Tumor- and Neoantigen-Reactive T-cell Receptors Can Be Identified Based on Their Frequency in Fresh Tumor

The presence of lymphocytes infiltrating into the tumor stroma (tumor-infiltrating lymphocytes; TIL) has been associated with a favorable prognosis in melanoma (1) and other cancer types, including ovarian (2), colon (3), and breast cancer (4). In melanoma, in vitro analysis of expanded TILs revealed a broad specificity of antigen recognition, including melanoma/melanocyte shared differentiation antigens (5–7), cancer germline antigens (8, 9), and mutated neoantigens unique to each patient's tumor (10–12).

Adoptive cell therapy using autologous TILs is an immunotherapeutic approach capable of inducing complete durable regression in 20% of patients with metastatic melanoma (13). However, TILs used for treatment undergo extensive in vivo and in vitro expansion, becoming highly differentiated cells with limited additional proliferative potential (13, 14). Control over which T-cell clonotypes expand in vitro is limited, so the TCR clonotypic repertoire present in the tumor can be altered, potentially leading to decreased frequencies of tumor-reactive clonotypes.

To overcome these problems, we focused our attention on the TCR clonotypes present in the tumor before any in vitro expansion. In melanoma, tumor-specific clonotypes are highly enriched in the fresh CD8⁺PD-1⁺ TIL subset (15, 16), which we hypothesize could be due to the oligoclonal expansion that occurs when T cells encounter their specific antigen in the tumor microenvironment in vivo (17), leading to the presence of predominant clonotypes within this population. Thus, the frequency of a clonotype within the TIL repertoire may indicate its tumor reactivity. To test this hypothesis, we analyzed the TCR repertoire of TILs in freshly resected tumors from 12 patients with metastatic melanoma and found that many of the most frequent TCR clonotypes present in the CD8⁺PD-1⁺ TIL subset recognized the autologous tumor and either mutated or nonmutated tumor antigens. Thus, it may be possible to efficiently identify tumor-reactive TCRs based solely on their frequency and PD-1 expression in the tumor. This strategy could provide an efficient means to obtain tumor-reactive TCRs that could be genetically engineered into autologous cells with high proliferative potential for use in cell therapy.

Twelve metastatic melanoma samples were obtained from patients who were not undergoing therapy at the time of sample collection. Patients had undergone a wide range of prior therapies, including surgery, chemotherapy, radiotherapy, immunotherapy, or none of the above. Peripheral blood lymphocytes (PBL) were obtained by either leukapheresis or venipuncture, prepared over Ficoll–Hypaque gradient (LSM; ICN Biomedicals Inc.), and cryopreserved until analysis. After surgical resection, tumor specimens were processed as previously described (18). Briefly, tumor specimens were minced, enzymatically digested overnight at room temperature or for several hours at 37°C [RPMI-1640 with l-glutamine (Lonza), 1 mg/mL collagenase IV (Sigma-Aldrich), 30 U/mL DNAse (Genentech), and antibiotics], and the tissue was separated mechanically using gentleMACS (Miltenyi Biotech). Tumor single-cell suspensions were cryopreserved.

Genomic DNA purification, library construction, exome capture of approximately 20,000 coding genes, next-generation sequencing of fresh tumor embedded in optimal cutting temperature (Sakura Finetek), and a matched normal pheresis sample were performed as previously described (19). An mRNA sequencing library was prepared from fresh tumors using the Illumina TruSeq RNA library prep kit, as previously described (20). Putative nonsynonymous mutations are defined by ≥3 exome variant reads, ≥8% variant allele fraction (VAF) in the exome, and ≥10 reads in the matched normal sample. Putative mutations with a variant allele frequency of >10% in the tumor exome, as well as mutations that were identified in both transcriptome and exome analysis are initially selected for screening. For some samples (i.e., 3903), the mutations selected based on exome were prioritized only by selecting those with >10 variant reads to increase the confidence of mutation calling.

Fluorescently conjugated antibodies were purchased from eBioscience [MIH-4, Anti-Human CD279 conjugated to allophycocyanin (APC) and anti-mouse TCRβ-fluorescein isothiocyanate (FITC)], Miltenyi (4B4-1, anti-human CD137-PE or -APC), BioLegend [anti-human CD8-phycoerythrin (PE)-Cy7, anti-human CD3-APC-Cy7]. For phenotypic characterization and cell sorting of CD8^+/−, CD8⁺PD-1^+/− T-cell tumor samples were thawed and rested overnight without cytokines (15). The T cells were sorted by flow cytometry with a modified FACSAria instrument or a BD Jazz instrument (BD Biosciences), gates were set according to isotype and fluorescence minus one controls. The sorting strategy is shown for two representative fresh melanoma samples (3903 and 3998) in Supplementary Fig. S1.

The T cells were sorted by flow cytometry in comparable numbers for each subset: 100,000 cells for the tumor single-cell suspension bulk TILs, 10,000 cells for the CD8⁺ and CD8⁻ subsets, and 1,000 to 3,000 cells for the CD8⁺PD-1⁺ and CD8⁺PD-1⁻ subsets. The cells were pelleted and snap frozen. The samples were sent to Adaptive Technologies for genomic DNA extraction and ImmunoSEQ TCRB survey sequencing. Tumor samples were sent to Adaptive Technologies for pairSEQ (21), 1 × 10⁶ total cells from tumor single-cell suspension were pelleted in a table top centrifuge at 6,000 rpm for 30 minutes, resuspended in 200 μL of RNAlater (Invitrogen) and snap frozen.

Single-cell sorting was performed using a modified FACSAria instrument or BD Jazz instrument (BD Biosciences) on CD8⁺PD-1⁺ TILs; for samples 1913, 2650, 3713, and 3784 CD8⁺-expanded TILs were used due to limited availability of tumor samples. TCR sequences from the sorted single cells were obtained by a series of two nested PCR reactions. Cells were sorted into RT-PCR buffer. For the first, reverse transcription and amplification reaction were performed with a One-Step RT-PCR kit (Qiagen) using multiplex PCR with multiple Vα and Vβ region primers and one primer for Cα and Cβ regions each (final concentration of each primer is 0.6 μmol/L). The RT-PCR reaction was performed according to the manufacturer's instructions using the following cycling conditions: 50°C for 15 minutes; 95°C for 2 minutes; 95°C for 15 seconds, 60°C for 4 minutes × 18 cycles; 4°C. For the second amplification reaction, 4 μL from the first RT-PCR was used as a template in total 25 μL PCR mix with HotStarTaq DNA polymerase (Qiagen) and multiple internally nested Vα and Vβ region primers and one internally nested primer for Cα and Cβ region each (final concentration of each primer was 0.6 μmol/L). The cycling conditions were 95°C for 15 minutes; 94°C for 30 seconds, 50°C for 30 seconds, 72°C for 1 minute × 50 cycles; 72°C for 10 minutes; 4°C. The PCR products were purified and sequenced by the Sanger method with internally nested Cα and Cβ region primers by Beckmann Coulter. All primers are listed in Supplementary Table S1.

In both pairing methods (single-cell RT-PCR and pairSEQ), cDNA is used as template for multiplex PCR using TCRA and TCRB gene–specific primers. The resulting PCR product contains the 3′ end of the variable region and the full CDR3 region of matching TCRA and TCRB genes. These partial TCR sequences were analyzed with IMGT/V-Quest tool (http://www.imgt.org/) which identified the TRAV and TRBV families with the highest likelihood to contain the segment found with our pairing methods. Utilizing the IMGT database, we reconstructed the full-length TRAV and TRBV regions for each pair. In regard to the constant regions, we used modified murine TRAC and TRBC sequences to improve stability and avoid mismatches with the endogenous human TCR after transduction into human T cells (22). Full TCR genes were synthetized, and a 2A peptide (23) was introduced between the TCRB and TCRA chains to ensure a comparable expression efficiency of the two chains. The resulting TCRB–TCRA gene blocks were cloned into either a gamma-retroviral expression vector (24, 25) or for the following TCR pairs: 2650-1, 2650-3, 2650-4, 2650-5, 2650-6, 2650-7, 2650-9, 3903-3A1, 3903-3A2, 3992-1, 3992-2, 3992-3, 3992-4, 3992-5, and 3998-1 into a nonviral Sleeping Beauty transposon system (26, 27). The expression of the TCRB was evaluated with an anti-murine TCRB Ab.

Melanoma tumor cell lines (TC 1913, TC 2630, TC 2650, TC 3678, TC 3713, TC 3759, TC 3784, TC 3903, TC 3922, TC 3926, TC 3977, TC 3992, and TC 3998) were established from tumor fragments or from mechanically or enzymatically separated tumor cells and cultured in RPMI 1640 plus 10% FBS (Sigma-Aldrich) supplemented with 100 U/mL penicillin and 100 μg/mL streptomycin at 37°C in 5% CO₂. COS-7 cells and COS-7 cells stably transduced with HLA molecules were maintained in DMEM containing 10% FBS (Sigma-Aldrich) supplemented with 100 U/mL penicillin and 100 μg/mL streptomycin at 37°C in 5% CO₂. TC 1913 and 1913 tumor–specific neoantigen recognition was previously reported (28). Generation of tandem minigene (TMG) constructs and autologous antigen presenting cells (dendritic cells and CD40L stimulated B cells) was done as previously described (11, 29). Briefly, up to 14 nonsynonymous mutations identified by whole-exome sequencing and RNAseq, each flanked by 12 amino acids of nonmutated protein, were genetically fused together to generate a TMG construct. These constructs were codon optimized, synthesized, and cloned into pcDNA3.1/V5 by Genescript. Autologous antigen-presenting cells were peptide pulsed for 2 hours with 1 μg/mL short peptides (9–10 mers) and overnight with 10 μg/mL of long peptides (25 mers) before coculture.

CD137 upregulation was used to measure target recognition by transduced T cells. CD137 is upregulated transiently in response to TCR stimulation, regardless of the effector cytokines produced or the differentiation state of the cell (30). We used the coexpression of murine TCR constant chain (identified as mTCR) and CD137 to identify the population of transduced antigen-reactive T cells (to be considered reactive the CD137 upregulation had to be greater than 1%, 3 times the background, and inhibited at least 50% by pan–MHC I blocking antibody, clone W6-32). Cells were stained with anti-CD3, anti-CD8, anti-CD137, and anti-murine TCRB antibodies after coculture and acquired by Fortessa (BD Biosciences). Data were analyzed with FlowJo software (Treestar).

Wilcoxon signed-rank test was used to determine the statistical significance of the data. P values of 0.05 or less were considered statistically significant. Statistical calculations were performed with Prism program 6.0 (GraphPad Software Inc.).

All patient samples were obtained in the course of a National Cancer Institute Institutional Review Board–approved clinical trial. Patients provided informed consent.

To characterize the TIL TCR clonotypic repertoire and identify tumor-reactive TCRs, we developed a multistep strategy (Fig. 1A). We first assessed the composition of TILs from 12 fresh metastatic melanoma lesions by flow cytometry (Table 1). The samples varied considerably in the frequency of CD8⁺ and CD8⁻ lymphocytes (P = 0.15), although the frequency of PD-1 expression, which is a marker for T-cell activation (16, 31), was usually higher on the CD8⁺ TILs (P = 0.003). TCRB deep sequencing is a robust method for quantifying the frequency of each T-cell clonotype present in different sample types (32–34), which we used to determine whether the CD8⁺PD-1⁺ TIL subset displayed evidence of clonal expansion. Genomic DNA extracted from bulk melanoma TILs and from sorted subsets (CD8⁺, CD8⁻, CD8⁺PD-1⁺, and CD8⁺PD-1⁻) was deep sequenced to determine the number of unique productive (35) TCRB CDR3 sequences that do not contain stop codons or frame shifts (Fig. 1B) in the 10 patients from whom all samples were available. These unique sequences represent a single unique clonotype independent of its frequency in the samples. Different samples can thus have comparable numbers of total reads (Fig. 1C), but it is the different number of unique sequences that determines the level of clonality of each sample. We found a significantly lower number of unique productive sequences present in the CD8⁺ compared with the CD8⁻ subsets (P = 0.002, Fig. 1B). Within the CD8⁺ lymphocytes, the PD-1⁺ population contained a lower number of unique productive sequences compared with PD-1⁻ cells (P = 0.002, Fig. 1B). We also compared the levels of clonal diversity (measured by Shannon entropy; ref. 36) for the samples studied and we found more diversity in the CD8⁻ subset compared with the CD8⁺ (P = 0.002) and within the CD8⁺ lymphocytes more diversity in the PD-1⁻ cells compared with the PD-1⁺ subset (P = 0.002, Supplementary Fig. S2A). We also compared the number of nonsynonymous mutations with the percentage of CD8⁺PD-1⁺ TILs and TCR clonal diversity in different TIL subsets (Supplementary Fig. S2B–S2E), but found no significant correlation. Each highly expressed individual clonotype in the CD8⁺PD-1⁺ subset was much less frequent in the PD-1⁻ group (P = 0.0003; Supplementary Fig. S3A), confirming that PD-1 is a marker that separates TILs into two separate subsets with different TCR repertoires (15). The same clonotypes that were highly expressed in the CD8⁺PD-1⁺ subset were found at low frequency in the total CD8⁺ subset (Supplementary Fig. S3B, P = 0.001). The highest expressers in the PD-1⁺ subset though were often also high ranked in the bulk TILs (Supplementary Table S2).

To reconstruct a functional TCR, the most frequent TCRB chains present in the CD8⁺PD-1⁺ TIL must be paired with the appropriate TCRA chains. One possible approach was to match the most frequent TCRB clonotype with the most frequent TCRA clonotype. However, in the 6 patient samples for which we ultimately matched the correct TCRA with the most frequent TCRB and determined their functionality, the most frequent TCR chains paired together in 3 cases, and the first TCRB in the other 3 cases paired with the 7th, 18^th, and 47th TCRA (Supplementary Table S3). Tumor recognition by the first TCRB with the first TCRA occurred in the 3 cases in which the first TCRB clonotype was present in >20% of the PD-1⁺ population. Discordance of pairing based on frequency of the TCRA was likely due to the presence of more than one α-chain in some cells (37) as well as the variable efficiency of primers used in the TCRA sequencing. Thus, we decided to identify the productive TCR pairs with two different approaches. After identifying the most frequent TCRBs in the CD8⁺PD-1⁺ population, we identified the corresponding TCRA with single-cell RT-PCR on CD8⁺PD-1⁺ FACS-sorted TILs or CD8⁺ TILs expanded in vitro. Alternatively, we used the pairSEQ approach (21) on single-cell suspensions from unsorted fresh tumors. The efficiency of the single-cell RT-PCR was between 26% and 90%, depending on the sample. Using this method, we identified a median value of 29 (range, 9–43) unique TCRA–TCRB pairs in each of the CD8⁺ or CD8⁺PD-1⁺ samples. Using pairSEQ on unsorted fresh tumors we identified a median value of 217 (range, 11–883) unique pairs for each sample. A total of 93 (median value 6, range 0–21) TCRA–TCRB pairs from 12 metastatic melanoma patients were identified using both methods (congruent pairs in Table 2; method of identification for each pair in Supplementary Table S4). We generated expression vector constructs encoding the 83 of these pairs ranked within the top 10 CD8⁺PD-1⁺ clonotypes and linked them to murine constant chain sequences to improve stability and avoid mismatches with endogenous human TCRs (22), and then introduced them into fresh PBLs. The frequency of T cells that expressed the recombinant TCRs after either retroviral transduction or transfection with a Sleeping Beauty transposon construct ranged between 24.4% and 97.6% (Supplementary Table S5).

We then evaluated the antitumor activity of T cells expressing those 83 TCRA–TCRB pairs (Table 2). The TCRs obtained from 10 of the 12 patients were evaluated for response to candidate neoepitopes identified by whole-exome sequencing of autologous tumor (TCRs from samples 2650 and 3977 were evaluated only against the TC line; Table 1). All the TCR pairs were also tested against autologous or HLA-matched antigen-presenting cells transfected with full-length RNA encoding the melanoma/melanocyte shared differentiation antigens MART-1, gp100, and tyrosinase (TYR) and the cancer-germline antigens NY-ESO-1, MAGEA3, and SSX2.

Evaluation of response against the corresponding autologous TC and/or autologous antigen-presenting cells that had either been pulsed with mutated tumor-specific neoantigen minimal epitopes or transfected with TMG (10–12) constructs provided evidence for tumor antigen reactivity in 11 of the 12 patients who were evaluated.

For example, for patient 3998 we initially evaluated the reactivity of some of the top eight most frequent TCR pairs based on the frequency of TCRB (3998-1, 3998-2, 3998-3A1, 3998-3A2, 3998-4, 3998-6, 3998-7, and 3998-8) against the autologous TC (Fig. 2A and C). Six of the TCR pairs tested (3998-1, 3998-2, 3998-4, 3998-6, 3998-7, and 3998-8) showed MHC-restricted recognition of the autologous tumor. The TCRB clonotype ranking third in frequency in the CD8⁺PD-1⁺ TILs was associated with two productive TCRA chains but neither of the two combinations (3998-3A1 and 3998-3A2) were tumor reactive. For tumor sample 3998, 345 nonsynonymous mutations were identified (Table 1). We next evaluated the reactivity of the 6 TCR pairs (3998-1, 3998-2, 3998-4, 3998-6, 3998-7, and 3998-8) against 115 mutated antigens encoded by 7 TMGs (Fig. 2B) and six shared melanoma/melanocyte differentiation antigens and cancer-germline antigens (MART-1, gp100, SSX2, TYR, NY-ESO-1, and MAGEA3; Supplementary Fig. S4A and S4B). The 115 mutated antigens were selected for screening from the 345 nonsynonymous mutations based on RNAseq data of their expression. Two TCR pairs (3998-7 and 3998-8) were reactive to TMG-1 (Fig. 2B). Further testing identified MAGEA6^E168K as the specific mutation recognized within the antigens encoded by TMG-1 (Fig. 2D and Supplementary Fig. S4C). Reactivity against one shared antigen (NY-ESO-1) was found for TCR pair 3998-5 (Table 3 and Supplementary Fig. S4B).

Figure 3 summarizes the findings of all 12 samples. Representative cocultures for all of the tumor samples are shown in Supplementary Figs. S4–S15. For example, in sample 1913 the TCRB rankings 2, 3, and 4 were specific for the autologous TC line, and the clonotypes ranking 2 and 4, as previously found (15, 28), also recognized a mutation in the HLA-11 gene (Table 3). Moreover, the most frequent TCR clonotype was found to be tumor reactive for seven samples (2650, 3759, 3903, 3922, 3926, 3977, and 3998). For all but patient 3992, up to five tumor-reactive TCRs were found among the five most frequently expressed TCRs in the CD8⁺PD-1⁺ TILs. Reactivity against autologous neoantigens was found in 5 of the 10 patients whose TCRs were screened against putative autologous mutations. In summary, we found that 36 TCR pairs were reactive against autologous tumor and 11 were directed against mutated tumor-specific neoantigens. This finding indicates that it is possible to identify tumor-reactive TCR pairs in the majority of melanoma samples simply based on their frequency in the CD8⁺PD-1⁺ TIL compartment.

We have shown that adoptive cell therapy with TILs that appeared to predominantly recognize patient-specific tumor neoantigens (11, 12) or T cells genetically engineered to express TCRs targeting cancer-germline antigens (8) can mediate complete response in patients with metastatic melanoma (10, 12) and in a patient with metastatic cholangiocarcinoma (29). Those studies used a labor-intensive screening approach with tandem minigenes or long peptides representing all known mutations and could identify T cells with reactivity against mutated neoantigens. Tumor-reactive TCRs expressed by mutation-reactive T cells can be isolated, cloned into expression vectors, and can potentially be transferred into autologous cells with high proliferative capacity (13, 38) for use in cell transfer therapy. Here, we demonstrate the identification of tumor and mutation-reactive TCRs from fresh melanoma samples, based on PD-1 expression and on TCRB frequencies as a guide to tumor reactivity.

The development of next-generation TCRB sequencing has allowed a study of the total TCR repertoire in different T-cell compartments in healthy individuals (32) as well as in tumor samples from colorectal (33) and ovarian carcinomas (34). In the present study, we have analyzed the TCR diversity in different subsets of TILs from freshly resected human metastatic melanomas and attempted to determine whether the rank frequency of TCRs was related to their ability to recognize the autologous cancer. This strategy can be used to prospectively identify tumor-reactive TCRs without the prior need to know their antigen specificity.

The major obstacle encountered in the evaluation of the antitumor reactivity of individual TCRs in the fresh tumor prior to in vitro expansion was the technical difficulty in pairing each high-frequency TCRB chain with the correct TCRA. In the present study, we utilized two independent approaches to identify the TCRA–TCRB pairs, single-cell RT-PCR on a specific subset (CD8⁺PD-1⁺ TILs or CD8⁺-expanded TILs) and pairSEQ on unsorted tumors. The single-cell RT-PCR allowed us to directly extract cDNA from single cells and sequence the TCR gene after several rounds of PCR with specific primers using the Sanger method (39), followed by TOPO TA cloning in the event the T cell expressed two different TCRA chains at the same time. This technique, based on a 96-well platform, has also been successful when applied to MHC-multimer sorting on antigen-specific cells (40, 41), although MHC-multimer sorting is feasible only when the HLA restriction element and the minimal epitope of interest are known. In this study, we used a different approach that does not require the knowledge of antigen specificity by sorting specific TIL subsets. Other higher- throughput approaches have been proposed, such as emulsion PCR (42) and “TCR gene capture” that utilizes an RNA-bait library to specifically target the genomic sequence encoding TCR genes (43). Recently, pairSEQ (21), a new high-throughput technology for pairing TCRA and TCRB sequences, has become available, and we tested its feasibility in fresh, unsorted melanoma samples. PairSEQ uses a statistical model for pairing TCRA and TCRB chains and, because it is based on next-generation sequencing, no extra steps are required (such as TOPO TA cloning) to identify two different TCRA genes expressed by the same cell. Combining the single-cell RT-PCR on specific TIL subsets and pairSEQ on unsorted tumors, we successfully identified the majority, but not all, of the top 10 most frequent TIL CD8⁺PD-1⁺ TCR pairs in 12 patients and tested them against autologous tumor tissue culture lines and autologous antigen-presenting cells expressing tandem minigenes encoding shared cancer antigens, mutated tumor neoantigens, and/or pulsed with the corresponding mutated peptides (11, 29). Multiple reactive pairs could be identified in the top-ranking TCRs in 11 of 12 patients with metastatic melanoma, including 11 TCRs specific for mutated neoepitopes (Fig. 3). We included the six most common melanoma/melanocyte and cancer-germline antigens commonly recognized by patient TILs in our screening. With this limited screening panel, we identified two TCR pairs that were specific for MART-1 (3922-1) and NY-ESO-1 (3998-5). For the other 25 TCR pairs, we were able to demonstrate MHC-restricted reactivity against the autologous tumor cell line, although their specificity remains undefined.

Nonreactive pairs need to be considered with caution. Our approach is based on PCR; therefore, it is subjected to potential errors that could have altered the original sequence of the TCRA–TCRB pairs. Incorrect pairing is also a possible explanation.

Despite these limitations, we identified tumor-reactive TCRs based on their TCRB frequency in the TIL CD8⁺PD-1⁺ population. The major advantage of this approach is the rapidity in finding functional TCRs without the need for further screening. However, once the high-frequency TCRs are identified, they can be tested in vitro in an overnight assay versus the fresh tumor suspension as well as normal autologous PBL to further identify their specificity. This finding opens the possibility for a highly personalized cell transfer cancer therapy in which patients can be treated with autologous, genetically engineered T cells with high proliferative potential. This new approach can potentially be applied to malignancies other than melanoma and is currently under study.

B. Howie has an ownership interest in patents pertaining to pairSEQ Technology. H. Robins is CSO at Adaptive Biotechnologies for which he has ownership interest, including patents. No potential conflicts of interest were disclosed by the other authors.

Conception and design: A. Pasetto, A. Gros, S.A. Rosenberg

Development of methodology: A. Pasetto, A. Gros, D.C. Deniger, R. Matus-Nicodemos, D.C. Douek

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): A. Pasetto, P.F. Robbins, T.D. Prickett, M.R. Parkhurst, J. Gartner, K. Trebska-McGowan, J.S. Crystal

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): A. Pasetto, A. Gros, P.F. Robbins, D.C. Deniger, T.D. Prickett, B. Howie, H. Robins, J. Gartner, J.S. Crystal, S.A. Rosenberg

Writing, review, and/or revision of the manuscript: A. Pasetto, A. Gros, P.F. Robbins, D.C. Deniger, T.D. Prickett, D.C. Douek, H. Robins, J. Gartner, K. Trebska-McGowan, J.S. Crystal, S.A. Rosenberg

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): J.S. Crystal, S.A. Rosenberg

Study supervision: A. Gros

Other (performed multiple cell assays and analyzed their results): K. Trebska-McGowan

We thank Sanja Stevanović, Eric Tran, William Lu, Mojgan Ahmadzadeh, and Cyril Cohen for helpful discussions; Yang-Li, Mona El-Gamil, and Lien Ngo for technical advice and support; and Arnold Mixon and Shown Farid for flow cytometry technical support and sorting. We also thank the Adelson Medical Research Foundation for their generous support for this study.

This research was supported by the Intramural Research Program of the NIH at the National Cancer Institute.

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.