed with cold FACS buffer (2% FBS in PBS). Red blood cells were lysed by resuspending cell pellets in ACK Lysis Buffer (Gibco) for 2 minutes, followed by washing with cold FACS buffer.
Freshly prepared cell suspensions were blocked with 1% mouse serum (Jackson ImmunoResearch), 1% rat serum (Jackson ImmunoResearch), and 2% mouse FcR Blocking Reagent (Miltenyi Biotec) for 15 minutes at 4°C. Fluorophore-conjugated primary antibodies against cell-surface antigens were added to the cell suspensions and incubated for 30 minutes at 4°C. Cells were then washed with cold PBS and stained with LIVE/DEAD UV (Invitrogen) according to the manufacturer's instructions. Cells were then washed with cold FACS buffer and fixed with Foxp3/Transcription Factor Staining Buffer Set (eBioscience) according to the manufacturer's protocol. Fixed cells were again blocked with 1% mouse serum, 1% rat serum, and 2% mouse FcR Blocking Reagent for 15 minutes at 4°C and then stained with fluorophore-conjugated primary antibodies against intracellular antigens. Cells were then washed, and data were acquired on an LSR II Flow Analyzer (BD Biosciences). Details on the antibodies used are provided in Supplementary Table S5. Data acquired were analyzed by using FlowJo software (BD Biosciences).
Whole-cell lysates were prepared in a modified RIPA buffer (50 mmol/L Tris-HCl, pH 8.0, 150 mmol/L NaCl, 2 mmol/L EDTA, 1% NP-40, and 0.1% SDS, without sodium deoxycholate), supplemented with protease [40 μg/mL phenylmethylsulfonylfluoride (PMSF), 2 μg/mL antipain, 2 μg/mL pepstatin A, 20 μg/mL leupeptin, and 20 μg/mL aprotinin] and phosphatase (10 mmol/L NaF, 1 mmol/L Na3VO4, 10 mmol/L β-glycerophosphate, and 10 mmol/L sodium pyrophosphate) inhibitors. After removal of insoluble material by centrifugation in a microfuge, protein levels were quantified by using the DC Protein Assay Kit (Bio-Rad). Proteins were resolved by standard SDS-PAGE and transferred in 1× transfer buffer (25 mmol/L Tris base, 200 mmol/L glycine) and 15% methanol. Membranes were incubated with the appropriate primary and secondary antibodies labeled with IRDye (680 and 800 nm) and then visualized by the LI-COR system.
Frozen tumor nodules were homogenized in Cell Lysis Buffer (20 mmol/L Tris-HCl pH 7.5, 150 mmol/L NaCl, 1 mmol/L Na2EDTA, 1 mmol/L EGTA, 1% Triton, 2.5 mmol/L sodium pyrophosphate, 1 mmol/L B-glycerophosphate, 1 mmol/L Na3VO4, 40 μg/mL PMSF, 2 μg/mL antipain, 2 μg/mL pepstatin A, 20 μg/mL leupeptin, and 20 μg/mL aprotinin. Cytokines and chemokines were measured by Mouse Cytokine 32-plex Assay (Millipore) or Human Cytokine 41-plex Assay (Millipore) implemented on the Bio-Plex 200 system (Bio-Rad). Concentrations (pg/mL) of each protein were derived from 5-parameter curve-fitting models. Lower and upper limits of quantification (LLOQ/ULOQ) were derived from standard curves for cytokines/chemokines above or below detection.
Tumor-bearing mice were perfused with PBS, fixed in 10% formalin for 48 hours, washed in 70% ethanol, and embedded for histologic analysis. Five-micron–thick sections were cut from paraffin blocks and stained with hematoxylin and eosin or with Akoya Biosciences Opal multiplex automation kit reagents (Leica; catalog no. ARD1001EA) on a Leica BondRX autostainer, according to the manufacturers' instructions. Briefly, all slides underwent sequential epitope retrieval with Leica Biosystems epitope retrieval-1 (ER1; citrate-based, pH 6.0; catalog no. AR9961) or ER2 solution (EDTA-based, pH 9; catalog no. AR9640), primary and secondary antibody incubation with Opal fluorophores (Op570 Akoya, catalog no. FP1488001KT or Op690 Akoya, catalog no. FP1497001KT), and tyramide signal amplification. Primary antibodies against mouse CD3 (Bio-Rad, catalog no. MCA1477T), p65 (Cell Signaling Technology; catalog no. 8242S), and CXCL1 (Thermo; catalog no. PA5-115328) and horseradish peroxidase–coupled secondaries (Rabbit-on-Rodent HRP polymer; Biocare; catalog no. RMR622) or Rat 1-step HRP polymer (sBiocare; catalog no. BRR4016H) were removed during sequential heat retrieval steps, whereas the Opal fluorophores remained covalently attached to the epitope. Rodent-on-mouse Block M (Biocare; RBM961) was used prior to staining sections with the rat CD3 antibody. Sections were counterstained with DAPI and mounted with Prolong Gold Antifade Reagent (Invitrogen; P36930). Semiautomated image acquisition was performed on a Vectra Polaris multispectral imaging system. After whole-slide scanning at 20×, the tissue was outlined manually to select fields for spectral unmixing using InForm version 2.4.11 software from Akoya Biosciences.
RNA Extraction, cDNA Synthesis, and qPCR
Isolated tumor cells or trypsinized cancer cell lines were washed with PBS, and total RNA was extracted from cell pellets by using the miRNeasy Mini Kit (Qiagen). cDNAs were generated by using the SuperScript IV First-Strand Synthesis System (Invitrogen). qRT-PCR was performed with Fast SYBR Green Master Mix (Applied Biosystems), following the manufacturer's protocol, in 384-well format in a C1000 Touch Thermal Cycler (Bio-Rad). Differential gene-expression analysis was performed with CFX Manager (Bio-Rad) and normalized to GAPDH expression. Primers used are listed in Supplementary Table S6. Raw Ct values for qPCR are provided in Supplementary Table S7.
RNA-seq was performed on total RNA from isolated tumor cells by the PCC Genome Technology Center Shared Resource (GTC). Libraries were prepared using the Illumina TruSeq Stranded Total RNA Sample Preparation Kit and sequenced on an Illumina NovaSeq 6000 machine using 150-bp paired end reads. Sequencing results were demultiplexed and converted to FASTQ format using Illumina bcl2fastq software. Subsequent data processing and analysis were performed by the PCC Applied Bioinformatics Laboratories (ABL). Promoter–enrichment analysis was performed on bulk RNA-seq data by using Enrichr (69, 70).
scRNA-seq and Data Analysis
Single-cell suspensions isolated from treated tumors are individually barcoded with hashtag antibodies (anti-mouse TotalSeq-C antibodies; BioLegend). Each sample was washed four times with PBS + 2% BSA before pooling. Three replicates from same treatment groups were pooled as one sample (52,000 cells each sample). Specimens were then filtered through 70-μm strainers (Fisher). Cell concentration, singularity, and viability were confirmed with a hemocytometer before submission for scRNA-seq (10X Genomics). Experiments were performed with DNA LoBind 1.5 mL tubes (Eppendorf). scRNA-seq was performed by the GTC, with subsequent data processing and analysis performed by the ABL. Quality controls included calculation of the number of genes, UMIs, and the proportion of mitochondrial genes for each cell. Cells with a low number of covered genes (gene count < 500) or high mitochondrial counts (mt genes > 0.2) were filtered out, and the matrix was normalized on the basis of library size. A general statistical test was performed to calculate gene dispersion, base mean, and cell coverage. Genes with high coverage (top 500) and high dispersion (dispersion > 1.5) were chosen to construct the gene model (1,890 genes). The iCellR R package (v1.5.5; https://CRAN.R-project.org/package=iCellR) was used to perform principal component analysis and batch alignment on this model. t-SNE, UMAP, and K-nearest-neighbor–based Network graph drawing Layout (KNetL) were then performed. KNetL map has a zoom option that allows users to see variable levels of detail (more or fewer subpopulations in cell communities); in the studies here, we used a zoom of 650. The network layout used in KNetL map is force based (71), and the zoom option changes the force in the system. Force-directed graph drawing algorithms assign attractive (analogous to spring force) and repulsive forces (usually described as analogous to the forces in atomic particles) to separate all pairs of nodes. The network analysis used in KNetL map has long been used for single-cell analysis and clustering (72); here, the nodes of the network layout are extracted and UMAP is performed to create the final plot (“KNetL map”). PhenoGraph (72) clustering was then performed on the KNetL map results, and marker genes were found for each cluster and visualized on heat maps, bar plots, and box plots as indicated. Marker genes were then used to assign cell types. Imputation was used for some data visualizations only and not for the analysis. For imputation, we used KNN to average the expression of 10 neighboring cells per cell, using iCellR's “run.impute” function on KNetL data.
Data are expressed as mean ± SEM. Statistical significance was determined using Student t test, one-way ANOVA, or log-rank test, as indicated. Statistical analyses were performed in Prism 9 (GraphPad Software). Significance was set at P = 0.05.
All the raw sequencing reads, processed files, and metadata are deposited in Gene Expression Omnibus with the accession number GSE180964.
J. Jen reports other support from American Association for Cancer Research outside the submitted work. J.A. Zebala reports a patent 10,993,953 issued, a patent 10,660,909 issued, a patent 8,969,365 issued, and a patent for RE47,415 issued. D.Y. Maeda reports a patent 10,993,953 issued, a patent 10,660,909 issued, a patent 8,969,365 issued, and a patent for RE47,415 issued. J.G. Christensen reports personal fees from Mirati Therapeutics during the conduct of the study; personal fees from BCTG Acquisition Corp outside the submitted work; in addition, J.G. Christensen has a patent 20,180,072,723 pending, a patent 10,125,134 issued, a patent 20,190,062,330 pending, a patent 20,190,144,444 pending, a patent 10,633,381 issued, and a patent 10,689,377 issued. P. Olson reports other support from Mirati Therapeutics during the conduct of the study; in addition, P. Olson has a patent for PCT/US19/050224 WO2020/055755 pending to Mirati, a patent for PCT/US19/050227 WO2020/055756 pending to Mirati, a patent for PCT/US19/050233 WO2020/055758 pending to Mirati, a patent for PCT/US19/050238 WO2020/055760 pending to Mirati, a patent for PCT/US19/050240 WO2020/055761 pending to Mirati, a patent for PCT/US19/64707 WO2020/118066 pending to Mirati, and a patent for PCT/US20/52185 WO2021/061749 pending to Mirati. A. Athanas reports other support from Mioceros Biosystems during the conduct of the study. K. Wong reports grants from Mirati during the conduct of the study; grants from Takeda, BMS, Dracen, AstraZeneca, Merus, Alkermes, Ansun, and Tvardi, grants and personal fees from Delfi, Zentalis, Pfizer, Janssen, personal fees from Recursion, Navire, and Prelude and grants from Ono outside the submitted work; and G1 Therapeutics equity holder/founder. B.G. Neel reports personal fees and other support from Northern Biologics, LTD, Navire Pharma, and Jengu Therapeutics, and other support from Recursion Pharma and Amgen, Inc. outside the submitted work. Clinical trials of the Novartis SHP2 inhibitor TNO0155 and the Amgen (Sotorasib) and Mirati (Adagrasib) KRAS inhibitors have been performed or are under way at Perlmutter Cancer Center, which B.G. Neel directs. No disclosures were reported by the other authors.
K.H. Tang: Conceptualization, data curation, formal analysis, supervision, funding acquisition, validation, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing. S. Li: Conceptualization, resources, validation, investigation. A. Khodadadi-Jamayran: Data curation, software, formal analysis, visualization. J. Jen: Validation, investigation. H. Han: Validation, investigation. K. Guidry: Resources, validation. T. Chen: Supervision. Y. Hao: Formal analysis, visualization. C. Fedele: Resources. J.A. Zebala: Resources. D.Y. Maeda: Resources. J.G. Christensen: Resources, data curation, formal analysis. P. Olson: Resources, data curation, formal analysis. A. Athanas: Data curation, formal analysis. C.A. Loomis: Supervision, investigation, methodology. A. Tsirigos: Formal analysis, supervision. K.-K. Wong: Conceptualization, resources, data curation, supervision, funding acquisition, methodology, writing–original draft, project administration, writing–review and editing. B.G. Neel: Conceptualization, resources, data curation, supervision, funding acquisition, methodology, writing–original draft, project administration, writing–review and editing.
We thank the Precision Immunology Laboratory and Experimental Pathology Laboratory shared resources, which are partially supported by the NYU-Perlmutter Cancer Center Support Grant P30CA016087, for technical support. The Akoya/PerkinElmer Vectra Polaris multispectral imaging system was acquired through Shared Instrumentation Grant S10 OD021747. A. Tsirigos is supported by the NCI/NIH P01CA229086 and NCI/NIH R01CA252239. We also thank the PCC Genome Technology Center (GTC) shared resource for expert library preparation and sequencing, and the Applied Bioinformatics Laboratories (ABL) shared resources for bioinformatics support and help with data analysis and interpretation. GTC and ABL are also supported by P30CA016087. This work used computing resources at the NYU Grossman School of Medicine High Performance Computing Facility. We thank Drs. Toshiyuki Araki, Kiyomi Araki, Abhishek Bhardwaj, Mitchell Geer, Connor Foster, Jiehui Deng, and Ms. Wei Wei for helpful advice and discussions. Carmine Fedele is currently affiliated with Novartis Institutes for Biomedical Research. This work was supported by NIH grants CA49152 (to B.G. Neel), CA248896 (to B.G. Neel and K.-K. Wong), and P30 CA016087 (to B.G. Neel).
Note: Supplementary data for this article are available at Cancer Discovery Online (http://cancerdiscovery.aacrjournals.org/).