Image analysis and manual scoring of PD-L1 in melanomas using physical and virtual double staining
Henrik Thirstrup1, Andreas Schønau1, Søren Nielsen2, Cigdem Babi2, Freja Stubkjær Bukalo2, Anne Vishoff Petersen2
1 Visiopharm, Hoersholm, Denmark
2 Aalborg University Hospital, Aalborg, Denmark
Objective To evaluate the diagnostic potential of a programmed death-ligand 1 (PD-L1)/SOX10 double stain in melanomas compared to PD-L1 single stained specimens by manual assessment and digital image analysis (DIA) methods.
Methods Melanoma tissue samples were stained with PD-L1 and SOX10 to clearly distinguish tumor and non-tumor cells for PD-L1 assessment. SOX10 positive/negative nuclei and PD-L1 positive/negative cells were quantified using DIA and classified as SOX10+/PD-L1+ based on presence of both markers and absence of adjacent SOX10-/PD-L1+ cells. Negative and PD-L1 stained samples were analyzed in corresponding SOX10 positive regions using VirtualDoubleStaining™.
Results In 28 samples (n=28), each method found the following numbers of <1%, 1-5% and >5% positive:
DIA PD-L1/SOX10: 11, 5, 12
DIA PD-L1: 1, 4, 23
MANUAL PD-L1/SOX10: 23, 2, 3
MANUAL PD-L1: 13, 9, 6
Conclusion Differences in PD-L1/SOX10 and PD-L1 scoring indicates that PD-L1 expression is diminished in double stained samples. However, tumor-infiltrating immune cells may be contributing to the assessment in the absence of SOX10 tumor identification. DIA scored PD-L1/SOX10 higher compared to manual PD-L1/SOX10 scoring in samples with low-positivity and cytoplasmic staining. DIA of PD-L1/SOX10 matches manual PD-L1 scoring more closely than DIA of PD-L1 and may be advantageous for disregarding tumor-infiltrating immune cells.