Abstract: FR-OR49
Complement Convertases in Glomerulonephritis: An Explainable Artificial Intelligence-Assisted Renal Biopsy Study
Session Information
- Molecular Pathology, Pathogenesis, and Animal Models: A Correlative Approach to Kidney Diseases
November 05, 2021 | Location: Simulive, Virtual Only
Abstract Time: 04:30 PM - 06:00 PM
Category: Pathology and Lab Medicine
- 1600 Pathology and Lab Medicine
Authors
- Wiech, Thorsten, Universitatsklinikum Hamburg-Eppendorf, Hamburg, Hamburg, Germany
- Noriega, Maria de las Mercedes, Universitatsklinikum Hamburg-Eppendorf, Hamburg, Hamburg, Germany
- Schmidt, Tilman, Universitatsklinikum Hamburg-Eppendorf, Hamburg, Hamburg, Germany
- Wulf, Sonia, Universitatsklinikum Hamburg-Eppendorf, Hamburg, Hamburg, Germany
- Koch, Timo, HS Analysis, Karlsruhe, Germany
- Marquardt, Philipp, HS Analysis, Karlsruhe, Germany
- Biniaminov, Sergey, HS Analysis, Karlsruhe, Germany
- Hoxha, Elion, Universitatsklinikum Hamburg-Eppendorf, Hamburg, Hamburg, Germany
- Tomas, Nicola M., Universitatsklinikum Hamburg-Eppendorf, Hamburg, Hamburg, Germany
- Huber, Tobias B., Universitatsklinikum Hamburg-Eppendorf, Hamburg, Hamburg, Germany
- Zipfel, Peter F., Leibniz-Institut fur Naturstoff-Forschung und Infektionsbiologie eV Hans-Knoll-Institut, Jena, Thüringen, Germany
Background
Complement activation is crucial in the pathogenesis of C3 glomerulopathy (C3GP). It is likely also involved in other forms of glomerulonephritis (GN), however, here, intensity, significance, and predominant activation pathways are less clear.
Methods
Proximity ligation assays (PLA) were used to visualize C3/C5 complement convertases in renal biopsies. Close proximity of C3b and Bb or C2 and C4b was interpreted as assembled alternative or classical/lectin C3/C5 convertases, respectively. For quantification we used deep learning based on explainable artificial intelligence (xAI) in a two-stage workflow: 1. detection of the glomeruli and 2. detection of the PLA signals. Signal densities were calculated as numbers of signals per glomerular area [signals/sqmm]. Cases of C3GP (n=10), immune complex-mediated membranoproliferative GN (IC-MPGN; n=10), IgA nephropathy (IgAN; n=10), postinfectious GN (PIGN; n=10), and membranous nephropathy (MN; n=10) were analyzed and compared with thin basement membrane disease (n=10) as control group, in which no local complement activity is expected.
Results
In C3GP and PIGN a clear predominance of the alternative convertase (mean 8410 and 14483 signals/sqmm) was detected as compared to the control group (mean 798 signals/sqmm) whereas IC-MPGN and MN cases showed higher densities of the classical/lectin convertase (3039 signals/sqmm and 5015 signals/sqmm) as compared to the control group (mean 176 signals/sqmm). Interestingly, cases with IgAN revealed increased densities for the alternative convertase (mean 2088 signals/sqmm) but only very slightly increased densities for the classical/lectin convertase (225 signals/sqmm).
Conclusion
This work shows the applicability of human-machine collaboration based on xAI to characterize and quantify local complement activity. The results reveal insights into the role of complement in the pathogenesis of different forms of glomerulonephritis. Moreover, it opens up the possibility to assess the local activity of the alternative and the classical/lectin pathway C3/C5 convertases in individual patients. Since several novel anti-complement agents are under clinical investigation, this method might become helpful for a more individual complement targeting therapy.
Funding
- Government Support – Non-U.S.