Analytical Validation of the nCounter^®-based Lymphoma Subtyping Test (LST) for Cell-of-Origin (COO) Identification in Formalin-Fixed Paraffin-Embedded (FFPE) Diffuse Large B-Cell Lymphoma (DLBCL) Specimens

Introduction: Diffuse large B-cell lymphoma (DLBCL) is an aggressive form of non-Hodgkin's lymphoma with two distinct molecular subtypes defined by Cell-of-Origin (COO): germinal center B-cell (GCB) and activated B-cell (ABC). COO is being evaluated as a predictive biomarker in multiple DLBCL clinical trials hence the need for an accurate and precise companion diagnostics (CDx). NanoString LST is an investigational multiplexed digital gene expression assay performed on the nCounter^® Dx Analysis System that identifies COO from FFPE tissue with a turnaround time of 2-3 days (Wallden B et al, JCO 2015; Storhoff J et al, Blood 2015). The test is based on the Lymph2Cx gene expression assay which profiles 15 classifier genes and 5 housekeeping genes to compute a linear predictor score (LPS) and determine the COO subtype (Scott D et al, Blood 2014). To establish a CDx, both analytical and clinical validations are required using the locked algorithm and assay procedure. Clinical validity of the LST is currently being evaluated in a global Phase 3 study in which the assay is used to prospectively select patients with ABC-type DLBCL to receive in a randomized fashion R2-CHOP or R-CHOP (Nowakowski G et al, JCO 2016). The analytical validation studies described herein were designed to evaluate the analytical performance of the test across multiple laboratory sites, operators, instruments, reagent lots, and RNA input levels.

Design: Reproducibility was measured across 3 sites and 6 operators by testing replicate tissue sections from 43 FFPE DLBCL tumor blocks following independent pathology review at each site. The testing included both core needle and surgical biopsy specimens. The standard deviation (SD) of the LPS output was analyzed as a continuous variable using a linear mixed model to estimate assay variance, and the site-to-site concordance was evaluated. Analytical precision was measured across 3 sites and 6 operators by testing 5 pooled DLBCL tumor RNA samples representing each COO subtype including samples near the subtype thresholds. The sensitivity of the assay was evaluated by testing 4 RNA levels within the recommended input range (62.5 ng, 125 ng, 500 ng, 1000 ng) and 2 input levels outside of this range (50 ng and 1250 ng) using 14 DLBCL tumor RNA samples and 2 reagent lots. The percentage of test samples passing all QC metrics was determined at each input level, and the subtype concordance was evaluated compared to the nominal test input level (500 ng). Interference of human genomic DNA was assessed by omitting DNase from the assay procedure. The impact of including non-tumor tissue was assessed by omitting the tissue macrodissection step from the assay.

Results: In the multi-site reproducibility study, the average site-to-site LST subtype concordance with independent pathology review was >97% with no ABC-to-GCB discordances (or vice versa). The overall range of LPS spanned approximately 5000 units. The total SD was <100 LPS units including all sources of variation (site/pathologist, operator, and between-run/residual). No differences in variance were observed between excisional and core needle biopsies and <1% of the variance was attributed to sites and operators. In the multi-site precision study, the total SD of LPS was 46 including all sources of variation (site, operator, between-runs/day, input level, and within-run/residual). There were no significant differences between sites, operators or days of run. A separate precision study showed that reagent lot and instrument do not significantly contribute to assay variability. In the RNA input study, all input levels inside and outside the specified operating range had 100% assay pass rate. The estimated rate of samples switching from ABC to GCB was <0.00001% for all input levels. In the interference study, genomic DNA contamination resulted in a negative bias in LPS (mean bias = -186), indicating the necessity of DNase treatment in this assay.

Conclusions: The investigational NanoString LST has been analytically validated for identifying COO subtypes across multiple testing laboratories. The test provides a highly precise, sensitive and rapid method for measuring COO on FFPE DLBCL tumor specimens, including both excisional and core needle biopsies.