PURPOSE: Rapidly expanding medical literature challenges oncologists seeking targeted cancer therapies. General-purpose large language models (LLMs) lack domain-specific knowledge, limiting their clinical utility. This study introduces the LLM system Medical Evidence Retrieval and Data Integration for Tailored Healthcare (MEREDITH), designed to support treatment recommendations in precision oncology. Built on Google's Gemini Pro LLM, MEREDITH uses retrieval-augmented generation and chain of thought . METHODS: We evaluated MEREDITH on 10 publicly available fictional oncology cases with iterative feedback from a molecular tumor board (MTB) at a major German cancer center. Initially limited to PubMed -indexed literature (draft system), MEREDITH was enhanced to incorporate clinical studies on drug response within the specific tumor type, trial databases, drug approval status, and oncologic guidelines. The MTB provided a benchmark with manually curated treatment recommendations and assessed the clinical relevance of LLM-generated options (qualitative assessment). We measured semantic cosine similarity between LLM suggestions and clinician responses (quantitative assessment). RESULTS: MEREDITH identified a broader range of treatment options (median 4) compared with MTB experts (median 2). These options included therapies on the basis of preclinical data and combination treatments, expanding the treatment possibilities for consideration by the MTB. This broader approach was achieved by incorporating a curated medical data set that contextualized molecular targetability. Mirroring the approach MTB experts use to evaluate MTB cases improved the LLM's ability to generate relevant suggestions. This is supported by high concordance between LLM suggestions and expert recommendations (94.7% for the enhanced system) and a significant increase in semantic similarity from the draft to the enhanced system (from 0.71 to 0.76, P = .01). CONCLUSION: Expert feedback and domain-specific data augment LLM performance. Future research should investigate responsible LLM integration into real-world clinical workflows.

Epithelial ovarian cancer (EOC) remains a highly-lethal gynecological malignancy, characterized by frequent recurrence, chemotherapy resistance and poor 5-year survival. Identifying novel predictive molecular markers remains an overdue challenge in the disease’s clinical management. Herein, in silico analysis of TCGA-OV highlighted the tRNA-derived internal fragment (i-tRF-GlyGCC) among the most abundant tRFs in ovarian tumors, while target prediction and gene ontology (GO) enrichment analysis predicted its implication in key biological processes. Thereafter, i-tRF-GlyGCC levels were quantified in a screening EOC (n = 98) and an institutionally-independent serous ovarian cancer (SOC) validation cohort (n = 100, OVCAD multicenter study). Disease progression and patient death were used as clinical endpoints for the survival analysis. Internal validation was performed by bootstrap analysis and the clinical net benefit was estimated by decision curve analysis. The analysis highlighted the significant association of i-tRF-GlyGCC with advanced FIGO stages, suboptimal debulking and most importantly, with early progression and poor overall survival of EOC patients. The OVCAD validation cohort corroborated the unfavorable predictive value of i-tRF-GlyGCC in EOC. Ultimately, evaluation of i-tRF-GlyGCC with the established/clinically used prognostic markers offered superior patient risk-stratification and enhanced clinical benefit in EOC prognosis. In conclusion, i-tRF-GlyGCC assessment could aid towards personalized prognosis and support precision medicine decisions in EOC.

A crucial mode of action of trastuzumab is the labeling of HER2-positive (HER2+) tumor cells for the eradication by natural killer (NK) cells, a process called antibody-dependent cellular cytotoxicity (ADCC). However, despite widespread HER2 expression among cancer entities, only a fraction, with robust HER2 overexpression, benefits from trastuzumab therapy. ADCC requires both sufficient lymphocytic infiltration and close binding of the immune cells to the antibody-tagged tumor cells. We hypothesized that the chemokine CX3CL1 could improve both processes, as it is synthesized as a membrane-bound, adhesive form that is eventually cleaved into a soluble, chemotactic protein. Here, we show that CX3CL1 overexpression is a positive prognostic marker in breast cancer. CX3CL1 overexpression attracted tumor-suppressive lymphocytes, including NK cells, and inhibited tumor growth and lung metastasis in the syngeneic 4T1 breast cancer mouse model. In HER2+ SKBR3, MDA-MB-453, and HT-29 tumor cells, CX3CL1 overexpression increased NK cell–mediated cytotoxicity in vitro and acted synergistically with trastuzumab. Even though CX3CL1 did not further improve trastuzumab efficacy in vivo in the trastuzumab-sensitive MDA-MB-453 model, it compensated for NK-cell depletion and prolonged survival. In the HER2 low–expressing HT-29 model, however, CX3CL1 overexpression not only prolonged survival time but also overcame trastuzumab resistance in a partly NK cell–dependent manner. Taken together, these findings identify CX3CL1 as a feasible pharmacologic target to enable trastuzumab therapy in HER2 low–expressing cancers and render it a potential predictive biomarker to determine therapy responders.

CX3CL1 is a multifunctional chemokine that is involved in numerous biological processes, such as immune cell attraction and enhanced tumor immune cell interaction, but also in enhancing tumor cell proliferation and metastasis. The multifarious activity is partially determined by two CX3CL1 isoforms, a membrane-bound and a soluble version generated by proteolytic cleavage through proteases. Here, we investigated the impact of CX3CL1 overexpression in MDA-MB-453 and SK-BR-3 breast cancer cells. Moreover, we evaluated the therapeutic capacity of Matrix-Metalloproteinases-inhibitors TMI-1 and GI254023X in combination with the anti-HER2 antibody trastuzumab in vitro and in vivo. TMI-1 and GI254023X caused a reduced shedding of CX3CL1 and of HER2 in vitro but without effects on tumor cell proliferation or viability. In addition, trastuzumab treatment did not retard MDA-MB-453 cell expansion in vitro unless CX3CL1 was overexpressed upon transfection (MDA-MB-453^CX3CL1). In humanized tumor mice, which show a coexistence of human tumor and human immune system, CX3CL1 overexpression resulted in a slightly enhanced tumor growth. However, trastuzumab treatment attenuated tumor growth of both MDA-MB-453^CX3CL1 and empty vector transfected MDA-MB-453 transplanted mice but showed enhanced efficiency especially in preventing lung metastases in CX3CL1 overexpressing cancer cells. However, TMI-1 did not further enhance the trastuzumab treatment efficacy.

Triple-negative breast cancer (TNBC) is an aggressive breast cancer subtype with poor patient prognosis and limited therapeutic options. A lack of prognostic biomarkers and therapeutic targets fuels the need for new approaches to tackle this severe disease. Extracellular matrix degradation, release, and modulation of the activity of growth factors/cytokines/chemokines, and the initiation of signaling pathways by extracellular proteolytic networks, have been identified as major processes in the carcinogenesis of breast cancer. Members of the kallikrein-related peptidase (KLK) family contribute to these tumor-relevant processes, and are associated with breast cancer progression and metastasis. In this study, the clinical relevance of mRNA expression of two members of this family, KLK10 and KLK11, has been evaluated in TNBC. For this, their expression levels were quantified in tumor tissue of a large, well-characterized patient cohort (n = 123) via qPCR. Although, in general, the overall expression of both factors are lower in tumor tissue of breast cancer patients (encompassing all subtypes) compared to normal tissue of healthy donors, in the TNBC subtype, expression is even increased. In our cohort, a significant, positive correlation between the expression levels of both KLKs was detected, indicating a coordinate expression mode of these proteases. Elevated KLK10 and KLK11 mRNA levels were associated with poor patient prognosis. Moreover, both factors were found to be independent of other established clinical factors such as age, lymph node status, or residual tumor mass, as determined by multivariable Cox regression analysis. Thus, both proteases, KLK10 and KLK11, may represent unfavorable prognostic factors for TNBC patients and, furthermore, appear as promising potential targets for therapy in TNBC.