Tumor cell p38 inhibition to overcome immunotherapy resistance

Patients with tumors that do not respond to immune-checkpoint inhibition often harbor a non-T cell-inflamed tumor microenvironment, characterized by the absence of IFN-γ-associated CD8+ T cell and dendritic cell activation. Understanding the molecular mechanisms underlying immune exclusion in non-responding patients may enable the development of novel combination therapies. p38 MAPK is a known regulator of dendritic and myeloid cells however a tumor-intrinsic immunomodulatory role has not been previously described. Here we identify tumor cell p38 signaling as a therapeutic target to potentiate anti-tumor immunity and overcome resistance to immune-checkpoint inhibitors (ICI). Molecular analysis of tumor tissues from patients with human papillomavirus-negative head and neck squamous carcinoma reveals a p38-centered network enriched in non-T cell-inflamed tumors. Pan-cancer single-cell RNA analysis suggests that p38 activation may be an immune-exclusion mechanism across multiple tumor types. P38 knockdown in cancer cell lines increases T cell migration, and p38 inhibition plus ICI in preclinical models shows greater efficacy compared to monotherapies. In a clinical trial of patients refractory to PD1/L1 therapy, pexmetinib, a p38 inhibitor, plus nivolumab demonstrated deep and durable clinical responses. Targeting of p38 with anti-PD1 has the potential to induce the T cell-inflamed phenotype and overcome immunotherapy resistance.

Immunotherapy with immune-checkpoint inhibition (ICI) has improved outcomes for 61 patients with cancer however most do not benefit. The T cell-inflamed tumor microenvironment 62 (TME), characterized by CD8 + T cell infiltration, type I/II interferon (IFN) gene expression and 63 antigen-presentation machinery 1,2 , is an important cancer immunotherapy biomarker 3 that can 64 also facilitate the identification of molecular correlates of immune exclusion 4 . Wnt/β-catenin 65 signaling was the first described tumor-intrinsic mechanism of immune exclusion, initially in 66 melanoma 5 then across tumors 6 . We described a molecular atlas of tumor-intrinsic mechanisms impact of HPV in HNSCC, we focused on the 395 HPV-negative tumors (Fig. 1a). Using a T 85 cell-inflamed gene expression signature 6,7,14,15 , we observed 36% versus 30% of tumors 86 expressing the T cell-inflamed versus non-inflamed phenotype, respectively. We detected 67 87 pathways significantly activated in non-inflamed relative to inflamed tumors (z-score≥1.95, 88 P<0.05) from causal network analysis 16 (Fig. 1a). Functional annotation of the pathways 89 revealed two main themes, with 43 pathways representing the MAPK, HIF-1, and T cell-receptor 90 regulation of apoptosis, and 24 representing IL6 pathway and oxidative stress (Fig. 1b).

91
Our previous pan-cancer analysis demonstrated that tumors with reduced T-cell 92 inflamed gene expression showed higher numbers of co-occurred activated pathways 7 . We 93 observed a similar trend in HPV-negative HNSCC, where pathway score was significantly and 94 inversely correlated with T cell-inflamed gene expression (R 2 =0.726, P<0.05) (Fig. 1c). Of 67 95 pathways, 59 were independently validated in an oral cancer cohort from The International 96 Cancer Genome Consortium 17 (ICGC-ORCA; P<0.05) (Fig. 1d). Twenty-four programs were 97 predicted to be activated by transcriptional factors such as HIF1A, the master regulator of 98 cellular response to hypoxia, whereas the other 35 were driven by kinases, hormones, or other 99 non-transcriptional factors (Fig. 1d). Consistent with our prior work [18][19][20] , CTNNB1 was strongly 100 associated with non-T cell-inflamed tumors. In addition, we observed p38 (MAPK14), as a drug 101 target where clinical therapeutics have been previously developed that was also strongly 102 associated with the non-T cell-inflamed phenotype. Multiple molecular targets were identified 103 with associated drugs (Table S1). We chose to focus specifically on p38 given that our research 104 group had immediate access to pexmetinib, a p38 inhibitor, for re-purposing in immunotherapy 105 combination clinical trials.

106
To investigate the cellular source of these immune exclusion signaling pathways, we

114
For each pathway, we computed an expression score in individual cells by averaging the levels 115 of all downstream target molecules involved in a pathway taking into consideration the change 116 of expression direction. Among the tumor, stromal, and immune cell populations, malignant 117 epithelial cells showed the highest pathway scores (Fig. 2b). A previous study reported p38 118 activation in fibroblasts as driving absence of T cell infiltration in triple-negative breast cancer 12 .

119
In our study, the tumor cell-expressing p38 pathway score was significantly and dominantly 120 higher than that from fibroblasts and other cells, suggesting that tumor cells are the major 121 contributor to the p38 pathway expression in TME (Fig. S1).

122
We then sought to understand differences in tumor cell-specific pathway expression for 123 the low-T cell-infiltrated versus high-T cell-infiltrated groups. For initial clarity, we focused on the 124 extreme phenotypes based on T cell fraction from lower to higher out of all cells sequenced in 125 the TME. Out of 11 patients, five had tumors of at least 40 malignant epithelial cells (ranging 126 113-330 per sample) and were included in further analysis. We detected elevated expression in 50 out of 59 pathways from tumors of lower levels of T cell infiltrate, with 49 showing significant 128 changes at FDR 0.10 ( Fig. 2c; Table S2). Tumor cell-expression of CTNNB1 and p38 pathways 129 were significantly upregulated in low-T cell-infiltrated relative to high-T cell-infiltrated tumors 130 (FDR-adjusted P=0.0435 and 0.0036, respectively) (Fig. 2d). We independently validated the 131 single-cell findings in a separate cohort of HPV-negative HNSCC patients (cohort B 22 ) and 132 observed the same results ( Fig. S2a-S2d; Table S3). For cohort B, we verified our scRNAseq-

139
To prioritize druggable targets for potential combination immunotherapy beyond those 140 previously identified, we developed an accumulative scoring system to nominate candidates 141 integrating both bulk tissue and single-cell data. From the 59 pathways, we computed a 142 combined relative rank for each pathway as the geometric mean of three values: the relative 143 rank in TCGA (z-score higher to lower), the relative rank in ICGC for anti-correlation with T cell-144 inflamed gene expression (coefficient high to low negative), and the relative rank in scRNAseq 145 tumor cell-expressing pathway scores of low-vs high-T cell-infiltrated tumors (Fig. 2e, 2f; Fig.   146 S2e, S2f; Table S4). With this approach we integrated results from TCGA, ICGC and both 147 HNSCC scRNAseq cohorts in order to narrow to six pathways (Fig. 3a), noting that VEGFA, a 148 known mediator of immune-checkpoint inhibitor (ICI) resistance 23 , is a downstream target 149 shared by all pathways (Fig. 3a). Further, these pathways form a core biomolecular protein-  Analysis Consortium 24 (CPTAC). We observed an inverse correlation between CD8A protein 155 abundance and CTNNB1 or p38 pathway target molecules (Spearman's ρ=−0.47 and −0.27, 156 respectively) (Fig. 3c, 3d). This inverse correlation was only detected in tumor cells (P<0.01)

157
and not in adjacent normal tissue (P>0.25), further supporting that activation of CTNNB1 and 158 p38 causally leads to immune exclusion in some tumors.

186
Our prior work demonstrated CTNNB1 activation as driving the absence of T cell 187 infiltrates in multiple tumor types 5,6 and we were interested in investigating whether p38 188 presents a similar mechanism of immune exclusion across human solid tumors. We generated a  (Fig. S2), and skin cutaneous melanoma (SKCM) 28 (Fig. S4). Comparing low-vs high-

196
but not in the SKCM scRNAseq cohort (Fig. S4) suggesting that certain tumor cell-intrinsic 197 signaling pathways may impact immunotherapy outcomes in different cancers.

198
To functionally examine the causal impact of tumor cell-expressing p38 on immune 199 exclusion, we generated shRNA knockdown of MAPK14 (encoding p38 alpha isoform) gene

212
The potential for pharmacologic inhibition of p38 MAPK to enhance ICI was explored  Table S7. Fifteen subjects were included in the study with 14 having at least one  (Table S8).

241
Subjects received a median of 3 (range 1 to 26) cycles.

250
These were managed per standard algorithms with steroids. These irAEs were observed to be 251 relapsing in some subjects with pexmetinib start/stop and some subjects having multiple distinct 252 episode irAEs (i.e. -transaminitis then gastritis etc).

254
As pexmetinib dose increased, exposure increased with dose ( Fig. S8a-c), half-life appeared to 255 be dose-dependent, yet dose-normalized C max and total clearance were not impacted ( Fig. S8d-256 f). In line with the short half-life, there was no accumulation with continued dosing, and 257 exposure was similar on days 29 and 1. AR00451575 metabolite exposure was a quarter to a 258 third that of the parent pexmetinib, and the half-life appeared slightly longer. The metabolic ratio 259 suggested a possible decreasing trend with dose ( Fig. S9a-b, P=0.238 and 0.043 for AUC and 260 C max , respectively). Evaluation of pexmetinib exposure by occurrence of DLT suggested that 261 higher AUC may be (P=0.058) associated with DLT ( Fig. S9c-d).

262
All but one subject had been previously treated with anti-PD1. All PD1 previously treated

277
We report a tumor-cell intrinsic role of p38 activation in driving immune exclusion using