Diffuse large B-cell lymphoma (DLBCL) is the most common B-cell neoplasm throughout the world, representing 30–35% of all B-cell lymphomas. DLBCL is a biologically aggressive and heterogeneous disease, with a cure rate of approximately 60% [1
]. DLBCLs comprise different molecular entities including DLBCL not otherwise specified (DLBCL NOS) and primary DLBCL of the central nervous system (DLBCL CNS) [4
]. Primary mediastinal (thymic) large B-cell lymphoma (PMBL) is a closely related entity, which the World Health Organization (WHO) separates from other large B-cell lymphomas on the basis of its peculiar clinical, immunophenotypic and molecular features [5
]. DLBCL NOS has been classified according to gene expression profiling (GEP) into two distinct molecular cell-of origin (COO) subtypes: germinal center B-cell (GCB) and activated B-cell (ABC). About 10–15% of cases cannot be included in either of these subtypes and remain unclassified (UNCL) [6
]. The COO subtypes are associated with very different prognoses (worst for ABC-subtype). The microarray-based GEP technique used for COO determination was originally described using RNA extracted from frozen tissue. The limited access to frozen tissues and the long turnover time have, however, limited its application, and other techniques using formalin-fixed paraffin-embedded (FFPE) tissue have been developed during the last decade. These include immunohistochemistry (IHC)-based classifications [9
], quantitative reverse transcription PCR (qRT-PCR) [11
] and digital GEP (Lymph2Cx) assays [15
]. These methods, in particular, IHC-based classifications, have shown rather mixed results as prognostic tools [16
]. For clinical DLBCL risk stratification, additional quantitative methods are, thus, required to improve outcome and identify personalized therapeutic targets. One such approach relies on protein levels rather than mRNA transcripts from FFPE samples by using “state-of-the-art” custom forward phase protein arrays (FPPA) [17
]. Antibody microarrays are a new tool for the evaluation of protein abundance in a parallel and highly multiplex manner [18
]. Through FFPA technology our aim was to identify proteins implicated in DLBCL pathophysiology, correlating with clinical aggressiveness and possibly representing novel therapeutic targets.
2. Materials and Methods
2.1. Primary Tumor Specimens
We initially evaluated 47 patients with newly diagnosed DLBCL at the Veneto Institute of Oncology (IOV) from July 2011 to November 2016 with sufficient tissue for proteomics analysis. For validation studies, we evaluated another 49 DLBCL patients. Written informed consent was obtained from all subjects, and the analysis was approved by local ethics committee. All experiments conformed to the principles set out in the WMA Declaration of Helsinki. Histological diagnosis was reviewed according to the 2016 WHO classification of lymphoid neoplasms [5
]. We also included three cases of primary DLBCL CNS.
2.2. Forward Phase Protein Arrays
Antibody microarrays were produced and used according to protocols and strict quality control procedures, as reported earlier [17
]. For the analysis, a set of 82 target proteins that represent the translational products of transcripts implicated in DLBCL pathobiology, such as those defining COO, oncogenes and drug targetable targets [11
]. Antibodies targeting 82 unique proteins were purchased from different sources or provided by collaborating partners. A complete list of binders is provided in the Supplementary Materials (Supplementary Table S1)
. In this pilot study, FFPE tissue material from 47 patients with DLBCL (including three cases of primary DLBCL CNS) and 3 reactive lymph nodes (LNF) were obtained. The samples were labelled at an adjusted protein concentration with scioDye1 and scioDye2 and washed and hybridised to antibody microarrays in a dual-colour approach using a reference-based design (Sciomics). For competitive dual-colour incubations, a reference sample was produced by pooling the same amount of all protein samples. The same reference sample was used throughout the analysis. After 3 h incubation, slides were washed and subsequently dried with nitrogen before being scanned using a Powerscanner (Tecan, Austria). Differences in protein abundance between samples or sample groups were represented as log-fold changes calculated for the base 2.
2.3. Histological Evaluation and Immunohistochemical Analysis
All cases (discovery and validation cohort) were retrieved from the archives of the Surgical Pathology & Cytopathology Unit of Padua University Hospital (Padua, Italy). Each case was re-evaluated and assigned to a COO subtype, according to the Hans algorithm (immunostain for CD10, Bcl6 and MUM1). Representative histological samples were selected for further phenotypic characterization (i.e., assessment of PIM2 and ETV6 expression). In detail, IHC analysis was performed on 4 μm-thick FFPE sections with the Bond Polymer Refine Detection kit in an automated immunostainer (BOND-MAX system; Leica Biosystems—Newcastle upon Tyne, UK), as previously described [21
]. Immunostains were performed on whole tissue sections. Where necessary, tissue microarrays were prepared as described previously [22
]. TMA blocks were prepared using the Galileo TMA CK3500 (Integrated System Engineering, Milan, Italy; Padova University Hospital) arrayers. Appropriate positive and negative controls were also included. Phenotypic studies on benign palatine tonsils with reactive lymphoid hyperplasia (n
= 5) were run in parallel to assess ETV6 and PIM2 expression in normal B-cell subsets. We defined the following four-tiered scoring system for ETV6 and PIM2 expression: (i) score 0: no staining or weak positivity in <20% of tumor cells; (ii) score 1+: weak positivity in ≥20% of tumor cells; (iii) score 2+: moderate positivity in ≥20% of tumor cells; (iv) score 3+: strong positivity in ≥20% of tumor cells. The scoring system was based on the nuclear expression of both markers, and intensity scores were defined by comparison with positive controls (i.e., squamous epithelium of palatine tonsils). Specifically, strong (score 3+) positivity was attributed to DLBCL cases with protein expression comparable to that of squamous epithelia of palatine tonsils, moderate (score 2+) positivity to cases with protein expression slightly fainter than controls and weak (score 1+) positivity to cases with barely detectable protein expression. To allow comparison among groups, DLBCL cases were lumped together based on ETV6 and PIM2 positivity scores as follows: (i) low expressing cases (immunohistochemical score 0 and 1+); (ii) high expressing cases (immunohistochemical score 2+ and 3+).
The following primary antibodies were used: anti-CD10 (clone 56C6, Menarini Diagnostics, Florence, Italy); anti-Bcl6 (clone LN22, Leica Biosystems, Milan, Italy); anti-MUM1 (clone MUM1p, Dako, Glostrup, Denmark); anti-PIM2 (clone D-8, Santa Cruz Biotechnology, Dallas, TX, USA); and anti-ETV6 (HPA000264, Sigma-Aldrich, St. Louis, MO, USA).
2.4. Bioinformatical Software and Analyses
Gene Set Enrichment Analysis (GSEA) using WEB-based GEne SeT AnaLysis Toolkit (WebGestalt [23
]) with Wikipedia cancer pathway as enrichment categories was performed on all proteins significantly associated with OS (n
= 41). Network Topology-based analysis (NTA; another module present in WebGestalt) for 41 prognostic proteins using the TCGA RNA Seq data for DLBCL samples (n
= 48) as functional database was used to identify relevant connecting sub-networks. GeneMANIA [24
] is a web interface that uses large sets of functional association data to identify single genes related to a set of input genes. Association data include protein and genetic interaction pathways, co-expression, co-localization and protein domain homology. GeneMANIA was used to contruct the ETV6, PIM2 and ETV6-PIM2 biological network. The list of identified co-expressed/interacting genes of ETV6 was used to run an over-representation analysis (ORA) using Reactome pathways as functional database to identify if they associate into certain pathways.
2.5. Gene Expression Data
Gene expression data of DLBCL patients analyzed with HGU133+2.0 Affymetrix GeneChip arrays (n
= 223) was obtained from Gene Expression Omnibus (GSE873371) [25
]. PIM2 and ETV6 expression levels were extracted and used to generate Kaplan–Meier survival plots. Gene expression changes in the B-cell lymphoma cohort [26
] were obtained from cBioPortal (a tool developed by the Computational Biology Center at Sloan Kettering) [27
], and the results were presented as OncoPrint format data and used to generate Kaplan–Meier survival plots. Gene expression data for normal and malignant B-cells (DLBCL) were extracted from GSE56315 [29
]. Genetic alterations and gene expression levels associated with selected genes in B-lymphoma cell lines [30
] were also obtained by using cBioPortal.
2.6. Western Blotting
For Western blotting, protein samples were separated on 4–12% gradient Tris-Glycine or 12% Tris-Glycine SDS-PAGE Gels (Invitrogen, Thermo Fisher Scientific, Waltham, MA, USA) and transferred to PVDF membrane (Millipore, Billerica, MA, USA). Antibodies against tubulin (TU-02; Santa Cruz Biotechnology), PIM2 (MAB4355, R&D Systems, Minneapolis, MN, USA or HPA000285, Sigma-Aldrich), ETV6 (HPA000264, Sigma-Aldrich), β-actin (#4970; Cell Signaling Technologies, Danvers, MA, USA), XIAP (#14334; Cell Signaling Technologies), cleaved PARP-1 (#5625; Cell Signaling Technologies) and survivin/BIRC5 (#2808; Cell Signaling Technologies) were used. The BioRad ChemiDoc XRS Imager was used to capture signals from blots. We quantified each protein band using ImageJ software (National Institutes of Health, Bethesda, MD, USA) and normalized each target protein after background subtraction to its loading control (β-actin or tubulin).
2.7. Lentiviral Constructs and Viral Production
Human ETV6 knock-down (KD) was performed using pLKO.1-shETV6-puro constructs (#1: TRCN0000003854; #2: TRCN0000003856; Sigma-Aldrich). pLKO.1-control (SHC007, SHC002; Sigma-Aldrich) were used as controls (CTRL). For viral production, appropriate expression plasmids were transfected in HEK293T cells using JetPEI transfection reagent (Polyplus, Illkirch, France) together with packaging plasmids. The viral supernatant was collected 48 h after transfection, filtered and used to infect target cells. All infections were performed by spinoculation. After infection, DLBCL cell lines were selected for 5–7 days in puromycin before functional assays
2.8. Cell Viability Assays and Flow Cytometry
Cell viability in DLBCL cell lines treated with different concentrations of YM155 (Selleck Chemicals LLC, Houston, TX, USA) was analyzed after 48 h via the bioluminescent method Vialight plus (Lonza, Basel, Switzerland). This assay allows bioluminescent detection of cellular ATP as a measure of viability. We analyzed apoptosis 48 h post completion of puromycin selection by flow cytometry (FACS) after staining with Annexin-V-FLUOS Staining Kit (Roche, Basel, Switzerland) and SYTOX Red dead cell stain (Thermo Fischer Scientific). Apoptosis was defined as the sum of the percentage of Annexin V+ and Annexin V+/SYTOX Red+ cells. The samples were collected on a FACS Calibur (BD Biosciences, Milan, Italy) using Cell Quest software (BD Biosciences) and analysed with FlowJo™ Software (FlowJo LLC, Ashland, OR, USA).
2.9. Statistical Analyses
We performed statistical analysis by Student’s t-test and Mann–Whitney U test where appropriate. A non-parametric test (Chi-square test) was used to compare qualitative data, including clinical variables presented in Table 1
. All statistical tests were two sided and unpaired and p
< 0.05 was considered statistically significant. Linear regression analyses and Pearson’s correlation coefficients were conducted for calculating correlations (GraphPad Prism Software, San Diego, CA, USA). Statistical analysis of protein expression data to identify clinically relevant prognostic proteins was also performed using the Cox regression analysis (MedCalc Software bv, Ostend, Belgium; https://www.medcalc.org
; accessed on 9 November 2018). The Kaplan–Meier method was used to estimate the distributions of OS. OS was considered as the time from diagnosis to date of death or last follow-up. The log-rank test or Gehan–Breslow–Wilcoxon test was used to compare survival distributions.
Methods concerning cell lines, array production, sample labeling and hybridization, array data analysis and statistical testing, Nanostring Assay for COO, Western blotting and quantitative real time RT-PCR are detailed in Supplementary Materials and Methods
DLBCL is the most common type of B-cell lymphoma in the Western world with cure rates of approximately 60% by modern immune chemotherapy (R-CHOP). The remaining 40% of patients experience refractory/relapsing disease and usually succumb to the disease [33
]. The marked heterogeneity of DLBCL prognosis represents a continuous challenge to physicians and requires the identification of better treatments and outcome predictors either before or shortly after treatment initiation. In particular, it is imperative to identify poor-risk patients in order to offer them more effective therapies. Initial molecular GEP studies revealed that histologically uniform lymphoma subtypes are prognostically and molecularly heterogeneous and identified two biologically distinct groups on the basis of their COO: GCB-like and ABC-like subtypes [6
]. An additional small subgroup could not be classified into these entities (UNCL) [47
]. However, the failure of numerous clinical trials of targeted therapies selecting patients using COO implies that this classification, although highly useful, lacks sufficient granularity to serve any prognostic-therapeutic purpose per se. We tried to address this issue by a different perspective. The aim of our study was to identify a small group of proteins for which its expression predicts survival in patients with DLBCL and that can be readily measured using standard FFPE tissue. Univariate Cox analysis identified many proteins significantly associated with survival, including immune check-point molecules (PDCD1, PDCD2 and PD1L2), components of the PI3K-AKT-mTOR signaling pathway (PTEN, p85 regulatory subunit α), Toll-like receptor signaling (MYD88), JAK-STAT signaling (JAK2), BCR signaling (BLNK) and BCL2, suggesting that these may represent useful therapeutic targets in poor risk patients. However, multivariate Cox analysis disclosed that only PIM2 and ETV6 were independent prognostic factors in our cohort. PIM2 transcript is present in the ABC-signature and is linked to B-cell survival pathways, such as those involving cytokines (IL6, IL10 and IL13) and CD40, NFkB and p53 signaling [48
] (see Figure 2
). PIM kinases (especially PIM2) have already been proposed as therapeutic targets in DLBCL, especially in ABC-DLBCL cases with aggressive behavior after R-CHOP treatment [48
]. Unlike these reports, we found that lower levels of PIM2 protein may be associated with reduced OS. This discrepancy may be due to differences in detection methods and cutoff definitions.
What was rather unexpected were also findings regarding ETV6, which is an ETS family transcriptional factor with a crucial role in hematopoiesis and embryonic development [50
]. While ETV6
is frequently rearranged or fused with other genes in human myeloid and lymphoid leukemias [51
], it is only rarely altered in B-cell lymphoma [53
]. Recently, however, whole-exome sequencing studies have found a significant fraction of DLBCL samples (mainly of the ABC-subtype) harboring ETV6
]. More precisely, alterations of ETV6
are considered part of the MCD [36
], C5 [35
] and MYD88 [34
] genetic subgroups associated with poor prognosis. Generally, ETV6
is inactivated early during leukemogenesis and is considered a tumor suppressor gene [38
]. However, most of the mutations found in DLBCL have not been functionally classified (Figure S5D
), and ETV6
alterations do not generally impact ETV6 expression levels [56
]. Furthermore, integrated bioinformatics analysis identified ETV6
as one of the hub genes associated with the two DLBCL subtypes [32
]. Thus, the functional role and transcriptional pathways downstream of ETV6 in DLBCL are currently unknown. We found that high ETV6 protein (and transcript) levels are associated with poor survival in B-cell malignancies, especially DLBCLs. Furthermore, from our initial in silico analysis it seems that ETV6 may play a stronger prognostic role in DLBCL NOS compared to PMBL.
ETV6 KD experiments in ABC-type, GCB-type and BL cell lines consistently determined loss of viability, suggesting that acute depletion of ETV6 is highly cytotoxic. Although we do not elucidate the apoptotic program elicited following abrupt ETV6 loss, this event was often associated with BIRC5 protein depletion. Of note, ETV6 depleted cells that persisted in vitro showed a paradoxical increase in BIRC5 expression (data not shown), suggesting that only abrupt ETV6 loss is cytotoxic. Studying the relationship between ETV6 and BIRC5 in established high-grade B-cell lymphoma cell lines revealed an inverse relationship between these two proteins. Interestingly, we found numerous putative ETV6 binding sites (GGAA/T) in the BIRC5
promoter region, suggesting that ETV6 may regulate its expression. Consistently, cell lines exhibiting high ETV6 protein levels were found to be more resistant to the BIRC5 inhibitor, YM155. The small molecule YM155, although seemingly well tolerated, as a single-agent has demonstrated only limited activity in refractory DLBCL patients in a phase II clinical trial [57
]. More encouraging results, however, have been obtained by the combination with rituximab or bendamustine in preclinical models [58
]. Our study suggests that evaluation of ETV6 (possibly in combination with PIM2 to determine the ETV6/PIM2 ratio) may serve two purposes, as high ETV6 levels may correlate with poor prognosis, while low ETV6 levels may identify DLBCL patients, who could benefit from a different therapeutic intervention (e.g., inclusion of BIRC5 inhibitors, such as YM155).
Our study has limitations that may impact the obtained results and that need to be taken into account. These include its retrospective nature, the small sample size (n
= 47), its histological and therapeutic heterogeneity and problems related to COO stratification in a subset of samples. Thus, further clinical investigation of the prognostic significance of ETV6 expression needs to be performed in a larger prospective series of uniformly treated DLBCL patients, ideally in combination with targeted sequencing analyses [34
]. Furthermore, the tolerability and therapeutic potential of combination regimens incorporating YM155 in selected patients awaits testing.