Jeffrey Lin, Lisa Durkin, Kiran Madanahally Divakar, Sandeep S. Dave, Lilly I. Kong, Gurunathan Murugesan, Paula Carver, Eric D. Hsi, Angela M. B. Collie, Mitchell R. Smith, Elena Manilich, Thomas M. Daly, Brian T. Hill, Jork Nolling, Jeanna M Guenther-Johnson, and Tomas Radivoyevitch
Microarray gene expression profiling (GEP) has been used to identify molecular subtypes of diffuse large B-cell lymphoma (DLBCL) based on the similarity of GEP to a putative “cell of origin” (COO) and defines two molecular subtypes: activated B-cell-like (ABC) DLBCL and germinal centre B-cell-like (GCB) DLBCL (Alizadeh, et al 2000). This dichotomization has prognostic and biological significance, with the ABC subtype having a worse outcome and distinct pathobiology that includes activation of the B-cell receptor and nuclear factor (NF)-κB pathways (Lenz, et al 2008). Attempts to reduce this subclassification to practice using immunohistochemistry are fraught with technical and interpretive issues such that a need for practical quantitative molecular assays exists (Coutinho, et al 2013, de Jong, et al 2007, de Jong, et al 2009, Salles, et al 2011). Indeed, studies have refined this COO concept using limited gene sets, including a 14-gene model to assign ABC and GCB DLBCL subtype developed by Wright et al (2003) as well as a recently-described DLBCL subtyping assay based on parsimonious digital gene expression (Nanostring) technology (Scott, et al 2014). In order to support clinical trials for therapies targeting populations enriched in ABC DLBCL and to offer prognostic information for DLBCL patients as part of our clinical service, we developed a novel multiplex, single-tube, gene expression assay on the ICEPlex® system (PrimeraDx, Mansfield, MA), which allows differentiation between GCB and ABC DLBCL subtypes in formalin-fixed paraffin-embedded (FFPE) specimens using a Food and Drug Administration-cleared platform. Institutional review board approval was obtained for review of newly diagnosed DLBCL specimens at our institution. Paired frozen tissue was available for 30 patients and was used to isolate RNA for hybridization to U219 (27 samples) or U133plus2.0 (3 samples) Affymetrix microarrays. Microarray subtype classification was assigned based on the Wright algorithm, which calculates a linear predictor score (LPS) by multiplying a gene weight and Ct value for each gene and summing the resulting numbers for 14 genes (Wright, et al 2003). Samples with LPS above -150 were classified as GCB, scores between -150 and -200 were unclassified, and scores less than -200 were classified as ABC. A multiplex quantitative reverse transcription polymerase chain reaction (RT-PCR) DLBCL subtyping assay was developed on the ICEPlex system (PrimeraDx), targeting expression of the 14 genes used in the Wright algorithm, two reference genes and an internal control (enterobacteriophage MS2) (Table I) (Hlousek, et al 2012). Primers were designed to amplify short template mRNA regions of exon-spanning junctions. To allow discrimination of target-specific amplicons on the ICEPlex platform by fluorescent label and by size using capillary electrophoresis, PCR reverse primers were labelled with FAM or TYE fluorescent dyes, and forward and reverse PCR primers were equipped with 5′-nucleotide tags. RNA was extracted from a 10-μm FFPE slice of a DLBCL specimen using the Qiagen Allprep DNA/RNA FFPE kit (Germantown, MD). cDNA was generated using SuperScript III (Life Technologies, Grand Island, NY) and RT primers. PCR was performed in triplicate for each 10-μm FFPE slice on the ICEPlex system with Roche AptaTaq ΔExo polymerase (Indianapolis, IN), proprietary buffer (PrimeraDx), and primers. The resulting Ct values were normalized for each replicate and averaged for each gene. The integrity of the results was evaluated using a quality (Q) score calculated for each sample replicate [(mean Ct value of all targets)/(total number of target genes + number of target genes with Ct values less than 36)]. Reactions with a Q-score greater than 1.0 were rejected. Table I DLBCL Molecular Subtyping Assay Genes and Primers A subtyping (S) score was calculated using the normalized Ct data and gene weights of the Wright algorithm (Wright et al 2013). If there was no Ct value for a target, the value was set to a defined target-specific, upper-limit Ct value. Based on correlations with the microarray data and paired FFPE, samples corresponding to an S-score between -100 and 0 were unclassified, while S-scores less than -100 were classified as ABC and greater than 0 were classified as GCB. Thus, our training set of 30 DLBCL FFPE samples with matched frozen tissue microarray GEP data served as the gold standard against which initial S-scores were calibrated. The test set showed 96.7% and 90% agreement compared to microarray subtype for assignment of ABC cases and overall (ABC, GBC, and unclassified) cases, respectively (Figure 1 a). Figure 1 Comparison of microarray GEP LPS and DLBCL molecular subtyping assay S-score for test (circles) and validation (diamonds) DLBCL cohorts (n=53) To validate the S-score cut-off values, additional DLBCL specimens, described previously (Zhang, et al 2013), with paired frozen and FFPE samples were analysed. Microarray data from the Affymetrix Genechip Human Gene 1.0 ST (Santa Clara, CA) platform was obtained, and LPS were calculated (Wright, et al 2003). The DLBCL molecular subtyping assay was performed on FFPE-extracted RNA, and S-scores were obtained for 23 samples (Figure 1 a). The validation set performed similarly, with 91.3% and 87.0% agreement compared to microarray subtype for assignment of ABC samples and overall samples, respectively. This validated the chosen S-score cut-off values, giving a sensitivity of 95.2% and specificity of 93.8% for assignment of the ABC subtype for the combined test and validation sets. For the combined sets, the sensitivity was 95.2% and specificity was 90.6% for assignment of the GCB subtype. Inter-assay and intra-assay variation was determined using two DLBCL specimens (Figure 1 b). The coefficient of variance (CV) was 6.1% for the GCB DLBCL specimen and was 2.9% for the ABC DBCL specimen. The assay was performed on freshly-isolated RNA from an additional 45 FFPE DLBCL samples from up to 12 years prior. When included with other samples from our institution in the test and validation cohorts, the assay failed in only 7 out of 88 samples (8.0%). Dilution studies showed that samples with as low as 0.4 μg of RNA per 10-μm FFPE slice yielded acceptable Q-scores and appropriate subtype classification. In summary, we have successfully developed a multiplex quantitative expression profiling assay designed for DLBCL tumour classification into GCB or ABC subtypes from a single 10-μm FFPE section. This DLBCL molecular subtyping assay has the potential to provide rapid and accurate subclassification for DLBCL patients for prognostic implication as well as clinical trial patient selection in a Clinical Laboratory Improvement Amendments-certified laboratory environment. Finally, the assay, validated against GEP, could be used as an external reference for those clinical laboratories attempting to validate their immunohistochemical-based algorithms against a GEP-based standard.