Matthias Kloor

Purpose: Lynch syndrome (LS), caused by pathogenic germline variants (PGVs) in DNA mismatch repair (MMR) genes, confers increased risks of colorectal, endometrial, and extracolonic malignancies. Urothelial carcinoma (UC) contributes substantially to LS-associated morbidity and mortality. We assessed its frequency, clinical presentation, genotype-associated risk, and molecular features. Methods: We retrospectively analyzed genetically confirmed adult LS carriers in the nationwide German Consortium for Familial Intestinal Cancer registry. UC diagnoses identified by International Classification of Diseases, 10th Revision codes C65-C68 underwent manual clinical and pathologic validation. Available tumors underwent centralized pathologic review, MMR immunohistochemistry, and microsatellite instability (MSI) testing. Results: Among 3,274 LS carriers, 144 (4.4%) had 224 primary UCs. The median age at first UC was 55 years (range, 22-79 years), and UC was the first malignancy in 50 patients (34.7%). Of 202 tumors with known location, 114 (56.4%) arose in the upper urinary tract. Metachronous and synchronous primary UCs occurred in 23 patients (16.0%) and five patients (3.5%), respectively. MSH2 -associated PGVs were present in 111 patients (77.1%). In multivariable analysis, MSH2/EPCAM carrier status was associated with UC risk (hazard ratio, 4.43 [95% CI, 2.85 to 6.89]), whereas previous colorectal cancer and UC family history were not. Among 109 patients with diagnostic information, 26 (23.9%) presented with urinary tract–related symptoms, 20 (18.3%) were diagnosed incidentally, and 11 (10.1%) were diagnosed during urologic or oncologic surveillance. Of 31 lesions from 18 patients, 29 (93.6%) were mismatch repair–deficient (dMMR) and 30 (96.8%) were MSI-high; all patients had at least one dMMR tumor and at least one MSI-high tumor. Conclusion: UC in LS occurs relatively early, is strongly associated with MSH2/EPCAM carrier status, and represents the first malignancy in one third of patients. These findings support prospective evaluation of genotype-adapted surveillance.

Colorectal cancer (CRC) development in Lynch syndrome (LS) has long been regarded to follow an adenoma-carcinoma sequence accelerated in comparison to microsatellite-stable (MSS) CRC development. Yet, several clinical observations challenged this hypothesis, most notably the persistently high CRC incidence under colonoscopy surveillance, a non-measurable benefit from shorter screening intervals, and substantial differences in cancer risk between carriers of the different mismatch repair (MMR) genes. Advances in the last decade have produced new hypotheses, highlighting the distinct biology of LS CRC and unravelling several co-existing pathways to cancer. Alongside the genomic heterogeneity, the immunogenic load created by the accumulation of frameshift mutations imposing selective pressure and leading to immune evasion, appears to be a critical step for cancer manifestation. We discuss potential clinical implications of these advances in understanding CRC pathogenesis for cancer prevention in LS and outline open questions that remain.

Accurate grading of cervical biopsies on Hematoxylin and Eosin (H&E) stained whole slide images (WSIs) is essential for distinguishing high grade lesions from low grade changes, yet this process is subject to considerable inter-observer variability. In this study, we evaluate a foundation model-based multiple instance learning (MIL) pipeline for binary high-grade squamous intraepithelial lesion (HSIL) detection on H&E stained WSIs. We benchmark our Athena foundation model against four state-of-the-art pathology foundation models: H-optimus-0, Hibou-L, Midnight-12k and Virchow, across datasets from five different countries: Portugal, Cambodia, Germany, Poland and Scotland. Athena achieved the highest mean area under the curve (AUC) (0.931) with the lowest cross-country variability (STD = 0.022). Furthermore, we compared the model’s diagnostic performance to that of trained pathologists on a dataset with p16-confirmed ground truth. Our model improved sensitivity from 84% to 95% while maintaining comparable specificity (85% vs. 84%). Failure analysis revealed that the model’s errors were concentrated at the diagnostic boundary between low-grade and high-grade lesions, whereas pathologists’ errors spanned a broader range of misclassifications. These findings show the potential of foundation models for cervical cancer screenings worldwide.

Background/Objectives It is not established whether melanoma and lung cancer are part of the tumor spectrum of the Lynch syndromes (LS). Our first hypothesis was that, if melanoma is associated with LS, most prospectively observed cases would be carriers of pathogenic MSH2 variants (path_MSH2), as for other LS non-endodermal tumors. Our second hypothesis, which was derived from findings of the first study, was that the pattern of differences in the incidence of lung cancer in path_MMR carriers would mirror that for melanoma. Methods We used the Prospective Lynch Syndrome Database (PLSD) data and methods. Results Firstly, consistent with hypothesis, ten of 3154 path_MSH2 carriers followed for 26,309 years developed melanoma, compared with 6 of 5288 path_MLH1/MSH6/PMS2 carriers followed for 45,081 years (p = 0.04). Secondly, although the incidence of melanoma in carriers of path_MSH2 was 0.00041—similar to the populations in which the carriers resided—the average incidence rates for path_MLH1/MSH6/PMS2 carriers was 0.00012, lower than the corresponding population rates. Thirdly, the average lung cancer incidence rates in PLSD were 0.00112 in path_MSH2 carriers and 0.00032 in path_MLH1/MSH6/PMS2 carriers (p = 0.02), both lower than the population rates. Conclusions Melanomas and lung cancers—both of which are immunogenic through mechanisms independent of mismatch repair deficiency—may become even more immunogenic when they also exhibit microsatellite instability. This may lead to an increased likelihood that both tumors are eliminated by the immune system and, thus, show variable and lower incidence. These insights may help inform strategies for cancer prevention or treatment that are aimed at simultaneously interrupting multiple carcinogenic pathways.

Purpose:The Immunoscore predicts colorectal cancer prognosis but faces adoption barriers because of complex software and reimbursement issues. This study used open-source methods to explore a simplified prognostic model in nonmetastatic colorectal cancer by focusing on single T-cell markers.Experimental Design:A multicentric prospective cohort study in patients with nonmetastatic colorectal cancer assessed CD3+ and CD8+ tumor-infiltrating lymphocytes (TIL) in the invasive margin (IM) and tumor core (TC) using QuPath. An immune cell score (ICS), based on TIL densities (CD3-IM, CD8-IM, CD3-TC, and CD8-TC), was calculated similarly to the Immunoscore. A split sample approach (70:30) estimated adjusted HRs for cancer-specific survival in training and validation sets. Classification and regression tree analysis identified the most prognostic TIL, and its model was compared with an ICS model for performance (Brier score) and discrimination (concordance probability estimate).Results:Over a median follow-up of 9.0 years, 203 colorectal cancer–specific deaths occurred among 1,260 patients. Classification and regression tree–selected CD8-IM was the most prognostic TIL at a cutoff of 231 cells/mm2. Patients with high CD8-IM had better cancer-specific survival than low CD8-IM in both training (HR 0.58, 95% confidence interval, 0.40–0.84) and validation sets (HR 0.35, 95% confidence interval, 0.21–0.60). Brier scores of CD8-IM and ICS survival models were comparable in both training and validation cohorts, whereas the survival discrimination of CD8-IM slightly outperformed the ICS in the validation set (concordance probability estimate: CD8-IM: 0.748; ICS: 0.730).Conclusions:CD8-IM alone provided prognostic information comparable with the ICS. Simplified, cost-effective TIL assessments could improve clinical translation and guide adjuvant therapy in early-stage colorectal cancer.

Microsatellite instability (MSI) is associated with high immunogenicity in tumors due to the abundance of neoantigens, making MSI cancers particularly responsive to immune checkpoint blockade (ICB) therapy. However, a substantial proportion of patients with MSI gastrointestinal (GI) adenocarcinomas do not benefit from ICB, and non‐invasive biomarkers for monitoring treatment response are lacking. This study investigated the utility of MSI detection in extracellular vesicle (EV) DNA as a liquid biopsy‐based approach for therapy monitoring in patients with advanced MSI GI cancers undergoing ICB therapy and compared it with the analysis of cell‐free (cf) DNA. Blood ( n = 52) and tumor samples ( n = 16) were collected from 19 patients before and during therapy. Plasma from 18 patients and 30 healthy controls was analyzed for MSI using four diagnostic mononucleotide markers in both EV DNA and cell‐free DNA (cfDNA). MSI was detected in EV DNA in 8 out of 18 patients and in cfDNA in 9 out of 18 patients, with high concordance (93%) between the two approaches. None of the healthy controls showed MSI signals. MSI detection was more frequent before therapy initiation compared to during therapy time points, and a transition from MSI to non‐MSI status during treatment was associated with clinical benefit or objective response. This switch often occurred before the first staging at 3 months. Thus, MSI analysis in EV DNA is a promising minimally invasive tool for real‐time monitoring of ICB therapy response in MSI GI cancer patients, yielding results comparable to cfDNA while providing handling advantages, and warrants further validation in larger cohorts.

Microsatellite instability (MSI) is a key biomarker in colorectal cancer (CRC). Accurate distinction between MSI and microsatellite stable (MSS) tumors is critical for diagnosis, prognosis, and therapeutic decision-making. We developed an uncertainty-aware deep learning model for MSI prediction directly from H&E-stained whole-slide images. The approach is based on attention-based multiple instance learning, which can learn from slide-level labels. We extended the approach and included ensembling for more robust prediction. For training, we used 1,492 slides from three public archives and for evaluation 1,094 slides from five independent external cohorts. The approach provides predictions with confidence intervals and rejects uncertain cases. It achieved sensitivity over 97% on 59-78% of the cases, depending on the cohort. These results show that MSS cases can be identified with high confidence while uncertain cases can be rejected to avoid unreliable predictions, enabling clinically reliable MSS triage in colorectal cancer with AI.