Targeting of cancer neoantigens with donor-derived T cell receptor repertoires

See allHide authors and affiliations

Science  10 Jun 2016:
Vol. 352, Issue 6291, pp. 1337-1341
DOI: 10.1126/science.aaf2288

Outsourcing cancer immunotherapy

Successful cancer immunotherapy depends on a patient's T cells recognizing tumor-specific mutations and then waging a lethal attack. Despite tumors harboring many mutations, most individuals have very few T cells that respond to these so-called “neo-antigens.” Strønen et al. isolated T cells from healthy donors that responded to predicted neo-antigens expressed by melanomas taken from three patients, sometimes including neo-antigens that the patient's own T cells ignored (see the Perspective by Yadav and Delamarre). Testing whether such an outsourcing strategy could improve clinical outcomes will be an important next step.

Science, this issue p. 1337; see also p. 1275


Accumulating evidence suggests that clinically efficacious cancer immunotherapies are driven by T cell reactivity against DNA mutation–derived neoantigens. However, among the large number of predicted neoantigens, only a minority is recognized by autologous patient T cells, and strategies to broaden neoantigen-specific T cell responses are therefore attractive. We found that naïve T cell repertoires of healthy blood donors provide a source of neoantigen-specific T cells, responding to 11 of 57 predicted human leukocyte antigen (HLA)– A*02:01–binding epitopes from three patients. Many of the T cell reactivities involved epitopes that in vivo were neglected by patient autologous tumor-infiltrating lymphocytes. Finally, T cells redirected with T cell receptors identified from donor-derived T cells efficiently recognized patient-derived melanoma cells harboring the relevant mutations, providing a rationale for the use of such “outsourced” immune responses in cancer immunotherapy.

Accumulating data suggest that tumor regression induced by cancer immunotherapies that exploit the endogenous T cell pool (1, 2) relies on recognition of neoantigens that are formed as a consequence of tumor-specific DNA mutations. A striking observation in cancer patients and in mouse models is that neoantigen-specific T cell reactivity is generally limited to just a few mutant epitopes, even though the number of predicted epitopes is large (312). This scarcity of T cell–recognized neoantigens could potentially reflect immune editing of tumors by T cells (13). Alternatively, an effector T cell pool toward many tumor-expressed neoantigens may be absent because of ineffective priming or because of tolerization of these T cells. Recent work has shown that vaccination with neoantigen peptide–loaded dendritic cells can increase the breadth of mutant peptide-specific T cells in melanoma patients (14). In that study, it could not be established whether newly induced T cells could recognize autologous tumor cells. Nonetheless, these data provide a further incentive for the development of strategies that broaden neoantigen-specific T cell reactivity.

Here, we aimed to establish whether T cell receptors (TCRs) that are obtained outside of the autologous T cell repertoire can be used to engineer neoantigen-specific T cell immunity. To this end, we generated immune responses to HLA-A*02:01-restricted neoantigens from the nontolerized T cell repertoires derived from donors that express this allele. Using this approach, we evaluated (i) whether donor-derived T cells can recognize relevant tumor cells, (ii) whether such “outsourced” immune responses provide evidence for a neglected pool of neoantigens on human cancers, and (iii) which types of mutant peptides are best detected by the T cell–based immune system.

To determine the feasibility of using donor-derived T cell pools to induce neoantigen-specific T cell reactivity, we initially focused on an HLA-A*02:01pos stage IV melanoma patient. Whole-exome and RNA sequencing of tumor material revealed 249 nonsynonymous mutations within expressed genes, and 126 mutant epitopes were predicted to bind to HLA-A*02:01 (15). Of these 126 neopeptides, only two were detected by T cells grown from the same tumor lesion. To investigate whether a larger fraction of predicted neoepitopes could be recognized by a healthy donor immune system, we selected 20 candidate neoepitopes based on high predicted binding affinity to HLA-A*02:01 (table S1). Nonadherent peripheral blood mononuclear cells (PBMCs) from healthy donors were then cocultured with autologous monocyte-derived dendritic cells transfected with mRNA encoding the candidate epitopes in a tandem minigene configuration, or with a control minigene encoding known epitopes from cancer/testis (C/T) antigens and CD20 (16) that were recognized by relevant cytotoxic T lymphocytes (CTLs) (fig. S1). Analyses of resulting cell populations by peptide–major histocompatibility complex (pMHC) multimer staining revealed T cell reactivity toward 5 of 20 neoantigens from patient 1, whereas such reactivity was negligible in control cultures (Fig. 1, A and B). Analysis of T cell reactivity from three additional donors revealed three to five neoantigen-specific T cell responses in all cases (Fig. 1C). One of the T cell responses reproducibly induced in this system, to the neoantigen CDK4R>L, was also one of two responses detected among tumor-infiltrating lymphocytes (TILs) of patient 1.

Fig. 1 In vitro induction and functional activity of donor-derived neoantigen-reactive T cells.

Data depict T cell responses against predicted HLA-A*02:01-binding neoantigens from patient 1. (A) PBMCs (donor 3) stimulated with autologous antigen-presenting cells (APCs) transfected with mRNA encoding either predicted neoantigens (solid squares) or CT/CD20 control antigens (open circles) were stained with pMHC multimers complexed with predicted epitopes. Symbols indicate percentage of live CD8pos cells staining positively for pMHC multimers complexed with indicated peptides. Colored squares indicate populations sorted for further analysis. (B) Flow cytometry analysis. (C) Magnitude of multimerpos T cell populations for the indicated predicted neoantigens induced by APCs transfected with mRNA encoding either relevant neoepitopes (left) or control CT/CD20 epitopes (right) from four healthy donors. (D) Degranulation responses of CTL clones (donor 4) analyzed as shown in fig. S2. Each graph represents the reactivity of 7 to 16 clones to the indicated neoantigen. Controls are depicted only for MLL2L>H-reactive CTL clones; corresponding data for remaining clones are shown in fig. S3A. Graphs are representative of CTL clones from all donors tested and show means of triplicates. Error bars denote SD.

pMHC-multimerpos CD8 cells were sorted from donors 2, 3, and 4 to generate CTL clones. Resulting clones that stained positively with relevant pMHC multimers (>82% of clones) were then tested for functional activity using a live-cell bar-coding assay (fig. S2). Analysis of 185 CTL clones revealed reactivity of the majority of clones toward target cells pulsed with mutant peptide at concentrations down to 1 nM and below, with negligible recognition of the wild-type counterpart (Fig. 1D and figs. S2B and S3).

To assess recognition of a short-term melanoma line of patient 1, we selected 76 CTL clones that specifically recognized target cells pulsed with the corresponding neoantigens at low concentrations. All MLL2L>H-reactive CTL clones tested (n = 10) recognized the relevant melanoma cells. In contrast, no recognition of an HLA-A*02:01pos third-party melanoma was observed, unless pulsed with MLL2L>H peptide (Fig. 2, A and B). By the same token, all CDK4R>L-reactive CTL clones tested (n = 6) showed vigorous and specific reactivity toward CDK4 mutant melanoma cells (Fig. 2B). Among CTL clones reactive with ASTN1P>L and SMARCD3H>Y, 7 of 24 clones and 5 of 20 clones, respectively, showed recognition of cognate melanoma (fig. S4). None of the GNL3LR>C-reactive CTL clones tested (n = 16) recognized cognate or third-party melanoma unless pulsed with the relevant neoantigen (fig. S4).

Fig. 2 Tumor recognition by donor-derived neoantigen-reactive CTLs.

(A) MLL2L>H-reactive CTL clone 7 (donor 3) was incubated with a melanoma line derived from patient 1 that carried the mutated MLL2 gene (cognate tumor; left) or with a third-party HLA-A*02:01pos melanoma line carrying the wild-type MLL2 gene (right). Data depict the percentage of IFN-γpos and/or CD107a/bpos cells of live CD8 cells. (B) Neoantigen-reactive CTL clones (donors 3 and 4) incubated with the indicated melanoma lines were analyzed as described in (A). Where indicated, melanoma lines were pulsed with 1 nM corresponding neoantigen (shown for MLL2L>H-reactive CTL clones from donor 3). All MLL2L>H- and CDK4R>L-reactive CTL clones were selected for TCR sequencing. Graphs show means of duplicates; error bars denote SD.

T cell inductions were subsequently performed for predicted neoantigens from tumors of two additional patients. For patient 2, a set of 27 neopeptides (table S2) with a median predicted binding affinity to HLA-A*02:01 of 34 nM (range 2 to 140 nM) was selected from among 154 mutant peptides predicted to bind to HLA-A*02:01. No HLA-A*02:01–restricted neoantigen-specific T cell responses had been detected in TILs isolated from this patient when screening for reactivity to these 154 peptides. In contrast, responses to six predicted neoantigens were induced among T cells derived from four healthy donors (fig. S5, A and B). For patient 3, no T cell responses to 10 predicted neoantigens were detected (table S3). Predicted binding affinities of these potential neoantigens were considerably lower (median 225 nM) than those from patient 1 (median 41 nM) and patient 2 (median 34 nM).

From pMHC-multimerpos CD8 cells from donors 5, 7, and 8, we established CTL lines reactive with the USP28C>F, SNX24P>L, PGM5H>Y-462-470, and PGM5H>Y-465-473 mutant peptides identified in the tumor of patient 2. All CTL lines responded strongly to target cells pulsed with relevant mutant peptides, whereas responses to target cells pulsed with corresponding wild-type peptides were generally low or negligible (fig. S5C, top row). Viable tumor material from patient 2 was scarce, and a tumor cell line for use in functional analyses could not be established. However, all but one of the CTL lines specifically recognized target cells transfected with a minigene encoding the mutant peptides PGM5H>Y, USP28C>F, and SNX24P>L, flanked on both sides by 10 naturally occurring amino acids (fig. S5C, bottom row). Together, these data demonstrate that neoantigen-specific T cell responses can readily be induced in T cell repertoires from healthy donors, and that these T cells specifically recognize naturally processed neoantigens, including antigens expressed in matched tumor material.

Next, we investigated the feasibility of transferring donor-derived tumor-specific T cell reactivity by TCR gene transfer. TCRs from 28 CTL clones from three donors selected on the basis of reactivity toward melanoma cells of patient 1 were sequenced (table S4), yielding 11 unique TCR sequences, with one or two TCR sequences identified per antigen-donor combination. Nine of these were reconstructed and eight were successfully expressed in peripheral blood T cells, as confirmed by anti-TCRβ staining (fig. S6), targeting all four epitopes for which antitumor reactivity was seen. TCR-transduced PBMCs were then tested for degranulation and interferon (IFN)–γ production in response to cognate melanoma cells, with results for the five most responsive TCRs shown in Fig. 3.

Fig. 3 Tumor recognition by genetic transfer of donor-derived neoantigen-specific TCRs.

Healthy donor peripheral blood T cells transduced with indicated donor-derived TCRs [(A), (B), (D)] or a patient-derived TCR (C) were incubated with the cognate melanoma line, an HLA-A*02:01pos third-party melanoma line, or melanoma cells modified as indicated. Percentage of responding CD8 cells was analyzed as described in Fig. 2. (A) Melanoma cells in which the mutant MLL2 gene was knocked out (cognate tumor MLL2 KO), mock-treated cognate melanoma cells (cognate tumor Mock KO), or third-party melanoma cells stably transfected with DNA encoding the relevant mutant neoantigen (3rd party tumor + MLL2MUT DNA) were used as target cells. (B to D) TIndicated melanoma cells were used as target cells for indicated healthy donor–derived [(B) and (D)] and patient-derived (C) TCRs. Data for each TCR are representative of two or three independent experiments using T cells from different healthy donors. Graphs depict mean of duplicate samples; error bars denote SD. Values were corrected for transduction efficiency, measured as percentage of CD8pos cells staining positively with antibody to mouse TCRβ chain. Asterisk indicates TCRs for which the fraction of TCR-expressing T cells is underestimated by staining with antibody to mouse TCRβ chain constant domain (fig. S6).

The MLL-2L>H–reactive TCR 41 strongly recognized patient-derived melanoma cells carrying the mutant MLL2 gene, with low reactivity to a third-party melanoma line lacking this mutation, unless the mutant epitope was genetically introduced (Fig. 3A). Furthermore, when the mutant MLL2 open reading frame in the cognate melanoma was disrupted, recognition of the mutant tumor was comparable to that of the third-party tumor (Fig. 3A). In addition, three CDK4R>L-reactive TCRs were expressed in healthy donor T cells (TCR 53, 55, and 57), with two of these showing high recognition of the cognate melanoma that carries the mutant CDK4 gene as well as recognition of melanoma cell line Mel 526, which carries the previously described CDK4 R24C mutation (17) (Fig. 3B). Notably, recognition by TCRs 53 and 57 was comparable to that seen for the patient-derived CDK4R>L-reactive TCR 17, previously isolated from TILs of patient 1 (Fig. 3, B and C, and fig. S7). The ASTN1P>L-reactive TCR 65 also showed specific recognition of cognate melanoma (Fig. 3D). ASTN1P>L-reactive TCR 52 and SMARCD3H>Y-reactive TCRs 59 and 67 did not recognize cognate or third-party tumor unless the relevant neoantigen was introduced. In total, recognition of endogenously presented neoantigen on the cognate melanoma was observed for three of four antigens evaluated.

With neoantigens emerging as attractive targets in the development of personalized immunotherapies, strategies for the rapid identification of relevant neoantigens have become a major priority. We speculated that the use of outsourced immune responses could facilitate analysis of the rules that govern neoantigen recognition by T cells. In the current experiments, immunogenicity was evaluated for 57 peptides that had been selected on the basis of predicted binding affinity to HLA-A*02:01. Of these, 11 generated immune responses, and T cells reactive with 10 of these epitopes recognized endogenously presented antigen. The median predicted binding affinity for this set of T cell–recognized neoantigens was 28 nM (range 6 to 119 nM), compared to 54 nM (range 2 to 925 nM) for peptides that did not induce immune responses (Fig. 4A). Prior work has suggested that pMHC complex stability may form a particularly strong determinant of immunogenicity (18, 19). To test the added value of experimental analysis of pMHC off-rate, we developed a flow cytometry–based assay for pMHC stability (fig. S8, A and B, and tables S1, S2, S3, and S5). Analysis of pMHC off-rates for all 57 predicted neoantigens revealed that neopeptides that were recognized by donor-derived T cells displayed a significantly longer half-life relative to neopeptides for which no responses were observed (median t1/2 of β2-microglobulin signal: 14.3 versus 4.7 hours, P < 0.0001) (Fig. 4B). Using a t1/2 cutoff value of 5 hours, 11 of 32 (34%) candidate neoantigens were recognized by donor T cells (Fig. 4C). Furthermore, the significant added value of measured pMHC off-rates, as compared to the sole in silico prediction of peptide affinity, is also apparent from receiver operating characteristic (ROC) curves (fig. S8C).

Fig. 4 pMHC stability predicts neoantigen immunogenicity.

(A and B) Predicted binding affinity to HLA-A*02:01 (A) and experimentally determined half-life of peptide–HLA-A*02:01 complexes as measured by dissociation of β2-microglobulin (B) for the 57 predicted neoantigens from patients 1, 2, and 3 that do or do not induce a T cell response. Peptide sequences and predicted affinities are listed in tables S1 to S3. (C) Red bars represent predicted neoantigens that were shown to be immunogenic; gray bars represent predicted neoantigens for which no T cell response could be detected. Dotted line represents suggested cutoff value of t1/2 = 5 hours. Values in (B) and (C) represent means of triplicates. ***P < 0.0001 (Mann-Whitney U test), n.s., not significant.

Our results show that T cell repertoires from healthy donors provide a rich source of T cells that specifically recognize neoantigens present on human tumors. Responses to 11 different epitopes were observed, and for the majority of evaluated epitopes, potent and specific recognition of tumor cells endogenously presenting the neoantigens was detected. We draw three main conclusions from this work. First, these results demonstrate the existence of a repertoire of neoantigens on human tumors to which the endogenous T cell pool has not mounted a measurable response in vivo, but that can be the target of T cells from an independent source. Specifically, among the neoantigen-specific T cell populations capable of recognizing endogenously processed antigen, only one was also detected within the original TILs. This observation forms a strong incentive for the further development of immunotherapies that aim to broaden neoantigen-specific T cell reactivity (14, 20, 21), either from an exogenous source or from the endogenous T cell pool. (Note that the latter approach relies on the presence of patient T cells that still have the capacity to respond to these neglected epitopes, an issue that remains to be addressed.) Second, the ability to evaluate large series of predicted epitopes for recognition by T cells from multiple independent T cell repertoires makes it feasible to systematically examine the rules that control neoantigen recognition. Finally, the current results suggest the possibility of personalized neoantigen-directed immunotherapies that are independent of the status of the patient’s own immune system.

Supplementary Materials

Materials and Methods

Figs. S1 to S8

Tables S1 to S8

References (2243)

References and Notes

  1. See supplementary materials on Science Online.
  2. Acknowledgments: We thank L. Fanchi, R. Mezzadra, and F. Scheeren for advice and support; A. Vefferstad and the OUH flow cytometry core facility for exemplary technical assistance; I. M. Svane for sharing clinical material; and the Norwegian Bone Marrow Donor Registry for HLA typing. The data presented in the manuscript are tabulated in the main paper and in the supplementary materials. DNA and RNA sequencing data have been deposited in the European Genome-Phenome Archive (accession codes EGAD00001000243 and EGAD00001000325). A patent application (P32649NL00) has been filed that covers the technology of targeting cancer-specific amino acid sequences with donor-derived TCR repertoires (inventors J.O., T.N.S., E.S., and M.T.). All described TCRs are available under an MTA with the Netherlands Cancer Institute and Oslo University Hospital. Supported by Dutch Cancer Society Queen Wilhelmina Award NKI 2013-6122 and EU H2020 project APERIM (T.N.S.); the K. G. Jebsen Foundation (J.O., T.N.S., F.L.-J.); and the Research Council of Norway, Regional Authorities South-Eastern Norway, the University of Oslo and Oslo University Hospital, and the Norwegian Cancer Society (J.O.).
View Abstract

Stay Connected to Science

Navigate This Article