No abstract

Current methods for biomarker discovery and target identification in immuno-oncology rely on static snapshots of tumor immunity. To better capture the dynamic and compartmentalized nature of antitumor immune responses, we generated longitudinal “temporal atlases” of productive versus non-productive antitumor immune responses in murine tumor models. We utilized a 34-parameter full spectrum flow cytometry panel to comprehensively profile immune composition within tumors, draining and non-draining lymph nodes, and blood in and around key inflection points of tumor regression or progression. We leveraged two distinct preclinical models for this; the NPK-C1 ectopic prostate cancer model to map dynamics of spontaneous cancer immunoediting, and anti-PD-1 treated MC38 tumors to study response or non-response to immune checkpoint blockade (ICB). We utilized UMAP and FlowSOM algorithms for iterative dimensionality reduction and clustering, respectively, to reveal novel phenotypes associated with productive versus non-productive immunity across model systems, tissues, and time points. We discovered expression of KLRG1 within the intratumoral CD4 T cell compartment was highly associated with tumor progression and response to ICB. Specifically, both FoxP3+ Tregs and FoxP3- Tconv cells within tumors accumulated KLRG1 expression through disease progression, but this was not observed in CD4 T cell or other immune subsets residing in lymph nodes or circulating in blood. Among all intratumoral clusters, KLRG1+ Tconv were the only subset significantly correlated with tumor burden at each time point tested and across both models. KLRG1+ Tconv were significantly enriched in NPK-C1 tumors undergoing progression to escape versus those under immune-mediated equilibrium (p=0.0004) and were lost in animals undergoing curative responses to ICB (p=0.003). In the Treg compartment, unsupervised clustering revealed a KLRG1+Helios- tumor Treg subset that was positively correlated with transition from equilibrium to escape in the NPK-C1 model (p=0.005). Also indicating a potential functional significance, this phenotype was absent in tumors undergoing curative responses to ICB (p=0.0002). Systematic investigation of the functional characteristics, transcriptional programming, and translational significance of intratumoral KLRG1+ CD4 T cell subsets is ongoing. Together, these findings identify KLRG1+ CD4 T cell populations as subsets for further investigation in cancer and demonstrate the utility of longitudinal full spectrum flow cytometry profiling as an engine of dynamic biomarker and/or target discovery in immuno-oncology. Citation Format: Casey Ager, Matthew Chaimowitz, Shruti Bansal, Meri Rogava, Johannes Melms, Catherine Spina, Cory Abate-Shen, Charles G. Drake, Matthew Dallos, Benjamin Izar. KLRG1 marks tumor-infiltrating CD4 T cell subsets associated with immune escape and immunotherapy response. [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2023; Part 1 (Regular and Invited Abstracts); 2023 Apr 14-19; Orlando, FL. Philadelphia (PA): AACR; Cancer Res 2023;83(7_Suppl):Abstract nr 4115.

While immune checkpoint blockade therapy has transformed the treatment of metastatic melanoma, most patients exhibit resistance to these therapies, often by unknown mechanisms. We previously found an association of cancer cell autonomous loss of CD58, which encodes a co-stimulatory/adhesion molecule, with immunotherapy resistance in patients. Here, we investigate the mechanistic molecular underpinnings of this observation. In patient-derived melanoma cells co-cultured with autologous tumor infiltrating lymphocytes (TILs), we found that CD58:CD2 ligation is specifically required for TIL-mediated killing and promotes T cell cytokine production. Furthermore, CD58 co-stimulation of TILs enhances their proliferation and activation compared to traditional methods of co-stimulation via CD28. We additionally examined the role of CD58 in vivo. Given that there is no known mouse homolog for CD58, we utilized a humanized mouse model in which hIL-2-expressing NOD/Shi-scid/IL-2Rγnull mice were implanted with patient-derived WT or CD58 knockout (KO) melanoma cells, followed by adoptive cell transfer (ACT) of autologous TILs. We found that CD58 KO tumors were resistant to ACT and had significantly lower TIL infiltration and proliferation compared to WT tumors, and that these effects were rescued by re-expressing CD58. We had previously found that CD58 loss concurrently leads to higher expression of PD-L1, suggesting that enhanced co-inhibitory PD-1/PD-L1 signaling could additionally contribute to cancer immune evasion. To understand CD58/PD-L1 co-regulation, we first sought to delineate genetic and protein-protein regulation of CD58. We performed a genome-wide CRISPR/Cas9 KO screen with FACS enrichment, as well as unbiased CD58 co-immunoprecipitation mass spectrometry screens and identified CMTM6 as a key regulator of CD58. Importantly, CMTM6 was previously identified as a regulator of PD-L1 maintenance, positioning it as a potential candidate for regulating both co-stimulatory (CD58) and co-inhibitory (PD-L1) signals in cancer cells. Indeed, we find that both PD-L1 and CD58 are regulated by CMTM6 at the level of lysosomal degradation. Furthermore, in cells with CD58/CMTM6 double-KO, CMTM6 is required for CD58/PD-L1 co-regulation, which is restored with re-expression of CMTM6. We therefore propose a model in which CD58 regulates PD-L1 by modulating its level of binding to CMTM6; in the absence of CD58, additional CMTM6 is available to bind PD-L1 and thereby stabilize its expression. In summary, we define a central role of the CD58:CD2 axis in tumor immunity and propose that loss of CD58 confers cancer immune evasion through impaired T cell toxicity, tumor infiltration, and concurrent CMTM6-dependent PD-L1 upregulation. Citation Format: Patricia Ho, Johannes Melms, Meri Rogava, Shivem Shah, Benjamin Izar. CD58:CD2 - a multi-dimensional axis in cancer immunotherapy resistance [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2022; 2022 Apr 8-13. Philadelphia (PA): AACR; Cancer Res 2022;82(12_Suppl):Abstract nr 2165.

Brain metastases are the most frequent malignancies in the brain and are associated with significant morbidity and mortality. Melanoma brain metastases (MBM) occur in most patients with advanced melanoma and are challenging to treat. Our understanding of the treatment-naïve landscape of MBM is still rudimentary, and there are no site-specific molecular therapies available. To gain comprehensive insights into the niche-specific biology of MBM, we performed multi-modal profiling of fresh and frozen samples using single-cell RNA-seq, single-cell TCR-seq, single-nuclei RNA-seq, and spatial transcriptional profiling. We evolved single-nucleus RNA-seq processing methods to enable profiling of minute amounts of archival, frozen specimens and compared data quality and structure between matched fresh and frozen MBM. We curated a treatment-naïve single-transcriptome atlas of MBM, collected either fresh samples over 1 year or profiled frozen samples dating back more than 15 years, and compared these samples to extracranial melanoma metastases (ECMM). In total, we profiled 25 samples with more than 114,000 transcriptomes. We identified more than 20 different cell types, including diverse tumor-infiltrating T-cell subsets and rare dendritic cell types, and tissue-specific cell types, such as activated microglia. Tumor cells in MBM showed an increase in copy number alterations (CNAs) compared to ECMM, which we validated using an external dataset of whole exome sequencing (WES) data including both MBM and ECMM. MBM-derived tumor cells show enrichment of genes involved in neuronal development and function, and site-specific metabolic programs (e.g., oxidative phosphorylation). Comparison with an external bulk RNA-seq dataset validated enriched key genes in MBM and ECMM as putative dependencies. We recovered cell-cell interactions between tumor and brain-resident cells involved in brain development, homeostasis, and disease. Similar to ECMM, the tumor microenvironment of MBM contained CD8+ T cells across a spectrum of differentiation, exhaustion and expansion, which was associated with loss of TCF7 expression and adoption of a TOX+ cell state. CD4+ T cells included T regulatory, T helper and T follicular-helper-like expression profiles. Plasma cells showed spatially localized expansion and limited heterogeneity. Myeloid cells largely adopted pro-tumorigenic cell states, including microglia, the brain-resident myeloid cells, which showed an activation trajectory characterized by expression of SPP1 (osteopontin). Spatial transcriptional analysis revealed restricted expression of antigen presentation genes with only a subset of these locations showing a type I interferon response. In summary, this work presents a multi-modal single-cell approach to dissect and compare the landscape of treatment-naïve MBM and ECMM. Citation Format: Johannes C. Melms, Jana Biermann, Amit Dipak Amin, Yiping Wang, Somnath Tagore, Massimo Andreatta, Ajay Nair, Meri Rogava, Patricia Ho, Lindsay A. Caprio, Zachary H. Walsh, Shivem Shah, Daniel H. Vacarro, Blake Caldwell, Adrienne M. Luoma, Joseph Driver, Matthew Ingham, Suthee Rapisuwon, Jennifer Wargo, Craig L. Slinguff, Evan Z. Macosco, Fei Chen, Richard Carvajal, Michael B. Atkins, Michael A. Davies, Elham Azizi, Santiago J. Carmona, Hanina Hibshoosh, Peter D. Canoll, Jeffrey N. Bruce, Wenya L. Bi, Gary K. Schwartz, Benjamin Izar. Dissecting the ecosystem of treatment-naïve melanoma brain metastasis using multi-modal single-cell analysis [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2022; 2022 Apr 8-13. Philadelphia (PA): AACR; Cancer Res 2022;82(12_Suppl):Abstract nr 984.

Genomic and adaptive determinants of organ-specific metastasis are poorly understood. A model of sequential acquisition of divergent somatic mutations is insufficient to explain metastasis. Liver metastasis (LM) occurs frequently and is associated with a poor prognosis and reduced therapy response in several cancers, including in patients with melanoma and lung cancer. To identify drivers of metastatic niches, we used a syngeneic mouse melanoma model which recapitulates genomic, metastatic and therapy response patterns seen in patients. We performed a large-scale in vivo CRISPR-Cas9 knockout screen and identified perturbations that promote LM, but not primary tumor growth or metastasis to other organs (e.g. lung). The “top hit” in this screen associated with LM was loss Pip4k2c. We generated Pip4k2cKO cells and show that in otherwise isogenic melanoma cell lines, loss of Pip4k2c led to increased baseline and insulin-induced activation of the PI3K/AKT pathway, and increased invasive capacity. Rescuing Pip4k2cKO with full-length (Pip4k2cRec) or allosteric domain deficient (Pip4k2cAD) Pip4k2c ORFs, we show that hyperactivation of the PI3K/AKT pathway in is mediated by loss of the allosteric domain function, and not loss of the kinase domain of Pip4k2c. Treatment with different PI3K inhibitors effectively abrogated the pathway, but was partly bypassed in the presence of insulin in Pip4k2cKO and Pip4k2cAD, but not parental or Pip4k2cRec cells. Upon tail vein injection, Pip4k2cKO cells produced a significantly increased LM burden compared to parental cells, and this effect was rescued in Pip4k2cRec but not Pip4k2cAD, further affirming that loss of allosteric domain was required for this phenotype. We reasoned that Pip4k2cKO cells preferentially colonized the liver by co-opting the insulin-rich milieu in this organ. To test this, we used shRNA targeted against the insulin receptor (Insr) generated Pip4k2cKO/InsrshIR and showed that Insr was required but not sufficient to enhance LM burden. Given the promising in vitro activity of PI3K inhibitors, we next tested whether these could abrogate LM in vivo. Surprisingly, we found a substantial increase in LM burden in mice with Pip4k2cKO-bearing LM treated with PI3K inhibition compared to vehicle treated animals. We show that this paradoxical observation was due to host-mediated increased in glucose and insulin in response to PI3K inhibitor, which promoted a forward loop of increased liver metastasis. Breaking this loop with either ketogenic diet or treatment with a SGLT2 inhibitor in turn rescued increased these host responses and resulted in reduced LM burden in combination with PI3K inhibition. In summary, we identify a novel mechanism of metastatic liver organotropism and pharmacological and dietary combinations to reduced liver metastatic burden. Given the expanding use of PI3K inhibitors, our findings may have important clinical implications. Citation Format: Meri Rogava, Johannes C. Melms, Stephanie Davis, Clemens Hug, Bryan Ngo, Michael J. Lee, Patricia Ho, Amit Dipak Amin, Yiping Wang, Sean Chen, William Ge, David Liu, Thomas Tüting, Martin Röcken, Thomas K. Eigentler, Samuel F. Bakhoum, Andrei Molotkov, Akiva Mintz, Lewis C. Cantley, Peter K. Sorger, Benjamin Izar. A genetic-metabolic axis of metastatic liver organotropism [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2022; 2022 Apr 8-13. Philadelphia (PA): AACR; Cancer Res 2022;82(12_Suppl):Abstract nr 981.