Abstract 4100: Developmental deconvolution for classification of cancer origin

Cancer is a disease manifesting in abrogation of developmental programs, and malignancies are namedbased on their cell or tissue of origin. However, a systematic atlas of tumor origins is lacking. Here we map the single cell organogenesis of 56 developmental trajectories to the transcriptomes of over 10,000 tumors across 33 cancer types. We use this map to deconvolute individual tumors into their constituent developmental components. Based on these deconvoluted developmental programs, we construct a Developmental Machine Learning Perceptron (D-MLP) classifier that outputs cancer origin. The D-MLP classifier (ROC-AUC: 0.974 for top prediction) outperforms classification based on expression of either oncogenes or highly variable genes. We analyze tumors from patients with cancer of unknown primary (CUP), selecting the most difficult cases where extensive multimodal workup yielded no definitive tumor type. D-MLP revealed insights into developmental origins and diagnosis for most patient tumors. Our results provide a map of tumor developmental origins, provide a tool for diagnostic pathology, and suggest developmental classification may be a useful approach for otherwise unclassified patient tumors.

Citation Format: Enrico Moiso, Alexander Farahani, Hetal Marble, Austin Hendricks, Samuel Mildrum, Stuart Levine, Jochen Lennerz, Salil Garg. Developmental deconvolution for classification of cancer origin [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 4100.

Diagnostic Value of MAML2 Rearrangements in Mucoepidermoid Carcinoma

Mucoepidermoid carcinoma (MEC) is often seen in salivary glands and can harbor MAML2 translocations (MAML2+). The translocation status has diagnostic utility as an objective confirmation of the MEC diagnosis, for example, when distinction from the more aggressive adenosquamous carcinoma (ASC) is not straightforward. To assess the diagnostic relevance of MAML2, we examined our 5-year experience in prospective testing of 8106 solid tumors using RNA-seq panel testing in combinations with a two-round Delphi-based scenario survey. The prevalence of MAML2+ across all tumors was 0.28% (n = 23/8106) and the majority of MAML2+ cases were found in head and neck tumors (78.3%), where the overall prevalence was 5.9% (n = 18/307). The sensitivity of MAML2 for MEC was 60% and most cases (80%) were submitted for diagnostic confirmation; in 24% of cases, the MAML2 results changed the working diagnosis. An independent survey of 15 experts showed relative importance indexes of 0.8 and 0.65 for "confirmatory MAML2 testing" in suspected MEC and ASC, respectively. Real-world evidence confirmed that the added value of MAML2 is a composite of an imperfect confirmation test for MEC and a highly specific exclusion tool for the diagnosis of ASC. Real-world evidence can help move a rare molecular-genetic biomarker from an emerging tool to the clinic.

A Pathologist-Annotated Dataset for Validating Artificial Intelligence: A Project Description and Pilot Study

Purpose:

Validating artificial intelligence algorithms for clinical use in medical images is a challenging endeavor due to a lack of standard reference data (ground truth). This topic typically occupies a small portion of the discussion in research papers since most of the efforts are focused on developing novel algorithms. In this work, we present a collaboration to create a validation dataset of pathologist annotations for algorithms that process whole slide images. We focus on data collection and evaluation of algorithm performance in the context of estimating the density of stromal tumor-infiltrating lymphocytes (sTILs) in breast cancer.

Methods:

We digitized 64 glass slides of hematoxylin- and eosin-stained invasive ductal carcinoma core biopsies prepared at a single clinical site. A collaborating pathologist selected 10 regions of interest (ROIs) per slide for evaluation. We created training materials and workflows to crowdsource pathologist image annotations on two modes: an optical microscope and two digital platforms. The microscope platform allows the same ROIs to be evaluated in both modes. The workflows collect the ROI type, a decision on whether the ROI is appropriate for estimating the density of sTILs, and if appropriate, the sTIL density value for that ROI.

Results:

In total, 19 pathologists made 1645 ROI evaluations during a data collection event and the following 2 weeks. The pilot study yielded an abundant number of cases with nominal sTIL infiltration. Furthermore, we found that the sTIL densities are correlated within a case, and there is notable pathologist variability. Consequently, we outline plans to improve our ROI and case sampling methods. We also outline statistical methods to account for ROI correlations within a case and pathologist variability when validating an algorithm.

Conclusion:

We have built workflows for efficient data collection and tested them in a pilot study. As we prepare for pivotal studies, we will investigate methods to use the dataset as an external validation tool for algorithms. We will also consider what it will take for the dataset to be fit for a regulatory purpose: study size, patient population, and pathologist training and qualifications. To this end, we will elicit feedback from the Food and Drug Administration via the Medical Device Development Tool program and from the broader digital pathology and AI community. Ultimately, we intend to share the dataset, statistical methods, and lessons learned.

BIOM-54. A RAPID GENOTYPING PANEL FOR SENSITIVE AND SPECIFIC SEGREGATION OF CNS PATHOLOGIES

Primary central nervous system lymphoma (PCNSL) remains challenging to diagnose due to nonspecific clinical and radiologic features and low diagnostic yields of cerebrospinal fluid (CSF) studies. We sought to characterize the diagnostic approach of suspected PCNSL, in order to improve clinical workflow. We first reviewed 1,007 new brain lesions of unknown etiology that included PCNSL in the radiologic differential diagnosis. The most common final diagnoses included high-grade glioma (28.2%) and PCNSL (14.6%). Diagnostic biopsies were frequently performed for high-grade glioma (100%) and PCNSL (94.4%), while CSF was frequently sampled for PCNSL (78.7%). We next identified 159 patients with an established new diagnosis of PCNSL. CSF studies were non-diagnostic in 86.7% of cases, whereas biopsy was positive in 93%. However, intraoperative histopathology was inconclusive for PNCSL in 54.5%, likely contributing to 22% of patients undergoing surgical resection. These challenges resulted in 12 days median time to treatment initiation, and readmission for further workup or treatment initiation in 27% of patients. These results indicated the need for a rapid, sensitive and specific platform to segregate PCNSL and glioma using CSF and tissue samples. We developed a qPCR-based assay to genotype the MYD88 L265P hotspot mutation from CSF and plasma within 80 minutes of sample acquisition. Results were concordant with orthogonal DNA sequencing in extracts from 87 archived specimens, with detection limits of 490pg of input genomic DNA and 0.15% mutant allele frequency. When performed simultaneously with assays for TERT promoter, IDH1/2, H3F3A and BRAF point mutations, the resulting panel accurately segregated PCNSL and adult diffuse glioma molecular diagnoses in 87 archived specimens and 19 prospective liquid biopsies, including cases of lymphoma and glioma. We propose that inclusion of targeted analysis of these mutually exclusive recurrent molecular alterations characterizing gliomas and PCNSL will facilitate rapid, sensitive diagnosis from solid and liquid biopsies.