Biochemical and functional characterization of mutant KRAS epitopes validates this oncoprotein for immunological targeting

Activating RAS missense mutations are among the most prevalent genomic alterations observed in human cancers and drive oncogenesis in the three most lethal tumor types. Emerging evidence suggests mutant KRAS (mKRAS) may be targeted immunologically, but mKRAS epitopes remain poorly defined. Here we employ a multi-omics approach to characterize HLA class I-restricted mKRAS epitopes. We provide proteomic evidence of mKRAS epitope processing and presentation by high prevalence HLA class I alleles. Select epitopes are immunogenic enabling mKRAS-specific TCRαβ isolation. TCR transfer to primary CD8+ T cells confers cytotoxicity against mKRAS tumor cell lines independent of histologic origin, and the kinetics of lytic activity correlates with mKRAS peptide-HLA class I complex abundance. Adoptive transfer of mKRAS-TCR engineered CD8+ T cells leads to tumor eradication in a xenograft model of metastatic lung cancer. This study validates mKRAS peptides as bona fide epitopes facilitating the development of immune therapies targeting this oncoprotein.

T he RAS family of GTPases (KRAS, NRAS, HRAS) is mutated in approximately 20% of all human malignancies 1,2 . The vast majority of RAS genomic alterations are the result of missense mutations at codon positions G12, G13, or Q61 that involve the RAS protein GTP-binding domain, resulting in the activation of downstream effector substrates ERK and PI3-K leading to dysregulated cell growth and survival 3 . Among cancers in which KRAS mutations predominate, including pancreatic ductal adenocarcinomas (PDA), lung adenocarcinomas (LAC), and colorectal carcinomas (CRC), over 75% of amino acid substitutions occur at the G12 codon position 4 . The high frequency of G12 amino acid substitutions makes this codon position an ideal drug target; however, no small molecule inhibitors of G12 variants have been successfully developed aside from KRAS G12C that has only recently demonstrated clinical promise 5,6 .
Somatic gene mutations within cancer cells may be translated into peptides that are processed and presented on the surface of tumor cells 7,8 . These mutated peptides can serve as foreign epitopes, or neoantigens, that may be recognized by αβ T cells of the host immune system. Neoantigen-specific T cell responses have been well documented in patients for whom immune checkpoint blockade therapy has been successful, and they are believed to be the key mediators of anti-tumor activity 9 . Neoantigen-specific T cells can be isolated from the peripheral blood or tumor tissue of antigen-exposed cancer patients 10 , induced and expanded from the peripheral blood of cancer patients following neoantigen vaccination 11 , and generated in vitro utilizing the naive T-cell receptor (TCR) repertoire of healthy donors 12 . These observations have garnered interest in the development of neoantigentargeted cancer vaccines and adoptive T cell therapies.
Neoantigen-based treatment strategies are inherently highly personalized as somatic tumor mutations are rarely shared between patients 13,14 . The high prevalence and conserved mutational profile of KRAS affords a unique opportunity to develop a neoantigen-targeted therapy with broad generalizability. Mutant KRAS (mKRAS) has been previously studied as a potential target of cancer vaccines, and clinical studies have demonstrated the successful generation of CD4 + and CD8 + αβ T cell responses with reactivity against allogeneic or autologous mKRAS tumor cell lines [15][16][17][18] . More recently, mKRAS-specific T cells have been isolated and characterized from the peripheral blood of patients with mKRAS epithelial cancers 19,20 , and they have been induced in vaccinated HLA-transgenic mice 21 . The therapeutic potential of targeting mKRAS as a cancer neoantigen was highlighted in a case report demonstrating clinical benefit in a patient with KRAS G12D metastatic CRC following the adoptive transfer of KRAS G12D-specific T cells restricted to HLA-C*08:02 22 . However, mKRAS as an immunological target remains poorly characterized, and there is a dearth of evidence regarding tumor cell processing and presentation of mKRAS-derived epitopes to guide the development of targeted immunotherapies.
In this work, we aim to validate mKRAS as an immunological target. Utilizing computational epitope prediction, biochemical assays, and proteomic analysis, we predict and identify highaffinity and/or high-stability mKRAS G12 peptides to HLA-A*03:01, HLA-A*11:01, and HLA-B*07:02, and we confirm these epitopes as constituents of their corresponding HLA class I ligandome. The induction of mKRAS-reactive T cells from healthy donors by select peptides both confirms peptide immunogenicity and enables the isolation of mKRAS-specific TCRs directed against the following three epitopes: G12V/HLA-A*03:01, G12V/HLA-A*11:01, and G12R/HLA-B*07:02. Expression of these mKRAS-specific TCRαβ pairs in a NFATinducible reporter T cell line or TCRαβ null primary CD8 + T cells serves as sensitive probes to characterize the presentation of mKRAS G12 epitopes by tumor cell lines. TCR-redirected T cells display cytotoxic function against mKRAS tumor cell lines of various histologies without reactivity to wild-type (WT) KRAS peptides, thereby authenticating mKRAS G12V and G12R peptide ligands as bona fide neoantigens. Finally, the adoptive transfer of TCR redirected CD8 + human T cells specific for G12V/HLA-A*03:01 or G12V/HLA-A*11:01 leads to tumor eradication and prolongs survival in a mouse xenograft model of metastatic lung cancer. These findings provide further validation and strengthen the nomination of mKRAS as an immunological target.

Results
Identification and characterization of mKRAS neoepitopes. Although predictive peptide-HLA (p-HLA) binding affinity and stability are important criteria for neoantigen nomination [23][24][25] , only a small percentage of the human proteome is presented by HLA molecules 26 . Therefore, only a fraction of candidate neoantigens are constituents of the tumor HLA ligandome 27 , and predicting high-affinity/stability p-HLA complexes is not indicative of intracellular processing and presentation 28 . To improve the identification of bona fide mKRAS neoepitopes, we employed a comprehensive multi-omic approach consisting of (1) computational epitope prediction, (2) biochemical assessment of p-HLA binding / stability, (3) proteomic validation of antigen processing and presentation, and (4) characterization of immunogenicity ( Supplementary Fig. 1).
We utilized the neoantigen prediction tool antigen.garnish 24 to perform a comprehensive identification of candidate HLA class I allele-restricted mKRAS G12 epitopes with recurrent amino acid substitutions (G12C, G12D, G12R, G12V) embedded within the 24 amino acid sequence derived from the WT KRAS protein.
Given the high sensitivity of TCRs for p-HLA complexes, we sought to isolate TCRs directed against 7-16V and 10-19R epitopes in order to characterize their expression by tumor cell lines and validate these epitopes as neoantigens. To isolate TCRα/ β pairs, cell cultures containing CD8 + T cell responses to 7-16V/ HLA-A*03:01, 7-16V/HLA-A*11:01, and 10-19R/HLA-B*07:02 were expanded using peptide-pulsed artificial antigen presenting cells (APC). Antigen specificity was confirmed by p-HLA multimer staining (Fig. 2b-d), and multimer + cells were flow cytometrically sorted to >99% purity. TCRα/β sequences were determined by next-generation DNA and RNA sequencing 37 . A total of five TCRα/β pairs were identified, three of which were selected for further studies: TCRA3V (7-16V/HLA-A*03:01), TCRA11V (7-16V/HLA-A*11:01), and TCRB7R (10-19R/HLA-B*07:02) ( Table 1). Fig. 1 Identification, characterization, and validation of candidate mKRAS epitopes. a Heat maps displaying candidate WT and mKRAS epitopes with predicted HLA class I binding affinity (red) to the indicated alleles as determined by antigen.garnish, USA population HLA class I allele frequencies (blue) are as reported in the Allele Frequency Net Database, and mKRAS G12 variant frequencies (green) in pancreatic adenocarcinoma (PDA; n = 836), colorectal cancer (CRC; n = 866), and lung adenocarcinoma (LAC; n = 516) patients are as reported in the ICGC Data Portal. b Graphic depiction of the KRAS 24-mer containing candidate epitopes restricted by alleles shown in (a). The G12 codon is indicated in red. Epitopes are identified by colored lines, and amino acid positions as well as HLA class I restricting alleles are indicated. c Peptide affinity measurements of mKRAS epitopes to HLA-A*02:01, HLA-A*03:01, HLA-A*11:01, and HLA-B*07:02. p-HLA affinities were assessed by fluorescence polarization competitive peptide-binding assays 30 . High peptide affinity is classified as a log 10 [IC 50 ] < 3.7 nM and is indicated by the dashed line. d p-HLA complex stability measurements of mKRAS epitopes to HLA-A*02:01, HLA-A*03:01, HLA-A*11:01, and HLA-B*07:02. p-HLA stability measurements were determined by scintillation proximity assay of β2microglobulin dissociation 29 . For each HLA class I allele tested, β2-microglobulin association with MHC heavy chain in the absence of peptide served as negative controls. Displayed p-HLA affinity and stability values represent the mean of two independent tests. For affinity and stability assays, validated T cell epitopes for each allele were used as positive controls (open circles) and are listed in Supplementary Table 3. Error bars in panel (d) specify mean values ± SD of positive controls. mKRAS peptides with high measured binding affinity or stability are indicated as red circles. e Summary chart of MS/MS detected HLA class I restricted mKRAS epitopes processed and presented by TMC-engineered monoallelic cell lines (red). Validated viral epitopes (NLV, ILR, IVT, and TPR) encoded by the TMC construct were used as positive controls for epitope processing and presentation and are listed in Supplementary  Table 3. Source data are provided as a Source Data file. ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-24562-2 We initially characterized the expression and function of these mKRAS-TCRs using a cell reporter system developed in our laboratory designated J ASP90 reporter cells. The J ASP90 reporter cells are derived from CD8 + /TCRαβ null Jurkat E6.1 cells engineered to express NFAT-inducible eGFP as a readout for TCR signaling. mKRAS-TCRα/β chain pairs were cloned into lentiviral vectors and expressed in J ASP90 reporter cells. Custom p-HLA multimer staining was performed to validate TCRA3V, TCRA11V, and TCRB7R expression as well as confirm antigen specificity and HLA restriction (Fig. 3a).
To further assess antigen specificity and measure TCR avidity, TCR-engineered J ASP90 reporter cells were cocultured with HLA class I matched monoallelic K562 cells pulsed with WT or mKRAS synthetic peptides. TCRA3V, TCRA11V, and TCRB7Rexpressing J ASP90 reporter cells demonstrated specific reactivity to mKRAS cognate epitopes without reactivity to WT KRAS peptide-pulsed or unpulsed monoallelic K562 cells (Fig. 3b).
Using titrated peptide concentrations, the functional avidities of each TCR were determined based on the mean concentration required to achieve 50% NFAT activation (Fig. 3b). The avidities of TCRA3V, TCRA11V, and TCRB7R were determined to be 1.64, 0.15, and 3.38 nM, respectively. Furthermore, TCRA3V and TCRB7R cells exhibited no cross-reactivity to other mKRAS G12 epitopes, while TCRA11V cells demonstrated cross-reactivity to peptide 7-16C, albeit with decreased avidity as compared to the cognate peptide. These TCRs constituted high-affinity probes for the characterization of naturally processed and presented mKRAS epitopes by human tumor cells.
mKRAS-specific TCRs confer lytic activity against human tumor cells. To probe for p-HLA complexes on tumor cell lines, the anti-tumor activity of mKRAS TCRs was evaluated using engineered TCRαβ null primary CD8 + T cells. To this end, healthy donor primary CD8 + T cells were gene edited using CRISPR/   Fig. 9e). Secondary expansion of mKRAS-TCR T cells using peptide-pulsed artificial APCs led to enrichment of antigen-specific CD8 + T cells ( Supplementary Fig. 9f), yielding populations with high (60-90%) expression of TCRA3V, TCRA11V, or TCRB7R as assessed by p-HLA multimer staining (Fig. 4a). In 4 h 51 Cr-release assays, TCRA3V, TCRA11V, and TCRB7R-expressing CD8 + T cells kill HLA class I matched K562 cells pulsed with exogenous cognate mKRAS peptide but not WT KRAS peptide (Fig. 4b). In addition, recognition of processed and presented mKRAS antigen by TCRA3V, TCRA11V, and TCRB7R-expressing CD8 + T cells was demonstrated using TMC-expressing K562 cells as targets (Fig. 4b). Lack of recognition of processed and presented WT KRAS was demonstrated using a WT KRAS-expressing cell line (Fig. 4b, c). Altogether, peptide-pulsing experiments validate antigen specificity while TMC experiments, provide support for proteomic results (Fig. 1e), and validate presentation of cell surface mKRAS p-HLA complexes.
In order to evaluate the recognition of p-HLA complexes on the surface of cancer cells by TCRs, we characterized the expression of HLA class I alleles in a panel of mKRAS G12 cancer cell lines of various tissue origins. Cell lines with reported KRAS WT (TPM 23), G12V (TPM 56-308), and G12R (TPM 16-130) expression, as well as cell lines expressing target HLA class I alleles but alternative KRAS mutations were selected for further study (Supplementary Table 4). Aside from G12V/HLA-A*03:01, no cell lines expressing endogenous G12V/HLA-A*11:01 or G12R/HLA-B*07:02 were identified. When applicable, KRAS G12V + cell lines were modified to express either HLA-A*03:01 or HLA-A*11:01, and KRAS G12R + lines were modified to express HLA-B*07:02. As shown in Fig. 4c, d, TCRA3V and TCRA11Vexpressing CD8 + T cells exhibited specific cytotoxicity, irrespective of tissue of origin, against all HLA-A*03:01 or HLA-A*11:01-matched KRAS G12V + cancer cell lines, respectively. Of note, neither TCR conferred cytotoxic activity against HLA  Quantitation and recognition of mKRAS peptides presented by human tumor cells. To gain further insights into the processing and presentation of 8-16V and 7-16V peptides on the surface of cancer cell lines, we performed targeted MS and absolute peptide quantitation to enumerate p-HLA complexes on the surface of CORL23 lung tumor cells engineered to express HLA-A*03:01 (CORL23-A3) or HLA-A*11:01 (CORL23-A11) 40 . CORL23 was chosen for further studies given its ability to recapitulate a model of metastatic lung cancer (see below) 41 . In the context of HLA-A*03:01, we detected 8 complexes of 8-16V and 27 complexes of 7-16V per cell as compared to 43 complexes of 8-16V and 78 complexes of 7-16V per cell in the context of HLA-A*11:01 ( Fig. 5a and Supplementary Fig. 11). This result confirms processing and presentation data obtained using TMC-engineered HLA class I monoallelic cell lines (Supplementary Figs 5 and 6) and suggests that KRAS G12V + tumors present both 9-mer (8-16V) and 10-mer (7-16V) epitopes in the context of HLA-A*03:01 and HLA-A*11:01. Although these cells were genetically engineered to express the restricting HLA class I allele, HLA-A*03:01 and HLA-A*11:01 cell surface levels were noted to be within the range of expression comparable to other nonengineered HLA-A*03:01 + and HLA-A*11:01 + mKRAS tumor cells lines (Supplementary Fig. 12a, b). Importantly, enumerated p-HLA complexes are within the ranges reported for other mKRAS/HLA complexes expressed by non-engineered cancer cell lines 36 .
The cytotoxic activity of TCRA3V-and TCRA11V-engineered CD8 + T cells against CORL23 tumor cell lines was then characterized via live cell imaging and cellular impedance. These complimentary measurements of cell death allow for visualization of the loss of GFP-labeled tumor cells and the accumulation of Annexin V-CF594 dye (Supplementary Fig. 13 and Supplementary Movie 1), as well as cell shrinkage and fragmentation measurements over a 6-day period. TCRA3V cells promoted rapid apoptotic cell death of CORL23-A3 tumor cells at a 10:1 E:T ratio compared to control T cells, with slower killing kinetics at lower E:T ratios of 3:1 and 1:1 (Fig. 5b). TCRA11V cells promoted rapid cell death of CORL23-A11 cells at all E:T ratios (Fig. 5c). The kinetics to reach 50% cytolysis (KT50) was significantly faster for TCRA11V cells compared to TCRA3V cells at all E:T ratios (Fig. 5d), a finding consistent with the higher avidity of TCRA11V (Fig. 3b) and the higher number of 7-16V complexes detected in the context of HLA-A*11:01 expressed on the surface of CORL23-A11 cells.
We next evaluated the sensitivity of mKRAS TCRs to recognize p-HLA complexes in a non-engineered cell line. We identified the LAC cell line NCI-H441 as being both KRAS G12V + and HLA-A*03:01 + (Supplementary Table 4), highlighting it as a potential target of TCRA3V. We confirmed this cell line retained HLA class I expression and expressed low levels of HLA-A*03:01 (Supplementary Fig. 12a). In both 4 h 51 Cr-release and cellular impedance assays, TCRA3V cells exhibited specific lysis against parental (non-engineered) NCI-H441 cells, while TCRA11V cells, which recognizes the identical 7-16V epitope but restricted by HLA-A*11:01, exhibited no cytotoxic activity (Fig. 5e, f). Together, we conclude that TCRA3V and TCRA11V can "sense" low numbers of p-HLA complexes per cell and represent valuable reagents to immunologically target mKRAS.
In vivo anti-tumor activity of mKRAS-specific TCR+ CD8+ T cells. CORL23-A3 and CORL23-A11 tumor cells engineered to express click beetle red (CBR) luciferase were infused into NOD/ scid/γcnull (NSG) mice via tail vein injection, and the establishment of pulmonary tumors was confirmed by bioluminescence imaging. Four days post tumor inoculation, mice were treated with TCRA3V-or TCRA11V-engineered CD8 + T cells. As controls, tumor-bearing mice received either no treatment (Mock) or TCRαβ null (TCR KO) CD8 + T cells. Mice treated with TCRA3Vor TCRA11V-engineered CD8 + T experienced complete eradication of CORL23-A3 (Fig. 6a, c) or CORL23-A11 (Fig. 6b, d) tumors, respectively, relative to control groups (p < 0.01). The treatment effect was long-lasting with no evidence of tumor outgrowth by day 39 in mice treated with mKRAS-TCR therapy, while control mice experienced persistent tumor growth. Consequently, mice treated with mKRAS-TCR therapy experienced prolonged survival relative to control groups (p < 0.0002, Fig. 6e, f). Altogether, these results demonstrate that the transfer of T cells engineered with high avidity mKRAS TCRs promotes tumor rejection and prolongs survival.

Discussion
Epitope discovery is an essential step in designing immunotherapies such as cancer vaccines, bi-specific engagers, and TCR therapies. The ubiquity of mKRAS along with its role as a recurrent clonal driver makes this oncoprotein an ideal target [42][43][44] . To identify mKRAS epitopes, we undertook a comprehensive multi-omics approach (e.g., bioinformatic, biochemical, proteomic and immunological) to address the intricate processes involved in the generation of immunogenic peptides. By subjecting each candidate peptide to a series of assays that interrogate critical properties of antigen processing and presentation, we have identified a curated, albeit partial set of mKRAS peptides that represent candidate epitopes. Using this approach, we identified three epitopes (7-16V/HLA-A*03:01, 7-16V/HLA-A*11:01, and 10-19R/HLA-B*07:02) that fulfilled criteria of strong measured p-HLA binding affinity, high complex stability, and proteomically validated antigen processing and presentation. Furthermore, we determined that these epitopes were immunogenic allowing for the identification of mKRAS TCR sequences that were used as sensitive probes to validate epitope presentation by human cancer cell lines. Direct quantification of mKRAS p-HLA complexes on CORL23 cells highlights the sensitivity of mKRAS TCRs to recognize low abundance p-HLA complexes and eliminate mKRAS tumors cells in a xenograft mouse model of metastatic cancer. Fig. 4 mKRAS TCRαβ gene transfer confers HLA-restricted lytic activity against human tumor cells. a FACS plot demonstrating p-HLA multimer binding to TCRA3V-, TCRA11V-, and TCRB7R-engineered TCRαβ null primary CD8 + T cells gated on viable/CD3 + population. The following p-HLA multimers were used as staining controls: gp17-25/A3, gp17-25/A11, NY60-72/B7 (Supplementary Table 3). b 4 h 51 Cr-release cytotoxicity assay demonstrating TCRA3V, TCRA11V, and TCRB7 cell tumoricidal activity against TMC-expressing (red), mKRAS peptide-pulsed (blue), or WT KRAS peptide-pulsed (black) monoallelic K562 target cells. Percent-specific lysis of triplicates is shown for each data point. Data are presented as mean ± SD. p < 0.05 or p < 0.01 for all E:T ratios ≥0.3:1 comparing TMC-expressing or mKRAS peptide-pulsed to WT KRAS peptide-pulsed K562 targets, respectively; one-way ANOVA followed by post-hoc pair-wise Student's t-test with multiple comparison adjustment. c, d.  b Cellular impedance to determine the kinetics of tumor cell death upon TCRA3V lytic recognition of CORL23-A3 vs TCR Ctrl (TCRA11V). Normalized cell index values over time for E:T ratios of 10:1, 3:1, and 1:1 are shown compared to tumor cells cultured in media alone. Data for cells exposed to triton-X to represent maximal impedance loss are displayed. Data are presented as mean values ± SD. p < 0.001 at 150 h for all E:T ratios for the TCRA3V group compared to media; one-way ANOVA followed by post-hoc pair-wise Student's t-test with multiple comparison adjustment. Data are representative of two independent experiments. c Cellular impedance data following TCRA11V lytic recognition of CORL23-A11 vs TCR Ctrl (TCRA3V). Data are presented as mean values ± SD. p < 0.0001 at 150 h for all E:T ratios for the TCRA11V group compared to media; one-way ANOVA followed by post-hoc pair-wise Student's t-test with multiple comparison adjustment. Data are representative of two independent experiments. d Comparative kinetics of the time to reach 50% cell death (KT50) of CORL23-A3 and CORL23-A11 cells upon TCRA3V and TCRA11V recognition, respectively, at E:T ratios of 10:1, 3:1, and 1:1. Data are presented as mean values with independent replicates displayed. *p < 0.05, **p < 0.01, ***p < 0.001; multiple t-test analysis. Data are representative of two independent experiments. e 4 h 51 Cr-release cytotoxicity assay demonstrating TCRA3V tumoricidal activity against NCI-H441 cells (G12V + /HLA-A*03:01 + ) compared to TCR Ctrl (TCRA11V) cells. Data are presented as mean values ± SD. p < 0.01 for all E:T ratio ≥ 1:1; one-tailed Student's t-test. Data are representative of two independent experiments. f Cellular impedance data following TCRA3V lytic recognition of NCI-H441 vs TCR Ctrl (TCRA11V). Data are presented as mean values ± SD. p < 0.0001 at 130 h for all E:T ratios for the TCRA3V group compared to media; one-way ANOVA followed by post-hoc pair-wise Student's t-test with multiple comparison adjustment. Data are representative of two independent experiments. Source data are provided as a Source Data file. complexes, and only the 5-14V peptide demonstrated high HLA class I affinity in competitive binding assays. By contrast, Mishto et al. 45 reported mKRAS linear (5-14V) and spliced (5-6/8-14V, lacking the V at position 7 of the canonical peptide) peptides as products of in vitro proteasome digests and have demonstrated these peptides to have high affinity for HLA-A*02:01. Given the high prevalence of HLA-A*02:01 in the population, further studies are needed to evaluate the discordant results between processing of the 5-14V mKRAS peptide by cell lines and in vitro proteasome digest assays.
The availability of TCRs directed against 7-16V/HLA-A*03:01 (TCRA3V), 7-16V/HLA-A*11:01 (TCRA11V), and 10-19R/ HLA-B*07:02 (TCRB7R) permitted their use as sensitive probes demonstrating the presence of these p-HLA complexes on the cell surface of various cancer cell lines from multiple histologies. Our observation that NCI-H441 can be recognized by TCRA3Vengineered CD8 + human T cells is noteworthy given this KRAS G12V mutant cell line naturally expresses the HLA-A*03:01 restricting allele. Further insights were gained by the quantification of p-HLA complexes on CORL23 cell lines demonstrating expression of both 9-mer and 10-mer peptides in the context of HLA-A*03:01 and HLA-A*11:01. As TCRA3V and TCRA11V recognize the same 7-16V peptide, we hypothesized differences in TCR avidity as well as p-HLA abundance by CORL23 lines may contribute to the distinct kinetics of in vitro killing. Indeed, higher expression of 7-16V/A*11:01 complexes (78 complexes/ cell) relative to 7-16V/A*03:01 (27 complexes/cell) correlates with the more rapid elimination of CORL23-A11 relative to CORL23-A3 tumor cells. Importantly, the adoptive T cell transfer of either TCRA3V or TCRA11V engineered CD8 + human cells efficiently eliminated CORL23 tumor cells in vivo and proved curative for a majority of treated mice.
In summary, epitope identification and validation strategies reliant on a contemporary toolbox of bioinformatic, biochemical, genomic, and immunological assays have recently been employed by several groups in the development of personalized melanoma vaccines targeting passenger mutations 11,47,48 . Here, we have employed a similar strategy to characterize epitopes presented by high prevalence HLA class I alleles targeting a recurrent clonal driver-mKRAS. Our findings provide further evidence highlighting this oncogenic protein as a clinical target of immunebased therapies.

Methods
Primary cells. Peripheral blood mononuclear cells and purified CD8 + T cells were provided by the Human Immunology Core after cell isolation from apheresis products of HLA class I-and class II-typed healthy donors enrolled on the Institutional Review Board-approved research protocol 705906 at the University of Pennsylvania after providing informed consent. Computational prediction of mKRAS G12 neoepitopes. We utilized the tumor neoantigen prediction and multi-parameter quality analysis tool antigen.garnish 24 to identify HLA class I binding peptide ligands of mKRAS G12. The input file consisted of the amino acids 1-24 of WT KRAS as well as the most prevalent mKRAS somatic G12 variants (G12C, G12D, G12R, and G12V) screened against HLA class I alleles with a prevalence >1% of the USA population. Nonamer and decamer peptide ligands with ensemble predicted binding affinities <500 nM to target HLA class I molecules were designated candidate mKRAS binding ligands. HLA prevalence among Caucasians, African Americans, Hispanics, and Asians represented in the USA population were gathered from the Allele Frequency Net Database 49 Table 4. Tumor cell lines were transduced with lentiviral particles expressing HLA class I/β2-microglobulin single-chain dimer constructs (HLA-SCD, eGFP + ) to generate mKRAS G12/HLA class I-matched cell lines. HLA class I cell surface expression was assessed by flow cytometry using APC-conjugated anti-human HLA-A, B, C (clone W6/32). HLA-allele-specific antibodies conjugated to PE, APC, or biotin were used to assess target HLA expression.
Lentiviral vector constructs and production. All lentiviral constructs were generated using the third generation lentiviral transfer vector pTRPE-eGFP-T2A-mCherry (generously provided by Dr Michael C. Milone, University of Pennsylvania, Philadelphia, PA). The HLA-A*02:01-SCD was PCR amplified from a HLA-A*02:01-SCD plasmid 52 and used to replace the mCherry moiety on pTRPE-eGFP-T2A-mCherry vector backbone. All other HLA class I-SCD constructs were generated by exchanging HLA-A*02:01 with HLA-A*03:01, HLA-A*11:01 or HLA-B*07:02. To generate the TMC, eGFP within the lentiviral transfer vector backbone was replaced with ubiquitin to generate pTRPE-Ubiquitin-T2A-mCherry. A synthetic DNA vector construct was synthesized (TWIST Bioscience, San Francisco, CA) encoding the first 25 amino acids of WT and mKRAS (G12C, G12D, G12R, G12V) expressed in tandem without linker sequences (Supplementary Fig. 4). The TMC also included viral epitopes embedded within natural 20 amino acid viral protein sequences restricted to HLA-A*02:01 (NLVPMVATV), HLA-A*03:01 (ILRGSVAHK), HLA-A*11:01 (IVTDFSVIK), and HLA-B*07:02 (TPRVTGGGAM). The TMC was PCR amplified and inserted between ubiquitin and T2A sequences. Synthetic TCR DNA vector constructs were synthesized (TWIST Bioscience, San Francisco, CA) to include the TCRα chain followed by the TCRβ chain separated by a T2A sequence. TRAC and TRBC regions of synthetic TCRs were codon altered to make them resistant to Cas9 protein riboprobes targeting endogenous TRAC and TRBC1/TRBC2. TCR constructs were PCR amplified and cloned into a lentiviral vector backbone. High titer lentiviral vector production was performed 53,54 .
Monoallelic KRAS-tandem minigene construct (TMC) cell lines. HLA class I monoallelic cell lines were generated by transduction of K562 or 721.221 cell lines using lentiviral particles encoding HLA-SCD/eGFP 52 . Cells were subsequently transduced with lentiviral particle encoding TMC/mCherry then sorted to >99% purity by flow cytometry based on coexpression of GFP, mCherry, and HLA class expression as measured by APC-conjugated anti-human HLA-A, B, C (clone W6/ 32) staining.
Proteomic validation of mKRAS antigen processing and presentation Sample preparation. Monoallelic KRAS-TMC K562 and 721.221 cell lines were expanded to 1-2 × 10 8 total cells and HLA class I immunopurification (IP) was performed 31 using MHC class I (W6/32) antibody non-covalently linked to agarose beads (Santa Cruz Biotechnology, Dallas, TX). Peptides were eluted from HLA class I molecules using 0.1% trifluoroacetic acid (TFA). Immunoprecipitation eluent was passed through a 10,000 Da Amicon molecular weight cut off filter (Merck Millipore) via centrifugation at 10,000 × g for 10 min. Filtered eluent was dried and resuspended in 100 μl of 0.1% TFA. Stage tip C18 columns (Harvard Apparatus) were equilibrated with 200 μl of acetonitrile and 200 μl of 0.1% TFA, and samples were loaded onto columns for desalting. Samples were then washed with three rounds of 200 μl of 0.1% TFA then eluted in 70% acetonitrile in 0.1% formic acid (FA) and dried.
Data-dependent acquisition (DDA). DDA analysis of monoallelic KRAS-TMC K562 and 721.221 cell lines was performed by the Quantitative Proteomics Resource Core at the University of Pennsylvania. DDA experiments were carried out using half of each enriched sample by nano LC-MS/MS using an Orbitrap Fusion (Thermo Scientific) coupled to an Easy-nLC system (Thermo Scientific). Samples were resuspended in 10 μl of 0.1% TFA and 5 μl was injected for each analysis. Samples were separated with an in-house packed column with ReproSil-Pur C18 AQ 3 μm resin with dimensions 75 μm × 20 cm (Dr Maisch GmbH, Ammerbuch, Germany) at a flow rate of 400 nl/min. Using 0.1% FA as buffer A and 80% ACN 0.1% FA as buffer B, peptides were eluted with a gradient of 4% buffer B to 35% buffer B in 50 min, then to 65% buffer B in 5 min, and finally a 5 min wash at 95% buffer B. Peptides were ionized in a Nanospray Flex Ion Source (Thermo Scientific) at 2.3 kV. An MS1 scan was acquired using a resolution of 120,000 FWHM, AGC target of 4e5, and maximum inject time of 50 ms. Precursors were isolated with 1.6 m/z isolation window and fragmented with an HCD collision energy of 30 eV. An MS2 scan was acquired using a resolution of 30,000 FWHM, AGC target of 5e4, and maximum ion inject time of 54 ms in 3 s cycle time scan mode. DDA data were processed with PEAKS X+ version 10.5 with 10 ppm parent mass tolerance, 0.02 Da fragment mass error tolerance, and no enzyme specificity. Data were searched against the SwissProt Human proteome appended with the custom KRAS peptide sequences.
Parallel reaction monitoring (PRM). PRM analysis of monoallelic KRAS-TMC K562 and 721.221 cell lines was performed by the Quantitative Proteomics Resource Core at the University of Pennsylvania. Targeted Analysis by PRM was performed on the remaining half of each sample to increase the sensitivity of mKRAS epitope detection. Samples were analyzed on a Q-Exactive HF-X (Thermo Scientific) coupled to an Ultimate 3000 nano UHPLC system (Thermo Scientific) using a PRM strategy 32 . Samples were resuspended in 10 μl of 0.1% TFA and 5 μl was injected for each analysis. Samples were separated with an in-house packed column with ReproSil-Pur C18 AQ 3 μm resin with dimensions 75 μm × 20 cm (Dr Maisch GmbH, Ammerbuch, Germany) at a flow rate of 400 nl/min. Using 0.1% FA as buffer A and 80% ACN 0.1% FA as buffer B, peptides were eluted with a gradient of 4% buffer B to 30% buffer B in 42 min, then to 65% buffer B in 6 min, followed by a 7 min wash at 95% buffer B, and a 5 min equilibration at 4% buffer B. Peptides were ionized in a Nanospray Flex Ion Source (Thermo Scientific) at 2.3 kV. An MS1 scan was acquired using a resolution of 120,000 FWHM, AGC target of 1e5, and maximum inject time of 50 ms. An inclusion list of 33 unique ions was used to sequentially isolate and fragment each target peptide with an isolation window of 1.6 m/z and an HCD collision energy of 28 eV. An MS2 scan was acquired using a resolution of 15,000 FWHM, AGC target of 1e5, maximum ion inject time of 100 ms, and loop count of 10. Results were analyzed with Skyline 55 (version 20.1) and reference spectra of synthetic peptides (New England Peptide, Gardner, MA) were used to validate fragmentation patterns. Spectra were manually annotated with IPSA 56 . Synthetic peptides were analyzed on a Q-Exactive HF (Thermo Scientific) coupled to an Ultimate 3000 nano UHPLC system (Thermo Scientific) with DDA. Each synthetic peptide was injected at 1 pmol/μl and analyzed with similar liquid chromatography conditions. An MS1 scan was acquired at a resolution of 120,000 FWHM, AGC target of 3e6, maximum inject time of 32 ms. The top 20 intense ions were isolated and fragmented with a dynamic exclusion of 45 s. For each peptide, an MS2 scan was acquired using a resolution of 15,000 FWHM, AGC target of 2e5, maximum ion inject time of 32 ms, and isolation window of 1.4 m/z. Ions were filtered for charges 2-5. Raw files were searched using Proteome Discoverer 2.2 (Thermo Scientific) against a database of targets with non-trypic digestion, precursor mass tolerance of 10 ppm, and fragment mass tolerance of 0.02 Da. Search results were filtered with the target decoy approach 40 . The validation of the precursor ion identity was based on isotopic dot product (idotp) and dot product (dotp) values generated by Skyline 33 . Both values provide a measurement of similarity between eluted and expected or reference peptide. The idotp value provides a measure to assess the precursor isotope distribution and its correlation between observed (eluted) and expected (synthetic) peptide while the dotp value provides a measure of similarity between observed and library spectrum (Proteome Discoverer Database) peak areas. Values range from a best of 1.0 to a worst of 0.0.
Proteomic quantitation of mKRAS peptides presented by human tumor cells Sample preparation. CORL23-A3 and CORL23-A11 cells were expanded to 1-2 × 10 8 total cells. HLA class I IP was performed by Cayman Chemical (Ann Arbor, MI). Cells were resuspended in MHC lysis buffer consisting of DPBS (0.25 % Sodium deoxycholate, 200 μM iodoacetamide, 1% N-Octyl-β-D-thioglucoside, 1 mM EDTA) containing protease inhibitor (1 ml of lysis buffer per 1 × 10 8 cells). Cell pellets were resuspended by pipetting and vortexing six times for 3-4 s each at 5-min intervals. Lysate was centrifuged at 800 × g for 5 min and the supernatant was transferred to a 50-ml tube. Additional lysis buffer was added to the supernatants (to double the volume) prior to centrifugation at 20,000 × g for 60 min. The supernatants were transferred to a new 15 ml conical tube. Then, 200 μl from each tube was saved for analysis (Pre-IP). The cell lysate was then mixed with 100 μl of W6/32 resin and incubated overnight at 4°C with gentle rotation. The IA resin (W6/32 antibody DMP crosslinked to protein A resin) was washed with lysis buffer before use. The mixture was centrifuged at 800 × g for 5 min and the supernatant was transferred to a new tube (500 μl of the supernatant (Post-IP) was retained for analysis). The IA resin pellets were washed in 2.5 ml of lysis buffer, centrifuged at 800 × g for 5 min at 4°C, and the supernatant was aspirated. Two additional washes were carried out using 2.5 ml each of wash buffers 2 (20 mM Tris-HCL, 400 mM Nacl, pH 8.0) and 3 (20 mM Tris-HCL, 150 mM Nacl, pH 8.0). The IA resin was transferred to a low protein binding Eppendorf tube for a final wash. 0.75 ml of wash buffer 4 (20 mM Tris-HCL, pH 8.0) was added, mixed, centrifuged at 800 × g for 5 min and the supernatant was aspirated. Peptides were eluted from MHC class I molecules using 1 ml of MHC-I elution buffer (0.1 M Acetic acid, 0.1% TFA), incubated for 5 min at 37°C and centrifuged at 800 × g for 5 min at 4°C. The eluate was collected into fresh low protein binding tubes, and was centrifuged at 20,000 × g for 2 min to clear the sample of any remaining resin. The eluate was collected into a new low protein binding tube, flash frozen in liquid nitrogen, and stored at −80°C. A quality assurance ELISA assay was performed to quantify the enrichment of HLA class I protein in Pre-and Post-IP samples.
Data-dependent acquisition (DDA). DDA analysis of CORL23 cell lines was performed by MS Bioworks (Ann Arbor, MI). DDA experiments were carried out using half of each enriched sample by nano LC-MS/MS using a Waters M-Class system interfaced to a ThermoFisher Fusion Lumos mass spectrometer. Peptides were loaded on a trapping column and eluted over a 75-μm analytical column at 350 nl/min; both columns were packed with Luna C18 resin (Phenomenex). A 2 h reverse phase gradient was employed. The mass spectrometer was operated in a combined data-dependent EThcD/CID mode, with MS and MS/MS performed in the Orbitrap at 60,000 FWHM resolution and 15,000 FWHM resolution, respectively. The instrument was run with a 3-s cycle for MS and MS/MS. DDA data were processed with PEAKS X+ version 10.5 with 10 ppm parent mass tolerance, 0.02 Da fragment mass error tolerance, and no enzyme specificity. Data were searched against the SwissProt Human proteome appended with the custom KRAS peptide sequences. Peptide 8-16V (VVGAVGVGK) and 7-16V (VVVGAVGVGK) identifications were confirmed with analysis of synthetic peptide standards (New England Peptide).
Quantification of mKRAS epitopes by parallel reaction monitoring (PRM). The quantification of mKRAS epitopes expressed by HLA-engineered CORL23 cell lines was performed by targeted MS using PRM (MS Bioworks, Ann Arbor, MI). Peptides were enriched as described above. Synthetic stable labeled peptides VVGAVGVGK^and VVVGAVGVGK^were purchased from New England Peptide, where^is Lysine ( 13 C 6 15 N 2 ). In all, 200 fmols of each stable labeled peptide was added to the enriched samples for analysis. Each enriched sample was analyzed in duplicate (50% of the sample per injection). PRM was performed with a Waters M-Class HPLC system interfaced to a ThermoFisher Fusion Lumos mass spectrometer. Peptides were loaded on a trapping column and eluted over a 75-μm analytical column at 350 nl/min; both columns were packed with Luna C18 resin (Phenomenex). The mass spectrometer was operated in PRM mode with the Orbitrap operating at 17,500 FWHM resolution. Collision-induced dissociation data were collected for the (M+2H) 2+ charge state ions of the target peptides VVGAVGVGK, VVGAVGVGK^, VVVGAVGVGK, and VVVGAVGVGK^. Extracted ion chromatograms for each of the target peptides were generated manually using the XCalibur QualBrowser software version 4.1.31.9 (Thermo-Fisher) with a 20 ppm mass tolerance for product ions. For the calculation of 8-16V and 7-16V epitopes expressed by HLA-A*03:01 and HLA-A*11:01 complexes, eluted and internal standard peptide peak area data were used to calculate the number of moles of peptide present in the sample. This number was doubled (half the IP was analyzed in a single injection) and converted to molecules by multiplying by Avogadro's number. The result was divided by the number of input cells (164M and 163M for the CORL23-A3 and CORL23-A11 cell lines, respectively) to give number of peptide molecules/cell.
IFN-γ ELISPOT assay. CD8 + T cell reactivity to peptide antigen was assessed by interferon-γ (IFN-γ) ELISPOT assay 58 . The spot number was determined in an independent blinded fashion (ZellNet Consulting, New York, NY) using the highresolution automated KS ELISPOT reader (Zeiss, Thornwood, NY) and KS ELI-SPOT 4.9.16 software with reading parameters established per International Harmonization Guidelines 59 . A positive response was recorded if the number of spots in the peptide-exposed wells was two times or more higher than the number of spots in the unstimulated wells and if there was a minimum of 20 (after subtraction of background spots) peptide-specific spots per 5 × 10 5 CD8 + cells.
p-HLA multimer assay. mKRAS-specific CD8 + T cell frequencies were determined by staining with PE-conjugated p-HLA dextramers (Immudex), followed by addition of APC-conjugated anti-human CD8 antibody (Invitrogen). Cells were washed and resuspended in FACS buffer. One million events in the CD8 + gate were collected using a hierarchical gating strategy. Data were acquired using BD FACSDiva software (version 8.0.2) and analyzed using FlowJo software (version 10.7.2).
TCRα/TCRβ sequencing. DNA was extracted using the Gentra Puregene cell kit following the manufacturer's directions (Qiagen, Valencia, CA, Cat. No. 158388). Bulk DNA TCR Vβ and Vα libraries were prepared for sequencing on the Illumina MiSeq platform using a cocktail of 23 Vβ families from framework region 2 (FR2) forward primers, and 13 Jβ region reverse primers, modified from the BIOMED2 primer series 60 and a cocktail of 39 Vα from FR3 primers and 50 Jα region reverse primers, respectively. Libraries were generated using the QIAGEN Multiplex PCR and Illumina Nextera XT index kits 61  Data analysis. Raw sequences were quality filtered 62 and clone assemblies were processed with MiXCR (v. 3.0.7) 63 and VDJtools (v1.2.1) 64 .
Jurkat reporter system to assess TCR antigen specificity and avidity. Using Cas9 protein and riboprobes as guides against TRAC (5'-TGTGCTAGACATGA GGTCTA-3') and TRBC1/TRBC2 (5'-GGAGAATGACGAGTGGACCC-3') 38 , a Jurkat E6.1 TCRα/β-negative cell line was generated. To generate J ASP90 reporter cells, TCR-negative Jurkat E6.1 cells were then transduced with lentiviral particles encoding human CD8αβ heterodimer (generously provided by Dr Jim L. Riley, University of Pennsylvania, Philadelphia, PA) and the Uni-Vect reporter system that features constitutive mCherry and NFAT-inducible eGFP expression as a means to assess TCR signaling (generously provided by Dr Daniel J. Powell Jr., University of Pennsylvania, Philadelphia, PA). J ASP90 reporter cells were flow cytometrically sorted to purity based on TCR -(anti-human TCRα/β antibody clone IP26, BioLegend, San Diego, CA), CD8 + (anti-CD8 antibody, ThermoFisher Scientific, Waltham, MA), mCherry + expression, and low basal NFAT (eGFP + ) signal. The TCRα and TCRβ chains of TCRA3V, TCRA11V, and TCRB7R were expressed via lentivirus in J ASP90 using the pTRPE vector to generate the TCRA3V, TCRA11V, and TCRB7R J ASP90 reporter cell lines. These cell lines were sorted to yield a unimodal p-HLA multimer + cell population and expanded for use in functional assays. Sorted TCRA3V, TCRA11V, and TCRB7R J ASP90 reporter cell lines were mixed 1:1 with HLA-SCD-expressing K562 cells pulsed with titrated peptide concentrations (10 μM-1 pM). After 16-20 h, cells were analyzed by flow cytometry to determine the percentage of eGFP + cells in each sample. Cells activated with PMA (50 ng/ml) and Ionomycin (750 ng/ml) were included as positive controls (%GFP max ), and J ASP90 reporter cells cultured in media alone were used as negative controls (%GFP min ). Data were fitted to a dose-response curve by linear non-regression analysis using GraphPad Prism version 7.0.
TCRs as probes for recognition of mKRAS epitopes. Tumor cell lines (2.5 × 10 5 cells/well) were cocultured with TCR transduced J ASP90 cells at a 1:1 ratio in 48well flat bottom plates in a final volume of 500 μl of media (RPMI, 10% FBS, L-Glutamine, Penicillin/Streptomycin). Each tumor cell line was tested in its parental (non-HLA matched) and HLA-engineered (HLA-matched) form using lentiviral particles encoding HLA-SCD constructs of interest and flow cytometric sorting to >99% purity based on the coexpression of the target HLA class I allele and eGFP. G12R + tumor cell lines were pre-cultured with IFN-γ (500 U/ml) for 48 h prior to co-culture with TCRB7R transduced J ASP90 cells. After 16-20 h of incubation at 37°C, cells were analyzed by flow cytometry to determine the percentage of live, CD8 + mCherry + eGFP + cells in each sample. Cells activated with PMA and Ionomycin were included as positive controls.
Expression of mKRAS-specific TCRs in primary CD8 + T cells Primary T cell expansion. On day 0, human primary CD8 + T cells were activated and expanded using anti-CD3/CD28 antibody-conjugated paramagnetic microbeads (Life Technologies). On day 1 the cells were transduced with lentiviral vector particles containing TCRA3V, TCRA11V, or TCRB7R genes at an MOI of 5. Cell cultures and maintained in media supplemented with IL-7 (5 ng/ml; R&D Systems) and IL-15 (5 ng/ml; R&D Systems). On day 5, T cell cultures were harvested and paramagnetic microbeads were magnetically removed prior to electroporation.
Secondary T cell expansion. T cell cultures were harvested and re-stimulated with peptide-pulsed (500 ng/ml) irradiated (10,000 Rads) HLA-SCD/4-1BBL expressing K562 artificial APCs at a 1:1 ratio. Cultures were supplemented with IL-7 and IL-15 every 48 h until culture termination. Engineered TCR expression was then assessed by CD3, CD8, and TCRαβ expression as well as p-HLA multimer binding then cryopreserved for use in subsequent assays.
Cytotoxicity assays 51 Cr-release assay. K562 cells and KRAS-mutated tumor cell lines were labeled with 25 μCi 51 Cr in the presence or absence of peptide (10 μM) for 1 h at 37°C, washed and tested as targets in a standard 4 h 51 Cr-release assay. Effector cells consisted of primary gene-edited (TCRα-/TCRβ-) mKRAS-specific TCR CD8 + T cells, designated TCRA3V, TCRA11V, TCRB7R. Transgenic TCR expression was assessed by p-HLA multimer assay. Assays were performed, in triplicate, at various effector: target ratios. Data were collected using a MicroBeta2 LumiJET Microplate Counter (PerkinElmer). Data are represented as percent-specific lysis reported as mean ± standard deviation (SD).
Real-time apoptotic cell death analysis. Real-time apoptotic cell death analysis (live cell imaging with cellular impedance) was performed to assess extended cytotoxic activity using the xCELLigence Real Time Cell Analysis eSight system (ACEA Biosciences). Target tumor cells were plated (1 × 10 4 cells/well) and allowed to adhere for 24 h. Effector T cells were added at E:T ratios 10:1, 3:1, and 1:1, and the media was supplemented with Annexin V-CF594 (Biotium, Fremont, CA). Cell index (relative cell impedance) was monitored every 15 min for 5 days and normalized to the maximum cell index value immediately prior to effector-cell plating. Shaded lines reflect the mean of duplicate wells ± SD. Concurrent time lapse video monitoring was performed with acquisition of brightfield, green (GFP), and red (CF594) windows every hour.
CORL23 lung xenograft mouse model. All animal procedures were performed according to the approved University of Pennsylvania Institutional Animal Care and Use Committee (IACUC) protocol 804226. NSG (NOD/scid/γcnull) mice were purchased from the University of Pennsylvania Stem Cell Xenograft Core (SCXC) and housed in micro-isolator cages under sterile conditions in the SCXC AAALAC accredited animal facility at the University of Pennsylvania (temperature: 24°C, humidity: 50-60%, dark/light cycle: 12 h/12 h). The CORL23 xenograft model was established by intravenous tail injections of 5 × 10 5 CORL23-A3 or CORL23-A11 cells expressing CBR luciferase for bioluminescence imaging (BLI). On day 4 post tumor inoculation, mice were treated with 1 × 10 7 mKRAS-TCR engineered CD8 + T cells (TCRA3V for CORL23-A3 Cohort, TCRA11V for CORL23-A11 Cohort). Control mice either received no treatment (Mock) or 1 × 10 7 TCR KO CD8 + T cells. BLI was used to monitor tumor growth by intraperitoneal injections of 3 mg D-luciferin/150 ul PBS per mouse (~120 mg/kg) and imaged after 10 min using an IVIS Spectrum imaging system. Data were analyzed using Living Image Version 4.5.2 software (PerkinElmer).
Statistics and reproducibility. Statistical analysis of multiple comparisons was performed using one-way ANOVA with Tukey's HST post-test, and comparisons between just two groups were performed using Student's unpaired t-test. Significance of overall survival was determined via Kaplan-Meier analysis with logrank (Mantel-Cox) analysis. Data were analyzed using Microsoft Excel (version 16.16.27), and all statistical analyses were performed with Graphpad Prism 7.0 (GraphPad), except analysis of tumor growth and survival curves, which were performed using linear mixed-effects model with Tukey's HSD post-test using the Ime4 and the survival package in R. Error bars represent the SD, and p < 0.05 was considered statistically significant. *p < 0.05, **p < 0.01, ***p < 0.001, and ****p < 0.0001 unless otherwise indicated. ns denotes not significant. All data presented are representative of two or more independent experiments.