Duane A. Mitchell

There is an increasing interest in developing artificial intelligence (AI) systems to process and interpret electronic health records (EHRs). Natural language processing (NLP) powered by pretrained language models is the key technology for medical AI systems utilizing clinical narratives. However, there are few clinical language models, the largest of which trained in the clinical domain is comparatively small at 110 million parameters (compared with billions of parameters in the general domain). It is not clear how large clinical language models with billions of parameters can help medical AI systems utilize unstructured EHRs. In this study, we develop from scratch a large clinical language model-GatorTron-using >90 billion words of text (including >82 billion words of de-identified clinical text) and systematically evaluate it on five clinical NLP tasks including clinical concept extraction, medical relation extraction, semantic textual similarity, natural language inference (NLI), and medical question answering (MQA). We examine how (1) scaling up the number of parameters and (2) scaling up the size of the training data could benefit these NLP tasks. GatorTron models scale up the clinical language model from 110 million to 8.9 billion parameters and improve five clinical NLP tasks (e.g., 9.6% and 9.5% improvement in accuracy for NLI and MQA), which can be applied to medical AI systems to improve healthcare delivery. The GatorTron models are publicly available at: https://catalog.ngc.nvidia.com/orgs/nvidia/teams/clara/models/gatortron_og .

PURPOSE: Immunologic targeting of tumor-specific gene mutations may allow precise eradication of neoplastic cells without toxicity. Epidermal growth factor receptor variant III (EGFRvIII) is a constitutively activated and immunogenic mutation not expressed in normal tissues but widely expressed in glioblastoma multiforme (GBM) and other neoplasms. PATIENTS AND METHODS: A phase II, multicenter trial was undertaken to assess the immunogenicity of an EGFRvIII-targeted peptide vaccine and to estimate the progression-free survival (PFS) and overall survival (OS) of vaccinated patients with newly diagnosed EGFRvIII-expressing GBM with minimal residual disease. Intradermal vaccinations were given until toxicity or tumor progression was observed. Sample size was calculated to differentiate between PFS rates of 20% and 40% 6 months after vaccination. RESULTS: There were no symptomatic autoimmune reactions. The 6-month PFS rate after vaccination was 67% (95% CI, 40% to 83%) and after diagnosis was 94% (95% CI, 67% to 99%; n = 18). The median OS was 26.0 months (95% CI, 21.0 to 47.7 months). After adjustment for age and Karnofsky performance status, the OS of vaccinated patients was greater than that observed in a control group matched for eligibility criteria, prognostic factors, and temozolomide treatment (hazard ratio, 5.3; P = .0013; n = 17). The development of specific antibody (P = .025) or delayed-type hypersensitivity (P = .03) responses to EGFRvIII had a significant effect on OS. At recurrence, 82% (95% CI, 48% to 97%) of patients had lost EGFRvIII expression (P < .001). CONCLUSION: EGFRvIII-targeted vaccination in patients with GBM warrants investigation in a phase III, randomized trial.

Immunosuppression is frequently associated with malignancy and is particularly severe in patients with malignant glioma. Anergy and counterproductive shifts toward T(H)2 cytokine production are long-recognized T-cell defects in these patients whose etiology has remained elusive for >30 years. We show here that absolute counts of both CD4(+) T cells and CD4(+)CD25(+)FOXP3(+)CD45RO(+) T cells (T(regs)) are greatly diminished in patients with malignant glioma, but T(regs) frequently represent an increased fraction of the remaining CD4 compartment. This increased T(reg) fraction, despite reduced counts, correlates with and is sufficient to elicit the characteristic manifestations of impaired patient T-cell responsiveness in vitro. Furthermore, T(reg) removal eradicates T-cell proliferative defects and reverses T(H)2 cytokine shifts, allowing T cells from patients with malignant glioma to function in vitro at levels equivalent to those of normal, healthy controls. Such restored immune function may give license to physiologic antiglioma activity, as in vivo, T(reg) depletion proves permissive for spontaneous tumor rejection in a murine model of established intracranial glioma. These findings dramatically alter our understanding of depressed cellular immune function in patients with malignant glioma and advance a role for T(regs) in facilitating tumor immune evasion in the central nervous system.

Brain tumors are horrific diseases with almost universally fatal outcomes; new therapeutics are desperately needed and will come from improved understandings of glioma biology. Exosomes are endosomally derived 30-100 nm membranous vesicles released from many cell types into the extracellular milieu; surprisingly, exosomes are virtually unstudied in neuro-oncology. These microvesicles were used as vaccines in other tumor settings, but their immunological significance is unevaluated in brain tumors. Our purpose here is to report the initial biochemical, proteomic, and immunological studies on murine brain tumor exosomes, following known procedures to isolate exosomes. Our findings show that these vesicles have biophysical characteristics and proteomic profiles similar to exosomes from other cell types but that brain tumor exosomes have unique features (e.g., very basic isoelectric points, expressing the mutated tumor antigen EGFRvIII and the putatively immunosuppressive cytokine TGF-beta). Administration of such exosomes into syngeneic animals produced both humoral and cellular immune responses in immunized hosts capable of rejecting subsequent tumor challenges but failed to prolong survival in established orthotopic models. Control animals received saline or cell lysate vaccines and showed no antitumor responses. Exosomes and microvesicles isolated from sera of patients with brain tumors also possess EGFR, EGFRvIII, and TGF-beta. We conclude that exosomes released from brain tumor cells are biochemically/biophysically like other exosomes and have immune-modulating properties. They can escape the blood-brain barrier, with potential systemic and distal signaling and immune consequences.