Fit-for-purpose quantitative liquid biopsy based droplet digital PCR assay development for detection of programmed cell death ligand-1 (PD-L1) RNA expression in PAXgene blood samples

Abstract

Development of a clinically applicable liquid biopsy-based test for PD-L1 mRNA expression would be beneficial in providing complementary evidence to current immunohistochemistry assays. Hence, we report the development of a fit-for-purpose assay for detection of blood PD-L1 mRNA expression using droplet digital polymerase chain reaction (ddPCR). TaqMan® assays were selected based on coverage of the PD-L1 gene and were tested for linearity and efficiency using real-time quantitative PCR. Four reference genes were analyzed in positive control cell line (A549 treated with interferon gamma, [IFN γ]) genomic DNA. The PD-L1 primer/probe sets were also evaluated in ddPCR for limit of blank, limit of detection, and precision. Finally, thirty-five healthy volunteer samples were evaluated to establish a baseline level of PD-L1 expression. In ddPCR, the limit of blank was determined to be 0 copies and the limit of detection was determined to be less than or equal to 19 copies of PD-L1. The average intra-run coefficient of variation in the ddPCR assay was 7.44% and average inter-run CV was 7.70%. Treatment of A549 cells with IFN γ resulted in a 6.7-fold increase in PD-L1 expression (21,580 copies in untreated cDNA versus 145,000 copies in treated cDNA). Analysis of healthy human samples yielded a median value of 1659 PD-L1 copies/μL with a range of 768-7510 copies/μL. The assay was transferred to an external service provider and results from our in-house experiments and those conducted externally shows a correlation of 0.994. In conclusion, a fit-for-purpose liquid biopsy-based, purely quantitative ddPCR assay for the detection of PD-L1 mRNA expression was developed and validated using PAXgene RNA blood samples. Linearity, reproducibility, limit of blank and limit of detection were measured and deemed suitable for clinical application. This ultra-sensitive liquid biopsy ddPCR assay has promising clinical potential in screening, longitudinal monitoring and disease progression detection.

Fig 1. IFN γ treatment effects on reference gene expression.

cDNA from either A549 untreated cells or A549 + IFN γ treated cells were titrated over four input points from 100ng to 1.56ng and tested in qPCR with four reference genes, B2M, GUSB, TBP, and RPLP0. B2M showed significant difference between untreated and treated samples whereas GUSB, TBP, and RPLP0 showed very little difference between the two samples. GUSB was ultimately chosen as the optimal reference gene and is shown outlined in the black boxes.

Fig 2. PD-L1 Assay 1, PD-L1 Assay 2 and PD-L1 Assay 3 standard curves.

Dilutions of the PD-L1 cDNA construct were run on PCR and checked in ddPCR to confirm copy numbers for accurate extrapolation of results. The lines indicate the regression line that is obtained from individual models. Furthermore, a confidence band of the regression line was added (95% pointwise confidence intervals of the mean response). On the top left side of this plot, the formula of the regression line, with X = log10(Copy Number), and the coefficient of determination (R²) are displayed. In our given model, we can utilize R² as a measurement of linearity, as it can be interpreted as fraction of variability that may be described by a straight line. Therefore, a value close to 1 (or 100%) indicates a good linear fit.

Fig 3

The relative expression of PD-L1 in 5 healthy volunteer samples normalized to A549 untreated (control) cell line as measured in qPCR (left) and the distribution of relative expression from three PD-L1 assays (right). The box plot (right) shows the mean expression in the horizontal bold line. The median relative expression is rather consistent between the three PD-L1 assays however the mean expression level shows more variation. Finally, PD-L1 assay 2 shows the tightest distribution of expression among the five healthy samples.

Fig 4

The absolute quantification of PD-L1 expression in 5 healthy volunteer samples as quantified in ddPCR (left) and the distribution of the absolute expression (right). The box plot (right) shows the mean expression in the horizontal bold line. In this analysis PD-L1 assay 3 showed lower overall expression than Assay 1 or Assay 2.

Fig 5. The correlation between EMD Serono (sponsor) and an external service provider’s (vendor) results when performing ddPCR on eight sponsor provider blinded samples was determined and visualized as shown below.

Samples tested include two commercially procured, healthy volunteer samples and A549 cell lines treated and untreated with IFN-γ. Treated cell line was tested neat, at 0.2x, 0.4x, and 0.8x dilution. Untreated sample was tested neat and at a 0.5x dilution. Data points for all samples are shown. A correlation coefficient of 0.994 was calculated.

Fig 6. Comparison of PD-L1 and GUSB copy numbers reported by two operators at an external service provider.

The continuous horizontal line shows the mean copy number and the median is shown in the box plot. As shown in the box and whiskers plot, the assay is highly replicable and concordant between the two operators. One sample produced very high PD-L1 copy numbers and also very high GUSB copy numbers.

Fig 7. Schematic showing the PD-L1 gene, highlighting regions which the three PD-L1 TaqMan® assays span.