Introduction/Background*RAD51 protein has been proposed as a functional readout of homologous recombination (HR) status using formalin-fixed paraffin embedded (FFPE) tumour tissue blocks. Recently, few laboratories have assessed the performance of RAD51 as a predictive biomarker. However, the robustness of the test when applied in different laboratories has not been systematically investigated so far. In this study, we performed an interlaboratory (n = 4) analytical validation to determine the interobserver variability and the effect of (subtle) differences in the co-immunofluorescence (co-IF) protocol and microscope technicalities on RAD51 scores.

Methodology The RING trial cohort comprised of 12 high-grade serous ovarian cancer cases. On unstained serial sections of FFPE tumour tissue blocks, a co-IF staining with RAD51 and geminin was performed: (1) centrally in Vall d’Hebron Institute of Oncology and (2) locally in participating centers. The centrally stained slides were distributed among participating centers for local RAD51 scoring. For the scope of the RING trial, a predefined and uniform scoring methodology was applied. Scoring was performed blinded for genetic and clinical data. Specific features in the analysis of the co-IF, including the number of RAD51 foci per nucleus and the presence of RAD51 subclonality, i.e., distinct RAD51 positive and negative areas, were incorporated in the RAD51 scoring form. For non-normally distributed data, variability was analyzed using the median, 25^th percentile (Q1) and 75^th percentile (Q3).

Result(s)*Median variability in RAD51 scores between observers in centrally stained slides was 21% (Q1: 15%; Q3: 24%) (figure 1). For the majority of cases (n = 10/12), a limited interobserver variability, defined as ≥ 3 observers with scores in a narrow range, was detected. In contrast, in cases where observers noted granular pannuclear RAD51 staining or RAD51 background, there was a substantial variability in scores (figure 1; case 6 and 8). Median variability in RAD51 scores between centrally and locally stained IF slides was 7.7% (Q1: 4.1%; Q3: 11.7%).

• Download figure
• Open in new tab
• Download powerpoint

Abstract 376 Figure 1

Violin plot showing the varaiability in RAD51 scores (foci cut-off: n=3) between four observers. RAD51 scores were based on centrally stained slides. Lines in the plot indicate the median, min and max and RAD51 scores

Conclusion*This is the first cross-European interlaboratory assessment of the performance of RAD51/Geminin co-IF. We show that subtle local protocol differences do not impact final RAD51 scores. Furthermore, we elucidated features that may negatively impact RAD51 score accuracy.