Wnt7b/β-catenin elevation marks inherent and acquired resistance to αPD1 therapy.
αPD1 refractory melanoma and hepatocellular carcinoma (HCC) exhibit elevated Wnt7b/β-catenin 26. To understand if this pathway is dysregulated in αPD1-treated GBM patients, we performed an analysis of a publicly available RNA-seq dataset (4). Almost all patients (28 out of 29) across two cohorts, either receiving neoadjuvant (N) + adjuvant (A) or adjuvant pembrolizumab only, had detectable levels of Wnt7b and Ctnnb1 (fig S1a). Interestingly, Wnt7b was significantly elevated, and Ctnnb1 trending towards higher expression in the cohort of patients receiving adjuvant pembrolizumab only (fig S1a-b). We then assessed whether this high Wnt7b and Ctnnb1 expression post-αPD1 treatment was associated with a poor outcome. Using the median Wnt7b and Ctnnb1 expression, the patients were divided into Wnt7blo, Wnt7bhi, Ctnnb1lo, and Ctnnb1hi groups. Patients from the Wnt7bhi group had a significantly worse PFS and the Ctnnb1hi group had significantly worse PFS and OS as compared to the Wnt7blo and Ctnnb1lo groups, respectively (Fig. 1a). To conclude, Wnt7b and Ctnnb1 expression is elevated in αPD1-treated GBMs and associates with poor outcome.
To validate these findings, we next identified the major Wnt ligands expressed in GBMs in the Glioma Longitudinal Analysis (GLASS) Consortium gene expression dataset comprised 168 GBM patients (121 IDH wildtype and 40 IDH mutant) obtained from 37 hospitals worldwide 27. We used CIBERSORTx 28 to deconvolute the GLASS dataset and reference cell-state signatures derived from 55,284 single-transcriptomes from 11 adult patients spanning glioma subtypes and time points 29. This deconvolution revealed Wnt7b, Wnt4, and Wnt6 as the Wnt ligands expressed in 168 patients. Wnt7b was mostly expressed in stem-cell-rich tumors followed by differentiated tumors and proliferative stem cell tumors (Fig. 1b). Wnt4 and Wnt6 were mainly expressed in fibroblasts and pericytes (fig. S1d). Single cell RNA-seq data from 28 IDH-wildtype GBMs showed that Wnt7b is expressed in malignant cells, macrophages and CD8 T cells (fig. S2a) 30. Our TCGA analysis showed that the Wnt pathway genes are enriched in the Mesenchymal and Pro-neural GBM subtypes. Wnt7b staining in a tissue array comprising 70 GBM patient tumor tissues and 10 normal cerebrum tissues revealed that GBM tissues had elevated Wnt7b (Fig. 1c-d). In addition to the array, we stained for Wnt7b protein in 15 GBM biopsies from the Brigham and Women’s Hospital. All 15 patients showed elevated Wnt7b levels as compared to the 10 normal cerebrum tissues of the array. Representative images of 3 patients and the % Wnt7b + cells in whole tumors are shown in (fig. S2b-c). Collectively, these results indicate that Wnt7b/β-catenin elevation is associated with both inherent and acquired resistance to αPD1 therapy in GBM patients.
Wnt7b is essential in the maintenance of GSC005 – a stem cell – rich murine GBM model.
We next checked Wnt7b levels in human GBM cell lines: U87 (a widely used GBM cell line), and two cell lines generated from patients at MGH, MGG4 (stem-like) 31 and MGG8 (diffusely invasive) 31. Wnt7b was higher in MGG4 than MGG8 or U87. To capture this feature in a preclinical model, we used a stem-cell-derived murine GBM model 005GSC that harbors elevated expression of Wnt7b (Fig. 2a). To investigate the role of Wnt7b in tumor cell survival in vitro and in vivo, we performed CRISPR/Cas9-mediated deletion of Wnt7b and clonal selection. Western blot analysis revealed that as compared to the wild type 005GSC, Wnt7b protein was almost absent, and pLRP6-Ser1490 and Cyclin D1 (a known canonical Wnt target) were lower in the knockout clone (Fig. 2b, c). MTT assay performed at 120 h showed a 30% reduction in proliferation of the Wnt7b-/- clone (Fig. 2d).
We previously showed that 005GSC is an αPD1 resistant IDH wildtype murine model 14. Additionally, similar to human GBM, 005GSC is rich in stem cells, invasive and resistant to temozolomide (TMZ) 32, 33. These combined features make 005GSC a compelling model to study the role of Wnt7b in inherent and acquired resistance to αPD1.To this end, we implanted 5,000 wild type (n = 8) and 5,000 knockout cells (n = 7) orthotopically in mice. The median survival of the 005GSCWnt7bWt -bearing mice was 68 days post implantation whereas that of the 005GSCWnt7b-/- -bearing mice was significantly higher 84.5 days (p < 0.05; Fig. 2e). Moreover, 25% of the mice from the 005GSCWnt7b-/- group survived more than 135 days. (Fig. 2e). Supporting the role of Wnt pathway in acquired resistance to αPD1, our time-matched IHC analysis of 005GSC implanted mouse GBM tissues showed that αPD1 treatment led to elevated Wnt7b and β-catenin (nuclear and perinuclear) levels (Fig. 2f-j).
Wnt inhibition reduces resistance to αPD1 in 005GSC GBMs.
To assess whether Wnt-signaling contributes to αPD1 therapy resistance, we employed WNT-signaling, we used WNT974 - an inhibitor of porcupine (PORCN), an ER-resident O-acyltransferase that mediates Wnt palmitoylation, an essential step in Wnt biosynthesis 34. Firstly, we repeated our published findings in that GSC005 is resistant to αPD1 monotherapy (Fig. 3a). To investigated whether WNT974 in combination with αPD1 can improve the survival of tumor-bearing mice, we implanted 005GSC orthotopically, monitored tumor growth and randomized the mice into 4 groups – control (vehicle for WNT974 + IgG for αPD1), WNT974, αPD1 and WNT974 + αPD1. The median survival of the control group was 25 days. WNT974 improved the median survival significantly to 36 days, and WNT974 + αPD1 increased the survival to 59 days, with 2 out of 8 mice (25%) surviving more than 135 days (Fig. 3a). Of interest, tumors did not form in the long-term survivors challenged with 005GSC cells in the contralateral hemisphere suggesting that these mice had developed a memory response and were cured (Fig. 3b). By contrast, WNT974 alone or the combination with αPD1 did not produce any survival benefit in CT-2A model 35 (Fig. 3c). These differences in survival benefit are consistent with the low protein level of Wnt7b in CT-2A and high protein level of Wnt7b in 005GSC in vivo (n = 3) (Fig. 3d). Further, we observed that lymphoid enhancer binding factor (Lef-1), a downstream protein in the Wnt pathway was much higher in CT-2A as compared to 005GSC (Fig. 3d). Therefore, it is not expected that tumors with Lef-1 activation independently of canonical Wnt ligand signaling, such as CT-2A, would respond to WNT974.
Importantly, WNT974 has been shown to be minimally effective in tumors with downstream activation of lef-1: a mediator of Wnt signaling in EMT and brain metastasis, adenomatous polyposis coli (APC) or β-catenin mutations 36, 37, since these tumors are not driven by aberrant Wnt ligand expression. We profiled the 19 Wnt ligands in 005GSC and CT-2A in vitro and found that the former has a higher expression of Wnt ligands than the latter (fig. S3a). Among the Wnt ligands in 005GSC, Wnt7b, Wnt7a, Wnt5a had the highest expression. We then implanted both GBM models (n = 3 per group) in immune competent mice. The mRNA expression in vivo was also elevated similar to in vitro levels. Specifically, 005GSC had higher Wnt5a, Wnt7a and Wnt7b expression compared to CT-2A, emphasizing that CT-2A resistance to αPD1 is not related to the canonical Wnt pathway 14, 38. Note that Wnt7b expression is 4-fold higher than Wnt7a and Wnt5a in 005GSC (fig. S3b: y-axis in log scale). At the protein level, Wnt5a was not detectable in 005GSC.
Since WNT974 decreased the viable cells by 30% in vitro at 5µM and 10µM concentrations (fig. S4a), we used ApopTag to measure apoptosis in the WNT974, αPD1, and the combination groups but did not observe any difference as compared to control (fig. S4b-c). Thus, WNT974 modestly decreases proliferation of 005GSC, but has no effect on apoptosis.
Response to therapy is caused by reduced levels of Wnt7b/β-catenin in 005GSC tumor cells and increased antigen presentation in the TME.
To dissect the mechanism of response, we interrogated whether WNT974 alone and/or in combination with αPD1 attenuated the Wnt pathway proteins and associated oncogenic signaling in tumor cells in vivo. We found that WNT974 monotherapy modestly decreased Wnt7b but did not affect beta-catenin levels in the tumor tissues (Fig. 2g-i). Intriguingly, WNT974 addition to αPD1 led to two strikingly different responses. In some tumors WNT974 attenuated αPD1-induced Wnt7b and beta-catenin (Response 1). In others, WNT974 + αPD1 did not change Wnt7b or β-catenin levels as compared to the control (Response 2) (Fig. 2g-j). On analyzing the phosphoproteome of the tumor tissues further, we identified that pro-oncogenic pathway proteins such as mTOR and MAPK/ERK were heavily phosphorylated at p-mTORSer2448 and pP44/42Thr202/Tyr204 sites, respectively, particularly in Response 2 (fig. S5c). Activation of these pathways has been implicated in activating β-catenin and downstream signaling. This is evident from the increased Lef-1 protein levels associated with Response 2 (fig. S5c).
To elucidate the mechanisms that drives tumor cell death and elimination, we analyzed the tumor immune microenvironment (TIME) after treating mice with WNT974 and αPD1. We found that combined treatment with WNT974 and αPD1 maintained the overall hematopoietic immune cells marked by CD45. However, the DC3-like dendritic cells marked by expression of CCR7+, CD80+, or CD40 + increased (Fig. 4a) 39, 40, 41, 42. To ensure this phenotype is related to the function, we implanted 005GSCs in Batf3−/−mice that have reduced number of conventional dendritic cells (cDC1). The survival benefit observed in wild type mice by treatment with WNT974 + αPD1 was abrogated in Batf3−/−mice (Fig. 4c, Fig. 3a), suggesting that antigen presentation within the TME is a key driver of response.
Reprogramming of MDSCs mediates resistance to WNT974 + αPD1
Beyond DCs, we examined various subpopulations of CD45 + cells and found that Iba1 + cells, CD45 + CD11b + myeloid cells, CD45 + CD11b + Ly6G-Ly6C-F4/80 + tumor-associated macrophages, CD45 + CD11C + or CD45 + CD11C + CD103 + dendritic cells were not altered (fig. S6a-c, Fig. 4b). Within MDSCs, however CD45 + CD11b + Ly6G + Ly6Clo% gMDSCs decreased in the combination group as compared to control or αPD1 group (Fig. 5a). A double-positive population of Ly6G + Ly6C + cells increased in the combination group as compared to the control group as well as the αPD1 group (Fig. 5a). By investigating the potential players in mediating the reduction in gMDSCs via bulk RNA-seq analysis of tumors, we found that αPD1 treatment (a) increased the expression of Arginase1, Mrc1 or CD206, and Mgl2 -- genes associated with immune suppression of MDSCs and tumor-brain interfacial macrophages/microglia 43, 44, 45, (b) increased Ccl24 and (c) led to a trend towards increased Ccr3, a receptor of Ccl24 involved in eosinophil, T cell and neutrophil chemotaxis (Fig. 5b). WNT974 addition to αPD1 alleviated these αPD1-induced changes in expression. Given the role of the vascular cell adhesion molecule 1 (VCAM1) in neutrophil infiltration 46, 47, 48, we performed a Western blot for VCAM1 in tumor tissue lysates (n = 3 mice per group). We found that WNT974 decreased VCAM1 levels to a greater extent than αPD1. The combination of WNT974 and αPD1 showed a reduction comparable to that of WNT974 as compared to control mice, indicating that blocking Wnt signaling is the main reason for the decrease in VCAM1 (fig. S7a). This pattern is consistent with the effect of WNT974, αPD1 and the combination on gMDSCs (Fig. 5a).
Our time-matched flow analysis indicated that the CD45 + CD11b + Ly6C + cells (that includes Ly6GloLy6Chi, Ly6G + Ly6G + and LyGhiLy6Chi cells) increased in the WNT974 + αPD1 group as compared to the control (Fig. 5a). Further, after eliminating the Ly6G + and Ly6Ghi cells, a purer mMDSC population CD45 + CD11b + Ly6ChiLy6Glo% increased in the WNT974 + αPD1 group as compared to the control or the WNT974 monotherapy group (Fig. 5a). CD4 + and CD8 + T cell % did not change with treatment in the TME (fig. S8).