Robust Assessment of Genomic Imbalance in Diffuse Large B-Cell Lymphoma Confirms Inferior Outcome Is Associated with Genomic Complexity and Identifies Potential Therapeutic Pathway Targets

Despite excellent initial responses of diffuse large B-cell lymphoma (DLBCL) patients to current frontline immunotherapy (RCHOP), only about 40% of patients are ultimately cured, with most relapses occurring within the first 2-3 years. Apart from clinical features, very few biomarkers are available and validated for risk stratification of DLBCL patients, other than cell-of-origin (COO) subtyping and MYC rearrangement. Genomic copy number changes identified using a variety of genomic profiling technologies including massively parallel sequencing, have variously been reported to have prognostic value in DLBCL. Our previous studies used a common analytical approach across three different clinical datasets to identify and define 32 common regions (CR) of overlapping genomic imbalance (17 gain, 15 loss), comprising 36 minimal common regions (MCR) observed in newly-diagnosed DLBCL. We now report the establishment of standardized calling criteria for a total of 50 aberrations within the 32 CR, taking into consideration GISTIC-defined peaks based on copy number data from three publicly available datasets: IS-172 (GSE11318, n=170), IS-51HR (E-MEXP-3463, n=51 with high IPI), and one similarly analyzed dataset IS-180 (GSE34171, n=180). As a form of validation of the scoring criteria, the frequencies of the aberrations detected in IS-172 in the known GCB and ABC COO subtypes were compared with those previously reported (PMID:18765795). The relative imbalance of the aberrations between the subtypes were re-capitulated including six occurring at higher frequency in the GCB subtype and five in the ABC subtype.

Integration of the expression profiles of 162 samples of IS-172 by univariate t-test with the 50 aberrations revealed 569 unique positively-correlated Refseqs mapping to 24 of the MCRs with at least 1.2 fold change in expression (Univariate t-test with P ≤ 0.05 after Benjamini-Hochberg false discovery rate [FDR]). Of these, 27 were located within overlapping peaks of gain/loss. Ingenuity Pathway Analysis determined five significant pathways (P ≤ 0.001, FDR<0.10): p53 signaling, PKC theta signaling in T lymphocytes, geranylgeranyl-diphosphate biosynthesis, B-cell receptor signaling, and RANK signaling in osteoclasts.

Association of each of the defined 50 aberrations with overall survival (OS) was performed with the Kaplan-Meier method and log-rank statistic in three RCHOP datasets: IS-124 (GSE15127, n=124), a subset of 70 of IS-180 (IS-70), and a dataset of 41 for which array-CGH was performed in-house using a targeted oligonucleotide (Agilent Technologies) with DNA extracted from sections of formalin–fixed paraffin-embedded de novo DLBCL (IH-41). Ten aberrations significantly associated (P ≤0.05) with poor outcome in at least one of the datasets: gain of 12q (12q13 and 12q14), 16q24, 19q13, and loss of 2q24, 2 sites at 6q21, 8p22, 9p21, 15q15, and 17p13. Of these 10 aberrations, loss of 17p significantly associated with gain of 16q, and losses of 6q, 8p, and 15q (P <0.05), while loss of 9p21 correlated with loss of 2q, 6q, and 8p as evaluated using the exact t-test. Gain of the 2 sites on 12q did not significantly correlate with any of the other 8 aberrations.

Genomic complexity was assessed by two methods. In the first, the median number of MCR aberrations was determined to be 1, across three datasets (IS-172, IS-180, and IS-124). Specimens in RCHOP-treated datasets IS-124 and IS-70 were then called “complex” if 2 or more MCR aberrations were detected: 31% and 67% respectively. In both datasets, a complex genome was significantly associated with overall unfavorable outcome (P=0.037, 0.006 respectively). Specimens in IS-172 and IS-70 were additionally classified as “complex” based on the presence of at least one of nine aberrations involving the CDKN2A-TP53-RB-E2F axis (PMID:22975378) where 33% and 57% of cases were scored as complex respectively. Using this approach, significant association with outcome was only found for IS-70 (P<0.001). Of note, between “complex” scoring approaches, 11% and 13% discordance was evident.

In summary, establishment of robust scoring criteria for copy number changes afforded correlative analyses with clinical features to reveal that genomic complexity is indeed associated with overall inferior survival in DLBCL and that integrated expression analysis identified biological pathways in DLBCL that may represent potential therapeutic targets.