ECP 2022 Abstract Book

Virchows Archiv (2022) 481 (Suppl 1):S1–S364 13 Results: The mean age in IgAN was 34±10 years; fourteen were males; mean s. creatinine 3.1±1.8mg/dl and 24-hour proteinuria 2.5±0.8gm/day. Mean circulating MPs levels in IgAN, LN and healthy were 8.1x105/μl, 6.7x103/μl and 2.4x105/μl, respectively. MMPs in IgAN constituted 54% of total MPs. Mean urinary sCD163, KIM-1 and MCP-1 in IgAN, LN and healthy controls were 11.4ng/ml, 27ng/ml and 0.18ng/ml; 2.5ng/ml, 1.84ng/ml and 0.37ng/ml; 2.3ng/ ml, 7.05ng/ml and 0.13ng/ml respectively. CD163+ mac- rophages in IgAN were 4.8±5.1/glomeruli and correlated significantly with presence of endocapillary hypercellularity (E1) and crescents (C2). The mean number of CD163+ cells in tubulo-interstitium were 69±35/hpf. Urinary sCD163 lev- els correlated significantly with number of CD163+ cells in glomeruli. Conclusion: We found monocyte activation and M2 (CD163+) macrophage tissue infiltration in IgAN. M2 macrophage tissue infiltration and urinary sCD163 levels correlate with prolifera- tive glomerular changes suggesting its role in the early active stage of renal disease. Urinary sCD163 may act as non-invasive biomarker in assessing active proliferative lesions in IgAN. Funding: SERB-DST Grant (CRG/2018/003042) OFP-05-012 Podocyte injury – aristolochic acid nephropathy in mice D. Miljkovic*, J. Grujic-Milanovic, I. Capo, M. Popovic, T. Kravic-Stevovic, J. Popovic, D. Lalosevic *Department of Histology and Embryology, Faculty of Medicine, University of Novi Sad; Center for Medical and Pharmaceutical Research and Quality Control (CEMPhIC), Faculty of Medicine, University of Novi Sad, Serbia Background & objectives: Aristolochic acid nephropathy is a chronic tubulointerstitial renal disease in which important symp- toms can be proteinuria and albuminuria. In this study, we exam- ined glomerular morphometric features and protein excretions in NMRI mice treated with aristolochic acid I. Methods: Experimental animals were treated intraperitoneally with 10 mg/kg aristolochic acid I for seven consecutive days, vehicle control received 2.5% polyethylene glycol 400, and the control received saline only. The experiment lasted 60 days, with several different euthanasia time points for light and transmission electron microscopy glomerular injury assessment. Nestin and WT1 were used as immunohistochemical markers for identify- ing podocytes. Results: For every euthanasia time point, mean mesangial score in glomeruli between aristolochic acid treated mice and control groups showed no significant difference. Furthermore, glomeruli of aristolochic acid treated mice had a decreased number of WT1 positive podocytes, lower cytoplasmic nestin expression and area fraction than mice that received 2.5% polyethylene glycol 400 and saline. In addition, ultrastructural changes of podocytes in the aristolochic acid treated group, observed under a transmis- sion electron microscope, indicate foot process effacement, kary- opyknosis, and thickening of the glomerular basement membrane with electron-dense deposits. Significant albuminuria occurred in experimental animals from later experiment phases compared with control groups. Conclusion: Our findings suggest that exposure to aristolochic acid I induce glomerular damage by reducing the number of podocytes and affecting the normal functioning of the glomerular filtration barrier, thus serving as valuable data in further research related to the treatment of aristolochic acid nephropathy. OFP-05-013 Deep learning-based histopathologic segmentation of peritu- bular capillaries in kidney transplant biopsies D. van Midden*, M. Hermsen, E. Steenbergen, L. Hilbrands, J. van der Laak *Radboud University Medical Center, The Netherlands Background & objectives: Peritubular capillaritis scoring is an important feature for diagnosing antibody-mediated rejection (ABMR). This task suffers from interobserver variability and might benefit from automation. As a first step towards automatic peritubular capillaritis quantification, we developed a peritubular capillary (PTC) segmentation algorithm. Methods: Kidney transplant biopsies (n=54) were 1) stained with periodic-acid Schiff (PAS), 2) scanned into whole-slide images (PAS WSI), 3) re-stained using CD34-antibody, and 4) scanned again (CD34 WSI). Guided by the CD34 WSI, a pathologist manually annotated approximately 19.000 PTCs on the PAS WSI. The dataset was used to train (n=40) and test (n=14) a deep learning (DL)-based network. Results: We developed a U-net DL network architecture, with an Efficientnetb2 backbone and a pre-trained encoder using ImageNet. The network was trained using 12,000 patches (512 x 512 pixels) per epoch. Various techniques were applied to prevent overfitting and to improve the model’s generalization. Training the network on a resolution of 0.5 μm/pixel using a non-PTC/PTC ratio of 3:1 yielded an F1 score of 0.74, with a precision and recall of 0.78 and 0.70, respectively. We observed reduced performance on cases with prominent interstitial alterations, as PTCs become less recogniz- able, while certain pathologies mimic PTCs (e.g. atrophic tubules, matrix deposition). Conclusion: This study presents a DL-based algorithm for the segmentation of PTCs in PAS-stained kidney transplant biopsies. This is a first step towards a more accurate, reproducible scor- ing of peritubular capillaritis using DL. The results highlight the applicability of DL for clinical use to guide pathologist in routine diagnostics. Next steps will include incorporation of this algorithm in the development of a fully automated Banff classification algo- rithm, as part of our DIAGGRAFT project, funded by the Dutch Kidney Foundation. OFP-05-014 Training a deep learning model for quantification of fibrosis in non-neoplastic kidney biopsies - a feasibility study N. Mola, E. Hodneland, H. Weishaupt, S. Leh* *Haukeland University Hospital, Norway Background & objectives: Interstitial fibrosis is a key prognos- tic marker of kidney disease. Accurate quantification is therefore important. The study aim is to develop a deep learning algorithm for quantification of interstitial fibrosis that can be used on hae- matoxylin-eosin (HE) stained kidney sections. Methods: To create annotated training data, tissue sections were first stained with HE and then - after destaining - with sirius red. Masks of the sirius-red stained fibrosis areas were created using conventional image analysis. A deep learning algorithm was trained with these masks to measure fibrosis in the identical HE stained slide. The model performance was validated with the F-statistics. Results: The advantage of this approach is that time and resource consuming manual annotations of the fibrosis areas are avoided but supervised learning still can be performed. A deep neural network based on U-Net was used for image segmentation. HE and the mask images were divided into tiles (512 x 512 pixels). Feasibility of the method was tested in a pilot study with 10 representative renal biopsies with varying degrees of interstitial fibrosis. The deep S22