The Validation of the BCR/ABL1-like Predictor across Laboratories Shows Reproducibility of Results

Background

BCR/ABL1-like (alias Ph-like) acute lymphoblastic leukemia (ALL) is a clinically challenging subgroup. Its incidence varies from 10% in children to 30% in adult ALL. The underlying genomic background, usually characterized by the presence of kinase activating lesions, opens the way to targeted treatment. However, one of the major issues of this subset is represented by the identification of a standardized assay for their recognition at diagnosis. Indeed, while gene expression profile (GEP) and next-generation sequencing (NGS) are efficient techniques to identify these cases, they are difficult to be implemented in the diagnostic routine. Our group previously developed a rapid, simple and cost-effective algorithm based on a quantitative real time-polymerase chain reaction (Q-RT-PCR) named "BCR/ABL1-like predictor". Briefly, the expression levels of 10 genes - NUDT4, SEMA6A, ADGRE5, SOCS2, JCHAIN, CRLF2, TP53INP1, CD99, IFITM1 and IFITM2 - are used to generate a score: cases with a score ≥-0.3 are classified as BCR/ABL1-like.

Aims

After developing the BCR/ABL1-like predictor, we sought to:

1. further refine the ability of this model to correctly predict BCR/ABL1-like ALL by evaluating NGS and RNA-sequencing of BCR/ABL1-like and non-BCR/ABL1-like cases, and to define the incidence of specific lesions;

2. evaluate the reproducibility of the predictor when performed in external laboratories or when samples with a known genetic background were analyzed in house by the predictor.

Results

NGS and RNA-sequencing were carried out in 28 BCR/ABL1-like cases: CRLF2 overexpression, was found 35.7% of cases, with 3 harboring a CRLF2 rearrangement and 1 with a concomitant CRLF2 mutation. Furthermore, 13 JAK/STAT pathway mutations - JAK1, n=5; JAK2, n=3; IL7R and CRLF2, n=2 each, JAK3, n=1 - were identified in 33.3% of cases. Finally, RNA-sequencing and/or FISH experiments of the BCR/ABL1-like ALL cases revealed 11 lesions: 5 ABL-class fusion genes (3 NUP214/ABL1, 1 ZC3HAV1/ABL2 and 1 EBF1/PDGFRB), 2 BCR/JAK2, 3 CRLF2-r and 1 DDX3X/USP9X.

In order to verify the reproducibility of the predictor, a collaboration was started across Europe with the Institute of Hematology and Blood Transfusion (ÚHKT, Prague), Josep Carreras Leukaemia Research Institute (Barcelona) and Munich Leukemia Laboratory (Munich), through the exchange of material and/or data. The first collaboration was carried out with the ÚHKT and Josep Carreras Leukaemia Research Institute laboratories: 11 cDNA samples (from 1 μg of total RNA), previously studied by our laboratory and classified as BCR/ABL1-like (n=5) and non-BCR/ABL1-like (n=6) ALL were shipped blindly and evaluated for the model in these laboratories. The technical set-up was sent to each laboratory, consisting of experimental procedures, PCR protocol and the threshold for Q-RT-PCR analysis. We received the raw data (i.e. 2^(-ΔCt) values) and uploaded them in the on-line BCR/ABL1-like predictor tool. We had 100% concordance with the ÚHKT and 81.8% with the Josep Carreras Leukaemia Research Institute. The main discrepancies were with 2/5 BCR/ABL1-like cases resulted as non-BCR/ABL1-like in the external laboratory: 1 was a p210 BCR/ABL1-positive case, used as control, which proved borderline, and the second was a BCR/ABL1-like case that, on the other hand, was not studied for the genetic features. A further collaboration was carried out with the MLL laboratory: we received 12 RNA samples - already analyzed by NGS and FISH - for evaluation by the BCR/ABL1-like predictor. The BCR-ABL1-like predictor classified 5 cases as BCR/ABL1-like and 7 as non-BCR-ABL1-like: all 5 BCR/ABL1-like carried a typical signature, consisting of a CRLF2 rearrangement plus JAK/STAT pathway mutations. Discordant cases were represented by 2 non-BCR-ABL1-like cases: one had a CRLF2 rearrangement plus JAK/STAT mutations and the other an ABL1 rearrangement. It is important to underline that the latter case had a borderline score (-0.482). Thus, the concordance rate was 83.3%.

Conclusions

This study shows that the BCR/ABL1-like predictor is a valid and reproducible tool across laboratories to identify rapidly these cases, with the goal of introducing genetic-driven therapeutic approaches upfront. One critical aspect is represented by borderline cases: in these patients, NGS experiments are required to define the underlying genomic lesion. Further investigations are currently underway.