Mutations in early follicular lymphoma progenitors are associated with suppressed antigen presentation

Frequent Cooccurring Mutations of Chromatin-Modifying Genes in FL.

To define recurrently mutated genes in FL, we performed exome sequencing of purified tumor B cells and matched germ-line DNA from tumor-infiltrating T cells of 28 FL tumors taken before treatment at the time of original diagnosis (SI Appendix, Fig. S1 and Table S1). Mutations identified in this cohort as well as those from prior studies were combined within a list of 284 gene candidates totaling 2.26 Mbp that we sequenced in tumors from 110 additional patients. These included 75 tumors for which malignant B cells were purified and 63 tumors studied as archival formalin-fixed paraffin-embedded (FFPE) specimens (SI Appendix, Table S2). Because FL tumors often contain substantial numbers of nonmalignant cells, we measured recurrence and cooccurrence of mutations within the 75 cases with purified tumor B cells available. FFPE tumors were used only as an extension cohort to compare the recurrence frequencies with those observed when interrogating these routinely used clinical specimens. The lower sensitivity for detecting mutations in FFPE sections compared with purified B cells is highlighted by variant frequencies in these samples (SI Appendix, Fig. S2). In total, we identified 28 genes that had somatic mutations detected in the exome sequencing cohort with matched germ-line DNA, that were targeted more frequently by coding mutations than silent mutations, that had detectable expression in normal or malignant B cells, and that were mutated in ≥5% of FL tumors or previously implicated in lymphoma (Fig. 1 and SI Appendix, Table S3). Among these were four novel recurrently mutated genes, including a component of the switch/sucrose nonfermentable (SWI/SNF) nucleosome remodeling complex (SMARCA4, 5%), a translation elongation factor (EEF1A1, 9%), and two subunits of the vacuolar ATPase proton pump (ATP6AP1, 12%; ATP6V1B2, 22%; SI Appendix, Fig. S3).

We performed hypergeometric enrichment analysis of recurrently mutated genes to define the hallmarks of FL and confirmed chromatin modification to be the most significantly perturbed biological process [false discovery rate (FDR) = 0.008; SI Appendix, Table S4]. Bromodomain-containing proteins were also significantly enriched (FDR = 0.062). Recurrently mutated chromatin-modifying genes (CMGs) included histone methyltransferases (KMT2D, 76%; KMT2C, 13%; EZH2, 12%), histone acetyltransferases (CREBBP, 68%; EP300, 9%), linker histone proteins (HIST1H1E, 12%; HIST1H1C, 4%), and components of the SWI/SNF complex (ARID1A, 9%; SMARCA4, 5%). Notably, 96% (72/75) of tumors from which purified B cells were interrogated contained one or more CMG mutation, and 71% (53/75) contained two or more CMG mutations (Fig. 1). This cooccurrence was not statistically significant because of the high frequency of KMT2D and CREBBP mutations across the cohort, but contrasts the significant mutual exclusivity seen for mutations in other genes with related functions such as receptor tyrosine kinase signaling genes in solid tumors (19). We also observed a significant association between CREBBP mutation and low histologic grade (P = 0.004; SI Appendix, Fig. S4).

Changes in the FL Genome During Disease Progression.

To evaluate the genomic evolution of FL during disease progression, we interrogated 59 tumors from 22 patients at various points and therapy milestones by single nucleotide polymorphism microarray and exome sequencing (SI Appendix, Table S1 and SI Appendix, Fig. S5). BCL2-IGH translocation breakpoints were assessed by nested PCR and identified in 19/22 patients, with the same breakpoint maintained throughout the course of disease (Fig. 2A and SI Appendix, Fig. S6). Somatic copy number alterations were frequently unstable during the course of disease, with frequent loss of alterations in sequential biopsies and complete loss of karyotypic complexity observed in three cases (Fig. 2A). In contrast, we observed an increasing burden coding somatic nucleotide variants (cSNVs) and insertions/deletions (InDels) at relapse compared with diagnosis and a trend toward increasing fractions of C > A mutations (Fig. 2A). However, these were not related to either the time between biopsies or type of intervening treatment (Fig. 2B). Although no gene showed significantly higher frequencies of mutation at relapse compared with diagnosis (Fisher P > 0.05), mutations in genes such as EZH2 (4/6), TP53 (2/3), IRF8 (2/3), TNFAIP3 (2/3), CARD11 (3/5), and TNFRSF14 (9/16) were more frequently detected in only the relapse tumor and not at initial diagnosis. Interestingly, mutations that were specific to relapse tumors occurred significantly more frequently within motifs recognized by either activation-induced cytidine deaminase (consensus WRGY) or apolipoprotein B mRNA editing enzyme catalytic polypeptide (10.38% of relapse specific mutations compared with 9.2% of all mutations; chi-square P < 0.001). However, apolipoprotein B mRNA editing enzyme catalytic polypeptide motifs were independently more significantly enriched within relapse-specific mutations (P = 0.018) than activation-induced cytidine deaminase motifs (P = 0.070). Tumors from the same patient shared a core set of mutations that made them more similar to each other than to tumors from other patients (SI Appendix, Fig. S6B). However, on average, 49.0% (range 2.9–94.3%) of the somatic mutations detected within a given tumor were not uniformly detected across all other tumors from the same patient. Notably, this was not a result of lack of sensitivity of exome sequencing, as these observations were validated at high depths of coverage, using our targeted sequencing approach. Specifically, there was 96.2% concordance between exome sequencing and high-depth targeted sequencing (median, 243×) for detecting the presence or absence of mutations in recurrently mutated genes, and for CMGs there was 100% concordance (81/81) for detecting the presence or absence of mutations (SI Appendix, Fig. S7). Only a single somatic mutation in GNAI2 was detected by high-depth targeted sequencing and not by exome sequencing.

Chromatin-Modifying Gene Mutations in Common Progenitors.

BCL2-IGH translocations were maintained with the same breakpoint throughout the course of disease (SI Appendix, Table S1), indicating that this lesion is present within a shared common progenitor to the serial tumor biopsies (Fig. 3 and SI Appendix, Fig. S6). To identify other somatic mutations within tumor cell progenitors, we evaluated the representation of mutations over the course of disease, with a special focus on CMGs. Using the somatic mutation hierarchy algorithm (20), we inferred the most parsimonious evolutionary phylogenies for tumors acquired from the same patient on the basis of the union of all somatic mutations detected across all tumors, regardless of predicted function (range, 232–2,633 mutations per patient). Mutations that are uniformly detected across all tumors from a single patient are inferred to originate from the EIP (green nodes; Fig. 3 and SI Appendix, Fig. S8). In 10 cases for which three or more tumors per patient were analyzed, greater resolution could be gained and secondary precursors can be inferred that contain sets of mutations that are shared between two or more tumors from the same patient, but not all tumors (purple nodes; Fig. 3). Those mutations that were private to a single tumor were inferred to be late events during genome evolution and not inherited from a common progenitor (yellow nodes; Fig. 3). Although this model does not take allelic frequencies into account when building evolutionary hierarchies, we observed that variants that were inferred to be present with the EIP showed allelic frequencies that are indicative of clonal representation, whereas mutations inferred to be acquired as later events in evolution primarily possessed lower allelic frequencies that are indicative of subclonal representation (SI Appendix, Fig. S8B). In addition, we also observed that subsets of shared and private mutations that are later in the evolutionary hierarchy can have allele frequencies that appear close to heterozygous or homozygous (SI Appendix, Fig. S8B). This is in line with our prior observations that allelic frequencies within the bulk tumor are poorly predictive of clonal origin (12).

The average mutational burden of the EIP was 221 mutations (range, 59–447 mutations). These accounted for a total of 33% of the coding mutations (591/1806) detected across all tumors in this series. Among coding mutations, the mutations in recurrently mutated genes were significantly enriched within the EIP (63/135; Fisher test FDR < 0.001; Fig. 3D and SI Appendix, Table S5), highlighting their importance in disease biology. However, among mutations of recurrently mutated genes, there was no significant enrichment in either CMG mutations as a whole (41/72; Fisher test FDR = 0.411) or KMT2D mutations (16/29; Fisher test FDR = 0.874) within EIPs. This analysis assumes that chromatin-modifying gene mutations, including those in KMT2D, are advantageous to tumor cell clones and are not lost during clonal evolution. In contrast, 94% (16/17) of CREBBP mutations were shared between all tumors of a given patient and inferred to be present within the EIPs of the respective cases (Fig. 3D and SI Appendix, Fig. S8B). This represents a significant enrichment of CREBBP mutations within EIPs compared with all coding mutations (Fisher test FDR < 0.001), mutations in recurrently mutated genes (Fisher test FDR < 0.001), CMG mutations (Fisher test FDR = 0.012), and KMT2D mutations (Fisher test FDR = 0.196).

Decreased MHC Class II Expression in CREBBP Mutant FL.

CREBBP mutations clustered within the lysine acetyltransferase domain as previously described, but we observed 34% (27/80) of mutations to affect a single amino acid, arginine 1408/1143 (R1408 in isoform b, R1446 in isoform a; SI Appendix, Fig. S9). This amino acid contacts the substrate of CREBBP, resulting in decreased histone acetylation (21), and other recurrently mutated residues in the lysine acetyltransferase domain also reside within the substrate-binding pocket (SI Appendix, Fig. S9). As changes in histone acetylation are likely to have broad effects on transcription, we next investigated the transcriptional signature associated with CREBBP mutations to define their functional consequences. Because of prior associations between CREBBP and the function of the BCL6 and p53 in diffuse large B-cell lymphoma (11), we evaluated expression of the targets of these transcription factors by gene set enrichment analysis but found no significant enrichment (SI Appendix, Fig. S10). We therefore used differential gene expression analysis to define those genes that were significantly altered in B cells from tumors with CREBBP mutations compared with those without. This signature consisted of 334 genes with significantly increased expression and 278 genes with significantly decreased expression in tumor B cells from cases with CREBBP mutation (Fig. 4 and SI Appendix, Table S6). Hypergeometric gene set enrichment analysis of this signature found it to be strikingly enriched for genes involved with antigen processing and presentation (FDR < 0.001; SI Appendix, Table S4). This was confirmed by conventional gene set enrichment analysis and found to be driven by decreased expression of multiple MHC class II genes (HLA-DRA, HLA-DRB1, HLA-DMA, HLA-DMB, HLA-DPA1, HLA-DQA1, HLA-DQB1) and the CD74 invariant chain within the mutant tumors (SI Appendix, Fig. S10 D–F). There was no significant difference in MHC class I expression associated with CREBBP mutation (FDR > 0.25), and the gene expression signatures associated with other CMGs showed no significant difference in MHC class II expression (SI Appendix, Fig. S10G). To ensure this effect was not the result of cosegregation between CREBBP mutations and deletions of the MHC class II locus, we interrogated somatic copy number data from our exome series and additional publicly available high-resolution data (12, 13, 22–24). Although we detected deletions of the MHC class II locus in a subset of patients (SI Appendix, Fig. S11), this only included a single tumor from our exome sequencing cohort (LPM011b) that did not possess a CREBBP mutation.

To confirm decreased expression of MHC class II on tumor B cells with CREBBP mutation, we measured cell surface protein levels of HLA-DR by flow cytometry in 14 tumors from diagnosis before any therapy. The mean fluorescence intensity of HLA-DR expression was on average 8.1-fold lower on tumor B cells from cases with CREBBP mutation (n = 9) compared with those without CREBBP mutation (n = 5; t test P value < 0.001). There was no significant difference in HLA-DR mean fluorescence intensity on tumor-infiltrating (T.I.) normal B cells of CREBBP mutant tumors compared with CREBBP wild-type tumors (t test P value = 0.338), indicating that this observation was not a result of sample handling or staining conditions. However, as MHC class II can be induced by a variety of stimuli that can vary from tumor to tumor, we assessed the relative expression of HLA-DR on tumor B cells and T.I. normal B cells from within the same tumor microenvironment. This showed consistently lower expression of HLA-DR on tumor B cells compared with T.I. normal B cells in tumors with CREBBP mutation, as opposed to moderately higher expression on tumor B cells in CREBBP wild-type tumors (Fig. 4 and SI Appendix, Fig. S12).

Expression of MHC class II genes are regulated at multiple levels, including immune signaling, cytokines, transcription factors, and epigenetic modifications (25, 26). We therefore investigated whether CREBBP-mutation-associated MHC class II deficit could be overcome by activation of TLR9 and CD40 signaling. Whether harboring heterozygous (LPJ117) or homozygous (LPM025, LPM019) CREBBP mutations, surface MHC class II protein levels on tumor B cells could be restored to levels comparable to those seen in stimulated CREBBP wild-type cells (LPM020; SI Appendix, Fig. S13).

CREBBP Mutant Tumors Harbor Muted Infiltrating T Cells.

We next evaluated the transcriptional profiles of T.I. T cells purified from 32 tumors and found significant differences between T cells isolated from tumors bearing CREBBP mutant tumor B cells compared with those isolated from tumors bearing CREBBP wild-type tumor B cells (Fig. 5A). Using gene set enrichment analysis (GSEA), we found these signatures to be indicative of reduced proliferation in T cells from tumors with CREBBP mutant tumor B cells, as shown by the significant enrichment of proliferation signatures within the T cells from tumors with CREBBP wild-type tumor B cells (Fig. 5B and SI Appendix, Table S7). No other CMG mutation was associated with signatures of suppressed T-cell proliferation by GSEA (FDR > 0.25), in line with their lack of association with decreased MHC class II expression (SI Appendix, Fig. S10G).

We therefore tested whether CREBBP mutant tumor B cells with lower MHC class II expression may be less capable of inducing proliferation of CD4 T cells. We accomplished this by sorting tumor B cells and T.I. normal B cells from tumors harboring wild-type or mutant CREBBP (Fig. 5B) and coculturing them with healthy donor CD4 T cells in the presence of toxic shock syndrome toxin-1 to link MHC class II and the T-cell receptor in an antigen-independent manner. Tumor B cells were identified by CD10 expression so as not to stimulate them through crosslinking of surface Ig, but this may result in minor contamination from a small fraction of CD10-postive normal B cells. Using dye dilution to measure proliferation of the cocultured healthy T cells, we assessed the ability of T.I. normal B cells and tumor B cells with identical HLA mismatches to stimulate CD4 T cells through cross-linking of MHC class II and T-cell receptor molecules (Fig. 5C). Across six cases (three CREBBP wild-type, three CREBBP mutant), we observed that the relative level of CD4 T-cell proliferation stimulated by tumor B cells compared with T.I. normal B cells from the same case was significantly lower in CREBBP mutant tumors compared with CREBBP wild-type tumors (P = 0.012). This is in line with the lower relative levels of MHC class II expression on tumor B cells from CREBBP mutant cases compared with T.I. normal B cells (Fig. 4 B and C). CREBBP mutation-associated differences in MHC class II expression are therefore related to the ability of these cells to stimulate proliferation of CD4 T cells.

As reduced T-cell proliferation is likely to result in altered T-cell frequencies, we enumerated the frequencies of T-cell subpopulations by flow cytometry in 32 tumors with known CREBBP status (Fig. 6 and SI Appendix, Table S8). CREBBP mutant tumors harbored significantly lower frequencies of total CD3⁺ T cells (P = 0.007; t test FDR = 0.033) and CD3⁺CD4⁺ helper T cells (P = 0.006; t test FDR = 0.033) than wild-type counterparts. We also noted lower levels of CD3⁺CD8⁺ cytotoxic T cells in these tumors, driven by decreased frequency of the CD3⁺CD8⁺CD45RO⁺ memory subset (P = 0.002; t test FDR = 0.014). There were no significant differences in CD3⁺CD4⁺CD25⁺ regulatory T cells, CD56⁺ natural killer cells, or CD14⁺ myeloid cells. Frequencies of T-cell subsets that were decreased in association with CREBBP mutation were not significantly decreased in association with mutations in other CMGs (P > 0.05; SI Appendix, Table S9), consistent with the lack of a significant difference in MHC class II expression on corresponding tumor B cells, nor were they significantly different in tumors with mutation of the immune gene TNFRSF14 (P > 0.05; SI Appendix, Table S9). Although EZH2 mutations were weakly associated with increased total T-cell number (P = 0.041) and memory cytotoxic T-cell number (P = 0.013), this is likely the result of the large degree of mutual exclusivity from CREBBP mutations in this dataset, as we also observed with MHC class II expression (SI Appendix, Fig. S10G). The observations were no longer significant (P > 0.05) when considering only CREBBP wild-type tumors, but associations between CREBBP mutation and T-cell frequencies remained significant (P < 0.05) when considering only EZH2 wild-type tumors. We found no significant differences between CREBBP mutant cases harboring homozygous compared with heterozygous allelic frequencies (P > 0.05), nor between CREBBP mutant cases harboring hotspot mutations compared with other variants (P > 0.05).

Patient Samples.

Follicular lymphoma (FL) tumor specimens were acquired as part of the Stanford University Lymphoma Program Project and cryopreserved. All specimens were obtained with informed consent in accordance with the Declaration of Helsinki and this study was approved by Stanford University's Administrative Panels on Human Subjects in Medical Research. The characteristics of patients in this study are described in SI Appendix, Table S1.

Cell Purification and Nucleic Acid Isolation.

BCL2 Translocation PCR.

The t(14;18)(q32;q21) translocation was detected using a nested PCR assay with multiplexed primers specific for the Ig heavy-chain J-region, major break region, minor cluster region, and internal cluster region (12). PCR products were visualized on a 2% (wt/vol) agarose gel, and size of the product used to determine the translocation breakpoint (SI Appendix, Fig. S6).

Next-Generation Sequencing and Somatic Mutation Hierarchy Construction.

For detailed methods, please refer to SI Appendix, Materials. In brief, next-generation sequencing (NGS) libraries were prepared from 1.5 µg sonicated genomic DNA, using TruSeq DNA sample preparation kits (Illumina) or KAPA Library Preparation kits (KAPA Biosystems). Libraries were enriched by hybrid capture for either the coding exome or a 284-gene custom-targeted panel, using SeqCap EZ Exome v3.0 or SeqCap EZ choice library capture reagents (NimbleGen). Criteria used for design of the 284-gene targeted capture panel can be found in the SI Appendix. Enriched libraries were sequenced with 101-bp paired-end reads on a HiSEq. 2000 instrument (Illumina). Average depths for exomes of tumor B cells and tumor-infiltrating T cells were 65× and 54×, respectively. Average depths for targeted capture were 267×.

Variants were called using Mutect (43), VarScan 2 (44), and GATK (12, 45) and annotated using SeattleSeq (46). Only those variants called twice within a single patient are reported. This approach allowed a sensitivity of 86.7% and a specificity of 92.9% (SI Appendix, Fig. S14). Evolutionary phylogenies were constructed using the somatic mutation hierarchy algorithm (20, 47) in 22 patients for whom multiple tumors were interrogated by exome sequencing, using all detected somatic mutations. Enrichment of mutations within EIPs was tested using a Fisher exact test with a Bonferroni correction for multiple hypothesis testing.

Protein Structural Models.

Suitable structural models for ATP6V1B2 (48) and CREBBP (49) were identified using the Protein Model Portal (50). Coordinate files of the models were downloaded from ModBase (51) and SWISS-MODEL (52), respectively. The figures were prepared with PyMOL software (Schrodinger).

Gene Expression Microarray Analysis.

Total RNA extracted from tumor B cells and tumor-infiltrating T cells was profiled using U133 plus 2.0 microarrays (Affymetrix). Raw cel files were RMA normalized and filtered for probes with maximum variance across tumor B cells and tumor-infiltrating T cells, and differential gene expression analysis was performed to identify the gene expression signatures of CMGs with five or more mutant tumors within the dataset (MLL2, CREBBP, MLL3, EP300, ARID1A, EZH2). For further detail, please refer to SI Appendix, Materials.

Gene Set Enrichment Analysis.

For GSEA of recurrently mutated genes, we used hypergeometric analysis (a statistical test to define the significance in overlap of two gene sets) for gene sets defined by gene ontology biological processes, using DAVID (53). For GSEA of gene expression profiling data, we used the GSEA-P tool (54). For further details, please refer to SI Appendix.

Modified Mixed Lymphocyte Reaction.

Cryopreserved tumor cell suspensions for three CREBBP wild-type and three CREBBP mutant cases in which CD10 expression had been previously determined to identify tumor cells (55) were thawed and counted. Cells were stained with CD5 (FITC), CD20 (PE), CD19 (APC), and CD10 (PE-Cy7) antibodies (BD Biosciences). Tumor B cells (CD5⁻CD19⁺CD10⁺) and tumor-infiltrating normal B cells (CD5⁻CD19⁺CD10⁻) were sorting using a FACS Aria II instrument (BD Biosciences). CD4⁺ T cells were obtained from peripheral blood of a single healthy male donor, using the RosetteSep Human CD4 T Cell Enrichment Mixture (Stemcell Technologies), and stained using the CellTrace Violet Cell Proliferation Kit (Life Technologies) according to the respective manufacturer’s instructions. Purified CD4 T cells were plated at 10,000 cells per well in a 96-well plate and cocultured with 20,000 cells per well of purified tumor B cells or tumor-infiltrating normal B cells in the presence of 1 µg/mL toxic shock syndrome toxin-1 (Toxin Technologies) in triplicate. The background of the assay was assessed using CD4 T cells with 1 µg/mL toxic shock syndrome toxin-1 and no B cells. The proliferative potential of isolated CD4 T cells was assessed using anti-CD3 (Clone OKT3; eBiosciences) and anti-CD28 (Clone CD28.2; eBiosciences) antibodies at concentrations of 0.5 and 5 µg/mL, respectively. Cells were incubated at 37 °C for 4.5 d and analyzed by flow cytometry on a Fortessa instrument (BD Biosciences), and proliferation of the CD4 T cells was assessed by dye dilution. The relative induction of proliferation by CREBBP wild-type and CREBBP mutant tumor B cells was compared statistically, using a Student’s t test of the median of triplicate experiments.

Flow Cytometry of HLA-DR Expression and T-Cell Subsets.

The measurement of HLA-DR expression on tumor and normal B cells and the enumeration of T-cell subsets were performed by flow cytometry of cryopreserved tumor cell suspensions using a Fortessa or LSRII instrument (BD Biosciences), respectively. Tumor specimens were thawed, counted, and stained with six to seven antibodies per panel. Exome sequencing cohort HLA-DR panel: HLA-DR (Pacific Blue), CD5 (FITC), Ig Lambda (PE), CD10 (PE-Cy7), CD19 (APC). Immunophenotyping cohort T-cell panel 1: CD3 (Pacific Blue), CD14 (FITC), CD10 (PE), CD5 (PE-Cy7), CD20 (PerCP-Cy5.5), an CD56 (APC). Immunophenotyping cohort T-cell panel 2: CD4 (Pacific Blue), CD3 (FITC), CD127 (PE), CD45RO (PE-Cy7), CD20 (PerCP-Cy5.5), and CD25 (APC). Immunophenotyping cohort T-cell panel 3: CD4 (Pacific Blue), CD8 (FITC), CD56 (PE), CD45RO (PE-Cy7), CD20 (PerCP-Cy5.5), CD137 (APC), and CD3 (Qdot 605). All antibodies were sourced from BD Biosciences except for HLA-DR (Pacific Blue) and CD3 (Qdot 605), which were sourced from BioLegend and Invitrogen, respectively. Before enumerating populations, all samples were gated for intact cells and nondoublets, using forward and side scatter. The relative HLA-DR expression in Fig. 3C was calculated by subtracting the mean fluorescence intensity of the T.I. normal B cells from the mean fluorescence intensity of the tumor B cells (SI Appendix, Fig. S12). The difference in population frequencies associated with CREBBP mutation status was tested by a Student’s t test with a Bonferroni correction for multiple hypothesis testing.