Page 125 - Haematologica - Vol. 105 n. 6 - June 2020

PTCL classification using RT-MLPA assay
Introduction
Peripheral T-cell lymphomas (PTCL) are a diverse group of neoplasms representing 10-15% of all lym- phomas worldwide, with large geographic variation. According to the 2017 revision of the World Health Organization (WHO) classification of lymphoid neo- plasms, PTCL comprise up to 30 entities derived from various subsets of mature T or natural killer (NK) cells.1 The heterogeneity and rarity of these tumors, combined with their complex immunophenotypic profile and par- tially overlapping features across different entities, make their diagnosis particularly challenging. In addition, there is a high variability in the diagnostic workup among pathologists, which may account for relatively poor reproducibility of the diagnoses.2-4 Although most cases can be ascribed to specific disease entities, approx- imately one-third of PTCL not fulfilling the criteria for other entities remain unclassifiable and are categorized “by default” as PTCL-not otherwise specified (NOS).
The classification of PTCL has undergone major changes over the past years with the incorporation of much new information on their genetic background and taking into account the notion that PTCL arise from dis- crete subsets of normal T cells. In recent years, the description of the signature and mutational landscape of PTCL has generated novel molecular biomarkers to refine the diagnostic criteria for some entities. Notably, the expression of TFH markers and the presence of genet- ic lesions associated with angio-immunoblastic T-cell lymphoma (AITL) (such as RHOA, TET2, DNMT3A, and IDH2 mutations), found in a significant proportion of PTCL-NOS,5-10 led to the reclassification of these as “nodal PTCL with a TFH phenotype” (TFH-PTCL) in the revised WHO classification.1 Among anaplastic large cell lymphoma (ALCL), the identification of recurrent rearrangements of the ALK gene led to ALK-positive ALCL being referred to as a definitive separate entity (ALCL ALK+), and to reconsider ALCL without ALK rearrangement as a distinct but genetically heteroge- neous group comprising subtypes characterized by alter- ations of the DUSP22/IRF4 or TP63 genes with distinct clinical, pathological and biological features.11 Among the remaining PTCL-NOS category, two molecular sub- groups defined by the expression of the TBX21 and GATA3 transcription factors have been proposed,12,13 with a worse prognosis suggested for GATA3-positive cases.13-16 In daily diagnostic practice, however, high- throughput technologies are difficult to integrate. Moreover, the immunohistochemical surrogates are not fully validated and require an increasingly large panel of antibodies, and their evaluation may be problematic or present limitations.3,17
Here, we designed a simple targeted mRNA expres- sion profiling assay based on reverse transcriptase-mul- tiplex ligation-dependent probe amplification (RT- MLPA), using a panel of molecular markers relevant to the characterization of PTCL. We first assessed the accu- racy of this assay in the classification of PTCL entities other than PTCL-NOS, and then used the assay to study the heterogeneity of PTCL-NOS. Our findings support this RT-MLPA assay as a robust and useful tool, suitable for the routine classification of PTCL and, therefore, promoting an optimal clinical management of PTCL patients.
Methods
Patients and tumor samples
A series of 270 lymphoma samples were selected within the framework of the multicentric T-cell lymphoma consortium (TENOMIC) of the Lymphoma Study Association (LYSA). All cases had been reviewed by at least two expert hematopathol- ogists, according to the criteria of the recently up-dated WHO classification.1 The series was enriched in nodal TFH-PTCL (TFH- PTCL) defined by the expression of at least two TFH markers among CD10, BCL6, CXCL13, PD1, ICOS and in PTCL-NOS defined as a diagnosis of exclusion of any well-defined entity. The design of the study is summarized in Online Supplementary Figure S1. Briefly, a classification cohort (n=230) was used to train a support vector machine (SVM) classifier and a diagnostic cohort (n=40) was used to evaluate its inter-laboratory repro- ducibility on formalin-fixed paraffin-embedded (FFPE) samples. The study was approved by the local ethics committee (CPP Ile de France IX 08-009).
RT-MLPA assay gene expression profiling
RNA extracted from frozen and/or FFPE tumor samples was applied to RT-MLPA, as described18 (Online Supplementary Methods). Briefly, this targeted multiplex assay consists of the hybridization and ligation of specific probes on cDNA, fol- lowed by PCR amplification. We designed 41 probes (Eurofins MWG Operon, Ebersberg, Germany) targeting 20 genes, select- ed for their relevance to PTCL classification (Table 1). RT-MLPA results were compared to Affymetrix HG-U133-plus-2.0 gene expression data in 72 previously reported cases.18,19
Bioinformatic analysis
A web interface was developed for the complete analysis of the RT-MLPA results (https://bioinfo.calym.org/RTMLPA). An SVM was developed to classify PTCL samples: two-thirds of the 184 PTCL of the classification cohort, which clustered in defined molecular branches according to the clustering (n=230), were ran- domly selected to train the classifier, which was validated in the remaining one-third of cases. A bootstrap resampling process was used to build 100 independent training and validation series. A definitive SVM predictor was thus developed using the 184 cases. This supervised learning model assigns a class to every PTCL sample. Therefore, we integrated the distance to the cen- troid of the predicted class for each sample to avoid classifying distant samples into the same group. The analytical process is detailed in the Online Supplementary Methods.
Histopathology and molecular validation
RT-MLPA signatures were correlated to immunochemical data, including expression of GATA3 and TBX21. The cut off for positive immunohistochemical staining was 10% of pre- sumed neoplastic cells (Online Supplementary Methods). Fluorescence in situ hybridization (FISH) for DUSP22/IRF4 rearrangement was performed in 20 ALCL. Mutations were val- idated using polymerase chain reaction (PCR) allele-specific and/or targeted deep sequencing.20,21 Technical details are pre- sented in the Online Supplementary Methods.
Data analysis
Affymetrix and RT-MLPA gene expression values were corre- lated using Spearman’s correlation test. Correlations between immunohistochemical results and RT-MLPA gene expression values were evaluated using Wilcoxon’s rank-sum test. Unsupervised hierarchical analysis was performed using the Ward method.
haematologica | 2020; 105(6)
1583