CD8⁺ T Cell–Dependent Antitumor Activity In Vivo of a Mass Spectrometry–Identified Neoepitope despite Undetectable CD8⁺ Immunogenicity In Vitro

Key Points

• MHC I–restricted neoepitope Jup^MUT was identified using mass spectrometry.

• Jup^MUT appears immunologically inert in vitro.

• However, Jup^MUT exerts tumor control in vivo in a CD8 T cell–dependent manner.

Abstract

Identification of neoepitopes that can control tumor growth in vivo remains a challenge even 10 y after the first genomics-defined cancer neoepitopes were identified. In this study, we identify a neoepitope, resulting from a mutation in the junction plakoglobin (Jup) gene (chromosome 11), from the mouse colon cancer line MC38-FABF (C57BL/6). This neoepitope, Jup mutant (Jup^MUT), was detected during mass spectrometry of MHC class I–eluted peptides from the tumor. Jup^MUT has a predicted binding affinity of 564 nM for the K^b molecule and a higher predicted affinity of 82 nM for D^b. However, whereas structural modeling of Jup^MUT and its unmutated counterpart Jup wild-type indicates that there are little conformational differences between the two epitopes bound to D^b, large structural divergences are predicted between the two epitopes bound to K^b. Together with in vitro binding data with RMA-S cells, these data suggest that K^b rather than D^b is the relevant MHC class I molecule of Jup^MUT. Immunization of naive C57BL/6 mice with Jup^MUT elicits CD8-dependent tumor control of a MC38-FABF challenge. Despite the CD8 dependence of Jup^MUT-mediated tumor control in vivo, CD8⁺ T cells from Jup^MUT-immunized mice do not produce higher levels of IFN-γ than do naive mice. The structural and immunological characteristics of Jup^MUT are substantially different from those of many other neoepitopes that have been shown to mediate tumor control.

Introduction

Neoepitopes arising from mutations in cancer cells are usually predicted using bioinformatics tools such as NetMHC (1–4), although the utility of such predictions is increasingly uncertain from studies in mouse models (5–8) as well as in humans (9, 10). The predicted MHC class I (MHC I)–binding neoepitopes that are expected to elicit CD8 responses mostly elicit CD4 responses in clinical trials (11). One source of uncertainty derives from the fact that prediction algorithms are unable to predict which of the candidate peptides are actually presented by MHC I on the cell surface. The most direct way to identify presented neoepitopes is by isolating the MHC I molecules from tumor cells and analyzing the eluted peptides by mass spectrometry (MS). Only a handful of neoepitopes have been discovered this way in humans (12) and mice (3, 4, 7, 13).

Recently, using a combination of genomic and bioinformatics tools (14), Brennick et al. (8) identified a number of neoepitopes of the C57BL/6 mouse colon carcinoma MC38-FABF that are effective in mediating tumor control. Nonsynonymous single-nucleotide variants of the MC38-FABF tumor were identified using multiple next-generation sequencing platforms and verified by Sanger sequencing. Long peptides containing the mutations were used to immunize mice that were then challenged with tumor cells (prophylactic models), or to treat tumor-bearing mice (therapeutic models), to test for antitumor activity. Nine tumor control-mediating neoepitopes (TCMNs) were discovered. Interestingly, none of the TCMNs could be detected by MS among the peptides eluted from MHC I molecules purified from MC38-FABF. We reinterrogated MC38-FABF using MS and identified one additional MHC I–presented neoepitope.

In this study, we describe the immunological and structural properties of this MHC I–presented neoepitope and characterize its ability to mediate Ag-specific tumor control. When interrogated with widely used techniques for measuring immune responses, this neoantigen appears to be immunologically inert. At the same time, it performs the defining function of a “vaccine,” which is to control the growth of the tumor in an immune-mediated manner.

Materials and Methods

Mice and cell lines

Female 6-wk-old C57BL/6 mice were purchased from The Jackson Laboratory (Bar Harbor, ME), and all experiments were performed with mice 6–10 wk of age. The MC38-FABF tumor cell line is a colon carcinoma (b haplotype) acquired from Alan B. Frey at New York University. RMA-S mouse lymphoma cells (TAP2-deficient, b haplotype) were purchased from American Type Culture Collection. Both cell lines were cultured in RPMI 1640 supplemented with 10% FBS.

Bioinformatics analysis of sequencing data

Sequencing was performed following the newer version of the Epi-Seq pipeline as previously published by our group (5). Epi-Seq is a multistep bioinformatics analysis pipeline that starts from the raw exome sequencing and RNA sequencing (RNA-seq) tumor reads and produces a set of predicted tumor-specific expressed epitopes. Exome and RNA-seq reads were aligned to the mm10 mouse reference genome using HISAT2. Single-nucleotide variants (SNVs) were called using the somatic variant caller version of SNVQ. Gene expression estimation from RNA-seq data were performed by using the IsoEM2 algorithm. A list of nonsynonymous SNVs was generated for those mutations with both exome and RNA coverage for each SNV position. A list of mutation-containing 21-mer peptides was generated for use in MS analysis.

Isolation of MHC-presented peptides from cells

MHC I/β₂-microglobulin/peptide complexes were isolated from MC38-FABF cells essentially as described by Bassani-Sternberg et al. (12). Ab clone M1/42 (Bio X Cell, catalog no. BE0077), a pan-murine MHC I Ab, was bound to protein G–Sepharose beads (Thermo Fisher Scientific, catalog no. 101243) by incubating the Ab and beads in PBS. Following the incubation, the beads were washed and resuspended in 0.2 M sodium borate buffer (pH 9). Cross-linking of the beads and the bound Ab was performed by incubation in sodium borate buffer in the presence of 20 mM dimethyl pimelimidate (Thermo Fisher Scientific, catalog no. 21667). Unreacted dimethyl pimelimidate was quenched with 0.2 M ethanolamine (pH 8) (Sigma-Aldrich, catalog no. E9508-100ML) following incubation, and the beads were washed and resuspended in PBS.

MC38-FABF cells were grown to a total number of ∼10⁹. Trypsin (0.05%) was used to remove the cells from culture flasks at ∼90% confluence. Viability was confirmed to be >90%. Cells were centrifuged at 400 × g for 5 min, washed with ice-cold PBS, and pellets were frozen at −20°C. While still in frozen pelleted form, cells were resuspended in ice-cold lysis buffer (20 mM Tris HCl, 150 mM NaCl, 1% Triton X-100, 0.1% octyl glucoside, and protease inhibitor mixture) and incubated for 30 min at 4°C. Lysate was centrifuged at 12,000 × g for 20 min at 4°C and loaded onto a protein G–Sepharose column (without bound Abs) to remove any existing Igs from the sample. The cleared lysate was collected and immediately loaded into the prepared protein G–Sepharose column with a covalently bound anti-MHC Ab. This column was capped and incubated for 1 h at 4°C on a rotary mixer. After the flowthrough was allowed to drain, the column was washed with 10 ml of wash buffer A (20 mM Tris-HCl, 150 mM NaCl). Next, it was washed with 10 ml of wash buffer B (20 mM Tris-HCl, 400 mM NaCl), 10 ml of wash buffer A again, and lastly 10 ml of wash buffer C (20 mM Tris HCl). Bound MHC I/β₂-microglobulin/peptide complexes were eluted in 0.5-ml fractions using 0.1 N acetic acid.

Eluted proteins were separated from peptides on a Sep-Pak cartridge (Waters, part no. WAT054960). The cartridge was first washed with 80% acetonitrile in 0.1% trifluoroacetic acid (TFA) and two additional times with 0.1% TFA. The eluates were applied to the cartridge and then washed with 0.1 TFA. Peptides were eluted from the cartridge in 30% acetonitrile in 0.1% TFA, whereas the proteins, MHC I, and β₂-microglobulin were eluted subsequently in 80% acetonitrile in 0.1% TFA. The peptides were vacuum dried at 37°C and stored at −20°C.

MHC-bound peptide analysis using ultra-performance liquid chromatography and high-resolution tandem MS

Dried, desalted peptides were resuspended in 0.1% formic acid in water and analyzed using nanoflow ultra-performance liquid chromatography (UPLC) coupled to tandem MS (MS/MS). One microliter of desalted peptides was loaded on a 75-µm × 25-cm Easy-Spray PepMap C18 analytical column (Thermo Scientific) held at 35°C and subjected to a 1-h, 300 nl/min flow linear gradient. Gradient conditions were as follows: 4% solvent B held for 10 min, ramped to 30% solvent B in 40 min, 30% solvent B to 90% solvent B in 10 min (solvent A, 0.1% formic acid in water; solvent B, 0.1% formic acid in acetonitrile) on a Dionex UltiMate RSLCnano UPLC system. Eluted peptides were directly ionized into a Q Exactive HF hybrid mass spectrometer (Thermo Scientific) using electrospray ionization and a +1.9 kV spray voltage.

The Q Exactive HF was operated in positive mode and implemented a data-dependent acquisition method comprised of a single full MS scan followed by 15 MS/MS scans. Full MS scans used the following parameters: mass range 300–1800 m/z, 60,000 resolution, default charge state 2, one microscan, 1e6 automatic gain control target. Data-dependent MS/MS scans used the following parameters: one microscan, 15,000 resolution, 1e5 automatic gain control target, maximum injection time of 40 ms, 2.0 m/z isolation window, 0.0 m/z isolation offset, normalized collision energy of 27, and dynamic exclusion set to 30 s.

Bioinformatics identification of peptide sequences analyzed using Byonic

Byonic v3.1 (Protein Metrics) was used to search the raw MS data against a custom proteome database comprised of the UniProt Mus musculus proteome (UP000000589, accessed February 2019) and manually added peptide sequences of the mutation-containing 21-mer peptides. The common proteomics contaminants Byonic database and decoy database were also searched. The following parameters were used: nonspecific enzyme specificity, 5 ppm precursor and 20 ppm fragment mass tolerances, oxidized Met and N-terminal acetyl variable modifications, 2000 Da maximum precursor mass, computes precursor and charge assignments from MS1, automatic score cut (5% peptide spectrum match false discovery rate [FDR] cuts) enabled, and no protein level FDR cuts. All other parameters were kept at default values. The Byonic-reported peptide hits were manually exported from Byonic Viewer and sorted by FDR 1D to identify pulsed peptide sequences ranked below 5% peptide spectrum match FDR. To increase the confidence of the identification, the MS/MS spectrum matched to SSVENIQRL was then compared with that for a synthetic peptide with identical sequence using the UPLC-MS/MS methods described above.

MHC–peptide binding

RMA-S cells were plated at 2 × 10⁵ cells/well in a 96-well plate. Different concentrations of each peptide were added (in triplicate), and cells were incubated for 4 h at 37°C. Cells were then put on ice and stained with either of two primary Abs, anti–H-2K^b (clone Y-3, Bio X Cell, catalog no. BE0172) or anti–H-2D^b (clone B22-249.R1, Invitrogen, product no. MA5-17992), in FACS buffer for 20 min, washed twice with FACS buffer, and secondary stained with polyclonal anti-IgG F(ab′)₂ fragment conjugated with Pacific Orange (Invitrogen, catalog no. P31585) for 20 min at 4°C. During this secondary incubation, a fixable viability dye was included (eFluor780, Invitrogen, catalog no. 65-0865-14). Cells were then fixed with a formaldehyde solution (Invitrogen, catalog no. 00-5123-43). Flow cytometry was performed on a MACSQuant Analyzer 10 (Miltenyi Biotec). Events were gated on forward and side scatter, then single cells using forward scatter height versus area, then viability dye exclusion, and the mean fluorescence index (MFI) of Pacific Orange was assessed.

Immunizations and tumor challenges

Immunizations for tumor growth experiments were performed as previously described (8). Neoepitope peptides used for immunization were 21-mer “long peptides” with the mutated amino acid in the center (i.e., the 11th residue). Bone marrow–derived dendritic cells (BMDCs) were pulsed with 100 µM peptide for 60 min, with mixing every 15 min, then washed with serum-free RPMI 1640, resuspended at ∼6 × 10⁷ cells/ml in serum-free RPMI 1640, and 100 µl was injected intradermally in the belly of each mouse. Two immunizations were performed a week apart. One week after the second immunization, MC38-FABF tumor cells were prepared at 3 × 10⁵ cells/ml in PBS. To implant tumors, 100 µl of this suspension (3 × 10⁴ cells) was injected intradermally in the rear flank. The growth of the tumor was measured twice a week. For checkpoint blockade therapy, 75 µg of anti–CTLA-4 (clone 9D9, Bio X Cell, catalog no. BE0164) was administered by i.p. injection coinciding with the second immunization, and then twice a week following the tumor challenge. For CD8⁺ T cell depletion experiments, mice were injected i.p. with anti-CD8α Ab (clone 2.43, Bio X Cell, catalog no. BE0061) 2 d prior to each immunization, and then once a week following tumor challenge.

Immunogenicity experiments

For experiments testing immunogenicity using BMDCs as an adjuvant, the same immunization protocol was used as in tumor challenge experiments. Spleens were harvested a week after the second immunization. For immunogenicity experiments using poly(I:C) as an adjuvant, 100 µg of peptide was mixed with 40 µg of poly(I:C) (InvivoGen, catalog code tlrl-pic) in a volume of 50 µl, which was injected intradermally in the rear flank. Spleens were harvested 10 d later. CD8⁺ T cells were enriched using magnetic negative selection (STEMCELL Technologies, catalog no. 19853A). In individual wells of a 96-well plate, 2 × 10⁵ naive splenocytes were pulsed with 10 µM peptide and cocultured for 12 h with 2 × 10⁵ CD8⁺ T cells from immunized mice, in the presence of brefeldin A (BioLegend, catalog no. 420601). Surface staining included anti-CD3 conjugated to FITC (clone 17A2, BioLegend, catalog no. 100204), anti-CD8 conjugated to allophycocyanin (clone 53-6.7, BioLegend, catalog no. 100712), and anti-CD44 conjugated to PerCP/Cy5.5 (clone IM7, BioLegend, catalog no. 103032). A fixable viability dye was included (eFluor780, Invitrogen, catalog no. 65-0865-14). Cells were fixed and permeabilized (Invitrogen, catalog no. 00-5521-00). Intracellular staining included anti–IFN-γ conjugated to PE (clone XMG1.2, BioLegend, catalog no. 505808). Events were gated on forward scatter and side scatter, then single cells using forward scatter height and area, then exclusion of the dead cell stain, then CD3 and CD8 double-positive cells; following this series of gates, cells double-positive for CD44 and IFN-γ were quantified as a percentage of these viable CD8⁺ T cells.

Structural modeling of junction plakoglobin mutant neoepitope and wild-type peptide/MHC complexes

Structural modeling of the junction plakoglobin (Jup) mutant (Jup^MUT) neoepitope and Jup wild-type (Jup^WT) peptide/MHC complexes was performed as described previously (15), using PyRosetta 3.0 in conjunction with the ref2015 score function and KIC loop modeling. Starting with coordinates from Protein Data Bank 2VAB for peptides presented by H-2K^b or Protein Data Bank 5WLI for those presented by H2-D^b (16, 17), mutations were made to the peptide chain with the PyMOL mutagenesis tool. These two starting structures were chosen based on their relative similarity to Jup^WT and Jup^MUT neoepitope along the nonameric peptide backbone conformations. Template structures were brought to local energy minima using Rosetta FastRelax with harmonic restraints of 0.05 kcal/mol which optimally balanced energy reduction with changes in root-mean-square deviation (RMSD) of the starting structures. After minimization, 1000 models or “decoys” were generated for each peptide/MHC variant, and the top five structures with the lowest ref2015 energy scores were chosen for analysis. The final five models for each peptide/MHC showed strong agreement in structure yielding Cα RMSDs between 0.15 and 0.27 Å for peptides presented by H-2K^b and 0.09 and 0.13 Å for those presented by H-2D^b. Representative structure models are shown for comparison in Fig. 2.

FIGURE 2.

Structural modeling of Jup^MUT and Jup^WT shows substantial differences when the peptides are presented by H-2K^b but not H-2D^b. (A) Superimposition of the lowest energy models of Jup^MUT and Jup^WT bound to H-2D^b. The epitopes are predicted to bind almost identically in the groove. Carbon atoms of Jup^MUT are blue; those of Jup^WT are green. (B) Superimposition of the lowest energy models bound to H-2K^b shows the predicted structural differences, beginning at Glu⁴ and extending through Arg⁸, with the position of the hydrophobic Ile⁶ pointing up with Jup^MUT but buried in the groove with Jup^WT.

Results

Identification of the Jup^MUT neoepitope using MS

Exome sequencing and RNA-seq of MC38-FABF tumor cells were performed. Mutations in expressed genes were identified by comparing them with the WT exome database using the Epi-Seq pipeline (5). To account for all possible presented 8- to 11-mer neoantigens for a given mutation, a list of 21 mers encompassing all identified nonsynonymous mutations were listed (Supplemental Table I).

Purification of MHC I molecules from MC38-FABF and isolation of MHC-bound peptides was performed as described (12). Mass spectra from the isolated peptides were searched against the list of all possible neoepitopes (Supplemental Table I), as deduced from the output of the Epi-Seq pipeline. A single mutated peptide derived from the Jup protein coding gene was identified in the MS spectra. Its mass spectrum is shown in the upper half of Fig. 1A, and its amino acid sequence is shown in Fig. 1B. This peptide, henceforth referred to as Jup^MUT, contains a mutation from valine to leucine at its C terminus (Fig. 1C). RNA-seq confirmed the expression of the Jup gene by MC38-FABF cells and, specifically, that the mutated allele is indeed expressed at levels similar to the WT protein (Fig. 1C). The Jup^MUT peptide was then synthesized and analyzed by MS in isolation to compare the spectra of the natural and synthetic peptides (Fig. 1A, lower half). The precise match between the spectra indicates that the mutated Jup^MUT peptide was indeed recovered from the cell-surface MHC I molecules of the MC38-FABF cell line.

FIGURE 1.

Neoepitope of MC38-FABF tumor cells identified using MS. (A) MHC I molecules were isolated from MC38-FABF cells using the M1/42 Ab as described in Materials and Methods. MHC-bound peptides were eluted and analyzed by MS. In the spectrum shown, the vertical axis is divided as above 0 and below 0; the upward-facing spectrum is derived from the neoepitope peptide (Jup^MUT) eluted from the MHC molecules. This 9-mer peptide was then synthesized, and the synthetic peptide was analyzed by the same method. The downward-facing spectrum is derived from the synthetic peptide. Peaks from N-terminal peptide fragments (b-type ions) are shown in blue; peaks from C-terminal peptide fragments (y-type ions) are shown in red. (B) The amino acid sequence of Jup^MUT is shown, with peptide fragments (denoted by blue and red marks) corresponding to the peaks from b-type and y-type ions indicated in (A). (C) The neoepitope sequence is compared with its WT counterpart, showing a single amino acid change at the C-terminal position (V→L). The number of RNA-seq reads obtained from each sequence is shown (the overall gene expression of Jup^MUT was 68.97 transcripts per million). (D) Predicted binding affinities of Jup^MUT and its unmutated counterpart Jup^WT with the H-2K^b and H-2D^b molecules, as calculated by the NetMHC 4.0 algorithm. The differential agretopic index (DAI) (5) is a metric for the difference in MHC binding affinity between mutated and WT peptides. (E) RMA-S cells were incubated with synthetic peptides for 5 h at four different concentrations (0.01, 0.1, 1, and 10 µM). Test peptides included the Jup^MUT neoepitope and its unmutated counterpart Jup^WT; the chicken OVA epitope SIINFEKL (OVA_257–264) was used as a control. Cells were stained with conformation-specific primary Abs Y-3 (against H-2K^b, left) or B22.249 (against H-2D^b, right), followed by secondary Ab staining. The mean fluorescence index (MFI) of Ab staining was determined by flow cytometry. MFI for Jup^MUT and Jup^WT were compared using a pairwise comparison t test. Background staining is indicated by the dashed lines. The data are representative of two or more experiments. *p < 0.05, **p < 0.01, ***p < 0.001.

H-2K^b is the relevant MHC I molecule for Jup^MUT

According to NetMHC 4.0, Jup^MUT is predicted to bind H-2D^b with an IC₅₀ value of 82 nM, thus categorizing it as an “intermediate binder” but not a “high binder” according to commonly accepted thresholds used for such categorization (18). The unmutated counterpart peptide also qualifies as a H-2D^b intermediate binder, albeit with a weaker affinity (146 nM). Predicted binding IC₅₀ of Jup^MUT to H-2K^b is 564 nM, designating it as a “low binder” (IC₅₀ > 500 nM). The unmutated counterpart Jup^WT is predicted to bind H-2K^b weakly with an IC₅₀ value of 1644 nM (Fig. 1D). A positive differential agretopicity index (DAI) for Jup^MUT in the context of both K^b (1.54) and D^b (0.84; Fig 1D) suggests that Jup^MUT may be a TCMN (5).

The binding of Jup^MUT to H-2D^b and K^b was tested experimentally using RMA-S cells. These cells, lacking a functional TAP2 gene, have a defect in the Ag presentation pathway that results in “empty” H-2 molecules on the cell surface (19). Due to the lack of stabilization in the absence of bound peptides, the surface H-2 molecules are rapidly degraded. Incubation with a peptide capable of binding H-2 induces the complex to assume its stably folded conformation, whereas incubation with a nonbinding peptide leaves the H-2 molecules unstable. The degree of H-2 stabilization can be determined by staining the cells with conformation-specific anti–H-2 Abs and flow cytometry. Ab clones Y-3 and B22.249 are specific for the peptide-dependent conformations of H-2K^b and H-2D^b, respectively (20, 21).

Incubation of synthetic Jup^MUT peptide with RMA-S cells induced concentration-dependent stabilization of H-2K^b as well as H-2D^b (Fig. 1E). Its unmutated counterpart induced similar, but lesser, stabilization of both H-2 molecules. The known strong H-2K^b binder SIINFEKL (chicken OVA 257–264, affinity 10 − 20 nM [22]) served as a positive control. As expected, based on the predicted affinity, Jup^MUT-induced stabilization of H-2K^b was less than that induced by SIINFEKL. These results strongly agree with all of the affinity predictions and confirm that Jup^MUT is capable of binding both H-2K^b and H-2D^b.

To further investigate the presentation of Jup^MUT, and in particular its restricting MHC protein, high-resolution structural modeling of the Jup^MUT neoepitope and its unmutated counterpart bound to both H-2K^b and H-2D^b was performed. Modeling was performed using our established peptide/MHC modeling procedures, which we previously used to identify structural and physical correlates with neoepitope immunogenicity (7, 8). Accordingly, we generated 1000 models (referred to as decoys) for each peptide/MHC complex using kinematic loop modeling in Rosetta modeling. Because the lowest energy decoys are not always reflective of the most accurate model (15), we selected the five lowest energy decoys of each complex for more comprehensive comparisons.

For H-2D^b, the modeling suggested that Jup^MUT and Jup^WT peptides adopt essentially identical conformations within the H-2D^b peptide-binding groove. For the five lowest energy decoys, the Jup^MUT and Jup^WT pair differed by an α carbon RMSD of only 0.22 ± 0.16 Å (average and SD of a pairwise comparison of the five decoys for each complex). The RMSD for all common atoms of the two peptides was 0.55 ± 0.48 Å. The predicted close similarities between the neoepitope and peptide WT presented by H-2D^b are shown in Fig. 2A.

The modeling predictions were quite different for H-2K^b. In this study, for the five lowest energy decoys, the α carbon and all atom RMSDs between Jup^MUT and Jup^WT were 1.4 ± 0.7 and 2.6 ± 1.1 Å, respectively. These results suggest that Jup^MUT and Jup^WT adopt distinct conformations when bound to H-2K^b, with different backbone and side chain positions. The predicted differences are illustrated in Fig. 2B. A key difference is seen for the placement of the isoleucine at position 6, which in the neoepitope bulges away from the groove and exposes the hydrophobic isoleucine side chain. As increased exposure of hydrophobic side chains have previously been correlated with neoepitope immunogenicity (23), the “structural difference from self” seen with H-2K^b but not H-2D^b suggests that H-2K^b is the relevant MHC I molecule for Jup^MUT.

Immunization with Jup^MUT elicits CD8-dependent antitumor activity

C57BL/6 mice were immunized with Jup^MUT peptide-pulsed BMDCs or unpulsed BMDCs as controls (n = 10 per group), as described in Materials and Methods (Fig. 3A), and as previously published (8). Immunization with Jup^MUT elicited significant tumor control in most of the immunized mice. Tumors grew progressively in all control mice; in Jup^MUT-immunized mice, 5 out of 10 mice underwent complete or near-complete tumor regression, two mice showed tumor growth inhibition after initial growth, and tumors in two mice grew progressively (p = 0.039, Fig. 3B, 3C). Analysis using tumor control index (TCI) scoring of the growth curves (24) confirmed that prophylactic immunization with Jup^MUT caused significant antitumor activity (Fig. 3D). In contrast, immunization with the unmutated Jup^WT peptide did not elicit significant antitumor activity (Supplemental Fig. 1).

FIGURE 3.

Prophylactic immunization with Jup^MUT controls tumor growth. (A) Experimental design of immunizations, tumor challenge, CTLA-4 blockade, and monitoring of tumor growth. (B) Growth of tumors implanted in mice immunized with unpulsed BMDCs (black) or Jup^MUT-pulsed BMDCs (red). Each line represents tumor growth in a single mouse. (C) The data from (B) are shown as averages of the tumor growth curves from each group. (D) Tumor control index (TCI) scores obtained from the tumor growth curves in (B) and (C); see Corwin et al. (24). The total TCI is shown, and the three components contributing to the total TCI are also shown individually. Statistics were calculated using two-way ANOVA; n = 20 mice/group. The data are representative of two or more experiments. **p < 0.01, ***p < 0.001. ns, not significant.

To test the CD8 dependence of tumor control elicited by immunization with Jup^MUT, mice were treated with an anti-CD8α Ab (clone 2.43) during the entirety of the immunization and tumor challenge. This Ab, which causes robust depletion of CD8⁺ T cells (25), was administered 2 d prior to both immunizations, 2 d prior to tumor challenge, and once a week during tumor growth and measurements (Fig. 4A).

FIGURE 4.

Antitumor activity of prophylactic immunization with Jup^MUT is CD8-dependent. (A) Experimental design including injections of either CD8-depleting (clone 2.43) or isotype control Abs, immunizations, tumor challenge, and CTLA-4 blockade. (B) Growth of tumors implanted in mice immunized with unpulsed BMDCs (top) or Jup^MUT-pulsed BMDCs (bottom) and having received CD8-depleting Ab (blue) or isotype control Ab (black). Each line represents tumor growth in a single mouse. (C) The data from (B) are shown as averages of the tumor growth curves from each group. (D) Tumor control index (TCI) scores obtained from the tumor growth curves in (B) and (C); see Corwin et al. (24). Statistics were calculated using two-way ANOVA; n = 20 mice/group. The data are representative of two or more experiments. *p < 0.05, **p < 0.01. ns, not significant.

Whereas mice treated with isotype control Ab showed the same difference between control and Jup^MUT-immunized mice as seen in previous experiments, the antitumor activity of Jup^MUT immunization was lost in mice treated with CD8-depleting Ab (Fig. 4B, 4C). TCI analysis of the growth curves confirmed a significant abrogation of the Jup^MUT antitumor activity in CD8-depleted mice (Fig. 4D). In contrast, depletion of CD4⁺ T cells did not significantly abrogate the antitumor activity of Jup^MUT immunization (Supplemental Fig. 2).

The CD8⁺ T cell response elicited by Jup^MUT is not detectable by flow cytometry

To test the immunogenicity of the Jup^MUT peptide, multiple methods of immunization were used to generate immune responses in mice, for example, the use of poly(I:C) as an adjuvant, and the injection of peptide-pulsed BMDCs (the same method that generated CD8-dependent antitumor activity). The CD8⁺ T cell response was then tested, in terms of IFN-γ expression by Ag-specific T cells, using intracellular cytokine staining flow cytometry and/or ELISPOT. In each case, immunization with the OVA peptide SIINFEKL (OVA_257–264) and subsequent response to that same peptide was used as a positive control. As shown in Fig. 5, a clear IFN-γ response can be seen against SIINFEKL following multiple methods of immunization. In contrast, immunization with Jup^MUT and subsequent restimulation with the same peptide did not result in a detectable response by either intracellular cytokine staining flow cytometry (Fig. 5) or ELISPOT (data not shown). In a separate but similar experiment, where production of TNF-α by CD8⁺ T cells was tested following immunization and restimulation, the results were similar to those of the IFN-γ experiments (Supplemental Fig. 3). Additionally, we attempted to expand Jup^MUT-specific T cells in vitro by stimulating splenocytes from immunized mice with peptide; no T cell expansion was observed (data not shown).

FIGURE 5.

CD8⁺ T cell response to Jup^MUT immunization is not detectable by IFN-γ flow cytometry. (A) Jup^MUT, SIINFEKL, or vehicle control was mixed with poly(I:C) and intradermally injected into the flanks of mice. CD8⁺ T cells were isolated from spleens of Jup^MUT immunized mice and then restimulated with Jup^MUT or not restimulated. As a positive control, CD8⁺ T cells from SIINFEKL-immunized mice were restimulated with SIINFEKL. (B) Mice were immunized twice, 7 d apart, with unpulsed BMDCs, Jup^MUT-pulsed BMDCs, or SIINFEKL-pulsed BMDCs. CTLA-4 blockade was performed immediately prior to the second immunization. CD8⁺ T cells from spleens were restimulated with the immunizing peptides. Representative dot plots are shown (left), and data from all mice are charted (right). The data are representative of two or more experiments.

Discussion

To date, 25 mouse neoepitopes have been identified by MS (Table I). Of these, 17 neoepitopes have been tested individually for their ability to elicit tumor control (Refs. 7, 13, and this study). Of the 17, only 5 neoepitopes have been reported to elicit tumor control. Jup^MUT is one of these only five antitumor neoepitopes. Of these five neoepitopes, three have high affinity (IC₅₀ < 50 nM) for MHC I alleles, one has intermediate affinity (IC₅₀ 157 nM), and two, Jup^MUT (this study) and Gtf2b^MUT (7), have weak affinity (IC₅₀ > 500 nM) (see Table I). TCMNs have been reported to have a wide range of affinities for their MHC I alleles, from as high as 3–4 nM IC₅₀ (4) to as low as 32,000 nM IC₅₀ (8). Jup^MUT occupies a relatively unique place in the sparse list of neoepitopes that have been shown to mediate tumor control in vivo. On a related note, previous studies inform us that it is not the absolute affinity of a neoepitope to MHC I but the difference in affinity between mutated and WT counterpart peptides toward MHC I that confers immunogenicity to neoepitopes (5, 8, 9). We formulated the DAI to describe this phenomenon (5). Jup^MUT adheres to the principle of DAI.

Table I. Mass spectrometry–identified neoepitopes from mouse cancers to date.

Refs.	Tumor Cell Line	Gene	Gene Expression (TPM)a,b	Neoepitope Sequencec	H-2	Predicted Affinity (IC50, nM)	CD8+ T Cell Response	Antitumor Activity In Vivo
Yadav et al. (3)	MC-38 (H-2b)	Dpagt1 Reps1 Adpgk Cpne1 Irgq Aatf Med12	15 61 28 36 5 54 15	SIIVFNLL AQLANDVVL ASMTNMELM SSPYSLHYL AALLNSAVL MAPIDHTTM DPSSSVLFEY	Kb Db Db Db Db Db Kb	7 16 4 182 7 90 40,630	Yes Yes Yes No No No Not tested	Yesd Yesd Yesd Not testedd Not testedd Not testedd
Gubin et al. (4)	d42m1-T3 (H-2b)	Alg8 Lama4	Data not provided	ITYTWTRL VGFNFRTL	Kb Kb	4 3	Yes Yes	Yesd Yesd
Ebrahimi-Nik et al. (7)	MethA (H-2d)	Rike Prpf19 Gtf2b Trib3 Pdpr Aebp1 Prpf19 Snw1	25 99 52 169 20 277 99 95	AYMKMLSSSL KYLQVASHVGL TGAARFDEF VGPEILSSL IGPRALDVL MAPVRTASM KYLQVASHV SFLPAPTHL	Kd Kd Dd Dd Dd Dd Kd Kd	22 40 499 1,451 157 13,767 9 61	No Yes Yes Yes Yes Not tested No No	Yes No Yes No Yes No No No
Hos et al. (13)	MC-38 (H-2b)	Aatf Adpgk Cpne Gtf2i Reps1 Rpl18 Tdg Wbp11	5,321 1,391 6,399 1,530 2,486 15,546 1,869 4,875	MAPIDHTTM ASMTNMELM SSPYSLHYL STYVIPRL AQLANDVVL KILTFDRL RALDKVHYYI DAVKNAQHL	Db Db Db Kb Db Kb Db Db	90 4 182 16 16 33 4,656 892	No Yes Yes No No Yes No No	No No No No No Yes No No
This study	MC38-FABF (H-2b)	Jup	69	SSVENIQRL	Kb	564	No	Yes

^aValues for Yadav et al. are given in reads per kilobase of transcript per million mapped reads.

^bValues for Hos et al. are published as depicted here, but are probably incorrect by two orders of magnitude, based on our own RNA-seq data for MC-38.

^cMutated amino acid residues are bolded and underlined.

^dFor Yadav et al. and Gubin et al., neoepitopes were pooled for immunization and were not tested individually for antitumor activity.

^eFull gene name is 1190007107Rik.

Analysis of Jup^MUT–MHC I binding reveals an interesting anomaly. The neoepitope Jup^MUT has a predicted binding affinity of 564 nM for the K^b allele. Although Jup^MUT is predicted to bind D^b with a higher affinity (82 nM), structural modeling of binding of Jup^MUT and its unmutated counterpart Jup^WT with K^b and D^b suggests that there is little difference between the binding of Jup^WT and Jup^MUT with D^b; on the contrary, binding of Jup^MUT with K^b is predicted to result in a very significant structural alteration as compared with binding of Jup^WT with K^b: isoleucine at position 6 of Jup^MUT bulges away from the groove and exposes the hydrophobic isoleucine side chain. Such structural differences from self have previously been associated with neoantigen immunogenicity, as they permit a mutated neoepitope to overcome tolerance toward the WT peptide (26). These observations suggest that K^b rather than D^b is the relevant MHC I molecule of Jup^MUT.

It is intriguing that although the tumor control elicited by Jup^MUT is entirely CD8-dependent in vivo, IFN-γ–secreting CD8⁺ T lymphocytes are not detectable by flow cytometry in immunized mice. Such anomaly has been observed before for immune response to other neoepitopes (5, 6, 8). It is conceivable that this anomaly results from the fact that neoepitopes, in contrast to foreign Ags such as OVA, or viral Ags have a self-counterpart from which they differ in a single amino acid change. Because the immune response to self is largely deleted—and when not deleted, heavily regulated—the T cell precursor frequency to neoepitopes may be highly abridged. This level of T cell activity, although sufficient to mediate effective tumor immunity in vivo, may be well below the threshold of detection in vitro with our standard assays. More sensitive tools for detecting T cell responses may bring clarity to this point.

Deployment of neoantigen vaccines as front-line anticancer therapy has been a long-sought avenue. In this pursuit, researchers have focused on Ags that elicit robust and measurable immune responses. Models developed for in silico neoepitope prediction have used training datasets constituted by such Ags; thus, these efforts have yielded prediction tools that accurately identify Ags that elicit measurable responses. However, based on data in this study and other recent work (5, 6, 8), tumor inhibition should be the criterion on which neoepitope prediction algorithms are built, regardless of whether measurable immune responses are elicited. Identification and study of additional anomalous epitopes/Ags will increase the breadth of targetable Ags, populate our training datasets with bona fide tumor-rejecting Ags, and eventually lead to better predictive in silico tools and thus better vaccines.

This article is featured in Top Reads, p. 1749

Disclosures

P.K.S. has interests in Agenus, iPeptide, and Life Science Pharmaceuticals. None of these interests relate to the subject of this publication. The other authors have no financial conflicts of interest.