ECP 2023 Abstracts

S6 Virchows Archiv (2023) 483 (Suppl 1):S1–S391 13 of these markers in vaVIN versus its wide differential of benign conditions. Methods: CK17 and SOX2 immunohistochemistry was evaluated on 13 vaVINs (lesion and adjacent normal mucosa) and 38 acanthotic mimickers including verruciform xanthoma (n=6), lichen simplex chronicus (LSC), lichen sclerosus, psoriasis & pseudoepitheliomatous hyperplasia (n=5 each), syphilis, granular cell tumour, and differenti- ated VIN (n=4 each). Stains were recorded as negative (absent=0/ patchy=1+) or positive (continuous moderate/strong, partial=2+ or full thickness=3+). Results: 13/13 (100%) vaVINs were positive for CK17 (92%=3+) while 6/7 (85%) were positive for SOX2 (17%=3+). The normal mucosa adja- cent to vaVIN was CK17-positive in only 1/10 (10%) cases, whereas it was SOX2-positive in 5/5 (100%) cases. Among mimickers, 28/38 (74%) were CK17-positive (43%=3+) and 20/38 (53%) SOX2-positive (25%=3+). CK17 showed higher rates of positivity than SOX2 in most categories including pseudoepitheliomatous hyperplasia (100% vs 0%), LSC (100% vs 40%), psoriasis (100% vs 40%), lichen sclerosus (83% vs 40%), and syphilis (50% vs 25%). CK17 sensitivity and specificity for vaVIN diagnosis was 100% and 26% if 2+/3+ were considered positive, and 92% and 69% if only 3+ was considered positive. Conclusion: CK17 and SOX2, portended as useful distinguishing vaVIN from non-dysplastic vulva, have limitations when acanthotic lesions are included in the differential. They are frequently positive in vaVIN but are also often expressed in benign lesions, although in the latter CK17 staining is more frequently partial-thickness. Full-thickness CK17 staining has good performance in vaVIN diagnosis, whereas negative CK17&SOX2 argues against it. CK17 also has value in margin assessment, as histologically normal mucosa next to vaVIN is consist- ently negative (unlike SOX2). OFP-02-002 Opening the black box: validating AI-based findings with spatial transcriptomics in high-grade serous ovarian carcinoma A. Laury*, S. Zheng, O. Youssef, N. Aho, S. Hänninen, J. Tang, O. Carpen *University of Helsinki, Finland Background & objectives: H&E images of high-grade serous ovarian carcinoma (HGSC) contain prognostic information detectable only by artificial intelligence (AI). These regions can predict patient outcome using H&Es alone, and now we aim to reveal the biology behind these regions via spatial transcriptomics. Methods: A deep learning neural network tool identified novel tumour regions most indicative of outcome in high-grade serous carcinoma. These novel (HC) regions, and background tumour (BG) tissue, were probed with 10x Visium for FFPE spatial transcriptomics. Gradient boosting machines were trained, and differential gene expression and pathway enrichment analyses were performed. Individual transcripts were validated by RNA in-situ hybridization. Results: Data was successfully obtained for 35/36 tumour samples from 16 patients, identifying 17,866 genes across 9,129 capture loca- tions. Gradient boosting machines confirm that prediction accuracy for outcome is higher using transcript profiles from HC regions (0.68) as compared to BG regions (0.44)(p = 0.0133). Pathway analyses show significant enrichment of inflammatory signal- ling pathways in poor outcome HC regions (interferon alpha response), while cell cycle control and replication pathways (Hallmark DNA repair) are enriched in improved outcome tumour HC regions. 1Differential gene expression analyses reveal that 142 genes (0.78%) show significant differential expression exclusively in HC regions. RNA-ISH validation of one such gene (JUN) confirms the transcriptomics finding. Conclusion: Artificial intelligence-based image-analysis (AI-IA) of diagnostic HGSC slides can identify morphologic patterns invisible to the human eye and guide selection of biologically meaningful regions. Spatial transcriptomics confirms improved outcome prediction in HC tumour regions, as well as biologically relevant pathway enrichment and differential gene expression, which can be validated using standard techniques. AI-IA together with spatial transcriptomics offers a promising toolkit to identify biological features associated with cancer behaviour, making AI-based findings more interpretable and clinically relevant. Funding: University of Helsinki Doctoral Programme in Clinical Research, Suomen Lääketieteen Säätiö, HUS Diagnostic Center, the European Union’s Horizon 2020 research and innovation programme under Grant Agreement No 965193 for DECIDER, Finnish Cancer Foundation, Sigrid Juselius Foundation and Finska Läkaresälskapet. OFP-02-003 Is it time to move towards a new classification of high grade endo- metrial stromal sarcoma? D. Arciuolo*, S. Patrizi, S. Vallese, R. Alaggio, S. Sfregola, B. Padial Urtueta, A. Travaglino, M. Valente, I. Giovannoni, G.M. Milano, L. Pedace, C. Nardini, L. Pedone Anchora, G.F. Zannoni, E. Miele *Pathology Unit, Department of Woman and Child’s Health and Pub- lic Health Sciences, Fondazione Policlinico Universitario Agostino Gemelli IRCCS, Rome, Italy Background & objectives: High-grade endometrial stromal sarcoma (ESS-HG), paediatric BCOR-rearranged sarcoma and clear cell sar- coma of the kidney share common molecular drivers. We investigated their methylation profile to better define the classification and the potential relationship among these entities with clinical impact. Methods: Fifty-one cases of morphologically and molecularly con- firmed tumours, including 6 ESS-HG, 10 undifferentiated uterine sar- comas (UUS), 27 paediatric BCOR-rearranged sarcoma (pBS) and 8 clear cell sarcoma of the kidney (CCSK), underwent DNA methylation profiling. For the hierarchical clustering we included 19 Ewing sarcomas as control group and an external cohort of 32 previously published pBS/ CCSK and low/high grade-ESS. Results: DNAmethylation profiling. All ESS-HGwere classified into ESS- HG class with calibrated scores (CS) >0.9 in 5/6 cases and pBS/CCSK as “sarcoma with BCOR alterations” with CS> 0.9 in 26/27 cases. UUS were classified into several groups with low CS except 2 that were classified as malignant rhabdoid tumour and undifferentiated sarcoma (CS>0.9). CNV analysis. Fifteen out of 27 pBS/CCSK and 1/16 uterine sarcoma cases showed a flat profile. The remaining cases exhibit several altera- tions, mostly 1q gain and loss of 9p- 13q- 10q. The hierarchical clustering revealed two main clusters, the “BCOR- altered family” composed of pBS/CCSK and ESS-HG and a second group with ESS-LG, UUS, and Ewing sarcoma Conclusion: ESS-HG and pBS/CCSK are epigenetically and closely related entities and could have a common progenitor cell. The variation in CNV could be related to the higher genetic instability in tumour of adult patients. These findings, combined with the histological and molecular similar- ity, allow us to speculate that in the future ESS-HG and pBS/CCSK might be grouped as BCOR-sarcoma. Moreover, the proximity between ESS-LG and UUS suggests the pos- sibility of common histogenesis to be further investigated. OFP-02-004 Is Mandard’s tumour regression score able to assess the response to chemo-radiotherapy in patients with locally advanced cervical cancer? G. Scaglione*, D. Arciuolo, A. Travaglino, A. Santoro, N. D’Alessandris, M. Valente, G. Angelico, F. Inzani, A. Muccilli, S. Sfregola, B. Padial Urtueta, S. Spadola, L. Pedone Anchora, M.G. Ferrandina, G.F. Zannoni