A platform-independent AI tumor lineage and site (ATLAS) classifier.

Histopathologic diagnosis and classification of cancer plays a critical role in guiding treatment. Advances in next-generation sequencing have ushered in new complementary molecular frameworks. However, existing approaches do not independently assess both site-of-origin (e.g. prostate) and lineage (e.g. adenocarcinoma) and have minimal validation in metastatic disease, where classification is more difficult. Utilizing gradient-boosted machine learning, we developed ATLAS, a pair of separate AI Tumor Lineage and Site-of-origin models from RNA expression data on 8249 tumor samples. We assessed performance independently in 10,376 total tumor samples, including 1490 metastatic samples, achieving an accuracy of 91.4% for cancer site-of-origin and 97.1% for cancer lineage. High confidence predictions (encompassing the majority of cases) were accurate 98-99% of the time in both localized and remarkably even in metastatic samples. We also identified emergent properties of our lineage scores for tumor types on which the model was never trained (zero-shot learning). Adenocarcinoma/sarcoma lineage scores differentiated epithelioid from biphasic/sarcomatoid mesothelioma. Also, predicted lineage de-differentiation identified neuroendocrine/small cell tumors and was associated with poor outcomes across tumor types. Our platform-independent single-sample approach can be easily translated to existing RNA-seq platforms. ATLAS can complement and guide traditional histopathologic assessment in challenging situations and tumors of unknown primary.

Radiotherapy-induced remodeling of the tumor microenvironment by stromal cells.

Cancer cells reside amongst a complex milieu of stromal cells and structural features known as the tumor microenvironment. Often cancer cells divert and co-opt functions of stromal cells of the microenvironment to support tumor progression and treatment resistance. During therapy targeting cancer cells, the stromal cells of the microenvironment receive therapy to the same extent as cancer cells. Stromal cells therefore activate a variety of responses to the damage induced by these therapies, and some of those responses may support tumor progression and resistance. We review here the response of stromal cells to cancer therapy with a focus on radiotherapy in glioblastoma. We highlight the response of endothelial cells and the vasculature, macrophages and microglia, and astrocytes, as well as describing resulting changes in the extracellular matrix. We emphasize the complex interplay of these cellular factors in their dynamic responses. Finally, we discuss their resulting support of cancer cells in tumor progression and therapy resistance. Understanding the stromal cell response to therapy provides insight into complementary therapeutic targets to enhance tumor response to existing treatment options.

Hypoxia-Induced Reactivity of Tumor-Associated Astrocytes Affects Glioma Cell Properties.

Glioblastoma is characterized by extensive necrotic areas with surrounding hypoxia. The cancer cell response to hypoxia in these areas is well-described; it involves a metabolic shift and an increase in stem cell-like characteristics. Less is known about the hypoxic response of tumor-associated astrocytes, a major component of the glioma tumor microenvironment. Here, we used primary human astrocytes and a genetically engineered glioma mouse model to investigate the response of this stromal cell type to hypoxia. We found that astrocytes became reactive in response to intermediate and severe hypoxia, similarly to irradiated and temozolomide-treated astrocytes. Hypoxic astrocytes displayed a potent hypoxia response that appeared to be driven primarily by hypoxia-inducible factor 2-alpha (HIF-2α). This response involved the activation of classical HIF target genes and the increased production of hypoxia-associated cytokines such as TGF-ß1, IL-3, angiogenin, VEGF-A, and IL-1 alpha. In vivo, astrocytes were present in proximity to perinecrotic areas surrounding HIF-2α expressing cells, suggesting that hypoxic astrocytes contribute to the glioma microenvironment. Extracellular matrix derived from hypoxic astrocytes increased the proliferation and drug efflux capability of glioma cells. Together, our findings suggest that hypoxic astrocytes are implicated in tumor growth and potentially stemness maintenance by remodeling the tumor microenvironment.

The Irradiated Brain Microenvironment Supports Glioma Stemness and Survival via Astrocyte-Derived Transglutaminase 2.

The tumor microenvironment plays an essential role in supporting glioma stemness and radioresistance. Following radiotherapy, recurrent gliomas form in an irradiated microenvironment. Here we report that astrocytes, when pre-irradiated, increase stemness and survival of cocultured glioma cells. Tumor-naïve brains increased reactive astrocytes in response to radiation, and mice subjected to radiation prior to implantation of glioma cells developed more aggressive tumors. Extracellular matrix derived from irradiated astrocytes were found to be a major driver of this phenotype and astrocyte-derived transglutaminase 2 (TGM2) was identified as a promoter of glioma stemness and radioresistance. TGM2 levels increased after radiation in vivo and in recurrent human glioma, and TGM2 inhibitors abrogated glioma stemness and survival. These data suggest that irradiation of the brain results in the formation of a tumor-supportive microenvironment. Therapeutic targeting of radiation-induced, astrocyte-derived extracellular matrix proteins may enhance the efficacy of standard-of-care radiotherapy by reducing stemness in glioma. SIGNIFICANCE: These findings presented here indicate that radiotherapy can result in a tumor-supportive microenvironment, the targeting of which may be necessary to overcome tumor cell therapeutic resistance and recurrence. GRAPHICAL ABSTRACT: http://cancerres.aacrjournals.org/content/canres/81/8/2101/F1.large.jpg.

Niche-derived soluble DLK1 promotes glioma growth.

Tumor cell behaviors associated with aggressive tumor growth such as proliferation, therapeutic resistance, and stem cell characteristics are regulated in part by soluble factors derived from the tumor microenvironment. Tumor-associated astrocytes represent a major component of the glioma tumor microenvironment, and astrocytes have an active role in maintenance of normal neural stem cells in the stem cell niche, in part via secretion of soluble delta-like noncanonical Notch ligand 1 (DLK1). We found that astrocytes, when exposed to stresses of the tumor microenvironment such as hypoxia or ionizing radiation, increased secretion of soluble DLK1. Tumor-associated astrocytes in a glioma mouse model expressed DLK1 in perinecrotic and perivascular tumor areas. Glioma cells exposed to recombinant DLK1 displayed increased proliferation, enhanced self-renewal and colony formation abilities, and increased levels of stem cell marker genes. Mechanistically, DLK1-mediated effects on glioma cells involved increased and prolonged stabilization of hypoxia-inducible factor 2alpha, and inhibition of hypoxia-inducible factor 2alpha activity abolished effects of DLK1 in hypoxia. Forced expression of soluble DLK1 resulted in more aggressive tumor growth and shortened survival in a genetically engineered mouse model of glioma. Together, our data support DLK1 as a soluble mediator of glioma aggressiveness derived from the tumor microenvironment.

CD44 Interacts with HIF-2α to Modulate the Hypoxic Phenotype of Perinecrotic and Perivascular Glioma Cells.

Hypoxia-inducible factors enhance glioma stemness, and glioma stem cells have an amplified hypoxic response despite residing within a perivascular niche. Still, little is known about differential HIF regulation in stem versus bulk glioma cells. We show that the intracellular domain of stem cell marker CD44 (CD44ICD) is released at hypoxia, binds HIF-2α (but not HIF-1α), enhances HIF target gene activation, and is required for hypoxia-induced stemness in glioma. In a glioma mouse model, CD44 was restricted to hypoxic and perivascular tumor regions, and in human glioma, a hypoxia signature correlated with CD44. The CD44ICD was sufficient to induce hypoxic signaling at perivascular oxygen tensions, and blocking CD44 cleavage decreased HIF-2α stabilization in CD44-expressing cells. Our data indicate that the stem cell marker CD44 modulates the hypoxic response of glioma cells and that the pseudo-hypoxic phenotype of stem-like glioma cells is achieved by stabilization of HIF-2α through interaction with CD44, independently of oxygen.

Vessel co-option mediates resistance to anti-angiogenic therapy in liver metastases.

The efficacy of angiogenesis inhibitors in cancer is limited by resistance mechanisms that are poorly understood. Notably, instead of through the induction of angiogenesis, tumor vascularization can occur through the nonangiogenic mechanism of vessel co-option. Here we show that vessel co-option is associated with a poor response to the anti-angiogenic agent bevacizumab in patients with colorectal cancer liver metastases. Moreover, we find that vessel co-option is also prevalent in human breast cancer liver metastases, a setting in which results with anti-angiogenic therapy have been disappointing. In preclinical mechanistic studies, we found that cancer cell motility mediated by the actin-related protein 2/3 complex (Arp2/3) is required for vessel co-option in liver metastases in vivo and that, in this setting, combined inhibition of angiogenesis and vessel co-option is more effective than the inhibition of angiogenesis alone. Vessel co-option is therefore a clinically relevant mechanism of resistance to anti-angiogenic therapy and combined inhibition of angiogenesis and vessel co-option might be a warranted therapeutic strategy.

The Initiator Methionine tRNA Drives Secretion of Type II Collagen from Stromal Fibroblasts to Promote Tumor Growth and Angiogenesis.

Expression of the initiator methionine tRNA (tRNAi(Met)) is deregulated in cancer. Despite this fact, it is not currently known how tRNAi(Met) expression levels influence tumor progression. We have found that tRNAi(Met) expression is increased in carcinoma-associated fibroblasts, implicating deregulated expression of tRNAi(Met) in the tumor stroma as a possible contributor to tumor progression. To investigate how elevated stromal tRNAi(Met) contributes to tumor progression, we generated a mouse expressing additional copies of the tRNAi(Met) gene (2+tRNAi(Met) mouse). Growth and vascularization of subcutaneous tumor allografts was enhanced in 2+tRNAi(Met) mice compared with wild-type littermate controls. Extracellular matrix (ECM) deposited by fibroblasts from 2+tRNAi(Met) mice supported enhanced endothelial cell and fibroblast migration. SILAC mass spectrometry indicated that elevated expression of tRNAi(Met) significantly increased synthesis and secretion of certain types of collagen, in particular type II collagen. Suppression of type II collagen opposed the ability of tRNAi(Met)-overexpressing fibroblasts to deposit pro-migratory ECM. We used the prolyl hydroxylase inhibitor ethyl-3,4-dihydroxybenzoate (DHB) to determine whether collagen synthesis contributes to the tRNAi(Met)-driven pro-tumorigenic stroma in vivo. DHB had no effect on the growth of syngeneic allografts in wild-type mice but opposed the ability of 2+tRNAi(Met) mice to support increased angiogenesis and tumor growth. Finally, collagen II expression predicts poor prognosis in high-grade serous ovarian carcinoma. Taken together, these data indicate that increased tRNAi(Met) levels contribute to tumor progression by enhancing the ability of stromal fibroblasts to synthesize and secrete a type II collagen-rich ECM that supports endothelial cell migration and angiogenesis.

Splice variants of SmgGDS control small GTPase prenylation and membrane localization.

Ras and Rho small GTPases possessing a C-terminal polybasic region (PBR) are vital signaling proteins whose misregulation can lead to cancer. Signaling by these proteins depends on their ability to bind guanine nucleotides and their prenylation with a geranylgeranyl or farnesyl isoprenoid moiety and subsequent trafficking to cellular membranes. There is little previous evidence that cellular signals can restrain nonprenylated GTPases from entering the prenylation pathway, leading to the general belief that PBR-possessing GTPases are prenylated as soon as they are synthesized. Here, we present evidence that challenges this belief. We demonstrate that insertion of the dominant negative mutation to inhibit GDP/GTP exchange diminishes prenylation of Rap1A and RhoA, enhances prenylation of Rac1, and does not detectably alter prenylation of K-Ras. Our results indicate that the entrance and passage of these small GTPases through the prenylation pathway is regulated by two splice variants of SmgGDS, a protein that has been reported to promote GDP/GTP exchange by PBR-possessing GTPases and to be up-regulated in several forms of cancer. We show that the previously characterized 558-residue SmgGDS splice variant (SmgGDS-558) selectively associates with prenylated small GTPases and facilitates trafficking of Rap1A to the plasma membrane, whereas the less well characterized 607-residue SmgGDS splice variant (SmgGDS-607) associates with nonprenylated GTPases and regulates the entry of Rap1A, RhoA, and Rac1 into the prenylation pathway. These results indicate that guanine nucleotide exchange and interactions with SmgGDS splice variants can regulate the entrance and passage of PBR-possessing small GTPases through the prenylation pathway.

SmgGDS regulates cell proliferation, migration, and NF-kappaB transcriptional activity in non-small cell lung carcinoma.

Non-small cell lung carcinoma (NSCLC) is promoted by the increased activities of several small GTPases, including K-Ras4B, Rap1A, Rap1B, RhoC, and Rac1. SmgGDS is an unusual guanine nucleotide exchange factor that activates many of these small GTPases, and thus may promote NSCLC development or progression. We report here that SmgGDS protein levels are elevated in NSCLC tumors, compared with normal lung tissue from the same patients or from individuals without cancer. To characterize SmgGDS functions in NSCLC, we tested the effects of silencing SmgGDS expression by transfecting cultured NSCLC cells with SmgGDS small interfering RNA (siRNA). Cells with silenced SmgGDS expression form fewer colonies in soft agar, do not proliferate in culture due to an arrest in G(1) phase, and exhibit disrupted myosin organization and reduced cell migration. The transcriptional activity of NF-kappaB in NSCLC cells is diminished by transfecting the cells with SmgGDS siRNA, and enhanced by transfecting the cells with a cDNA encoding SmgGDS. Because RhoA is a major substrate for SmgGDS, we investigated whether diminished RhoA expression mimics the effects of diminished SmgGDS expression. Silencing RhoA expression with RhoA siRNA disrupts myosin organization, but only moderately decreases cell proliferation and does not inhibit migration. Our finding that the aggressive NSCLC phenotype is more effectively suppressed by silencing SmgGDS than by silencing RhoA is consistent with the ability of SmgGDS to regulate multiple small GTPases in addition to RhoA. These results demonstrate that SmgGDS promotes the malignant NSCLC phenotype and is an intriguing therapeutic target in NSCLC.