Gene Expression Profiling Signatures Allow the Identification of Unclassifiable Leukemic B-Cell Lymphoid Neoplasms

Leukemic B-cell chronic lymphoproliferative disorders (B-CLPD) encompass a heterogeneous group of defined disease entities. However, around 10% of the cases cannot be reliably classified based on the current criteria and are considered B-CLPD not otherwise specified (B-CLPD NOS). Few recurrent mutations and genetic alterations have been reported in some entities, but none is specific.

We performed gene expression profiling (GEP) to develop a robust GEP-molecular classifier for leukemic B-CLPD. We analyzed purified blood of 189 B-CLPD, including 54 chronic lymphocytic leukemia (CLL), 54 mantle cell lymphoma (MCL), 12 follicular lymphoma (FL), 23 splenic marginal zone lymphoma (SMZL), 4 splenic diffuse red pulp lymphoma (SDRPL), 4 hairy cell leukemia (HCL), 4 HCL-variant (HCL-v), 6 lymphoplasmacytic lymphoma (LPL) and 28 B-CLPD NOS. We used a multiple step approach to build a GEP-array classifier and then analyzed an additional series as a validation cohort by quantitative PCR (qPCR). Mutational analysis of BRAF, MAP2K1, MYD88, NOTCH1, NOTCH2, SF3B1 and TP53 and copy number alterations were studied.

In the training set, a supervised analysis clustering revealed that each B-CLPD entity has a specific expression profile. By a multi-step GEP classifier using 43 genes (Table 1) we could classify CLL, SOX11-positive MCL (MCLc), HCL, FL, SOX11-negative MCL (MCLi) and HCL-v cases in six successive stages. However, we were not able to clearly identify distinct signatures for LPL, SMZL and SDRPL. Furthermore, we could classify 36% of B-CLPD NOS cases. Interestingly, the 43-gene signature identified in leukemic samples could also classify the 28 tumor splenic biopsies. Finally, we built a simple 8-gene predictive model using qPCR data (including: FMOD, KSR2, SOX11, MYOF, MME, CCND1, CXCR4, and CAMSAP2) that was used in an independent validation cohort and classified 14% B-CLPD NOS cases. The classification yield increased to 61% and 50%, for GEP-array and qPCR, respectively, when additional morphological, molecular and genetic features were considered.

Our findings support the use in a routine base of a simple test as a diagnostic tool that can be applied to help multiparameter interpretation in the classification of leukemic B-CLPD and specially B-CLPD NOS.

Table 1. Gene signatures (43 genes) and steps used for the GEP-array model.

Step model	B-CLPD	Specific signatures
1	CLL	FMOD, KSR2, ADTRP, CLNK, LEF1, FILIP1L, CTLA4, IGSF3, EBF1
2	MCLc	SOX11, PLEKHG4B, HDGFRP3, CNN3, PON2, SH3BP4, FCGBP, STMN1, FARP1, DBN1, NREP, NINL, MARCKSL1, MEX3D, CRIM1, KAZN
3	HCL	HPGDS, IL1R2, TJP1, PLOD2, EMP1, NOVA1
4	FL	MME, SLC2A5, SMAD1, PRDM15
5	MCLi	CCND1
6	HCL-v	TUSC1, LRP1B, KCNJ3, NRCAM, MS4A14, FAM129C, CXCR4
7	Miscellaneous LPL-SDRPL-SMZL	No specific genes

CLL: chronic lymphocytic leukemia; FL: follicular lymphoma, HCL: hairy cell leukemia;

HCL-v: hairy cell leukemia variant; LPL, lymphoplasmacytic lymphoma; MCLc: conventional

SOX11-positive mantle cell lymphoma; MCLi: indolent SOX11-negative mantle cell lymphoma;

SDRPL, splenic diffuse red pulp lymphoma; SMZL, splenic marginal zone lymphoma.

Bold letters indicate some of the genes included in the simple qPCR model.