Deregulation of COMMD1 is associated with poor prognosis in diffuse large B-cell lymphoma.

BACKGROUND: Despite improved survival for the patients with diffuse large B-cell lymphoma (DLBCL), the prognosis after relapse is poor. The aim was to identify molecular events that contribute to relapse and treatment resistance in DLBCL.
METHODS: We analysed 51 prospectively collected pretreatment tumour samples from clinically high risk patients treated in a Nordic phase II study with dose-dense chemoimmunotherapy and central nervous system prophylaxis with high resolution array comparative genomic hybridization (aCGH) and gene expression microarrays. Major finding was validated at the protein level immunohistochemically in a trial specific tissue microarray series of 70, and in an independent validation series of 146 patients.
RESULTS: We identified 31 genes whose expression changes were strongly associated with copy number aberrations. In addition, gains of chromosomes 2p15 and 18q12.2 were associated with unfavourable survival. The 2p15 aberration harboured COMMD1 gene, whose expression had a significant adverse prognostic impact on survival. Immunohistochemical analysis of COMMD1 expression in two series confirmed the association of COMMD1 expression with poor prognosis.
CONCLUSION: COMMD1 is a potential novel prognostic factor in DLBCLs. The results highlight the value of integrated comprehensive analysis to identify prognostic markers and genetic driver events not previously implicated in DLBCL. TRIAL REGISTRATION: ClinicalTrials.gov NCT01502982.

Introduction

Diffuse large B-cell lymphoma (DLBCL) is the most common lymphoid neoplasm. It is an aggressive lymphoma entity, and only 50% of patients can be cured with anthracycline-based CHOP (cyclophosphamide, doxorubicin, vincristine, and prednisone) or CHOP-like chemotherapy. However, following the addition of rituximab or etoposide to CHOP, or the administration of CHOP dose-densely at two-week intervals (CHOP-14), response rates and survival have significantly improved [1]–[5]. Despite these advances, 20–30% of patients experience disease relapses or have primary refractory disease. Such patients could benefit from alternative therapies if their clinical outcome could be more accurately predicted at the time of diagnosis. Therefore, the identification of biological prognostic factors that could identify high-risk DLBCL patients is a priority.

Genome-wide molecular profiling has revealed a high degree of complexity in DLBCL,and significantly accelerated the understanding of oncogenic mechanisms in lymphomagenesis [6], [7]. On the basis of gene expression profiling (GEP), DLBCL is classified into distinct molecular subtypes [8]–[11]. Three major DLBCL entities, showing germinal center B-cell (GCB), activated B-cell (ABC)-like, and primary mediastinal B-cell lymphoma signatures have been recognized. Many oncogenic mechanisms distinguish GCB and ABC subtypes. For example, chromosomal translocations involving BCL2 and the c-REL locus amplification on chromosome 2p occur predominantly in the GCB DLBCLs [10], [12]. In contrast, ABC DLBCLs are characterized by transcriptional overexpression of BCL2 and a constitutive activation of the NF-κB signaling pathway [10], [13]. According to the gene expression based classification, the patients in the molecular subgroups also have different outcomes in response to chemo- and chemoimmunotherapy [9], [10].

Over the past few years, progress in molecular genetics and sequencing technologies has also revealed several previously unrecognized genetic lesions and pathways that are involved in DLBCL [14]–[17]. For example, recurrent mutations inactivating histone and/or chromatin modifying genes, and genes involved in immune recognition have been identified. However, despite the rapidly growing number of genetic aberrations reported in DLBCL, association of these findings with treatment outcome remains to be shown.

We have integrated the information from high-resolution gene copy number and expression microarrays to identify the most likely “driver gene” candidates associated with DNA copy number aberrations (CNAs) and poor prognosis in DLBCL. Importantly, with our cohort of high-risk DLBCL patients treated homogenously in a phase II study with dose-dense chemoimmunotherapy and systemic CNS prophylaxis, we were able to identify a genomic region harbouring a gene that has a survival effect and thus is a candidate for a novel molecular marker for poor prognosis.

Materials and Methods

Ethics Statement

Written informed consent was obtained prior to treatment and sampling from all patients included in the NLG-LBC-04 study. Clinical protocol and sampling were approved in the participating countries at the national level by Regional Committee on Health Research Ethics in Glostrup, Denmark, the Hospital District of Helsinki and Uusimaa Regional Committee on Medical Research Ethics in Finland, Oslo Regional Committee for Medical and Health Research Ethics in Norway, and Lund Regional Ethics Committee in Sweden. The trial was registered at ClinicalTrials.gov, number NCT01502982. For the retrospectively collected validation cohort, approval was obtained from the National Authority for Medicolegal Affairs, Finland and Helsinki University Central Hospital, Finland.

Patients

The prospectively collected screening (aCGH) and tissue microarray (TMA) cohorts consisted of DLBCL patients who were less than 65 years old and had primary high-risk (age-adjusted International Prognostic Index (aaIPI) score 2–3) disease. They were treated in the Nordic phase II NLG-LBC-04 protocol with six courses of R-CHOEP14 (rituximab, cyclophosphamide, doxorubicin, vincristine, etoposide, and prednisone supported with G-CSF) followed by systemic CNS prophylaxis with one course of high-dose methotrexate and one course of high-dose cytarabine [18]. The original clinical study included 156 eligible patients. Histological diagnosis was established from surgical or needle biopsy of the pretreatment tumour tissue by local pathologists according to current criteria of the World Health Organization classification [19], and subsequently reviewed by expert hematopathologists on a national basis. The patient selection for this molecular study was based on availability of fresh frozen tissue containing adequate material for DNA extraction and aGCH (screening cohort; n = 51), and for RNA extraction and gene expression profiling (n = 38). The infiltration of lymphoma cells in the tissue was assessed from frozen tissue section using hematoxylin eosin and toluidine blue stainings. Formalin-fixed paraffin-embedded (FFPE) tissue containing adequate material was used for the preparation of TMAs (TMA cohort; n = 70).

To validate the findings, an independent series of 146 primary DLBCL patients treated with chemoimmunotherapy at the Helsinki University Central Hospital between 2001 and 2010 was used. The patients were treated with R-CHOP (n = 126), R-CHOEP (n = 11) or other regimen (n = 9). The cases were selected based on the availability of FFPE tissue and clinical information.

Samples

RNA and DNA were extracted with Qiaqen AllPrep DNA/RNA/Protein Mini kit. CNAs were analysed from the DNA of 51 tumour samples hybridized onto Agilent Human (4×) 180 K CGH arrays. Tumour samples from 38 patients were eligible for mRNA analyses using Affymetrix Human Exon 1.0 ST arrays. All hybridizations were performed at the Biomedicum Genomics (University of Helsinki) according to manufacturer’s instructions. Hybridization protocols and raw expression microarray data are available at ArrayExpress archive http://www.ebi.ac.uk/arrayexpress/experiments/E-MEXP-3488 and http://www.ebi.ac.uk/arrayexpress/experiments/E-MEXP-3463. All tissue samples were collected before treatment.

qRT-PCR

Expression of the COMMD1 (Hs04190004_m1) and XPO1 (Hs00418963_m1) genes were validated by quantitative real-time polymerase chain reaction (qRT-PCR) using TaqMan Gene Expression Assays (Assays-On-Demand, Applied Biosystems) and the ABI Prism 7500 Fast Sequence Detection System (Applied Biosystems) for 24 available tumour samples. Normalization for the quantity of DNA was done by performing simultaneous qRT-PCR for GAPDH (TaqMan Pre-Developed Assay Reagents, Applied Biosystems). Each assay was determined by a comparative cycle threshold method, using the arithmetic formula provided by the manufacturer. All assays were performed in triplicate.

Subgroup Classification by Gene Expression Profiling

Samples involved in exon array analysis were divided into germinal centre B-cell (GCB) and non-GCB subgroups by gene expression profiling (GEP). Briefly, we utilized the log ratios of 44 genes from the gene expression panel by Wright et al. [20] and agglomerative hierarchical clustering (complete linkage) to divide samples into two subgroups. The IGHM gene was not on our array and was subsequently dropped from the analysis. Following the Wright classification [20], samples in one main branch of the resulting cluster tree were categorized as GCB and samples in the other branch as non-GCB DLBCLs. In addition, all samples were classified into GCB and non-GCB phenotypes immunohistochemically (IHC) according to Hans algorithm [21].

Immunohistochemical Analyses of COMMD1

IHC stainings were performed on FFPE tissue sections on TMA slides containing 2–4 tissue cores/patient, with a core diameter of 1 mm (TMA cohort), or whole tissue sections (independent validation cohort). After deparaffinization, heat-induced epitope retrieval (121°C, 3 min), and blocking of endogenous peroxidase, the slides were incubated with anti-COMMD1 antibody (1∶200, Sigma-Aldrich, Prestige Antibodies) at 4°C overnight. Staining was completed with Vectastain ABC kit reagents (Vector Laboratories) according to the manufacturer’s instructions, and slides were counterstained with hematoxylin.

To score the stainings, COMMD1-positivity was evaluated from one to three high-power fields (hpf; x630 magnification) with the Leica DM LB bright-field microscope (Leica Microsystems GmbH) and a camera attached to it (Olympus DP50, InStudio 1.0.1 Software). The most representative areas with intense staining pattern were first selected with low magnification and further digitized with hpf, resulting in microscopic images with area size of 0.02 mm2. Images were subsequently scored using computerized image analysis system [22]. All scorings were performed blindly.

Quantitative image analyses were performed using Anduril [22]. The colour space of each image was categorized to four expected representative colour classes: Brown, blue, white and background. The background class included faint brown and blue colours considered to be unspecific staining. The colour values were selected by pointing at 15 example colours for each class. All images were subjugated to use the same class specification. Each pixel was assigned to a class by finding the nearest example colour value, and a final staining coverage calculated from the area of each class present in each image.

Data Analysis

In copy number profiling the aCGH data were first normalized with locally weighted scatterplot smoothing (LOWESS) normalization (four iterations, 30% window) and the data were denoised with circular binary segmentation (p<0.05, split undo = none) [23]. The background noise level of copy number arrays was estimated by calculating the median probe signal of all arrays. An aberration was called significant if it was two standard deviations apart from the median. We identified minimal common CNA regions in which a CNA overlapped in 10% of the samples with both transcriptome and copy number data, and subsequent analysis were restricted to these regions. The Database of Genomic Variants (DGV, version 10, Nov 2010) [24] was used to determine locus specific copy number variants (CNVs). In short, for each gene in our CNA regions, the number of overlapping CNVs in the DGV according to genomic locus was counted. Genes with more than 10 CNVs were excluded from the subsequent analyses. Exon array expression data were normalized and transformed to gene expression level data by the Multiple Exon Array Preprocessing (MEAP) algorithm [25].

In order to find genes with a significant association of expression and CNA, all genes in the expression data, which were located in a minimal common CNA region, were first matched with their respective segmented copy number values. The samples were split into two groups based on their CNA status, separately for gains and losses, and a signal-to-noise statistic on the expression of each gene was calculated [26]. A p-value for each signal-to-noise score was calculated with a permutation test. Gain or deletion was associated with expression for genes that displayed up- or downregulation in CNA samples but stable expression in non-CNA samples (p<0.05). Moreover, only genes that exhibited CNA in at least three patients were analysed.

A Chi square test was performed to evaluate the differences in the frequency for the prognostic factors. Categorical data were compared using the Fisher’s exact test (two-sided). Pearson correlation coefficient was calculated to evaluate the correlation between the expression values from microarray and qRT-PCR analyses. All genes with altered copy number levels in at least five samples were analysed for patient survival. Survival curves with corresponding p-values were calculated using Kaplan-Meier analysis with the log-rank test. Receiver operating characteristic (ROC) curve analysis was used to determine the ideal cutoff values for survival outcomes. Univariate analyses were performed according to the Cox proportional hazards regression model. The progression-free survival (PFS) was calculated as the period between the dates of registration and lymphoma progression or relapse. Otherwise, the patients were censored at the last date of follow-up. Patients in remission were censored at the last date they were known to be alive. Patients who died due to causes other than lymphoma were censored at the date of death. Overall survival (OS) was calculated as a period between registration date and date of death. Surviving patients were censored at the last date they were known to be alive. Lymphoma-specific OS was calculated as a period between registration date and the date of death due to lymphoma. P-values less than 0.05 were considered to indicate statistical significance. Data analyses were done with the computational framework Anduril [22], which is designed for systematic integration, analysis and result interpretation of large-scale molecular data, and with IBM SPSS Statistics 20.0.

Results

Clinical Characteristics of the Screening Cohort

The baseline characteristics of the screening (aCGH) cohort of 51 patients treated in the NLG-LBC-04 protocol [18] are shown in Table 1. Median age of the patients was 55 years (range, 20–65 years). The overall response rate (ORR; complete response (CR)+partial response (PR)) in this study population was 98%. Median follow-up of the patient cohort was 55 months (range 31–101 months), 15 patients had relapsed, three experienced CNS relapse and 12 had died. Five of the deaths were not lymphoma-related. Predicted 5-year PFS was 69%, lymphoma-specific OS 85%, and OS 76%.

Table 1

Patient characteristics in the screening cohort.

	All n (%)	51 (100)
Gender	Female	19 (37)
	Male	32 (63)
Age	Median (range)	55 (20–65)
	<60	35 (69)
	60–65	16 (31)
	>65	0 (0)
Histology	GCB	27 (53)
	Non-GCB	20 (39)
	Other/Unclassified	4 (8)
Performance status	0–1	33 (65)
	2–3	18 (35)
B-symptoms		31 (61)
Elevated LDH		49 (96)
Stage	I–II	1 (3)
	III–IV	37 (97)
aaIPI	0	0 (0)
	1	0 (0)
	2	36 (71)
	3	15 (29)
mRNA analysis		38 (75)

Gene Copy Number Aberrations

Genome-wide copy number analysis of 51 lymphoma samples with aCGH revealed several gains and losses. All patients had at least one abnormality with an average number of 17.5±9.8 CNAs per patient. The most frequently (≥10%) altered regions as well as their frequencies and possible target genes are shown in Table 2. Some samples exhibited narrower alterations than others, which caused the small variation in the CNA frequencies. Of the recurrent CNAs previously reported in DLBCL [27]–[30], gains in 3q, 7q22.1, and 19p13, and loss in 6q were observed in 9.8% of the patients.

Table 2

Genome-wide overview of recurrent gains and losses.

Band	Gain Freq %	Loss Freq %	Position (Mb)	Possible target genes
1q24.2	12	8	167.69–167.76	MPZL1
1q44	20	6	247.00–247.10	AHCTF1
2p16.1–p15	12	NA	60.68–63.27	BCL11A, PAPOLG, REL, PUS10, PEX13, KIAA1841, AHSA2, USP34, XPO1, FAM161A, CCT4, COMMD1, B3GNT2, TMEM17, EHBP1
9p21.3	NA	14–18	21.80–22.01	MTAP, CDKN2A, C9orf53, CDKN2B
14q11.2	2	12	22.938–22.939	TRDV3
18q12.2	12	NA	34.82–35.15	CELF4
18q21.1	12	4	44.06–44.34	LOXHD1, ST8SIA5
18q23	12–16	4	77.62–77.71	KCNG2, PQLC1
20q11.22	16	NA	33.13–33.15	MAP1LC3A

The association of CNAs with molecular subgroups is summarized in Table S1 in File S1. Notably, GCB type DLBCL was characterized by more frequent gain of 2p15 and 2p16.1 including the well-known proto-oncogenes REL and BCL11A as compared to non-GCB DLBCL (15–19% vs. 5%, p = ns). Instead, the most frequently altered genomic regions in the non-GCB DLBCL subgroup in comparison to GCB DLBCL patients were gains of 18q12.2 (20% vs. 4%, p = ns) and 18q23 (20% vs. 4–7%, p = ns), and loss of 9p21.3 (30–35% vs. 7%, p = 0.026). Of the other genomic imbalances (<10% of all patients), only the gain of 18q21.2–33 in the non-GCB subgroup was significantly more frequent when compared to GCB subgroup (20% vs. 0%, p = 0.027).

Copy Number Associated Gene Expression Changes

To identify genes with altered expression due to large genomic aberrations, we combined the CNA and gene expression data obtained from 38 patients for whom both data sets were available. Our analysis showed that copy number gains and losses of 31 genes were associated with a simultaneous and significant increase or decrease in gene expression. The majority (n = 29) of the genes were over-expressed due to copy number gains in chromosomes 2 (2p15 and 2p16.1) and 18 (18q21.2, 18q21.31–33, 18q23). In contrast, two genes were suppressed and located in regions of copy number losses at 9p21.3. The aforementioned CNA areas and target genes are presented in more detail in Table S2 in File S1 and at http://csbi.ltdk.helsinki.fi/pub/lymphoma. As an example, the patients with 2p15 amplification had elevated COMMD1 expression with a p-value and fdr <0.001.

Prognostic Significance of Chromosomal Alterations

In the whole series of 51 patients with the aGCH data, we found two chromosomal regions with genomic alterations associated with PFS and lymphoma-specific OS. Patients with amplification in chromosome 2p15 (n = 6; 12% of all patients) had inferior PFS in comparison to patients without this gain (p = 0.010; Figure 1A). In addition a non-significant difference towards poor lymphoma-specific OS was observed (p = 0.131; Figure 1B). Similarly, patients with amplification in 18q12.2 (n = 6) had worse PFS (p = 0.044) and lymphoma-specific OS (p<0.001) than patients without this gain (Figures 1C and 1D). The survival associated gain in 2p15 contained the genes B3GNT2 (UDP-GlcNAc:betaGal beta-1,3-N-acetylglucosaminyltransferase 2), FAM161A (family with sequence similarity 161, member A), CCT4 (chaperonin containing TCP1 subunit 4), COMMD1 (copper metabolism (Murr1) domain containing 1), and XPO1 (exportin 1), and the amplification was associated with their over-expression (Figure 2). Association of COMMD1 and XPO1 over-expression with 2p15 amplification was further confirmed by qRT-PCR (Figure S1 in File S1). The amplification in 18q12.2 contained the CELF4 gene but we found no correlation between the gain and CELF4 gene expression (Figure 2).

Figure 1

Survival of DLBCL patients according to genomic aberrations.

PFS (A and C) and lymphoma-associated OS (B and D) rates according to indicated genomic aberration.

Figure 2

Expression of genes associated with amplifications in 2p15 and 18q12.2 locuses.

Boxes contain expression values between the 25th and 75th percentile in the tumour subgroup. The extremes denoted by asterisks represent maximum and minimum expression values.

Consistent with previous studies on lymphomas [27], [28], [31], we identified REL and BCL11A located at 2p16.1, and BCL2 at 18q21.3 being among the genes, whose expression was linked with copy number gains (Table S2 in File S1). CDKN2A and MTAP genes, which have also been described in lymphomas, specifically in the chemoresistant and ABC type DLBCLs [27], [28], [30], [32], [33]}, were in turn located in the regions of copy number losses at 9p21.3. However, these genomic alterations were not associated with survival in our study population.

To further identify biomarker candidates located in the survival associated 2p15 amplification locus, we performed survival analysis for five genes whose expression values correlated with amplification. Using gene expression values as continuous variables in Cox univariate analyses, only COMMD1 expression was identified to have prognostic impact on PFS (p = 0.037). When Kaplan-Meier analysis was performed, patients with high COMMD1 expression had significantly inferior PFS as compared to patients with low expression (5-year PFS 65% vs. 100%, p = 0.033; Figure 3A). Association of COMMD1 expression with the survival was further validated using qRT-PCR and Cox univariate analysis with continuous variables (p = 0.009) and Kaplan-Meier analysis with categorical data (p = 0.031, Figure 3B). In comparison, the expression of XPO1, another selected gene for qRT-PCR validation, was not significantly associated with survival (p = 0.345). Correlation coefficients between the expression arrays and qRT-PCR were 0.641 (p<0.001) for COMMD1 and 0.494 (p = 0.037) for XPO1.

Figure 3

PFS according to COMMD1 expression.

A. PFS according to exon array based COMMD1 expression values. B. PFS according to quantitative PCR analysis based COMMD1 expression values. In both A and B, the ideal cutoff values have been calculated using ROC curve analyses. In A the estimated area under the curve (AUC) was 0.717 (p = 0.063, 95% CI 0.531–0.903). In B the AUC was 0.759 (p = 0.062, 95% CI 0.468–1.000).

COMMD1 Protein Expression is Associated with Outcome

Considering that our multi-level analysis revealed COMMD1 to be amplified, over-expressed and survival associated gene in DLBCL, we extended COMMD1 analyses to the protein level. IHC stainings were performed on a TMA consisting of 70 lymphoma samples from the patients treated in the NLG-LBC-04 protocol (Table 3). Overall, intensity of COMMD1 positivity was highly variable (Figure 4A–B). COMMD1 immunoreactivity was primarily localized as perinuclear, granular, cytoplasmic pattern in lymphoma cells (Figure 4B), but also in endothelial cells and macrophages with more uniform cytoplasmic staining pattern.

Table 3

Baseline characteristics according to COMMD1 expression in TMA and validation cohorts.

		TMA cohort(n = 70)			Validation cohort (n = 146)
	Number of patients	COMMD1 low	COMMD1 high	p	COMMD1 low	COMMD1 high	p
	All n (%)	41 (100)	29 (100)		93 (100)	53 (100)
Gender	Female	13 (32)	11 (38)	0.618	34 (37)	26 (49)	0.163
	Male	28 (68)	18 (62)		59 (63)	27 (51)
Subtype	GCB	20 (49)	16 (55)	1.000	34 (37)	28 (53)	0.115
	Non-GCB	10 (24)	9 (31)		52 (56)	24 (45)
	FL grade 3B	4 (10)	1 (3)		2 (2)	1 (2)
	Other/unknown	7 (17)	3 (10)		5 (5)	0 (0)
Age	<60	27 (66)	22 (76)	0.434	40 (43)	22 (42)	1.000
	60–65	14 (34)	7 (24)		10 (11)	8 (15)
	>65	0 (0)	0 (0)		43 (46)	23 (43)
B-symptoms		24 (59)	13 (45)	0.333	23 (27)	13 (27)	1.000
Elevated LDH		38 (93)	29 (100)	0.261	57 (62)	41 (77)	0.067
Performance status	0–1	32 (78)	13 (45)	0.006	73 (79)	41 (77)	1.000
	2–4	9 (22)	16 (55)		20 (21)	12(23)
Stage	I-II	0 (0)	1 (3)	0.414	46 (50)	21 (40)	0.301
	III-IV	41 (100)	28 (97)		47 (50)	32 (60)
IPI	0–2	17 (41)	9 (31)	0.455	57 (65)	30 (59)	0.586
	3–5	24(59)	20 (69)		31 (35)	21 (41)

Figure 4

COMMD1 protein expression and outcome.

Representative examples of low (A) and high (B) expression levels of COMMD1 in FFPE DLBCL tissue (original magnifications 100×, and 400×). C–D. Outcome according to COMMD1 expression in the trial specific TMA cohort. PFS in in the whole TMA cohort (C) and in the GCB subgroup (D). E–F. Outcome according COMMD1 expression in the validation cohort. PFS in in the whole validation cohort (E) and in the GCB subgroup (F). The cutoff point (staining coverage of 8.9%) for survival outcomes (COMMD1 low vs high) was selected by the ROC curve analysis in the training set, and then applied also to validation cohort.

The prognostic significance of COMMD1 expression and correlation with mRNA data were assessed by computerized image analysis of COMMD1 positivity in the tumour tissue. In the univariate analysis the increasing COMMD1 positivity was an adverse prognostic factor for PFS (p = 0.003). The cutoff point for survival outcomes was selected by ROC curve analysis, resulting in a staining coverage of 8.9% being the most discriminative value (median 7.3%, range 0–24%), with an area under the curve (AUC) value of 0.663 (95% CI 0.516–0.810, p = 0.027). In Kaplan-Meier analyses, the patients with high COMMD1 expression had a significantly worse PFS and a trend towards adverse lymphoma associated OS in comparison to the remaining patients with lower COMMD1 expression (5-year PFS 47% vs. 79%, p = 0.005 (Figure 4C) and 5-year OS 75% vs. 90%, p = 0.081). According to COMMD1 expression, the relative risk of relapse was 3.2 (95% CI 1.361–7.584, p = 0.008) and death 2.825 (95% CI 0.837–9.536, p = 0.094). In multivariate analysis with aaIPI, COMMD1 expression retained its prognostic value on PFS (RR 2.996; CI 1.210–7.418, p = 0.018). When clinical characteristics of the patients were compared according to COMMD1 expression, no differences in gender, subtype, age, LDH level or stage were observed between the subgroups (Table 3). However, low COMMD1 expression was associated with low performance status.

We also examined prognostic impact on COMMD1 expression according to molecular subtype, and found that a significant adverse prognostic impact of COMMD1 expression was restricted to the GCB subgroup (Fig. 4D; p = 0.029). The relative risk of relapse according to COMMD1 expression within the GCB subgroup was 3.434 (95% CI 1.056–11.164, p = 0.040). Overall, immunohistochemically defined molecular subgroup was not associated with survival.

In contrast to the results from genomic and transcriptomic levels, no correlation was found between COMMD1 protein levels and CNA or gene expression data (r = 0.236). The observation suggests that post-transcriptional mechanisms may be involved in the regulation of COMMD1 protein levels in DLBCL.

COMMD1 Expression in an Independent DLBCL Series

In order to further validate the importance of COMMD1 in DLBCL, we analysed the prognostic significance of COMMD1 expression in a larger independent cohort of 146 DLBCL patients treated with chemoimmunotherapy (Table 3). Median age of the whole cohort was 63 years (range, 16–84 years), median follow-up 64 months (range 20–133 months), predicted 5-year PFS 74%, lymphoma-specific OS 79%, and OS 71%. While high IPI score was a strong predictor for survival (p<0.001), immunohistochemically defined molecular subgroup was not associated with outcome.

The clinical features of the patients according to COMMD1 expression (low versus high, cut-off defined according to TMA cohort) are summarized in Table 3. Accordingly, no differences were observed between COMMD1 low and high subgroups. In Kaplan-Meier analyses, PFS at five years for the patients with high COMMD1 expression was 64% compared with 79% for those with lower expression levels (p = 0.034; Figure 4E). When adjusted for IPI, COMMD1 expression maintained its prognostic effect on PFS (p = 0.023). In addition, when adjusted for age (<60 vs. ≥60), COMMD1 expression remained predictor for PFS (p = 0.035). According to COMMD1 expression, the relative risk or relapse was 1.9 (95% CI 1.040–3.606, p = 0.037). In multivariate analysis with IPI, COMMD1 retained its prognostic value on PFS (RR 2.0; 95% CI 1.037–3.730, p = 0.038).

Finally, when adjusted for molecular subgroups, COMMD1 expression was marginally predictive for PFS (p = 0.066). When COMMD1-related PFS was analysed separately for the patients in different molecular subtypes, there was a non-significant difference in PFS between COMMD1 high and low subgroups in the GCB DLBCLs (Fig. 4F; p = NS). Collectively, the results in our two independent cohorts provide evidence that COMMD1 is a novel survival-associated marker in DLBCLs.

Discussion

Although the addition of rituximab to chemotherapy has considerably improved the survival rates of DLBCL, the patients with high IPI scores still have a poor prognosis. Here we have studied genomic alterations and gene expression in freshly frozen lymphoma tissue collected prospectively from patients included in a Nordic phase II study for young high-risk DLBCL patients. With this comprehensive approach we have identified copy number gain at 2p15 driving COMMD1 mRNA upregulation with impact on survival. The gain was more frequently seen in the GCB than ABC DLBCLs. Furthermore, we have validated the results at the protein level by IHC in the same patient cohort as well as in an independent larger series. These data show that COMMD1 is a novel biomarker candidate that may be useful in improving risk stratification for DLBCL patients.

In a previous prospective study of poor prognosis DLBCL patients treated uniformly with dose-escalated CHOP followed by high-dose chemotherapy with autologous stem cell transplantation, the amplification of 2p16.1 was suggested to be associated with poor treatment response [34]. Candidate genes mapped within this amplification include proto-oncogenes REL and BCL11A, with increased expression levels [31], [35]. However, the functional correlation of this amplification event has not been illustrated [36]. Although REL and BCL11A expression levels were integrated with chromosomal amplification in our cohort, they were not associated with survival. Instead, we observed that the increased copy number in region 2p15 was associated with adverse PFS and lymphoma-specific OS. The region covers five genes, CCT4, XPO1, COMMD1, FAM161A and B3GNT2, which are all potentially important regulators of cellular growth. None of the genes, however, have previously been associated with DLBCL biology. While further investigation of the roles of these genes in DLBCL pathogenesis is needed, our results demonstrate that COMMD1 is a candidate genetic prognostic biomarker in DLBCL.

COMMD1 is a pleiotropic factor that participates in multiple processes, including copper metabolism, sodium excretion, inflammatory responses, and adaptation to hypoxia [37]. Recent mechanistic studies have revealed that COMMD1 suppresses NF-kB- and HIF-mediated gene expression [38], [39]. COMMD1 is underexpressed in some carcinomas, and low COMMD1 expression has been associated with inferior clinical outcome in patients with endometrial cancer [39]. In lymphomas, the prognostic role of COMMD1 has not previously been established. In the Oncomine database [40] the expression of COMMD1 is increased in lymphomas in comparison to other cancers [41]. Furthermore, two independent studies showed that COMMD1 expression is higher in DLBCLs than follicular lymphomas [41], [42], whereas no differences in the COMMD1 expression were observed between molecular subtypes of DLBCL [9]. At the present time, it remains unclear how COMMD1 is involved in a variety of seemingly unrelated and even opposite cellular activities. However, in most instances including lymphomas, the mechanism is likely via protein-protein interactions and ubiquitination [39], [43], [44].

To confirm whether COMMD1 expression could be useful to recognize DLBCLs with a more aggressive clinical course, we studied COMMD1 protein expression by IHC, which is a method that can be easily incorporated into a routine diagnostic approach. The predictive value of COMMD1 positivity was first defined in a training cohort of clinically high-risk DLBCL patients, and subsequently confirmed in an independent, larger and more heterogeneous DLBCL cohort. Thus, COMMD1 expression seems to represent a potential novel prognostic marker preferentially in the GCB type molecular subgroup.

Recently Monti and colleagues investigated gene expression and copy number data in 168 DLBCL patients, with the focus on the role of p53/cell cycle pathway in patient survival [28]. Interestingly, even though they reported that the region harboring COMMD1 is the second highest region in their cohort, and COMMD1 among the top genes in the region (Table1S in their publication [28]), they did not study its survival association. Additionally, our findings are supported by reports showing amplification of 2p15-p16 with the concordant elevated gene expression in DLBCL [28], [31], [34], [35]. These indicate that while integration of copy number and expression data is known to be a powerful approach to find driver genes, carefully selected, homogenous patient cohort together with integrative analysis can produce clear and important findings that may not be evident in more heterogeneous cohorts.

Since CNA status was integrated with gene expression data, only the genes, whose expression correlated with the CNAs were identified. The results from the qRT-PCR also correlated with the exon array data. However, we found no correlation between COMMD1 protein and gene expression levels. The reason for this difference is currently unknown but based on the literature indicating a strong regulatory role for the processes downstream of transcription [45], [46], it is plausible to speculate that post-transcriptional mechanisms may have a role in the regulation of COMMD1 protein levels in DLBCL. The work demonstrating that COMMD1 cellular levels are tightly controlled by ubiquitination [47] provides additional evidence that the regulatory level may be posttranslational. Together with the CNA data the results indicate that COMMD1 expression is regulated at multiple levels.

Amplification in 18q12.2 was found to be another significant CNA associated with inferior outcome. The gain has not been previously associated with survival in DLBCL, but its deletion has been reported to correlate with poor outcome in colorectal carcinoma. The CNA was found to harbour a single gene, CELF4/CUGBP, coding for a member of a family of RNA binding proteins playing an essential role in post-transcriptional gene regulation. However, despite the gain in 18q12.2 locus we were not able to demonstrate the over-expression of the CELF4 gene. While more work is needed to establish the exact role of the 18q12.2 gain in DLBCL, our data supports the prognostic importance of this region in DLBCL.

Consistent with previous studies [27], [30], [32], [33], expression of CDKN2A was associated with deletion of 9p21.3 in our patient cohort, and especially in the non-GCB subtype. However, overall incidence of the deletion was lower and no correlation with survival was found. Considering that aCGH and exon arrays were not performed on purified tumour samples it is possible that the presence of background material (tumour infiltrating non-malignant cells) could to some extent dilute tumour specific genetic alterations and explain a lower incidence of 9p21 deletions in our material. However, a more likely explanation for the differences in the results between different studies is that clinical and histopathological features of the study populations are not identical. The differences in the treatments may also contribute.

In conclusion, we have integrated copy number alteration and transcriptomic data in a carefully chosen high-risk DLBCL patient cohort to identify biological markers that could be used in risk stratification. We found two profiles with increased copy number of genes in chromosomes 2p15 and 18q12.2 that predicted a poor outcome for a subgroup of DLBCL patients. Furthermore, we identified a novel potential genetic driver event with prognostic significance. Notably, the prognostic impact of COMMD1 on survival was also observed at the protein level. The strengths of our study are a prospectively collected and homogenously treated study population, the availability of copy number, gene expression and IHC data from the same patients, the possibility to correlate the findings with clinical outcome, and validate the findings in an independent cohort of DLBCL patients. Our results demonstrate that it is possible to use relatively small but carefully designed prospective cohorts as a hypothesis generating material to identify a list of putative targets, and then validate and extend the major results to the protein level. Taken together, the results presented herein are promising and novel, and emphasize the importance of integrated genetic information and multilevel analyses for both the optimal use of existing combination therapies and the development of novel treatments for DLBCL.

Table S1, The most recurrent CNAs distributed between molecular subgroups. Table S2, Gene expression changes associated with CNAs. Figure S1, Association of COMMD1 (A) and XPO1 (B) overexpression with 2p15 amplification according to qRT-PCR analysis based expression values.

(PDF)

Click here for additional data file.