Search

Aggressive B-cell non-Hodgkin lymphomas are haematological malignancies that account for significant morbidity and mortality worldwide. They encompass a wide range of histological and clinical features: Diffuse large B-cell lymphoma (DLBCL) represents the most frequent subtype and Burkitt lymphoma (BL) occurs less frequently. Recent advances in molecular profiling have demonstrated considerable heterogeneity at the molecular level, and further classified DLBCL into germinal centre B-cell-like (GCB), activated B-cell-like (ABC), primary mediastinal B-cell lymphoma (PMBL) and type III subgroups, where ABC associates with the most unfavourable prognosis. Gene-expression profiling (GEP) also confirmed a subgroup with features intermediate between BL and DLBCL, and these cases particularly that have concurrent chromosomal rearrangements of MYC and BCL2are often associated with poor prognosis. Despite the progress made from GEP, the classification still has limited influence on clinical treatment decision-making. In fact, as more heterogeneity beyond the subtypes above has been discovered by recent next-generation sequencing studies the approach of dividing cases into limited subgroups makes less sense in clinical practice. It is clear that the subtypes overlap to an extent in the affected signalling and regulatory pathways, and that small groupings within subtypes exhibit clear mechanistic differences and treatment responses. We describe an alternative approach, where a large database of aggressive B-cell lymphomas is used in a similarity search to identify those cases most similar at a molecular level to a query case. The hypothesis is that the most similar cases provide the best guide to prognosis and treatment outcome in the query case, independent of any need to place the query case into a particular subtype. We used both large public datasets and data from our Haematological Malignancy Research Network (www.HMRN.org) to explore genes associated with pathogenic pathways and an unfavourable prognosis. We also defined similarity between cases according to their molecular features and treatment responses. We then trained the similarity search method, by employing a distance metric learning approach that has been successfully used in similar machine learning applications, to test our HMRN dataset which contains detailed clinical data with treatment and outcome information. The cross-validation result on the public dataset achieved nearly 90% accuracy in recognizing cases with similar overall survival, and initial test results on HMRN data also shows that over 80% cases can be correctly represented by similar cases. In summary, we present a similarity learning method as an alternative to the current sub-type classification method. This mathematical method is able to accurately recognise cases with similar molecular features and provides important information on predicted treatment response for any given query case. Disclosures Smith: Novartis: Research Funding; Celgene: Research Funding; Jansen Cilag: Research Funding; Amgen: Research Funding.

This study tested the validity of whole-genome expression profiling (GEP) using RNA from formalin-fixed, paraffin-embedded (FFPE) tissue to sub-classify Diffuse Large B-cell Lymphoma (DLBCL), in a population based cohort of 172 patients. GEP was performed using Illumina Whole Genome cDNA-mediated Annealing, Selection, extension & Ligation, and tumours were classified into germinal centre (GCB), activated B-cell (ABC) and Type-III subtypes. The method was highly reproducible and reliably classified cell lines of known phenotype. GCB and ABC subtypes were each characterized by unique gene expression signatures consistent with previously published data. A significant relationship between subtype and survival was observed, with ABC having the worst clinical outcome and in a multivariate survival model only age and GEP class remained significant. This effect was not seen when tumours were classified by immunohistochemistry. There was a significant association between age and subtype (mean ages ABC - 72·8 years, GC - 68·4 years, Type-III - 64·5 years). Older patients with ABC subtype were also over-represented in patients who died soon after diagnosis. The relationship between prognosis and subtype improved when only patients assigned to the three categories with the highest level of confidence were analysed. This study demonstrates that GEP-based classification of DLBCL can be applied to RNA extracted from routine FFPE samples and has potential for use in stratified medicine trials and clinical practice.

Purpose To assess whether comparative gene network analysis can reveal characteristic immune response signatures that predict clinical response in Diffuse large B-cell lymphoma (DLBCL). Background The wealth of available gene expression data sets for DLBCL and other cancer types provides a resource to define recurrent pathological processes at the level of gene expression and gene correlation neighbourhoods. This is of particular relevance in the context of cancer immune responses, where convergence onto common patterns may drive shared gene expression profiles. Where existing and novel immunotherapies harness the immune response for therapeutic benefit such responses may provide predictive biomarkers. Methods We independently analysed publically available DLBCL gene expression data sets and a wide compendium of gene expression data from diverse cancer types, and then asked whether common elements of cancer host response could be identified from resulting networks. Using 10 DLBCL gene expression data sets, encompassing 2030 cases, we established pairwise gene correlation matrices per data set, which were merged to generate median correlations of gene pairs across all data sets. Gene network analysis and unsupervised clustering was then applied to define global representations of DLBCL gene expression neighbourhoods. In parallel a diverse range of solid and lymphoid malignancies including; breast, colorectal, oesophageal, head and neck, non-small cell lung, prostate, pancreatic cancer, Hodgkin lymphoma, Follicular lymphoma and DLBCL were independently analysed using an orthogonal weighted gene correlation network analysis of gene expression data sets from which correlated modules across diverse cancer types were identified. The biology of resulting gene neighbourhoods was assessed by signature and ontology enrichment, and the overlap between gene correlation neighbourhoods and WGCNA derived modules associated with immune/host responses was analysed. Results Amongst DLBCL data, we identified distinct gene correlation neighbourhoods associated with the immune response. These included both elements of IFN-polarised responses, core T-cell, and cytotoxic signatures as well as distinct macrophage responses. Neighbourhoods linked to macrophages separated CD163 from CD68 and CD14. In the WGCNA analysis of diverse cancer types clusters corresponding to these immune response neighbourhoods were independently identified including a highly similar cluster related to CD163. The overlapping CD163 clusters in both analyses linked to diverse Fc-Receptors, complement pathway components and patterns of scavenger receptors potentially linked to alternative macrophage activation. The relationship between the CD163 macrophage gene expression cluster and outcome was tested in DLBCL data sets, identifying a poor response in CD163 -cluster high patients, which reached statistical significance in one data set (GSE10846). Notably, the effect of the CD163-associated gene neighbourhood which correlates with poor outcome post rituximab containing immunochemotherapy is distinct from the effect of IFNG-STAT1-IRF1 polarised cytotoxic responses. The latter represents the predominant immune response pattern separating cell of origin unclassifiable (Type-III) DLBCL from either ABC or GCB DLBCL subsets, and is associated with a trend toward positive outcome. Conclusion Comparative gene expression network analysis identifies common immune response signatures shared between DLBCL and other cancer types. Gene expression clusters linked to CD163 macrophage responses and IFNG-STAT1-IRF1 polarised cytotoxic responses are common patterns with apparent divergent outcome association. Disclosures Davies: CTI: Honoraria; GIlead: Consultancy, Honoraria, Research Funding; Mundipharma: Honoraria, Research Funding; Bayer: Research Funding; Takeda: Honoraria, Research Funding; Janssen: Honoraria, Research Funding; Roche: Honoraria, Research Funding; GSK: Research Funding; Pfizer: Honoraria; Celgene: Honoraria, Research Funding. Jack:Jannsen: Research Funding.

Introduction: DLBCL subtypes may be classified by gene expression corresponding to germinal centre (GCB) or activated peripheral blood (ABC) B-cells. Treatment outcomes with R-CHOP therapy were inferior for ABCs in retrospective series, and this study investigated whether adding bortezomib could reverse the adverse prognosis. The trial used gene expression profiling (GEP) to stratify cases, with adaptive design to analyse the outcome by subtypes at predefined timepoints. Methods: Newly diagnosed patients with DLBCL underwent staging and commenced standard R-CHOP. During cycle 1, formalin-fixed paraffin-embedded (FFPE) tissue was used to extract messenger RNA for GEP using the Illumina DASL array platform. Cases were allocated to GCB, ABC or Unclassifiable (Unc) type before starting cycle 2, using an established algorithm based upon 20 genes. Patients with successful GEP were randomised 1:1 to receive R-CHOP +/- bortezomib 1.6 mg/m2 s/c on days 1+8 in cycles 2-6. The study was powered to detect a difference in progression-free survival (PFS) of 10% with bortezomib, with a 2-sided significance, 5% and 90% power. The adaptive design allowed for closure of randomization for GCB cases if 1-year PFS was Results: Between 6/2011 and 5/2015 1132 patients were enrolled from 109 sites, with 1078 samples analysed. Of these, 157 (15%) biopsies had inadequate material for GEP, but the remaining 921 were classified as 246 (27%) ABC, 476 (52%) GCB and 199 (22%) Unc. Successful classification was possible from both surgical and needle core biopsies. Median laboratory turnaround time was 12 working days and all results were available prior to the scheduled administration of cycle 2. Characteristics of the patients of different subtypes are shown in the table. Following both interim analyses the DMEC recommended continued recruitment of patients with a GCB phenotype. Table. ABC GCB Unc Age (years): median 67 63 63 Age (years) : range 23 to 86 20 to 82 20 to 85 % performance status 0-1 88 88 90 % at least one extranodal site 53 54 62 % bone marrow involved 15 14 23 % LDH>ULN 69 76 79 % IPI score 0/1 29 27 26 % IPI score 2/3 57 55 55 % IPI score 4/5 15 19 19 % B symptoms 46 43 49 % Bulk>10cm 17 26 21 Conclusions: This study has demonstrated the feasibility of GEP at diagnosis to subsequently guide therapy in a large multicentre trial. Although patients with ABC type lymphoma were in general slightly older, they did not appear to have other adverse prognostic features at diagnosis vs GCB. All patients will have completed therapy by the time of the meeting, allowing the initial response and toxicity data to be available for presentation. Disclosures Davies: GIlead: Consultancy, Honoraria, Research Funding; Mundipharma: Honoraria, Research Funding; CTI: Honoraria; Takeda: Honoraria, Research Funding; Bayer: Research Funding; GSK: Research Funding; Janssen: Honoraria, Research Funding; Roche: Honoraria, Research Funding; Pfizer: Honoraria; Celgene: Honoraria, Research Funding. Off Label Use: The addition of bortezomib to R-CHOP chemotherapy in diffuse large B-cell lymphoma. Pocock:Janssen: Honoraria. Jack:Jannsen: Research Funding. Johnson:Takeda: Honoraria; Pfizer: Honoraria; Janssen: Research Funding.

Germinal Center (GCB) and activated (ABC) B-cell subtypes of DLBCL can be identified by Gene expression profiling (GEP). These subgroups are biologically distinct, harboring mutations in different pathways. Patients classified as ABC more often have mutations of the NF-kB pathway and an inferior response to standard R-CHOP therapy. The REMoDL-B trial utilised GEP to stratify patients for the addition of bortezomib to R-CHOP, based on the hypothesis that this agent may selectively improve the outcome of the ABC subtype. GEP was performed on RNA extracted from diagnostic formalin-fixed paraffin-embedded (FFPE) biopsies using Illumina WG-DASLTM during the 1st cycle of R-CHOP. Patients were classified as GCB, ABC or Unclassified before cycle 2 using the cross-platform DAC classifier (Care et al, PLOS ONE 2013) and randomised to continue R-CHOP+/-bortezomib. Work is underway to compare data generated on Affymetrix arrays and targeted RNA-seq (Illumina TRex), as well as validation by targeted mutational analysis of 18 genes associated with DLBCL (TNFAIP3, CARD11, CD79A, CD79B, MYD88, TRAF3, TNFRSF11A, PRDM1, TP53, FAS, B2M, CD58, EZH2, MLL2, MEF2B, EP300, CREBBP, KDM2B) using Fluidigm multiplex PCR and Illumina MiSeq on DNA also from the FFPE blocks. The trial closed to recruitment in May 2015 and 1147 samples have been analysed. One hundred and fifty three (13%) biopsies were unsuitable for GEP (for insufficient tumor tissue, inappropriate block sent). The remaining samples were classified as ABC (n=261, 23%), GCB (n=471, 44%) and Unclassified (n=214, 19%), with only 11 samples (1%) failing to yield a GEP result. GEP was successful in a range of sample types, including needle and endoscopic biopsies, bone marrow trephines and formal biopsies, with results obtained from as little as 40ng of total RNA, all from FFPE samples. Mutational data were available in 199 samples, with 73% of these having a mutation detectable in 1 or more genes (range 0-5) at a AAF (alternative allele frequency) cutoff at 10%. MYD88 was most commonly mutated (in 30% of ABC and 7% of GCB). EZH2 mutations were restricted to the GCB category (26%) and MYD88, CD79a/b and PRDM1 were more commonly associated with the ABC group. MYD88/PRDM1 were the most frequently associated events, with MYD88/CD79a/b and MYD88/NF-kB being mutually exclusive. Where MYD88 was seen in GCB cases, coexisting mutations imply an origin from transformed follicular lymphoma. B2M mutations were commonly identified across all subtypes (n=26), but specifically enriched in Type III (unclassified) cases (25%), which supports the hypothesis that mutational immune escape may be a feature of DLBCL, in common with other tumor types. Cross platform validation is highly concordant using Affymetrix arrays from a pilot series (27/27 gave the same classifier output, with correlating confidences also seen between platforms). RNA-seq analysis is ongoing, however initial analysis shows 86% concordance with the DASL output. Comprehensive cross platform comparison data will be available for presentation at the meeting. This study demonstrates the feasibility of GEP classification of DLBCL at diagnosis in a large international trial. The molecular classification can also be replicated using different technologies. Mutational analysis confirmed the association between DLBCL subtype and specific mutational hotspots. Disclosures Sharon: Johnson & Johnson: Other: Funded the laboratory work for the REMoDL-B trial (ISRCTN 51837425). Davies:Takeda: Honoraria; Seattle Genetics: Research Funding. Jack:Jannsen: Research Funding. Johnson:Johnson & Johnson: Other: Funded the laboratory work for the REMoDL-B trial (ISRCTN 51837425).

The transcriptional repressor B lymphocyte-induced maturation protein-1 (BLIMP1) regulates gene expression and cell fate. The DNA motif bound by BLIMP1 in vitro overlaps with that of interferon regulatory factors (IRFs), which respond to inflammatory/immune signals. At such sites, BLIMP1 and IRFs can antagonistically regulate promoter activity. In vitro motif selection predicts that only a subset of BLIMP1 or IRF sites is subject to antagonistic regulation, but the extent to which antagonism occurs is unknown, since an unbiased assessment of BLIMP1 occupancy in vivo is lacking. To address this, we identified an extended set of promoters occupied by BLIMP1. Motif discovery and enrichment analysis demonstrate that multiple motif variants are required to capture BLIMP1 binding specificity. These are differentially associated with CpG content, leading to the observation that BLIMP1 DNA-binding is methylation sensitive. In occupied promoters, only a subset of BLIMP1 motifs overlap with IRF motifs. Conversely, a distinct subset of IRF motifs is not enriched amongst occupied promoters. Genes linked to occupied promoters containing overlapping BLIMP1/IRF motifs (e.g. AIM2, SP110, BTN3A3) are shown to constitute a dynamic target set which is preferentially activated by BLIMP1 knock-down. These data confirm and extend the competitive model of BLIMP1 and IRF interaction.

Cell of origin classification of diffuse large B-cell lymphoma (DLBCL) identifies subsets with biological and clinical significance. Despite the established nature of the classification existing studies display variability in classifier implementation, and a comparative analysis across multiple data sets is lacking. Here we describe the validation of a cell of origin classifier for DLBCL, based on balanced voting between 4 machine-learning tools: the DLBCL automatic classifier (DAC). This shows superior survival separation for assigned Activated B-cell (ABC) and Germinal Center B-cell (GCB) DLBCL classes relative to a range of other classifiers. DAC is effective on data derived from multiple microarray platforms and formalin fixed paraffin embedded samples and is parsimonious, using 20 classifier genes. We use DAC to perform a comparative analysis of gene expression in 10 data sets (2030 cases). We generate ranked meta-profiles of genes showing consistent class-association using ≥6 data sets as a cut-off: ABC (414 genes) and GCB (415 genes). The transcription factor ZBTB32 emerges as the most consistent and differentially expressed gene in ABC-DLBCL while other transcription factors such as ARID3A, BATF, and TCF4 are also amongst the 24 genes associated with this class in all datasets. Analysis of enrichment of 12323 gene signatures against meta-profiles and all data sets individually confirms consistent associations with signatures of molecular pathways, chromosomal cytobands, and transcription factor binding sites. We provide DAC as an open access Windows application, and the accompanying meta-analyses as a resource.

Abstract 2484 Diffuse large B cell lymphoma (DLBCL) is a heterogenous disease, which has been subclassified into germinal centre (GCB) and activated B-cell (ABC) type using gene expression profiling. This has been shown to separate DLBCL into distinct prognostic sub-groups in patients treated with either CHOP or CHOP-R therapy. A number of published immunohistochemistry algorithms have attempted to replicate this subclassification using a limited number of markers, however, despite being widely adopted, the reproducibility of the algorithms has proven difficult, possibly due to the subjective nature of interpreting immunohistochemistry results. The aim of this study was therefore to evaluate the most widely accepted immunohistochemistry algorithms, and validate the results using RQ-PCR on RNA extracted from paraffin sections in a large series of well characterised formalin fixed paraffin embedded (FFPE) biopsies of R-CHOP treated DLBCLs. RNA was extracted using the Ambion Recoverall extraction kit. Applied Biosytems Taqman probes were used to evaluate gene expression of CD10, BCL6, GCET1, FOXP1 and IRF4. RQ-PCR was run on an Applied Biosystems 7500Fast cycler, and results were calculated using the deltadeltaCt method, using PGK1 as the reference housekeeper gene and either RAJI cell line or commercial RNA as the standard. Using the Hans criteria, 130/277 (47%) presentation DLBCL biopsies were classified as GCB and 147/277 (53%) were ABC. Further classification of a subset of cases using the Choi algorithm showed concordant results in 48/61 (78.7%) cases, with 1 (1.6%) case classified as GCB using Hans and ABC using Choi, and 12 (19.7%) cases classified as ABC using Hans and GCB using Choi. RQ-PCR data showed excellent correlation with immunohistochemistry for all genes incorporated into the algorithms (CD10, p In univariate Kaplan-Meier survival analysis, there was no difference in outcome when comparing either Hans (n=170, p=0.7) or Choi (n=60, p=0.3) algorithms using immunohistochemistry, however using RQ-PCR data to define GCB and ABC subgroups, both algorithms showed a poorer survival in the ABC subgroup. For the Hans algorithm, 6 month overall survival (OS) was 76% in the GCB group compared to 55% in patients classified as ABC (p=0.06), and similarly for the Choi algorithm, 6 month OS was 77% in the GCB group and 53% for patients classified as ABC (p=0.03). This data supports the use of gene expression algorithms to classify DLBCL patients into clinically relevant prognostic groups, with patients classified as ABC exhibiting an inferior outcome compared to the GCB group. RQ-PCR provides a quantitative method to determine expression and eliminates the subjective element associated with interpreting immunohistochemistry. Using RQ-PCR to define prognostic subgroups in DLBCL provides a realistic alternative to gene expression profiling, which is currently not applicable to the majority of diagnostic laboratories. Patients classified as ABC type using this approach show early treatment failure with CHOP-R, and alternative therapies should be considered in this group. Disclosures: No relevant conflicts of interest to declare.

Abstract 2485 Diffuse large B cell lymphoma (DLBCL) is a heterogenous disease, which has been subclassified into germinal centre (GCB) and activated B-cell (ABC) type using gene expression profiling. This has been shown to separate DLBCL into distinct prognostic sub-groups in patients treated with either CHOP or CHOP-R therapy. Previous studies have required the use of fresh or frozen samples for the extraction of RNA of sufficient quality to permit whole genome expression analysis. The Illumina ‘DASL' platform allows for highly reproducible gene expression data to be generated from FFPE material, which opens up large series' of retrospective data for detailed expression studies. The aim of this study was therefore to determine whether the Illumina DASL platform could yield reproducible results on formalin fixed paraffin embedded (FFPE) biopsies from a large series of archival CHOP-R treated DLBCL samples. RNA was extracted from paraffin sections using the Ambion Recoverall extraction kit, with 179/206 (87%) of cases yielding >200ng of RNA sufficient for DASL analysis. The DASL assay was performed according to Illumina protocols. Using stringent exclusion criteria, 157/179 (88%) cases yielding results that were considered to be of sufficiently high quality to be included in the analysis. To fully assess the reproducibility of the assay, 35 cases were analysed on 2–8 occasions across multiple experimental days. Using Pearson's correlation, with full-linkage clustering, four discrete clusters were identified (n=28, 40, 46 and 43). Of important note, 95% of the samples were seen to cluster more tightly with their repeats than with any other sample, with all duplicated samples being called in the same cluster with 100% accuracy, suggesting that the technique is highly reproducible. Univariate Kaplan-Meier survival analysis showed that the clusters identified patients with very different outcomes. Two of the clusters showed identical survival curves and therefore these clusters were merged to give 3 clusters with 2-year overall survivals (OS) of 51% (n=71), 65% (n=46) and 77% (n=40), log rank p=0.03, with a 3.7 year follow-up. This data supports the use of gene expression profiling to classify DLBCL patients into clinically relevant prognostic groups. The Illumina DASL assay allows for highly reproducible gene expression data to be produced in valuable, archival data series, and also in the context of clinical trials, where the majority of the tissue available for study is FFPE. The patients identified in this study as having a sub-optimal response to CHOP-R should be considered for alternative therapies, which should be validated in the context of a clinical trial. Disclosures: No relevant conflicts of interest to declare.

Abstract 1466 Poster Board I-489 B-lymphocyte induced maturation protein 1 (BLIMP-1) has been defined as a key driver of the genetic reprogramming during differentiation of B-cells to plasma cells. Frequent inactivation of PRDM1, the BLIMP-1 gene, in diffuse large B-cell lymphoma (DLBCL) indicates that loss of function is an important event in lymphomagenesis. Only a limited set of direct BLIMP-1 target genes have been defined. In order to better understand the function of human BLIMP-1 in differentiation and malignancy we have established a more comprehensive set of occupied promoters. These data provide an extended view of the regulatory network controlled by BLIMP-1, and identify novel sets of targets involved in transcription and immune response. The composition of occupied promoters identifies complexity in BLIMP-1 binding motif selection, and substantial overlap between BLIMP-1 sites and Interferon regulatory factor (IRF) elements. Consistent with active competition between BLIMP-1 and IRFs, target genes associated with such overlapping motifs are found to be preferentially induced in response to BLIMP-1 knockdown. Finally BLIMP-1 targets are found to include key components of DLBCL gene expression signatures. This map of BLIMP-1 occupied promoters thus illuminates key aspects of function in normal and malignant cell biology. Disclosures: No relevant conflicts of interest to declare.

Background A number of methods that use both protein structural and evolutionary information are available to predict the functional consequences of missense mutations. However, many of these methods break down if either one of the two types of data are missing. Furthermore, there is a lack of rigorous assessment of how important the different factors are to prediction. Results Here we use Bayesian networks to predict whether or not a missense mutation will affect the function of the protein. Bayesian networks provide a concise representation for inferring models from data, and are known to generalise well to new data. More importantly, they can handle the noisy, incomplete and uncertain nature of biological data. Our Bayesian network achieved comparable performance with previous machine learning methods. The predictive performance of learned model structures was no better than a naïve Bayes classifier. However, analysis of the posterior distribution of model structures allows biologically meaningful interpretation of relationships between the input variables. Conclusion The ability of the Bayesian network to make predictions when only structural or evolutionary data was observed allowed us to conclude that structural information is a significantly better predictor of the functional consequences of a missense mutation than evolutionary information, for the dataset used. Analysis of the posterior distribution of model structures revealed that the top three strongest connections with the class node all involved structural nodes. With this in mind, we derived a simplified Bayesian network that used just these three structural descriptors, with comparable performance to that of an all node network.