Toward Personalized Lymphoma Immunotherapy: Identification of Common Driver Mutations Recognized by Patient CD8⁺ T Cells

A fundamental challenge in cancer genomics is to design effective treatments that exploit the mutational profile of tumors. Next-generation sequencing (NGS) studies have revealed that tumors can contain tens to thousands of mutations (1). While some mutations affect common cellular pathways, there is often extensive mutational heterogeneity within and between patients (2). Adding to the challenge, many mutations or aberrant pathways are difficult to target pharmacologically because of either fundamental biochemical obstacles or a lack of selective agents.

An alternative strategy is to use mutation profiles to design personalized T-cell–based therapies. This approach relies on the exquisite sensitivity and specificity of the adaptive immune system. T cells can detect even single amino acid changes in mutated proteins; therefore, any nonsynonymous mutation has the potential to generate a new T-cell epitope. Moreover, T cells can recognize mutations in both intracellular and extracellular proteins, regardless of protein structure or function. Finally, T cells can be used to target multiple mutant proteins simultaneously. Thus, T-cell–based therapies offer an attractive approach to target the tumor genome.

Driver mutations (mutations that give cells a proliferative or survival advantage) represent the most attractive targets for any therapeutic modality. Indeed, T cells that recognize the common drivers BCR-ABL and mutant KRAS have been identified in patient blood samples, and peptide-based vaccines against these mutations have shown promise in the clinic (3–6). With the advent of NGS, it is now feasible to extend this concept to additional driver mutations, even those that are unique to individual patients. Indeed, many cancer clinics are now using “panel” approaches wherein commonly mutated genes are sequenced to guide the design of individualized therapies.

Although immunologic targeting of mutations offers several potential advantages, there are two major issues to consider. First, for a given mutation to be recognized by the immune system, several criteria must be met: (i) a peptide containing the mutant residue must be generated by the antigen processing machinery; (ii) the mutant peptide must bind with sufficient affinity to one or more of the patient's HLA alleles; and (iii) the patient must harbor T cells that recognize the mutant peptide–HLA complex (7). Owing to these factors, studies in melanoma and other cancers have found that only about 0.5%–4% of nonsynonymous mutations spontaneously trigger T-cell recognition (8–11). A second major issue is immune editing, whereby the immune system eliminates cells expressing specific antigens (12). As driver mutations are generally present at tumor initiation, their prolonged exposure to the adaptive immune system may make them particularly susceptible to immune editing. If this were the case, the only driver mutations present at diagnosis would be those that are immunologically “silent” (i.e., not processed or presented by the patient's HLA complex or recognized by their T-cell repertoire).

To investigate these issues in human cancer, we assessed the immunogenicity of a panel of genes that are common sites of driver mutations in follicular lymphoma. Follicular lymphoma is considered an immunologically sensitive disease, as tumor-infiltrating lymphocytes (TIL) are associated with increased patient survival (13, 14) and up to 30% of patients experience spontaneous tumor regressions (15). Moreover, at least 28 commonly mutated genes have been identified in follicular lymphoma, involving such processes as histone modification (e.g., CREBBP, EP300, EZH2, MEF2B, MLL2), signal transduction (e.g., CARD11, CD79B, STAT6), cell-cycle regulation (e.g., CCND3, PIM1), and cellular differentiation (e.g., FOXO1, IRF4; refs. 16–26). For the majority of these genes, the specific mutations are unique to individual patients. Here we demonstrate that patients with follicular lymphoma harbor CD8⁺ T cells with exquisite specificity for driver mutations in their tumors. Our results support the concept of using NGS data to design individualized immunotherapies for follicular lymphoma and other malignancies.

All specimens and clinical data were collected with written informed consent under protocols approved by the University of British Columbia-British Columbia Cancer Agency Research Ethics Board. Tumor tissue was obtained from biopsy specimens and cryopreserved or fixed in formalin and processed in paraffin. Peripheral blood mononuclear cells (PBMC) were purified from annual 200 mL blood draws using Ficoll-Paque PLUS (GE Healthcare) and cryopreserved in nitrogen vapor freezers. The frequencies of major immune cell subsets were determined by flow cytometry (Supplementary Table S1).

Genomic DNA was isolated from tumor specimens and matched PBMC using the AllPrep DNA/RNA Mini Kit (Qiagen). Targeted exon capture and sequencing was performed at Canada's Michael Smith Genome Sciences Centre, Vancouver, Canada (Supplementary Methods). High-resolution HLA class I typing of patient samples was performed using sequence-based approaches (27). HLA class I typing of healthy donor samples was performed using PCR-SSOP (ProImmune).

Epitope predictions were performed using NetMHCpan, version 2.8 (28). For 13 patients with adequate PBMC availability and at least one predicted epitope, all peptides with a predicted IC₅₀ binding score <1,000 nmol/L (and wild-type counterparts as required) were synthesized and purified commercially (ThinkPeptides and Genscript), reconstituted in 80% DMSO, and stored at −80°C.

Monocyte-derived dendritic cells were pulsed with peptides and cultured with autologous PBMC to activate peptide-specific CD8⁺ T cells (Supplementary Methods). Stimulations were performed in 96-well plates with 1.5 × 10⁵ PBMCs/well and an average of 413 wells or 6.2 × 10⁷ PBMCs/patient. After a second round of stimulation using irradiated, peptide-pulsed autologous PBMC as antigen-presenting cells (APC), T cells were screened for reactivity by IFNγ ELISPOT (Supplementary Methods). The frequency of responding T cells was determined by counting the number of positive wells and dividing by the number of input cells. Positive cultures were subjected to the Rapid Expansion Protocol (REP; ref. 29). In some cases, T cells were cloned by limiting dilution following FACS-based enrichment of peptide-stimulated cells expressing the activation marker CD137 (4-1BB).

Mutant and wild-type EZH2 cDNA, as well as wild-type CCND3 and MEF2B cDNA were provided by the Genome Sciences Centre. A mini-gene encoding 96 amino acids of CREBBP fused with a C-terminal flag-tag was synthesized commercially (Life Technologies). All sequences were cloned into pENTR/D-TOPO (Invitrogen). To recreate patient-specific mutations, wild-type CCND3, CREBBP, and MEF2B sequences were subjected to site-directed mutagenesis using the Transformer Site-Directed Mutagenesis Kit (Clontech). All sequences were cloned into the pST1-Gateway destination vector (30) downstream of the T7 promoter (Invitrogen) and verified by Sanger sequencing (Genscript). Plasmids were linearized after the poly(A) sequence with PmeI (New England Biolabs) and subjected to in vitro transcription using the mMessage mMachine T7 Ultra Kit (Ambion). RNA integrity was confirmed by gel electrophoresis. Lymphoma lines [BCWM.1 (31) and Karpas 422 (32)] and CD40-activated B cells (Supplementary Methods; 3 × 10⁶ cells) were electroporated with 1 to 18 μg of in vitro-transcribed RNA (ivtRNA) using the Amaxa Cell Line Nucleofector Kit L (Lonza, program X-001). BCWM.1 does not have an endogenous MEF2B^D83A mutation (unpublished data), and Karpas 422 does not have an endogenous CREBBP^R1446C mutation (33). Recombinant protein expression was confirmed by Western blotting (Supplementary Methods).

T-cell lines were pulsed with the relevant minimal peptide (10 μmol/L; to serve as targets) and incubated at a 1:1 ratio with unpulsed T cells (effectors) overnight at 37°C. Cells were stained with APC-labeled anti-human CD137 (BD Pharmingen, clone 4B4-1) and FITC-labeled anti-human CD8 (BD Pharmingen, clone RPA-T8) and analyzed using a BD Influx flow cytometer and FlowJo software.

The patient cohort consisted of 53 follicular lymphoma patients, whose clinical characteristics are shown in Supplementary Table S2. Using genomic DNA from biopsy specimens, we performed exon capture and NGS of 10 genes frequently mutated in non-Hodgkin lymphomas: CARD11, CCND3, CD79B, CREBBP, EP300, EZH2, FOXO1, IRF4, MEF2B, and PIM1 (16–20, 34). Nonsynonymous mutations were identified in samples from 81% of patients (Supplementary Table S3), with a mean of 1.7 mutated genes per case (range: 1–5). In accord with previous studies (16), the most frequently mutated genes were CREBBP, EZH2, and MEF2B, with mutations occurring in 20, 17, and 10 patients, respectively (Fig. 1A). FOXO1 has been reported as recurrently mutated in DLBCL (35) but not follicular lymphoma, yet it was the fourth most commonly mutated gene in our cohort. This may be partly attributable to poor coverage of FOXO1 in previous NGS data owing to high guanine-cytosine content (R.D. Morin and R.D. Gascoyne; unpublished data). The relative abundance of each mutation within a given tumor was inferred from the variant allele frequency (VAF; Table 1).

Epitope predictions were performed for each patient's mutation(s) against all 6 of their HLA class I alleles (Supplementary Table S4). Using an IC₅₀ cut-off of 1,000 nmol/L, predicted epitopes were identified for 37 of 43 patients (Supplementary Fig. S1), with a mean of 3.1 predicted epitopes per mutation and 5.3 predicted epitopes per patient (range: 0–26 epitopes). On the basis of PBMC availability and the presence of at least one predicted epitope, 13 patients were selected for in vitro immunogenicity experiments. The clinical characteristics and treatment protocols for this subcohort were representative of the larger cohort (Supplementary Table S2; Supplementary Fig. S2).

To assess whether predicted epitopes (Fig. 1B) could elicit T-cell responses, patient PBMCs from all available time points were subjected to two rounds of in vitro stimulation with autologous APC pulsed with a pool of purified, synthetic peptides corresponding to each patient's predicted epitopes. Stimulations were performed in 96-well plate microcultures containing an initial aliquot of 150,000 PBMCs/well, and an average of 6.2 × 10⁷ PBMCs (413 wells) were screened per patient. By IFNγ ELISPOT, T-cell responses were detected in samples from 7 of 13 patients (Table 1). The frequency of responding T cells ranged from 1:5 × 10⁶ to 1:9.7 × 10⁷ PBMCs (Table 1). All positive T-cell cultures were expanded, and reactivity to mutant peptides was confirmed by ELISPOT.

For the seven patients demonstrating responses, each positive T-cell response was further evaluated to determine the underlying mutation, minimal peptide, relevant HLA allele, and extent of cross-reactivity to corresponding wild-type peptides. T-cell responses were directed against peptides corresponding to mutations in four genes: CCND3, CREBBP, EZH2, and MEF2B (Table 2, Fig. 2A). From these genes, a total of 15 minimal peptides were recognized, along with 7 extended peptides containing these minimal peptides. Each peptide was recognized by between one and seven T-cell lines. To assess specificity, T-cell lines were screened for reactivity against mutant and corresponding wild-type peptides at concentrations ranging from 0.01 to 10,000 nmol/L. Eleven of 15 T-cell lines exhibited exquisite specificity for the mutant peptide over the corresponding wild-type peptide (Table 2, Fig. 2B–E), whereas the remainder cross-reacted to wild-type peptides (Table 2). To determine HLA restriction, T-cell lines were incubated with peptide-pulsed B-lymphoblastoid cell lines (LCL) matched at 0–6 HLA class I alleles (Supplementary Methods). HLA restriction was successfully determined for 9 of 15 peptides, and in each case the actual and predicted HLA alleles matched (Table 2; Fig. 2F).

As surgery is not part of the standard care for most follicular lymphoma patients, sufficient autologous target cells for tumor recognition experiments were not available. Instead, we evaluated T-cell recognition of mutant proteins after ectopic expression in autologous B cells and HLA-matched lymphoma cell lines. Constructs encoding full-length wild-type and mutant versions of CCND3, EZH2, and MEF2B were generated. CREBBP was too large to be expressed as a full-length protein, so instead it was expressed as a 96 amino acid polypeptide with the mutated residue located centrally. Constructs were transfected as ivtRNA into autologous CD40L-stimulated B cells. T-cell lines were tested by IFNγ ELISPOT for recognition of cells expressing the relevant mutant gene, the wild-type counterpart, an irrelevant gene, or no exogenous gene. For some patients, multiple T-cell lines exhibited the same peptide reactivity pattern; in such cases, the T-cell line with the highest apparent affinity was evaluated. T-cell lines from 3 of 7 patients exhibited reactivity against B cells expressing the relevant mutant protein (Table 2; Fig. 3). In each case, no significant cross-reactivity was seen against the corresponding wild-type or irrelevant proteins (Table 2; Fig. 3). All of these T-cell lines had shown specificity for mutant over wild-type peptides in the experiments described above. Conversely, none of the T-cell lines that had exhibited cross-reactivity to wild-type peptides were able to recognize full-length proteins (Table 2). By flow cytometry, we confirmed that 2 of 2 tested bona fide mutation-specific responses were mediated by CD8⁺ rather than CD4⁺ T cells (Supplementary Fig. S3). The third bona fide response could not be tested by flow cytometry because of the low frequency of mutation-specific cells in the T-cell line. However, this T-cell line recognized a 9-mer peptide with a strong predicted binding affinity for HLA-B*07:02 (IC₅₀: 4 nmol/L), so it is also likely to represent a CD8⁺ T-cell response.

The three bona fide T-cell responses were rare in terms of both precursor frequency and temporal dynamics. For patient IROL017, the T-cell response to MEF2B^D83A was detected at a frequency of 1:2.9 × 10⁷ PBMCs (1:7.3 × 10⁶ CD8⁺ T cells) in a pretreatment blood sample and was not found in any of three post-treatment samples (Supplementary Fig. S2). For patient IROL054, the T-cell response to MEF2B^D83G was detected at a frequency of 1:3.4 × 10⁷ PBMCs (1:6.8 × 10⁶ CD8⁺ T cells) in a blood sample taken during clinical remission (3 years postchemotherapy, 1.5 years postmaintenance rituximab therapy) and was not found at earlier or later time points (Supplementary Fig. S2). Finally, for patient IROL046, the T-cell response to CREBBP^R1446C was detected at a frequency of 1:4.2 × 10⁶ PBMCs (1:1.4 × 10⁶ CD8⁺ T cells) in a blood sample taken during clinical remission (6 years postchemotherapy, 4 years postmaintenance rituximab therapy) and was not found at 3 earlier (posttreatment) time points (Supplementary Fig. S2). These low precursor frequencies suggest that the T cells may have been derived from the naïve repertoire.

To ensure that transfection resulted in physiologically relevant protein expression levels, RNA titration experiments were performed. IROL017 B cells were transfected with 1 to 18 μg of RNA encoding MEF2B^D83A. By Western blot analysis, 9 μg of RNA gave rise to a similar level of protein expression as seen for endogenous MEF2B in the DB lymphoma line (Fig. 4A). By ELISPOT, IROL017 T cells recognized cells transfected with as little as 3 μg of RNA (Fig. 4B), indicating that recognition was in the physiologic range. In an analogous experiment, IROL046 T cells could recognize autologous B cells transfected with as little as 1 μg of RNA encoding CREBBP^R1446C (Fig. 4C).

To determine whether lymphoma cells had the capacity to process and present the antigens of interest, we expressed the same constructs in HLA-matched lymphoma cell lines, where available. The bona fide epitope recognized by T cells from patient IROL017 corresponded to a peptide from MEF2B^D83A presented by HLA-B*40:01; therefore, the HLA-B*40:01–positive lymphoma cell line, BCWM.1 was transfected with RNA-encoding mutant MEF2B. Similarly, T cells from patient IROL046 recognized a peptide from CREBBP^R1446C presented by HLA-B*07:02; therefore, the HLA-B*07:02–positive lymphoma line Karpas 422 was transfected with RNA-encoding mutant CREBBP. In both cases, patient T cells recognized lymphoma lines expressing the appropriate mutant protein (Fig. 4D and E), demonstrating that these mutant epitopes can be processed and presented by the MHC class I pathway in lymphoma cells to confer recognition by CD8⁺ T cells.

We systematically investigated the extent to which autologous CD8⁺ T cells recognize patient-specific driver mutations in follicular lymphoma. By interrogating a panel of 10 genes in 53 follicular lymphoma cases, we identified mutations in 81% of cases. Of relevance to immunotherapy, T cells specific for bona fide mutant epitopes were identified in 23% (3/13) of patients. One of the corresponding mutations (CREBBP^R1446C) is a well-established driver mutation in follicular lymphoma (18). The other two mutations (MEF2B^D38A and MEF2B^D38G) involve a commonly mutated residue in MEF2B (19); these too are likely to represent driver mutations based on amino acid similarities to the well-characterized MEF2B^D83V driver mutation (36). Mutation-specific T cells were present at very low frequencies in peripheral blood, yet they could be expanded to large numbers using clinically feasible methods. Thus, driver mutations represent attractive target antigens for T-cell–based immunotherapy of follicular lymphoma and related malignancies.

Our study used an in vitro priming approach that, in theory, interrogated both the naïve and activated T-cell repertoires of cancer patients for mutation-specific responses. Encouragingly, our findings suggest that a higher proportion of mutations can be recognized by autologous T cells than previously thought based on analyses of pre-existing T-cell responses alone (37). For example, in anecdotal studies of melanoma patients who responded well to TIL therapy or ipilimumab, TIL were shown to recognize only about 2% of mutations identified by whole exome sequencing (8, 9, 38). Similarly, in a study of three ovarian cancer patients, we found that TIL recognized only 1.3% (1/78) of mutations (11). Finally, in a non–small cell lung cancer patient who responded to pembrolizumab, only one mutation was recognized out of 226 putative neoantigens (39). In contrast, by using in vitro priming in the current study, we identified bona fide T-cell responses to 11% (3/27) of mutant proteins in 23% (3/13) of follicular lymphoma patients tested. Likewise, in two patients with chronic lymphocytic leukemia, in vitro priming revealed T-cell responses against 12% (3/25) of mutations tested (although MHC class I processing and presentation was assessed for only one of these peptides; ref. 40). Thus, although the sample sizes are small in each of these studies, the combined results suggest that, by using sensitive methods that allow detection of naïve and/or rare T cells, one can identify T-cell responses to a larger proportion of mutations than previously appreciated. This conclusion is supported by murine studies, where vaccination with mutant peptides can prime CD4⁺ and CD8⁺ T-cell responses to approximately 30% of selected mutations (41, 42).

Although T-cell–based targeting of driver mutations offers several advantages over pharmacologic strategies, our findings highlight two major challenges that necessitate an individualized approach. First, despite the existence of commonly affected genes and pathways, follicular lymphoma demonstrates a considerable interpatient mutational heterogeneity. For example, by interrogating 10 commonly mutated genes, we identified mutations in 81% of cases. About half (46%) of these mutations were located in hotspots in 8 genes; however, the remaining mutations were unique to individual cases. Thus, NGS will be required to identify the relevant mutations in each patient's tumor. Second, as expected from basic immunologic principles, the patient's HLA status was a critical determinant of the immunogenicity of mutations (7). For example, of the two cases with a CREBBP^R1446C substitution, a corresponding T-cell response was detected in only the HLA-B*07-02–positive individual. The prediction algorithm used here worked well in the case of CREBBP, as it gave a better binding score for the HLA-B*07-02 allele (IC₅₀: 4 nmol/L) than for the HLA alleles from the other patient (IC₅₀ scores: 177–976 nmol/L). In fact, applying an IC₅₀ cut-off of 165 nmol/L would have correctly identified the 3 bona fide epitopes, while identifying only 30 irrelevant peptides (5 of which elicited responses to peptides that were not processed). The fact that all three bona fide epitopes had predicted IC₅₀ values ≤165 nmol/L indicates that, despite being rare, the mutation-reactive T cells identified are expected to have T-cell receptor (TCR) affinities within a therapeutically relevant range. Given the importance of HLA haplotypes, clinical trials will likely be more effective if candidate epitopes are subjected to in vitro immunologic assays before being selected for clinical use. The need for empirical testing should diminish as protein processing and epitope prediction algorithms continue to improve.

Despite the need for an individualized approach overall, our results also illustrate that recurrent mutations occasionally give rise to shared epitopes presented by common HLA alleles. For example, the CREBBP^R1446C mutation discussed above has a prevalence of 4.3%–13% in follicular lymphoma (Supplementary Table S3; ref. 18), and the corresponding HLA-B*07-02 allele is present in 30% of American Caucasians (43). Therefore, the mutant epitope we identified may be a relevant target in up to 4% of patients with follicular lymphoma. The two remaining bona fide epitopes were derived from substitutions at D83 of MEF2B, which occur in 4.2–5.7% of follicular lymphoma cases (Supplementary Table S3; ref. 19); although D83V is the most common substitution at this site, the alanine and glycine substitutions found here are also recurrent mutations (19). The two corresponding HLA alleles (HLA-B*40 and HLA-A*03) are present in 13% and 21% of American Caucasians, respectively (43), and are predicted to bind mutant epitopes containing D83V, D83A, and D83G with equivalent affinity. On the basis of these estimates, substitutions at D83 of MEF2B may give rise to CD8⁺ T-cell epitopes in approximately 1.9% of patients with follicular lymphoma. Thus, in this relatively small study, 3 of 3 bona fide epitopes identified have the potential to serve as shared epitopes with therapeutic relevance to additional patients with follicular lymphoma. With further study, it may be possible to create an inventory of shared mutant epitopes that could partly alleviate the need for entirely personalized mutation targeting.

Taking into consideration mutation frequency, epitope predictions, and in vitro stimulation results, we estimate that approximately 1 of 6 patients with follicular lymphoma would be amenable to T-cell–based therapy targeting this panel of 10 mutations. The number of eligible patients could be increased in several ways. There are at least 18 other common driver mutations in follicular lymphoma that could be added to the panel used here (16, 19). Moving beyond driver mutations, the average lymphoma contains approximately 100 mutations (1, 16), which could be readily identified by whole exome sequencing. Finally, the proportion of eligible patients could be increased by assessing both CD8⁺ and CD4⁺ (T helper 1, or Th1) T-cell responses. In the current study, we elected to focus on CD8⁺ T-cell responses owing to the greater accuracy of epitope prediction algorithms for MHC class I compared with class II. However, CD4⁺ T-cell responses also play important roles in antitumor immunity. Indeed, recent studies have shown a high prevalence of mutation-specific CD4⁺ T-cell responses in murine tumor models (42) and human melanoma patients (44). Moreover, infusion of TIL enriched for a mutation-specific CD4⁺ Th1 clone led to a dramatic clinical response in a patient with cholangiocarcinoma (10).

One can envision several therapeutic strategies for inducing immune responses to patient-specific driver mutations. Arguably the simplest strategy would be checkpoint blockade using antibodies to inhibitory proteins such as PD-1 and CTLA-4. In support of this, ipilimumab appears preferentially to induce new T-cell responses rather than to enhance preexisting responses (45). Moreover, immune checkpoint blockade appears to be more efficacious against tumors with high mutation load (39, 46). However, the increasing use of NGS in clinical practice offers an opportunity to generate more precise antitumor immune responses. For example, NGS data are currently being used to design personalized mutation-specific vaccines (47). However, the low T-cell precursor frequencies reported here suggest that it may prove difficult to generate robust, mutation-specific T-cell responses in patients through vaccination alone.

Adoptive T-cell therapy (10, 48) allows one to circumvent issues of low precursor frequency, cross-reactivity, and immune suppression by isolating and expanding T cells from peripheral blood to produce defined T-cell products with high specificity for target mutations. Recent adoptive cell therapy trials using TIL specifically selected for mutation reactivity (or retrospectively shown to be mutation-reactive) underscore the clinical feasibility and efficacy of this strategy (8–10). Moreover, in the setting of lymphoma, adoptive transfer of T cells engineered to express CD19-specific chimeric antigen receptors (CAR) is showing promise in early clinical trials (49, 50). Notably, CD19-CAR T-cell therapy can induce prolonged B cell aplasia, neurologic toxicities, and cytokine release syndrome, in part due to destruction of the normal B-cell compartment. These risks could potentially be mitigated by targeting tumor-specific mutations rather than the pan-B cell marker CD19. Finally, our results suggest that for the most common driver mutations in follicular lymphoma, such as CREBBP^R1446C, it may be practical to produce off-the-shelf mutation-specific TCR vectors for use in T-cell engineering. TCR engineering would also be required in cases where mutation-specific T cells fail to expand to sufficient quantities for ACT. Irrespective of the precise approach, our results provide proof-of-concept for T-cell–based targeting of common driver mutations in follicular lymphoma and other malignancies.

J. Nielsen, B. Nelson, N. Macpherson, R. Morin, and C. Sedgwick are co-inventors on US Provisional Patent No. 62162366 entitled Compositions and Methods for Immunotherapy of Cancer, which is owned by British Columbia Cancer Agency. R.D. Gascoyne reports receiving speakers bureau honoraria from Seattle Genetics, and is a consultant/advisory board member for Celgene and Janssen. No potential conflicts of interest were disclosed by the other authors.

Conception and design: J.S. Nielsen, N. Macpherson, B.H. Nelson

Development of methodology: J.S. Nielsen, C. Sedgwick, Z.L. Brumme, S.P. Treon, M.A. Marra, R.D. Morin, N. Macpherson, B.H. Nelson

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): J.S. Nielsen, C. Sedgwick, A. Shahid, Z.L. Brumme, S. Yu, L. Liu, D.R. Kroeger, J.M. Connors, R.D. Gascoyne, B.R. Berry, R.D. Morin, N. Macpherson

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): J.S. Nielsen, C. Sedgwick, Z.L. Brumme, J.M. Connors, R.D. Morin, B.H. Nelson

Writing, review, and/or revision of the manuscript: J.S. Nielsen, C. Sedgwick, A. Shahid, Z.L. Brumme, D.R. Kroeger, J.M. Connors, R.D. Gascoyne, B.R. Berry, M.A. Marra, R.D. Morin, N. Macpherson, B.H. Nelson

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): J.S. Nielsen, S. Yu, D.R. Kroeger, S.P. Treon, J.M. Connors, R.D. Gascoyne, N. Macpherson, B.H. Nelson

Study supervision: J.S. Nielsen, R.D. Morin, N. Macpherson, B.H. Nelson

Other (part of Dr. Zabrina Brumme Laboratory, Simon Fraser University. Performed HLA typing of the patients enrolled in this study and the responses above reflect on this contribution. Also reviewed the manuscript before submission): A. Shahid

Other (library construction and next generation sequencing): Z. Zong

Other (Her laboratory performed the HLA class I typing of study patients. The responses to the above specifically refer to this contribution. Also critically reviewed the manuscript): Z.L. Brumme

Other (provided study materials, and input on study design): S.P. Treon

The authors thank the study participants and referring clinicians for specimens and clinical data; Adria Devlieger, Tim Turcotte, and Natalie Kinloch for technical assistance; and Drs. Richard Moore, John Webb, and Robert Holt for advice. Canada's Michael Smith Genome Sciences Centre provided sequencing data.

This work was supported by the Canadian Cancer Society (grant #701789), the British Columbia Cancer Foundation (BCCF), and the Whittle Family Multiple Myeloma Research Fund. J.S. Nielsen received funding from the Lymphoma Foundation Canada. Z.L. Brumme is the recipient of a Canadian Institutes of Health Research (CIHR) New Investigator Award and a Michael Smith Foundation for Health Research Scholar Award. R.D. Gascoyne, M.A. Marra, and J.M. Connors are supported by a Terry Fox Research Institute New Frontiers in Cancer Program Project Grant (# 1023). J.M. Connors receives research funding from Genome British Columbia, Genome Canada, CIHR, and BCCF. RDM is supported by start-up funds provided by the BCCF Innovator Support Fund and a CIHR New Investigator Award.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.