Gene expression meta-analysis reveals immune response convergence on the IFNγ-STAT1-IRF1 axis and adaptive immune resistance mechanisms in lymphoma
- 2.9k Downloads
Cancers adapt to immune-surveillance through evasion. Immune responses against carcinoma and melanoma converge on cytotoxic effectors and IFNγ-STAT1-IRF1 signalling. Local IFN-driven immune checkpoint expression can mediate feedback inhibition and adaptive immune resistance. Whether such coupled immune polarization and adaptive resistance is generalisable to lymphoid malignancies is incompletely defined. The host response in diffuse large B-cell lymphoma (DLBCL), the commonest aggressive lymphoid malignancy, provides an empirical model.
Using ten publicly available gene expression data sets encompassing 2030 cases we explore the nature of host response in DLBCL. Starting from the “cell of origin” paradigm for DLBCL classification, we use the consistency of differential expression to define polarized patterns of immune response genes in DLBCL, and derive a linear classifier of immune response gene expression. We validate and extend the results in an approach independent of “cell of origin” classification based on gene expression correlations across all data sets.
T-cell and cytotoxic gene expression with polarization along the IFNγ-STAT1-IRF1 axis provides a defining feature of the immune response in DLBCL. This response is associated with improved outcome, particularly in the germinal centre B-cell subsets of DLBCL. Analysis of gene correlations across all data sets, independent of “cell of origin” class, demonstrates a consistent association with a hierarchy of immune-regulatory gene expression that places IDO1, LAG3 and FGL2 ahead of PD1-ligands CD274 and PDCD1LG2.
Immune responses in DLBCL converge onto the IFNγ-STAT1-IRF1 axis and link to diverse potential mediators of adaptive immune resistance identifying future therapeutic targets.
KeywordsImmune Checkpoint Immune Response Gene Focus Gene Classical Hodgkin Lymphoma Immune Checkpoint Blockade
activated B cell
cyclophosphamide, doxorubicin hydrochloride (hydroxydaunomycin),vincristine sulfate (Oncovin), prednisone
cell of origin
diffuse large B-cell lymphoma
false discovery rate
germinal centre B cell
Gene Expression Omnibus
HUGO Gene Nomenclature Committee
primary mediastinal B-cell lymphoma
Emergence of clinically detectable malignant disease is associated with escape from tumour immune surveillance . Two principal mechanisms may operate: on the one hand the immune systems loses the ability to detect the neoplastic population through changes in antigen presentation or editing of the antigen receptor repertoire; on the other hand initially effective immune responses may be rendered ineffective through development of an immune suppressive environment . In the latter scenario, local expression of immune checkpoint components can be viewed as subversion of a physiological mechanism, which acts during chronic infections to balance effective immunity with immune-mediated tissue damage .
In a range of cancers the density, location and functional polarization of tumour infiltrating lymphocytes are of prognostic value , providing evidence that the nature of immune evasion remains of importance after clinical detection. This is particularly relevant in the context of novel therapeutic strategies aimed at re-invigorating the “exhausted” anti-tumour immune response through immune checkpoint blockade [5, 6]. Gene expression analysis of bulk tumour tissue integrates expression profiles from multiple cellular sources, often allowing global assessment of the predominant vector of functional immune polarization. A paradigm has been proposed in which cancer-associated immune responses converge on a common “immunologic constant of rejection” characterized by a pattern of cytotoxic and T-cell immune responses and a dominant IFNγ-STAT1-IRF1 signalling axis [4, 7]. Linking the polarized pattern of interferon (IFN)γ-driven immune responses to the expression of immune checkpoints is the concept of “adaptive immune resistance” [5, 8]. In this model IFNγ signalling drives local feedback inhibition through the transcriptional regulation of ligands for the inhibitory receptor PD1 [5, 8]. The common association between cytotoxic responses and expression of IFN signatures and potential mediators of adaptive immune resistance has been further supported by analysis of solid tumour gene expression data from The Cancer Genome Atlas . Importantly, such feedback may be mediated both at the immediate interface between tumour cell and cytotoxic lymphocyte, and by the establishment of a wider immune suppressive milieu in the tumour microenvironment.
The combination of convergent IFN-polarized immune responses [4, 7], coupled to IFN-driven adaptive immune resistance [5, 8], provides a powerful model with which to explain common pathologic associations in carcinoma and melanoma. The recent success of therapies targeting CTLA4 and PD1 immune checkpoints [10, 11, 12], combined with an extended range of other therapeutic options , means that evidence to support prioritization of therapeutic combinations in different tumour settings is required. Lymphoma, which comprises immune system malignancies, provides an instance in which these pathways are incompletely studied. Classical Hodgkin lymphoma is the archetype in which host response elements dominate to the point of obscuring the neoplastic B-cell clone , and in classical Hodgkin lymphoma PD1 pathway blockade has recently been described as a promising therapeutic approach . Diffuse large B-cell lymphoma (DLBCL) is the commonest form of nodal lymphoma in the western world and represents an aggressive malignancy that frequently remains incurable. It is well established that this lymphoma type is associated with a varied extent of host response at diagnosis, which can include elements of IFN signalling . Since several large data sets are publicly available [15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25], this malignancy represents an empirical human model in which to test the association between immune polarization and adaptive immune resistance mechanisms.
The “cell of origin” (COO) classification provides the dominant paradigm for our current understanding of DLBCL [24, 26]. This classification relates the gene expression profiles in DLBCL to those of germinal centre B cells (GCBs) or activated B cells (ABCs), the latter representing the initial stage of B-cell terminal differentiation to plasma cells. Although the COO classification allows the division of DLBCL based on expression of a restricted set of classifier genes into the two principal classes , a subset of cases show patterns of classifier gene expression that do not allow confident assignment to either GCB or ABC subsets. Such cases are referred to as “type 3” [24, 26], or “unclassified” [27, 28]. To avoid ambiguity we refer to these cases as COO-unclassified DLBCL in the following. In a parallel “consensus cluster” classification developed by Monti et al. , it was shown that DLBCL could be divided into three categories characterized by preferential expression of genes linked to proliferation and B-cell receptor signalling, metabolic oxidative phosphorylation, or host response. The latter included multiple elements attributable to components of the immune system and supporting stromal cell types. It was noted that a greater proportion of COO-unclassified DLBCL belonged to the host/immune response cluster, which had increased numbers of intra-tumoral T cells and macrophages and a relative decrease in neoplastic B cells .
We reasoned that the potential association of COO-unclassified DLBCL with intense host responses provided a starting point for a meta-analysis of immune response elements in DLBCL. In originating from a prevailing paradigm this provided a wider biological and clinical context. Furthermore, by asking whether evidence supporting a common polarized immune response could be discovered from within the construct of the COO paradigm, we sought to avoid bias that might have arisen by focusing ab initio on components of the polarized immune response or immune checkpoints. With this approach we identify a distinct signature characterised by a pattern of cytotoxic T-cell and IFNγ-polarized immune response genes as a dominant pattern across ten DLBCL data sets encompassing 2030 cases. Using components of this polarized pattern we then explore the immune context of DLBCL in a fashion independent of COO class. We demonstrate the strong association with an IFNγ-STAT1-IRF1 axis and an expression hierarchy of immune checkpoints/modulators, consistent with adaptive immune resistance as a common feature operating in DLBCL.
Ten DLBCL data sets were downloaded from the Gene Expression Omnibus (GEO)  [GEO:GSE4475, GSE10846, GSE12195, GSE19246, GSE22470, GSE22895, GSE31312, GSE32918, GSE34171 and elsewhere [15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25]. GSE10846 was split according to treatment groups (CHOP [cyclophosphamide, doxorubicin hydrochloride (hydroxydaunomycin), vincristine sulfate (Oncovin), prednisone]/R-CHOP [rituximab-CHOP]), which were treated independently for analysis, thus giving a total of 11 data sets.
Normalisation and re-annotation of data
For each data set the probes were re-annotated with the latest version of HUGO Gene Nomenclature Committee (HGNC)-approved symbols . The complete HGNC list was downloaded (on 1 October 2014). Each probe was re-annotated to the latest approved symbol if an unambiguous mapping (i.e. single symbol mapping to approved symbol) could be determined, else the original gene name was maintained.
Each data set was quantile normalised using the R Limma package . The probes for each gene were merged by taking the median value for probe sets with a Pearson correlation ≥0.2 and the maximum value for those with a correlation <0.2 .
We used the COO classifications assigned by the DLBCL automatic classifier (DAC) classifier in our previous work .
See Additional file 1 for an outline of meta-profile generation using COO classification.
For each of the 11 data sets a linear model was fitted to the gene expression data using the R Limma package. Differentially expressed genes between the three classes were gauged using the Limma empirical Bayes statistic module, adjusting for multiple testing using Benjamini and Hochberg correction.
The absolute fold changes for all genes per data set were normalised between 0 and 1. The results were merged across data sets retaining only genes with an adjusted p value (false discovery rate, FDR < 0.05. A meta-profile was created for each contrast (e.g. upABC_GCB) by retaining all genes differentially expressed in six or more data sets. These were then used to draw Wordles  with each gene’s score set to (NumDataSets3) × NormalisedFoldChange.
Signature enrichment analysis
A data set of 14,104 gene signatures was created by merging signatures downloaded from SignatureDB , MSigDB v.4 (MSigDB C1--C7) , Gene Signature Database v.4 (GeneSigDB)  and the work of Monti et al.  and others [37, 38, 39, 40]. Enrichment of meta-profiles against signatures was assessed using a hypergeometric test, where the draw is the meta-profile genes, the successes are the signature genes and the population is the genes present on the platform.
Gene ontology analysis
Meta-profile gene lists were assessed for gene ontology (GO) enrichment using the Cytoscape BiNGO tool . GO and annotation files were downloaded from  (13 June 2014). The background reference was set to a non-redundant list of the genes present in the 11 data sets. The FDR rate (Benjamini and Hochberg) was set to ≤0.1.
Signature enrichment visualisation
See Additional file 2 for an outline of the process for integrating and visualizing analysis of gene signature and ontology enrichments.
The results from gene signature and gene ontology enrichment were used to create heatmap visualisations. For each meta-profile the top 100 most enriched signatures and 100 most enriched GO terms were used to construct a matrix of signatures against genes. This is a binary matrix with 1 s depicting an assigned signature/GO annotation. Using Python a row-wise (gene correlation) and column-wise (signature correlation) phi coefficient was calculated. These were then hierarchical clustered using GENE-E  with complete linkage.
Focus gene analysis
See Additional file 3 for an outline of the focus gene approach.
Per data set the genes were ordered by their variance across the patient samples, and the top 80 % were used to calculate Spearman’s rank correlations per row using the Python scipy.stats package. The resultant p value and correlation matrices were merged across the 11 data sets by taking the median values (across the sets in which the gene was contained), giving a final matrix of length 20,121. For a given focus gene the median rho and p values were reported along with a breakdown of the correlations and relative expression levels across the data sets (Additional file 4). For select focus genes a correlated gene set was created by taking all genes with a p > 0.45 present in six or more data sets. These correlated gene sets were then used for signature enrichment analysis and visualisation.
The Survival library for R was used to analyse right-censored survival data. Overall survival was estimated using the Kaplan-Meier method, modelled with Cox Proportional Hazards technique. Survival analysis was restricted to data sets of cases treated with the currently standard immunochemotherapy regimen R-CHOP.
Shared meta-profiles for COO-unclassified and COO-classified DLBCL
COO-unclassified DLBCL is enriched for features of a polarized immune response
To assess underlying biology in the COO-classified and COO-unclassified meta-profiles we developed an approach for integrated analysis of GO and gene signature enrichment (Additional file 2) which applies hierarchical clustering to reciprocally assess the relationships of enriched ontology and signature terms and associated genes contributing to enrichments (Additional file 6). The results are displayed as heatmaps of the hierarchically clustered correlations.
A cytotoxic and interferon polarized immune response as an independent molecular feature of DLBCL
To assess whether the 16-gene score also reflected the expression of other genes associated with the immune response in COO-unclassified DLBCL we added further components of the meta-profile. Expression of these genes followed the overall pattern of expression of the 16-gene score across all DLBCL data sets (Fig. 5b; Additional file 13). Thus, the 16-gene score provides a tool with which to identify the overall pattern of this polarized immune response in DLBCL.
Polarised immune response and COO-unclassified DLBCL do not overlap significantly with signatures of primary mediastinal B-cell lymphoma
COO-unclassified DLBCL cases lacking both polarized immune response and COO-classifier gene expression are distinct from the subset of cases in which the extent of the polarized immune response obscures the characterization of the neoplastic B-cell population. At least two principal explanations could be considered for this subgroup: on the one hand these might include cases in which gene expression was technically challenging with poor representation of tumour cell RNA; alternatively, they might include a subset of large B-cell lymphoma which fails to express COO-classifier genes at significant levels. Primary mediastinal B-cell lymphoma (PMBL) is a biologically distinct subgroup of large B-cell lymphoma, more common in women, with a mediastinal localization, distinct molecular genetics and possible derivation from a thymic B-cell population . This lymphoma class can be associated with a pattern of gene expression distinct from either GCB- or ABC-DLBCL. While many PMBL cases would be excluded on the basis of diagnosis from conventional DLBCL gene expression data sets, it was possible that some PMBL cases might contribute to the COO-unclassified DLBCL cases, in particular those lacking a polarized immune response signature. To address this we used the 23-gene PMBL signature described by Rosenwald et al. , and first tested for enrichment within the COO-classified and COO-unclassified meta-profiles, but this showed no evidence of significant enrichment, nor was a signature separating PMBL from Hodgkin lymphoma enriched (Additional file 6). We next used the 23-gene PMBL signature in place of the extended immune response gene list to reanalyse the DLBCL data sets by hierarchical clustering (Additional file 15). We found no evidence of distinct clusters of cases identifiable with the 23-gene PMBL signature amongst COO-unclassified DLBCL, although a few elements of the 23-gene signature, most notably PDCD1LG2, CD274 and BATF3, do correlate with the polarized immune response. In contrast, in several data sets small clusters of cases were identifiable with coordinated high expression of the 23 genes of the PMBL signature, but such cases were classifiable as GCB-DLBCL, suggesting a greater overlap of PMBL signature gene expression amongst cases otherwise classifiable as GCB-DLBCL rather than ABC-DLCBL or COO-unclassified DLBCL. Thus, we found no gene expression-based evidence for a significant contribution of PMBL-like gene expression patterns amongst COO-unclassified DLBCL in the data sets analysed. Inclusion of PMBL-like cases does not have a major impact on the detection of the polarized immune response signature, nor provide an explanation for the subset of COO-unclassified DLBCL that lacks both COO-classifier and polarized immune response gene expression.
A polarized immune response is associated with improved outcome in DLBCL
Polarization along an IFNγ-STAT1-IRF1 axis is a defining feature of the DLBCL immune response
As focus genes we selected two components of the 16-gene polarized immune response signature, TRAT1 and FGL2, to reflect origin from the two branches of the COO-unclassified meta-profile (Fig. 8b; Additional files 18 and 19). TRAT1 was selected as the most highly correlated gene from cluster 1 (Fig. 4), while FGL2 was selected as the second most highly correlated gene in cluster 2, and of more established immunologic interest than TC2N and less overt connection to immune response polarization than IFNG, the other two genes derived from cluster 2 that contribute to the 16-gene polarized immune response classifier.
Genes correlating with TRAT1 could be assigned to clusters of signatures and GO terms related to T-cell state, and T-cell signal transduction, cell motility and interferon response. For FGL2 as the focus gene a similar pattern emerged, including an expanded cluster of signature enrichments related to interferon responses, including some derived from models of viral infection, and an additional association with monocyte/macrophage-derived signatures.
IFNγ-STAT1-IRF1 axis and adaptive immune regulatory pathways in DLBCL
The common convergence of cancer immune responses onto patterns of cytotoxic and IFNγ-dominated pathways has been summarised in the concept of an “immune constant of rejection” [4, 7]. In parallel the model of adaptive immune resistance argues for the control of such immune responses via local feedback driven through IFN-mediated upregulation of immune checkpoints [5, 8]. Our analysis here provides extensive empirical evidence across currently available large DLBCL data sets that this combination of IFNγ polarisation and induction of adaptive immune resistance mechanisms is a feature of the immune response to DLBCL. Unbiased analysis of gene expression correlations moreover suggests a hierarchy of IFN-associated immune modulatory gene expression with LAG3, IDO1 and FGL2 as key elements in this context. Thus, adaptive immune resistance is likely to provide an important component of immune evasion in DLBCL.
Other mechanisms of immune evasion have been previously identified as playing an important role in the pathogenesis of DLBCL, including mutation and deletion of B2M and CD58, and amplification of genomic regions encompassing genes encoding PD1 ligands [48, 49]. Furthermore previous studies have demonstrated the presence of PD1 expression on infiltrating T-cell populations and PD-L1(CD274) on tumour cells and in the microenvironment of DLBCL and related neoplasms [50, 51]. In the context of gene expression profiling, morphologically defined T-cell and histiocyte-rich large B-cell lymphoma, which represents a relatively rare subcategory, has been characterized by evidence of an IFN-associated immune response, linked on the one hand with over-expression of PD1 (PDCD1) on infiltrating T cells when compared with classical Hodgkin lymphoma , or the expression of IDO1 when compared with nodular lymphocyte predominant Hodgkin lymphoma, another relatively rare lymphoma subtype . Indeed, expression of IDO1 has been defined as a feature associated with poor outcome in generic DLBCL in one patient series . Thus, the involvement of several pathways of immune modulation in large B-cell lymphomas is supported by prior studies.
Using the 16-gene polarized immune response score we have ranked DLBCL cases across multiple data sets, and demonstrate that a substantial fraction of cases regardless of COO class are linked to a polarized immune response. In the context of the COO classification, the dominance of this immune response at the expense of proliferating B cells provides the most common explanation for DLBCL cases that are “unclassifiable” as originally suggested by Monti et al. . Equally important is the identification of a distinct group of DLBCL characterized by an absence of host response elements, which is consistent with “immunological ignorance”, a feature which in other cancers is associated with poor response to immune checkpoint blockade . These cases are also consistent with a model of host tissue “effacement” proposed by Scott and Gascoyne  as distinguishing subsets of aggressive lymphomas. Immune evasion in DLBCLs can be associated with loss of MHC class I expression consequent on mutation and/or deletion of B2M, which may be further accompanied by inactivation of CD58 , and a prediction might be that such cases would be enriched in the subset characterized by apparent immunological ignorance. However, analogous lesions affecting B2M were recently identified as recurrent events positively associated with cytotoxic gene signatures in solid tumours . This suggests a model in which adaptive immune resistance mechanisms may be followed by somatic genetic alterations that further enhance tumour immune escape. Whether a similar positive association between cytotoxic response and B2M or CD58 mutation status exists in DLBCL is, to our knowledge, not established.
Across several cancer types the intensity of tumour infiltrating lymphocytes and their functional polarization has proved to be of prognostic significance in the absence of specific immune checkpoint blockade [4, 55, 56, 57]. Our analysis indicates that a trend toward an improved outcome in association with a more intense polarized immune response is generally maintained in the context of DLBCL treated with the current immunochemotherapy regimen, R-CHOP. However, this benefit is not equivalent across all DLBCL when considered in relation to COO class, and is most pronounced for GCB-DLBCL. Indeed, in the largest available data set of R-CHOP-treated DLBCL, GSE31312 , a substantial group of patients with both a GCB-DLBCL classification and a high polarized immune response score appeared curable with current therapy. As a statistically significant association is not consistently observed across all three data sets of DLBCL treated with R-CHOP, and there is a potentially confounding association with young age, the overall prognostic value of this association remains uncertain in the context of current therapy. Additional features of the host response, which did not emerge as principal discriminants between COO-classified versus COO-unclassified DLBCL, such as contributions from macrophage/monocyte lineage cells, may add value to immune response classifiers. These will need to be considered alongside the polarized immune response signature in future work. Nonetheless, the analysis presented here demonstrates a graded pattern of immune response in DLBCL, with one extreme characterized by minimal cytotoxic immune response signature and tendency to poor outcome, and another extreme characterized by intense polarized immune response and a tendency toward better outcome which is modified by COO class. In other settings the pattern of pre-existing immune response prior to immune checkpoint therapy has proved to be of predictive value [11, 12, 58, 59]. Based on this evidence, it is the subset of DLBCL cases with pre-existing polarized immune response which is most likely to benefit from immune checkpoint/modulatory therapy, and is readily identifiable in a quantitative fashion from gene expression data.
Immune checkpoint inhibitors are already under evaluation in the context of large cell lymphomas [60, 61]. Recent clinical trials with PD1 pathway blockade have shown substantial promise in Hodgkin lymphoma , as in other tumour types [11, 12, 62]. Combining immune checkpoint inhibitors may hold particular promise, and both LAG3 and IDO1 are therapeutic targets with novel agents in current clinical evaluation. Our analyses support these as high priority candidates for therapeutic evaluation in DLBCL alongside PD1 blockade. In addition to direct interventions specifically targeting immune checkpoints, signalling pathways that mediate survival of neoplastic B cells, and are the targets of novel therapeutic agents in lymphoma, overlap with pathways controlling T-cell responses. Such agents have the potential to de-repress cytotoxic T-cell populations and promote anti-tumour immunity . Thus, companion biomarkers evaluating the potential association between pre-existing immune response at diagnosis and treatment response should arguably also be included in the setting of lymphoma clinical trials where agents targeting lymphocyte signalling pathways are being evaluated.
A notable element of the DLBCL immune response is the consistent association with FGL2 expression. This encodes fibrinogen-like 2 prothrombinase, a protein that has dual roles as a pro-coagulant and immune modulator. FGL2 has been shown to act as an immune responsive coagulant in settings such as foetal loss driven by Th1 polarized immune responses  and fulminant hepatitis . Subsequently, FGL2 has been implicated as a repressor of T-cell activation both in the ability of recombinant FGL2 to block graft rejection  and in the context of Fgl2 knockout mice developing autoimmune glomerulonephritis . In several experimental models FGL2 has been associated with suppression of cytotoxic and Th1-polarized immune responses [67, 68, 69]. FGL2 effects in DLBCL could relate to both pro-coagulant and immune modulatory functions. In DLBCL FGL2 expression correlates with multiple elements of the IFNγ-STAT1-IRF1 axis; supporting direct regulation, FGL2 expression has previously been shown to be responsive to IFNγ in T cells [70, 71], and was shown to act downstream of IRF1 in Th1-driven foetal loss . Thus, the relationships in DLBCL suggest that FGL2 may provide an additional element of negative feedback and adaptive immune resistance, which is potentially suitable for therapeutic targeting [72, 73].
We note that some DLBCL cases with a prominent immune response may be associated with Epstein-Barr virus (EBV) infection and oncogenic drive. In the meta-analysis approach taken here the contribution of EBV cannot be systematically assessed from available data since EBV status is incompletely annotated, and not necessarily assessed using both immunohistochemistry for EBV LMP1 and RNA-FISH for EBERs. Immune surveillance is known to contribute to the control of EBV-mediated tumours , and the presence of high EBV loads can contribute to the establishment of an exhausted cytotoxic response . Indeed, there are significant overlaps between the gene expression profiles of the immune response in EBV-associated large cell lymphomas occurring in the post-transplant setting  and the polarized IFNγ-associated gene expression that is evident from our DLBCL meta-analysis. However, while the frequency of EBV infection in generically diagnosed DLBCL has been established at close to 10 % , significant expression of genes linked to the polarized immune response is more frequent across DLBCL data sets. An overlap of gene expression profiles between the immune response targeting EBV-driven and EBV-independent lymphomas is consistent with the model of convergent patterns of “immune rejection” across diverse immune contexts [4, 7]. It is arguable that the principal predictive factor of response to immune checkpoint inhibition will be the presence of a pre-existing polarized immune response and the mechanisms controlling its chronic activation/exhaustion rather than the nature of the initial triggering antigens whether viral or cancer-associated.
The analysis presented here supports the central importance of convergent patterns of immune response linked to the IFNγ-STAT1-IRF1 axis, coupled to IFN-driven feedback pathways in DLBCL. This argues for the generalisable nature of these interconnected mechanisms, and implicates a hierarchy of immune modulators, known to promote the establishment of an immunosuppressive microenvironment , in the process of IFNγ-driven adaptive immune resistance.
This work was supported by a Cancer Research UK senior clinical fellowship to RMT (C7845/A10066) and Cancer Research UK programme grant C7845/A17723. We thank Dr Gina Doody, Dr Andrew Jack and Dr Darren Newton for critical review.
- 8.Taube JM, Anders RA, Young GD, Xu H, Sharma R, McMiller TL, et al. Colocalization of inflammatory response with B7-h1 expression in human melanocytic lesions supports an adaptive resistance mechanism of immune escape. Sci Transl Med. 2012;4:127ra37. doi: 10.1126/scitranslmed.3003689.
- 14.Ansell SM, Lesokhin AM, Borrello I, Halwani A, Scott EC, Gutierrez M, et al. PD-1 blockade with nivolumab in relapsed or refractory Hodgkin's Lymphoma. N Engl J Med. 2014. doi: 10.1056/NEJMoa1411087.
- 20.Salaverria I, Philipp C, Oschlies I, Kohler CW, Kreuz M, Szczepanowski M, et al. Translocations activating IRF4 identify a subtype of germinal center-derived B-cell lymphoma affecting predominantly children and young adults. Blood. 2011;118:139–47. doi: 10.1182/blood-2011-01-330795.CrossRefPubMedGoogle Scholar
- 22.Monti S, Chapuy B, Takeyama K, Rodig SJ, Hao Y, Yeda KT, et al. Integrative analysis reveals an outcome-associated and targetable pattern of p53 and cell cycle deregulation in diffuse large B cell lymphoma. Cancer Cell. 2012;22:359–72. doi: 10.1016/j.ccr.2012.07.014.PubMedCentralCrossRefPubMedGoogle Scholar
- 23.Visco C, Li Y, Xu-Monette ZY, Miranda RN, Green TM, Tzankov A, et al. Comprehensive gene expression profiling and immunohistochemical studies support application of immunophenotypic algorithm for molecular subtype classification in diffuse large B-cell lymphoma: a report from the International DLBCL Rituximab-CHOP Consortium Program Study. Leukemia. 2012;26:2103–13. doi: 10.1038/leu.2012.83.PubMedCentralCrossRefPubMedGoogle Scholar
- 25.Barrans SL, Crouch S, Care MA, Worrillow L, Smith A, Patmore R, et al. Whole genome expression profiling based on paraffin embedded tissue can be used to classify diffuse large B-cell lymphoma and predict clinical outcome. Br J Haematol. 2012;159:441–53. doi: 10.1111/bjh.12045.CrossRefPubMedGoogle Scholar
- 27.Bea S, Zettl A, Wright G, Salaverria I, Jehn P, Moreno V, et al. Diffuse large B-cell lymphoma subgroups have distinct genetic profiles that influence tumor biology and improve gene-expression-based survival prediction. Blood. 2005;106:3183–90. doi: 10.1182/blood-2005-04-1399.PubMedCentralCrossRefPubMedGoogle Scholar
- 30.Gray KA, Yates B, Seal RL, Wright MW, Bruford EA. Genenames.org: the HGNC resources in 2015. Nucleic Acids Res. 2014; doi: 10.1093/nar/gku1071.
- 32.Care MA, Barrans S, Worrillow L, Jack A, Westhead DR, Tooze RM. A microarray platform-independent classification tool for cell of origin class allows comparative analysis of gene expression in diffuse large B-cell lymphoma. PLoS One. 2013;8, e55895. doi: 10.1371/journal.pone.0055895.PubMedCentralCrossRefPubMedGoogle Scholar
- 33.Wordle™. http://www.wordle.net/.
- 34.Signature database. http://lymphochip.nih.gov/signaturedb/.
- 35.The Molecular Signatures Database (MSigDB). http://www.broadinstitute.org/gsea/msigdb/index.jsp.
- 36.GeneSigDB. http://compbio.dfci.harvard.edu/genesigdb/.
- 40.Rosenwald A, Wright G, Leroy K, Yu X, Gaulard P, Gascoyne RD, et al. Molecular diagnosis of primary mediastinal B cell lymphoma identifies a clinically favorable subgroup of diffuse large B cell lymphoma related to Hodgkin lymphoma. J Exp Med. 2003;198:851–62. doi: 10.1084/jem.20031074.PubMedCentralCrossRefPubMedGoogle Scholar
- 42.Gene Ontology Consortium. http://www.geneontology.org.
- 48.Challa-Malladi M, Lieu YK, Califano O, Holmes AB, Bhagat G, Murty VV, et al. Combined genetic inactivation of beta2-Microglobulin and CD58 reveals frequent escape from immune recognition in diffuse large B cell lymphoma. Cancer Cell. 2011;20:728–40. doi: 10.1016/j.ccr.2011.11.006.PubMedCentralCrossRefPubMedGoogle Scholar
- 52.Chetaille B, Bertucci F, Finetti P, Esterni B, Stamatoullas A, Picquenot JM, et al. Molecular profiling of classical Hodgkin lymphoma tissues uncovers variations in the tumor microenvironment and correlations with EBV infection and outcome. Blood. 2009;113:2765–3775. doi: 10.1182/blood-2008-07-168096.CrossRefPubMedGoogle Scholar
- 53.Van Loo P, Tousseyn T, Vanhentenrijk V, Dierickx D, Malecka A, Vanden Bempt I, et al. T-cell/histiocyte-rich large B-cell lymphoma shows transcriptional features suggestive of a tolerogenic host immune response. Haematologica. 2010;95:440–8. doi: 10.3324/haematol.2009.009647.PubMedCentralCrossRefPubMedGoogle Scholar
- 60.Ansell SM, Hurvitz SA, Koenig PA, LaPlant BR, Kabat BF, Fernando D, et al. Phase I study of ipilimumab, an anti-CTLA-4 monoclonal antibody, in patients with relapsed and refractory B-cell non-Hodgkin lymphoma. Clin Cancer Res. 2009;15:6446–53. doi: 10.1158/1078-0432.CCR-09-1339.PubMedCentralCrossRefPubMedGoogle Scholar
- 61.Armand P, Nagler A, Weller EA, Devine SM, Avigan DE, Chen YB, et al. Disabling immune tolerance by programmed death-1 blockade with pidilizumab after autologous hematopoietic stem-cell transplantation for diffuse large B-cell lymphoma: results of an international phase II trial. J Clin Oncol. 2013;31:4199–206. doi: 10.1200/JCO.2012.48.3685.CrossRefPubMedGoogle Scholar
- 66.Chan CW, Kay LS, Khadaroo RG, Chan MW, Lakatoo S, Young KJ, et al. Soluble fibrinogen-like protein 2/fibroleukin exhibits immunosuppressive properties: suppressing T cell proliferation and inhibiting maturation of bone marrow-derived dendritic cells. J Immunol. 2003;170:4036–44.CrossRefPubMedGoogle Scholar
- 69.Khattar R, Luft O, Yavorska N, Shalev I, Phillips MJ, Adeyi O, et al. Targeted deletion of FGL2 leads to increased early viral replication and enhanced adaptive immunity in a murine model of acute viral hepatitis caused by LCMV WE. PLoS One. 2013;8, e72309. doi: 10.1371/journal.pone.0072309.PubMedCentralCrossRefPubMedGoogle Scholar
- 73.Urbanellis P, Shyu W, Khattar R, Wang J, Zakharova A, He W, et al. The Treg effector molecule fibrinogen-like protein 2 is necessary for the development of rapamycin-induced tolerance to fully MHC-mismatched murine cardiac allografts. Immunology. 2014; doi:10.1111/imm.12354.Google Scholar
- 75.Macedo C, Webber SA, Donnenberg AD, Popescu I, Hua Y, Green M, et al. EBV-specific CD8+ T cells from asymptomatic pediatric thoracic transplant patients carrying chronic high EBV loads display contrasting features: activated phenotype and exhausted function. J Immunol. 2011;186:5854–62. doi: 10.4049/jimmunol.1001024.PubMedCentralCrossRefPubMedGoogle Scholar
Open AccessThis article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.