Citations

Cancer is a complex disease involving the deregulation of intricate cellular systems beyond genetic aberrations and, as such, requires sophisticated computational approaches and high-dimensional data for optimal interpretation. While conventional artificial intelligence (AI) models excel in many prediction tasks, they often lack interpretability and are blind to the scientific hypotheses generated by researchers to enable cancer discoveries. Here we propose that hypothesis-driven AI, a new emerging class of AI algorithm, is an innovative approach to uncovering the complex etiology of cancer from big omics data. This review exemplifies how hypothesis-driven AI is different from conventional AI by citing its application in various areas of oncology including tumor classification, patient stratification, cancer gene discovery, drug response prediction, and tumor spatial organization. Our aim is to stress the feasibility of incorporating domain knowledge and scientific hypotheses to craft the design of new AI algorithms. We showcase the power of hypothesis-driven AI in making novel cancer discoveries that can be overlooked by conventional AI methods. Since hypothesis-driven AI is still in its infancy, open questions such as how to better incorporate new knowledge and biological perspectives to ameliorate bias and improve interpretability in the design of AI algorithms still need to be addressed. In conclusion, hypothesis-driven AI holds great promise in the discovery of new mechanistic and functional insights that explain the complexity of cancer etiology and potentially chart a new roadmap to improve treatment regimens for individual patients.

Abstract Background Efficient and accurate delineation of organs at risk (OARs) is a critical procedure for treatment planning and dose evaluation. Deep learning‐based auto‐segmentation of OARs has shown promising results and is increasingly being used in radiation therapy. However, existing deep learning‐based auto‐segmentation approaches face two challenges in clinical practice: generalizability and human‐AI interaction. A generalizable and promptable auto‐segmentation model, which segments OARs of multiple disease sites simultaneously and supports on‐the‐fly human‐AI interaction, can significantly enhance the efficiency of radiation therapy treatment planning. Purpose Meta's segment anything model (SAM) was proposed as a generalizable and promptable model for next‐generation natural image segmentation. We further evaluated the performance of SAM in radiotherapy segmentation. Methods Computed tomography (CT) images of clinical cases from four disease sites at our institute were collected: prostate, lung, gastrointestinal, and head & neck. For each case, we selected the OARs important in radiotherapy treatment planning. We then compared both the Dice coefficients and Jaccard indices derived from three distinct methods: manual delineation (ground truth), automatic segmentation using SAM's ’segment anything’ mode, and automatic segmentation using SAM's ‘box prompt’ mode that implements manual interaction via live prompts during segmentation. Results Our results indicate that SAM's segment anything mode can achieve clinically acceptable segmentation results in most OARs with Dice scores higher than 0.7. SAM's box prompt mode further improves Dice scores by 0.1∼0.5. Similar results were observed for Jaccard indices. The results show that SAM performs better for prostate and lung, but worse for gastrointestinal and head & neck. When considering the size of organs and the distinctiveness of their boundaries, SAM shows better performance for large organs with distinct boundaries, such as lung and liver, and worse for smaller organs with less distinct boundaries, like parotid and cochlea. Conclusions Our results demonstrate SAM's robust generalizability with consistent accuracy in automatic segmentation for radiotherapy. Furthermore, the advanced box‐prompt method enables the users to augment auto‐segmentation interactively and dynamically, leading to patient‐specific auto‐segmentation in radiation therapy. SAM's generalizability across different disease sites and different modalities makes it feasible to develop a generic auto‐segmentation model in radiotherapy.

Abstract Objective The pathological expansion of white adipose tissue (WAT) in obesity involves adipocyte hypertrophy accompanied by expansion of the collagen‐rich pericellular extracellular matrix (ECM) and development of crown‐like structures (CLS). Traditionally, WAT morphology is assessed through immunohistochemical analysis of WAT sections. However, manual analysis of large histological sections is time‐consuming, and the available digital tools for analyzing adipocyte size and pericellular ECM are limited. To address this gap, the authors developed the Adipose Tissue Analysis Toolkit (ATAT), an ImageJ plugin facilitating analysis of adipocyte size, WAT ECM, and CLS. Methods and Results ATAT utilizes local and image‐level differentials in pixel intensity to independently threshold image background, distinguishing adipocyte‐free tissue without user input. It accurately captures adipocytes in histological sections stained with common dyes and automates the analysis of adipocyte cross‐sectional area, total‐field, and localized region‐of‐interest ECM. ATAT allows fully automated batch analysis of histological images using default or user‐defined adipocyte detection parameters. Conclusions ATAT provides several advantages over existing WAT image analysis tools, enabling high‐throughput analyses of adipocyte‐specific parameters and facilitating the assessment of ECM changes associated with WAT remodeling due to weight changes and other pathophysiological alterations that affect WAT function.

TPS228 Background: Colorectal cancer (CRC) is the third most frequently diagnosed cancer and the second leading cause of cancer-related mortality worldwide. PD-1/PD-L1 inhibition, in combination with other modalities, has demonstrated significant benefit in patients (pts) with microsatellite instability-high or mismatch repair–deficient CRC. However, these agents have limited, if any, clinical benefit in pts with microsatellite stable (MSS) or mismatch repair–proficient (pMMR) CRC. RP2 is an enhanced potency oncolytic herpes simplex virus type 1 which expresses the fusogenic gibbon ape leukemia virus glycoprotein with the R sequence deleted (GALV-GP-R–), granulocyte-macrophage colony-stimulating factor (GM-CSF), and an anti–CTLA-4 antibody-like molecule; RP3 additionally expresses 4-1BB and CD40 activating ligands but does not express GM-CSF. Both agents have demonstrated preliminary safety and antitumor activity in pts with solid tumors. This study will evaluate the safety and activity of RP2 and RP3 in combination with atezolizumab (Atezo) plus bevacizumab (Bev) in pts with advanced MSS/pMMR CRC (NCT05733611). Methods: Pts with a histologic diagnosis of unresectable and/or metastatic CRC, with documented MSS or pMMR status, and previously treated with up to 3 standard-of-care systemic regimens will be enrolled in the RP2 + Atezo + Bev or RP3 + Atezo + Bev treatment groups (30 pts per group). Further key inclusion criteria include Eastern Cooperative Oncology Group performance status 0 to 1 and adequate hepatic, renal, and hematologic function. RP2/RP3 will be injected into tumors by direct or image-guided injection. Pts will receive 8 total doses of up to 10 mL of RP2 or RP3, with a first dose concentration of 1 × 106 plaque-forming units (PFU)/mL, followed by 3 doses of 1 × 107 PFU/mL every 2 weeks, and then 4 doses of 1 × 107 PFU/mL every 3 weeks. Pts may receive a second course of up to 8 injections of RP2/RP3 if study criteria are met. Bev will be administered starting on week 1; Atezo will be administered starting week 7. The primary endpoint is objective response rate; secondary endpoints are safety, overall survival, progression-free survival, duration of response, and complete response rate. Antitumor activity will be evaluated using Response Evaluation Criteria in Solid Tumors version 1.1 as modified for this study. Safety will be assessed by physical examination, clinical laboratory evaluations, vital signs, and monitoring for adverse events (AEs; including serious AEs). Clinical trial information: NCT05733611 .

175 Background: Alterations to mismatch repair (MMR) genes such as MLH1, MSH2, MSH6, and PMS2, can lead to microsatellite instability-high (MSI-H) tumors. MMR mutations can be inherited in Lynch syndrome (LS) but can also be a result of de novo alterations resulting in MSI-H malignancies. Treatment with immune checkpoint inhibitors (ICIs) have been shown to improve survival in such patients (pts) compared to systemic chemotherapy. However, there is a paucity of information in the literature with respect to outcomes of pts with germline vs somatic MMR mutations in MSI-H colorectal cancer (CRC). Furthermore, the presence of underlying BRAF mutation confers resistance to chemotherapy in pts with CRC, however data regarding outcomes with immunotherapy in such pts with MSI-H CRC is lacking. Methods: Pt records from the Mayo Clinic (AZ, MN, FL) between 2008-2023 denoted as MSI-H CRC, were included for retrospective review. Pt demographics, treatment courses, and genomic profiles were collected. Overall survival (OS) and progression-free survival (PFS) were estimated using the Kaplan-Meier method. Outcome differences between sub-groups were accessed with the log rank test and univariate analyses were assessed with Cox-regression. Results: A total of 81 pt records were identified (n = 18 LS, n = 63 non-LS). Stage IV disease was found in 28% of LS pts vs 59% of non-LS pts at diagnosis. Pembrolizumab was the most common ICI selected for metastatic CRC pts for both LS and non-LS pts (65% vs 94%) with median treatment duration for each group 11.7 vs 8.8 months. There were no differences in OS or PFS for any stage between LS vs non-LS. Median OS for all patients was 82 months. For all stage IV patients, median OS was 44 months, and median PFS was 19 months. Interestingly, for stage IV pts treated with immunotherapy, tumors that harbored BRAF V600E mutation (n = 21) vs BRAF wild type (n = 20) had a significantly lower median OS 19 vs 113 months (HR = 2.69, p = 0.043) as well as median PFS 12 vs 95 months (HR = 2.48, p = 0.041). Conclusions: In this study, we found that patient treated with ICIs have similar outcomes in the presence of germline vs. somatic MMR mutations in MSI-H CRC. The presence of an associated BRAF V600E mutation, which occurs in non-LS MSI-H CRC conferred worse outcomes.

170 Background: KRAS-mutant colorectal cancer (CRC) has a worse prognosis and greater resistance to therapy, attracting tremendous efforts to develop both allele-specific and pan-KRAS targeting agents including KRAS G12C and G12D inhibitors. CRC has demonstrated both primary and acquired resistance to KRAS G12C inhibitors, which can be better evaluated by circulating tumor DNA (ctDNA)-based next generation sequencing (NGS) methods without tissue sampling bias. Here, we utilized clinical ctDNA data to delineate the landscape of candidate alterations associated with primary resistance to KRAS G12D targeting in CRC. Methods: Samples from advanced CRC patients with Guardant360 data in 2021 were categorized into a KRAS G12D group and comparator groups of KRAS G12C and KRAS not detected (ND) according to KRAS mutation status. Demographics, MSI-H status, and blood tumor mutational burden (bTMB) were summarized and compared between groups using t-test for continuous variables and Fisher’s exact test for categorical variables. Co-occurrence patterns of alterations between KRAS and other genes were assessed. Primary resistance mutations were inferred by acquired resistance findings to KRAS G12C inhibitors reported by Awad et al. N Engl J Med. 2021 and Tanaka et al. Cancer Discov. 2021. Results: Among approximately 2500 samples, 16.8% had KRAS G12D, the most common KRAS mutation in CRC. The mean allele frequency of KRAS G12D (11.4%) is comparable to that of KRAS G12C (10.9%). Mean age (61.6 years), MSI-H frequency (3.8%), and mean bTMB (14.2 Mut/Mb) of KRAS G12D group is comparable to those of KRAS ND group. This contrasts with KRAS G12C with significantly lower mean age (60.2 years, p=0.01), lower MSI-H frequency (0.56%, p=0.03), and higher mean bTMB (16.3 Mut/Mb, p=0.02) compared to KRAS ND. Both KRAS G12D and G12C significantly co-occur with APC, PIK3CA, and SMAD4 alterations. FBXW7, AR, and KEAP1 alterations were unique co-occurrences to KRAS G12D, while KIT was unique to KRAS G12C. Candidate alterations associated with primary resistance in KRAS G12D CRC included simultaneously present mutations of KRAS Q61H (14.1%), G12V (6.4%), G12C (5.9%), G13D (3.6%), BRAF V600E (2.7%) and others, amplification of MET (2%), MYC (1.6%), and KRAS (1.4%), and FGFR3-TACC3 translocation (Table). Conclusions: KRAS G12D is associated with unique co-occurring molecular alterations compared to KRAS G12C in CRC. ctDNA-based NGS platforms can survey candidate alterations associated with primary resistance to KRAS G12D targeting in CRC. Further validation in preclinical models is needed. [Table: see text]

Background: Nanoparticles (NPs) have been extensively utilized as a drug delivery system to control the release of therapeutic agents to treat cardiac injuries. However, despite the advantages of utilizing NP-based drug delivery for treating heart diseases, the current delivery system lacks specificity in targeting the cardiac tissue, thus limiting its application. Methods: We created three linear peptides, each consisting of 16–24 amino acids. These peptides were conjugated on the surface of NPs, resulting in the formation of cardiac targeting peptide (CTP)-NPs (designated as CTP-NP1, CTP-NP2, and CTP-NP3). To assess their effectiveness, we compared the binding efficiency of these three CTP-NPs to human and mouse cardiomyocytes. Additionally, we determined their distribution 24 h after injecting the CTP-NPs intravenously into adult C57BL/6J mice. Results: When compared to control NPs without CTP (Con-NPs), all three CTP-NPs exhibited significantly increased binding affinity to both human and mouse cardiomyocytes in vitro and enhanced retention in mouse hearts in vivo. A thorough assessment of the heart sections demonstrated that the binding specificity of CTP-NP3 to cardiomyocytes in vivo was significantly greater than that of Con-NPs. None of the three CTP-NPs were proven to cause cardiomyocyte apoptosis. Conclusions: Biocompatible and safe CTP-NP3 can target the heart via binding to cardiomyocytes. This approach of targeting specific molecules-coated NPs may help in delivering therapeutic compounds to cardiomyocytes for the treatment of heart diseases with high efficacy and low toxicity to other tissues.

Abstract Background: Clinical, molecular, and genetic epidemiology studies displayed remarkable differences between ever- and never-smoking lung cancer. Methods: We conducted a stratified multi-population (European, East Asian, and African descent) association study on 44,823 ever-smokers and 20,074 never-smokers to identify novel variants that were missed in the non-stratified analysis. Functional analysis including expression quantitative trait loci (eQTL) colocalization and DNA damage assays, and annotation studies were conducted to evaluate the functional roles of the variants. We further evaluated the impact of smoking quantity on lung cancer risk for the variants associated with ever-smoking lung cancer. Results: Five novel independent loci, GABRA4, intergenic region 12q24.33, LRRC4C, LINC01088, and LCNL1 were identified with the association at two or three populations (P < 5 × 10−8). Further functional analysis provided multiple lines of evidence suggesting the variants affect lung cancer risk through excessive DNA damage (GABRA4) or cis-regulation of gene expression (LCNL1). The risk of variants from 12 independent regions, including the well-known CHRNA5, associated with ever-smoking lung cancer was evaluated for never-smokers, light-smokers (packyear ≤ 20), and moderate-to-heavy-smokers (packyear > 20). Different risk patterns were observed for the variants among the different groups by smoking behavior. Conclusions: We identified novel variants associated with lung cancer in only ever- or never-smoking groups that were missed by prior main-effect association studies. Impact: Our study highlights the genetic heterogeneity between ever- and never-smoking lung cancer and provides etiologic insights into the complicated genetic architecture of this deadly cancer.