The identification of effective tumor-suppressing neoantigens using a tumor-reactive TIL TCR-pMHC ternary complex

Neoantigens are ideal targets for cancer immunotherapy because they are expressed de novo in tumor tissue but not in healthy tissue and are therefore recognized as foreign by the immune system. Advances in next-generation sequencing and bioinformatics technologies have enabled the quick identification and prediction of tumor-specific neoantigens; however, only a small fraction of predicted neoantigens are immunogenic. To improve the predictability of immunogenic neoantigens, we developed the in silico neoantigen prediction workflows VACINUSpMHC and VACINUSTCR: VACINUSpMHC incorporates physical binding between peptides and MHCs (pMHCs), and VACINUSTCR integrates T cell reactivity to the pMHC complex through deep learning-based pairing with T cell receptors (TCRs) of putative tumor-reactive CD8 tumor-infiltrating lymphocytes (TILs). We then validated our neoantigen prediction workflows both in vitro and in vivo in patients with hepatocellular carcinoma (HCC) and in a B16F10 mouse melanoma model. The predictive abilities of VACINUSpMHC and VACINUSTCR were confirmed in a validation cohort of 8 patients with HCC. Of a total of 118 neoantigen candidates predicted by VACINUSpMHC, 48 peptides were ultimately selected using VACINUSTCR. In vitro validation revealed that among the 48 predicted neoantigen candidates, 13 peptides were immunogenic. Assessment of the antitumor efficacy of the candidate neoepitopes using a VACINUSTCR in vivo mouse model suggested that vaccination with the predicted neoepitopes induced neoantigen-specific T cell responses and enabled the trafficking of neoantigen-specific CD8 + T cell clones into the tumor tissue, leading to tumor suppression. This study showed that the prediction of immunogenic neoantigens can be improved by integrating a tumor-reactive TIL TCR-pMHC ternary complex.


Patient materials
Fresh frozen tumor tissue was obtained during the surgical procedure and collected in RPMI 1640 (Welgene, Daegu, Korea) supplemented with 10% fetal bovine serum (FBS, Gibco).Peripheral blood (PB) from the same patient was collected up to 20 mL in EDTA tubes (BD, USA).A portion of the tumor tissue was excised for whole exome sequencing (WES) and whole transcriptome sequencing (WTS), while the remaining tissue was used for single-cell RNA-sequencing (scRNA-seq) following dissociation.The tumor tissues were enzymatically dissociated into single-cell suspensions using the tumor dissociation kit (Miltenyi Biotec,   Germany).CD8 T cells were isolated from the single-cell suspensions using EasySep human CD8+ T cell isolation kit (STEMCELL Technologies, Canada).The isolated CD8 T cells were used for single cell RNA and TCR sequencing.PB was used for WES and in vitro immunogenicity screening assay.

WES, WTS, scRNA, scTCR sequencing
Genomic DNA samples (33 matched blood and tumor tissue) underwent library construction using the Twist Human Core Exome kit (Twist Bioscience, USA).The captured DNA libraries were sequenced with paired end reads of 150 bp on a NovaSeq6000 or NextSeq550 (Illumina, USA).
RNA capture with poly-A tail was performed using magnetic Oligo(dT) beads.RNA integrity was measured using Agilent 4200 TapeStation.The RNA libraries were constructed using the Illumina Stranded mRNA Prep Kit (Illumina).These libraries were sequenced with paired end reads of 150 bp on a NovaSeq6000 or NextSeq550.
The scRNA-seq libraries were prepared using the Chromium Next GEM Single Cell 5ʹ Kit v2 (Dual Index) of Chromium platform (10x Genomics, USA) following the manufacturer's instruction.cDNA library quality was determined using an Agilent Bioanalyzer (Agilent, USA).The scRNA libraries were sequenced on the Illumina NovaSeq 6000 sequencing platform using the following read lengths: read 1:50; i7 index:10; i5 index:10; read 2:100.
The scTCR libraries were enriched via PCR amplification using a Chromium Single-cell V(D)J library (v1.0 Chemistry).Around 12,000 cells were loaded per sample, with the targeted cell recovery of 8,500 cells according to the protocol.Single cells were isolated and lysed, mRNA was converted into barcoded cDNA through reverse transcription using provided reagents (10x Genomics).13 PCR cycles were used to amplify cDNA and generated barcoded cDNA libraries for single cell 5`library.Part of the amplified cDNA was targetenriched for TCRs and V(D)J library was obtained according to manufacturer protocol.cDNA library quality was determined using an Agilent Bioanalyzer (Agilent).Barcoded libraries (included VDJ libraries) were pooled and sequenced with paired end on the Illumina Novaseq6000, using the following read lengths: read 1:50; i7 index:10; i5 index:10; read 2:100.

Single cell data processing
The FASTQ sequencing reads were processed using Cell Ranger version 5.0.0 (10x Genomics) and the GRCh38 human transcriptome reference for alignment.Data preprocessing was performed using Seurat v4.3.0 1 , applying quality control criteria to filter out cells with a mitochondrial gene percentage greater than 20%.Only T cells detected in the TCR sequence from scTCR sequencing data were retained for downstream analysis.Specifically, T cells with CD8 expression without CD4 expression were selected for further analysis.
Then, the raw count data were normalized, and scaling was performed by regressing out the mitochondrial gene percentage.Principal component analysis (PCA) was conducted using the 3,000 most variable features by Seurat.To avoid clonotype bias, TCR genes were removed from the variable features.Clusters were manually annotated using CD8+ T cell subtype markers to identify specific subtypes.We analyzed the clonal expansion of putative tumor-reactive TILs using the STARTRAC package 2 .

Mapping, data processing and gene expression quantification
We used the Burrows-Wheeler Aligner MEM algorithm BWA-MEM (v0.7.17) 3 for WES and STAR aligner (v2.7.8a) 4 for WTS data to align to the reference genome (hg38).Alignment and data processing were carried out following the Genome Analysis Toolkit (GATK) Best Practices guidelines for data pre-processing 5 .

In silico HLA typing
The 4-digit human leukocyte antigen (HLA) class I typing was performed using OptiType (v1.3.2) 11 , a novel HLA genotyping algorithm based on integer linear programming, and the previously mapped DNA sequencing data from the tumor and blood samples.

in vivo Proof-of-Concept (PoC) of VACINUS platforms [VACINUSpMHC + VACINUSTCR] for optimized selection of immunogenic neoantigen
As the binding affinities between peptides and murine MHC class I molecules predicted higher compared to those in humans, our selection criteria for neoantigens have been adjusted.Criteria for selection of neoantigens include a binding affinity threshold of less than 200 nM for H2-Kb-peptide and less than 400 nM for H2-Dbpeptide interactions.
Key marker genes used were different from those of humans.In mice, tumor-reactive TILs included activated, exhausted, and proliferative T cells, as determined by the expression levels of Gzma and Gzmk, Tnfrsf9 and Pdcd1, and Mki67 and Pcna, respectively.
As pMTnet does not support the mouse tumor genomics data, we needed an equivalent model for the murine system.We obtained the actual pMHC-TCR structures from the RCSB PDB database of X-ray crystallography experimental data (https://www.rcsb.org/).TCRpMHCmodels (v1.0) 12 which takes the amino acid sequence of pMHC and TCR as input, was used to predict protein structure models.We then used FoldX (v5.0) 13 AnalyzeComplex to calculate the energy parameters for the pMHC-TCR protein structure.A logistic regression algorithm model was developed using the energy parameter values of protein structures to predict the binding between TCR and pMHC complexes.Using this model, the energy parameter values for tumorreactive TILs TCR and pMHC interactions were calculated.Based on the results, we categorized TCR-pMHC pairs with a model score of 0.7 or higher as tier 1 neoantigens.

Peptides
Synthetic 9-10mer and 27mer peptides covering tumor-specific mutations were synthesized to > 90%, 4 mg per neoantigen by Biostem (South Korea).Peptides were dissolved in distilled water (DW) or DMSO to create a stock solution with a concentration of 20 mM.A working solution of 1 mM was prepared by diluting the stock solution 20-fold.The stock solution was stored in a deep-freezer, while the working solution was kept refrigerated for up to one month.

PBMCs isolation
PBMCs (peripheral blood mononuclear cells) were isolated from PB of HCC patients using Lymphoprep (STEMCELL Technologies) following the manufacturer's recommendations and cryopreserved until neoantigen synthetic peptides become available for in vitro immunogenicity screening.

In vitro immunogenicity screening
Cryopreserved PBMCs from patients were thawed and suspended in AIM-V (Thermo Fisher, USA) supplemented with 10% human AB serum (Merck, USA).PBMCs were washed with PBS and centrifuged at 500g at RT for 10 minutes.After removing the supernatant, the pellet was resuspended in thawing media.
Cells were treated with DNase (10mg/ml, STEMCELL Technologies) for 4 to 6 hours at 37°C in a humidified 5% CO2 incubator, followed by harvesting under the same conditions.Cells were cultured at a density of 3ⅹ 10 5 cells per well in a 96-well plate and treated with a single neoantigen peptide (10 μg/ml, 9~10 mer).In addition, each well was treated with anti-PD-1 antibody (5 μg/ml, Biolegend, USA).The cells were cultured for 16 days and were treated with IL-2 (10 U/ml, PeproTech, USA) and IL-15 (5 ng/ml, PeproTech) cytokines on days 3, 6, 9, and 12.If the media color changed to yellow, half of the culture media was replaced with fresh media.On day 16, the human IFN-γ ELISpot PRO kit (ALP) (MabTech, Sweden) was used to determine the amount of cytokine-secreting T cells.T cells were seeded at 50,000 cells per well in a ELISpot plate followed by stimulation with individual neoantigen peptide (10 μg/ml) for 22 hours.All tests were performed in triplicate and included DMSO (0.05%) as negative controls and PHA (phytohemagglutinin) (Thermo Fisher) (1.25 mg/ml) as positive controls.Spots were visualized with a biotinylated-anti IFN-γ antibody following the manufacturer's instructions.Plates were scanned using an AID Classic ELISpot Reader (Germany).T cell responses stimulated with neoantigen peptides were compared to those treated with DMSO using a Mann-Whitney U test.

Statistics
Prism 8 (GraphPad) and SPSS were employed for statistical analysis.One-way analysis of variance (ANOVA) followed by Tukey's post hoc test was applied to assess differences among more than two groups.Kaplan-Meier analysis with the log-rank test was used to evaluate survival benefit.The error bars represent the standard error of the mean (SEM).Statistical significance was determined at p values < 0.001 and < 0.05.

Generation of B16F10 tumor bearing mice
All studies utilized male C57BL/6 mice of 6 weeks old.C57BL/6 were purchased from Orient Bio (South Korea).All mice were maintained in a specific pathogen-free (SPF) animal room at the Korea Institute of Science and Technology (KIST).All animal experiments were performed in accordance with the guidelines of the Institutional Animal Care and Use Committee (IACUC) of KIST.All mice were stabilized for at least one week prior to the experiment.Tumors were induced by injecting 1 x 10 6 cells of B16F10 subcutaneously in the right flanks of mouse.Tumor volume and body weights were measured every 4 days.Each experimental group comprised five mice housed together in a single cage.The mice were monitored daily for the duration of the experiment.

Acquiring the B16F10 tumor tissues and tail tips for VACINUS platform
When tumors reached a size of 100 mm 3 , tumor tissues from B16F10 tumor-bearing mice were harvested.For control samples, approximately 1 cm of the tail tip was excised from normal C57BL/6 mice.These control samples provided a baseline comparison for the tumor tissues.The harvested tumor tissues and control tail tip samples were collected on the same day of extraction.The harvested tumor tissues were enzymatically dissociated into single-cell suspensions using the tumor dissociation kit (Miltenyi Biotec) and the gentleMACS™ Octo Dissociator (Miltenyi Biotec).The dissociated single-cell suspensions from tumor tissues were filtered through a 40 μm strainer to remove large debris.Erythrocytes were then lysed using an RBC lysis buffer (BioLegend).After removing dead cells using the specific kit (Miltenyi Biotec), CD8 T cells were magnetically enriched from the single-cell suspensions using CD8 T cell microbeads (Miltenyi Biotec).
The purity of isolated CD8 T cells was confirmed by CD8 FACS analysis.The purified CD8 T cells were suspended in RPMI1640 media (Welgene) containing 10% FBS (Gibco), and subsequently used for scRNAseq.

Immunization of mice with selected neoepitopes (27-mer peptide)
Male C57BL/6 mice, aged 6-8 weeks and matched in age, were injected subcutaneously with selected neoantigens.Each flank received a combination of the selected neoantigens (100 μg) and adjuvant (50 μg poly(I:C) (InvivoGen, France) in PBS, with a total of 100 μg poly(I:C) per mouse.The number of neoantigen groups was 15, with 3 mice per neoantigen, resulting in a total of 45 mice.Mice were immunized on day 0 and again on day 7.An adjuvant-only control group received 100 μg poly(I:C) injection without neoantigens.
The injection solution was prepared by mixing a 10 mg/ml stock solution (10 μl) with a 1 mg/ml stock solution (50 μl), and diluting with PBS (140 μl) to a total volume of 200 μl per flank.

Immunogenicity of selected neoantigens (9-10 mer peptide)
On day 13, spleens were extracted from the immunized mice.A total of 2ⅹ10 6 splenocytes were incubated with 10 ug/ml CD8+ T cell epitopes of 9-10 mer of the injected neoantigens, along with protein transport inhibitors: 1:1,500 Monensin (Thermo Fisher Scientific), 1:1,000 Brefeldin A (BD Biosciences), 5μg/ml DNase I solution (GenDEPOT) for 6 hours.As a positive control, a cell stimulation cocktail (40.5 uM phorbol 12-myristate 13-acetate (PMA) and 670 uM ionomycin in ethanol (eBioscience, USA) was used.For a negative control, RPMI media was added to fill the appropriate volume for each group.

Therapeutic efficacy of neoantigens
Male C57BL/6 mice, aged 7-8 weeks, were subcutaneously injected with B16F10 cancer cells (2.5ⅹ10 5 cells) on the flank.The combination of 3 peptide neoantigens of the same tier (100 μg each) were mixed with 100 μg poly(I:C) in PBS for subcutaneous injection.The initial injection was administered when the average size of the cancer reached 30-70 mm³, followed by a second injection after 7 days.Furthermore, intraperitoneal injections of aPD-1 (200 μg/mouse) were given a total of 5 times at 3-day intervals, starting from the first injection.The study included 7 treatment groups, each consisted of 7 mice, resulting in a total of 49 mice.
Mice were monitored for tumor size, body weight, and survival rate at three-day intervals.Mice were sacrificed if the tumor size exceeded 2000 mm³.The percentage of tumor free was assessed in mice by absence of palpable tumor on side flank on day 25 post tumor cell inoculation.The treatment groups for B16F10 neoantigens included the following: G1, the non-treated group; G2, the poly(I:C)-only group; G3, the poly(I:C) + tier1 vaccine group; G4, the poly(I:C) + nontier1 vaccine group; G5, the aPD-1 + poly(I:C) group; G6, the aPD-1 + poly(I:C) + tier 1 vaccine group; and G7, the aPD-1 + poly(I:C) + nontier1 vaccine group.
The tier 1-specific CD8 T cells were sorted from the pooled splenocytes, stained with CD3 (BV421 anti-Supplementary Fig.

2 .
Evaluation of VACINUSpMHC using the TESLA dataset and comparison with the top-5 teams' neoantigen detection sensitivity.Data from 3 patients with melanoma and 2 patients with NSCLC in the TESLA dataset were used.Comparison of clonal expansion between tumor-reactive TILs and non-tumor reactive TILs (p <0.01).represent tumor-reactive T cells including exhausted T cells, activated T cells, proliferative T cells, and effector T cells.A total of 21,961 CD8+ TILs were analyzed.e Comparison of clonal expansion between tumorreactive TILs and non-tumor reactive TILs.Clonal expansion detected in tumor-reactive TILs.p values are shown; statistical comparisons were performed using two-way ANOVA.* p <0.05 vaccine was synthetic long peptides (SLP) consist of 27mer and contains 3 SLP of tier 1 epitopes.Vaccine was injected subcutaneously into right flank 2 times a week and anti-PD-1 was injected intraperitoneally with 5 times every 3 days after cancer cell inoculation.The concentration of peptide, poly(I:C) and anti-PD-1 was 100ug, 100ug, and 200ug per mouse, respectively.b The weight of B16F10 tumors bearing mice receiving the respective treatments (n = 5 mice per treatment group).c The survival of mice was measured at days 9, 12, 15, and 16. d The frequency of IFN-γ+CD8+ T cells in the spleen following restimulation with tier 1 neoepitopes.Numbers indicate mean +/-SD of pooled data.p values are shown; statistical comparisons were performed using two-way ANOVA.*p<0.05,**p<0.01,*

table 5 . Validation of predicted neoantigens in 8 patients with HCC.
**p<0.001 subsets defined by marker gene expression in splenocytes and tumor cells.e Ratio of naïve and effector CD8+ T cells in the spleen.ScRNA sequencing data were pooled 5 mice in all groups.