Contents
Antibodies
The following antibodies were used for flow cytometry: CD3-PerCP-Cy5.5 (clone SK7; eBioscience; used 1:20); CD4-FITC (clone RPA-T4; BD Biosciences; used 1:20), CD4-APC (clone RPA-T4; BD Biosciences; used 1:30), CD4-BV421 (clone SK3, Biolegend; used 1:100), CD8-BV421 (clone RPA-T8; BD Biosciences; used 1:50), CD14-APC-H7 (clone MoP9, BD Biosciences; used 1:100), CD16-APC-H7 (clone 3G8, BD Biosciences; used 1:100), CD19-FITC (clone 4G7, BD Biosciences; used 1:30), CD137-BV421 (clone 4B4-1; Biolegend; used 1:200), CD137-APC (clone 4B4-1; BD Biosciences; used 1:30), OX40-PE-Cy7 (clone Ber-ACT35, Biolegend), CD107-PE (clone H4A3, BD Biosciences; used 1:150) and PE-conjugated anti-mouse TCRβ constant domain (clone H57-597; BD Biosciences; used 1:150). The viability stain IR-Dye (Thermo Fisher, used 1:2,000) was used to identify live cells.
Generation of patient T cell products, Bcl-6/Bcl-xL-immortalized B cells and tumor organoids
Tumor tissue and PBMCs were collected from patients treated at the Netherlands Cancer Institute—Antoni van Leeuwenhoek Hospital (NKI-AVL) with written informed consent and in accordance with guidelines of the Medical Ethical Committee. The study protocol was approved by the Medical Ethical Committee of the NKI-AVL. Fresh tumor tissue obtained by surgical resection was mechanically disrupted and digested overnight in RPMI 1640 medium (Life Technologies) supplemented with 1 mg ml−1 collagenase type IV (BD Biosciences), penicillin-streptomycin (Roche) and 0.01 mg ml−1 pulmozyme (Roche).
For patients NKIRTIL027, NKIRTIL063 and ITO34, TIL cultures were generated by culturing tumor digest suspensions in T cell medium (RPMI 1640 medium supplemented with 10% human AB serum (Life Technologies), penicillin-streptomycin, l-glutamine (Life Technologies)), supplemented with 6,000 U ml−1 IL-2 (Proleukin, Novartis) for 2–4 weeks. Obtained TIL cultures were subsequently stained with IR-Dye and antibodies against CD3, CD4 and CD8, and single CD3+CD4+ and CD3+CD8+ T cells were sorted using a FACSAria Fusion cell sorter (BD Biosciences). Isolated CD4+ and CD8+ T cells were expanded using the rapid expansion protocol (REP), using 30 ng ml−1 anti-CD3 antibody (clone OKT-3; eBioscience) and 3,000 U ml−1 IL-2 in a 1:1 mixture of RPMI 1640 and AIM-V medium (Gibco) supplemented with 5% human AB serum, in the presence of irradiated (40 Gy) allogeneic PBMCs (200:1 feeder/T cell ratio). After 7 days of REP culture, medium was refreshed with medium and IL-2 every 2 days. Purity of the resultant CD4+ and CD8+ T cell populations was confirmed by flow cytometry at day 14 after start of REP (routinely >99%), and cells were subsequently either used directly in antigen discovery screens or cryopreserved in liquid nitrogen. Data from flow cytometry experiments was acquired using FACSDiva software and analyzed using Flowjo (BD Biosciences).
Immortalized patient B cell lines were generated by retroviral transduction with Bcl-6/Bcl-xL27. Patient PBMCs were isolated from peripheral blood by Ficoll-Paque density gradient separation and stained with IR-Dye and antibodies against CD3, CD14, CD16 and CD19. Single IR-Dye−CD3−CD14−CD16−CD19+ cells were sorted using a FACSAria Fusion cell sorter and stimulated for 36 h with irradiated (55 Gy) CD40L+ mouse L cells in B cell medium (IMDM medium (Gibco) supplemented with penicillin-streptomycin, 10% heat-inactivated fetal bovine serum (Sigma-Aldrich) and 50 ng ml−1 IL-21 (Peprotech)), followed by retroviral transduction of Bcl-6 and Bcl-xL. The Bcl-6/Bcl-xL-encoding vector also encodes GFP to allow evaluation of transduction efficiency. Bcl-6/Bcl-xL-immortalized (GFP+) B cells were cultured in B cell medium and were stimulated every week by addition of irradiated CD40L+ L cells. Medium and IL-21 were refreshed every 3–4 days.
For patient ITO66, tumor organoids were established20,21. Tumor-reactive patient T cells were generated by coculturing PBMCs and tumor organoids as follows. Following incubation with 200 ng ml−1 IFNγ (Peprotech) for 24 h, tumor organoids were dissociated into single-cell suspensions using TripLE Express (Gibco). Tumor organoid cells were mixed with patient PBMCs (20:1 PBMC/tumor cell ratio) and 1 × 105 PBMC were seeded in each well of a U-bottom 96-well plate precoated with 5 μg ml−1 anti-CD28 antibody (clone CD28.2; eBioscience). Coculture medium consisted of T cell medium supplemented with 150 U ml−1 IL-2 and 20 µg ml−1 anti-PD1 blocking antibody (clone 5C4; kindly provided by Merus). Coculture medium was refreshed every 2–3 days. PBMCs were harvested and restimulated every 7 days by replating with fresh tumor organoid cells.
Codon-optimized TCR α and β variable sequences (encompassing V-CDR3-J domains) of selected TCRs were gene-synthesized (Twist Biosciences) and subcloned into a modified pMP71 retroviral vector12. This vector contains mouse TCR constant regions to reduce mispairing of introduced and endogenous TCR chains, as well as the puromycin N-acetyltransferase resistance gene. Retrovirus was produced by transfecting FLY-RD18 packaging cells with pMP71-TCR plasmid DNA using Xtremegene 9 transfection reagent (Roche). In parallel, healthy donor PBMCs (Sanquin Blood Bank) were separated into CD8+ and CD8− (for transduction with MHC class I- and MHC class II-restricted TCRs, respectively) cells using the CD8+ T Cell Isolation Kit (Miltenyi Biotec). Isolated cell fractions were stimulated with CD3/CD28 Dynabeads (Life Technologies) in T cell medium with 150 U ml−1 IL-2. After 48 h, retroviral supernatants were collected and used to infect prestimulated CD8−/CD8+ PBMCs by spinoculation (2,000 g for 90 min) in Retronectin (Takara)-coated plates. Transduction efficiency was measured 72 h later by staining with an anti-mouse TCRβ constant domain antibody and analysis by flow cytometry. TCR-transduced T cells were then selected with 2.5 µg ml−1 puromycin (Gibco) for 48 h and received fresh medium and IL-2 every 3–4 days. After 12–14 days of culture, transduced T cells were expanded using the REP as described above.
T cell activation assays
Reactivity of TCR-transduced donor T cells was determined by coincubating T cells and target cells for 18–24 h in U-bottom 96-well plates (1:1 T cell/target cell ratio) in T cell medium. Incubation of T cells without target cells, and in the presence of 50 ng ml−1 phorbol 12-myristate 13-acetate (Sigma-Aldrich) and 1 µg ml−1 ionomycin (Sigma-Aldrich) served as negative and positive controls, respectively. Following incubation, cells were stained with IR-Dye and antibodies against CD3, CD4, CD8 and the activation markers CD137 or OX40 and analyzed by flow cytometry. When T cell reactivity towards tumor organoids was tested, IFNγ-pretreated organoids were incubated with T cells in the presence of 20 µg ml−1 anti-PD1 blocking antibody (Merus) in anti-CD28 antibody precoated plates.
The cytotoxic capacity of T cells was assessed by coincubating T cells and target cells for 72 h in 96-well plates at a T cell/target cell ratio of 5:1, unless indicated otherwise. Target cells cultured in the absence of T cells served as negative control. Following incubation, 7.46 µm AccuCount blank counting beads (Spherotech) were added to individual cultures to enable quantification of remaining live target cells. Cells were subsequently harvested, stained with 4,6-diamidino-2-phenylindole and anti-CD3 antibody, and measured by flow cytometry. When cytotoxicity against tumor organoids was assessed, IFNγ-pretreated organoids were incubated with T cells in the presence of 20 µg ml−1 anti-PD1 blocking antibody (Merus) and 10 µM Y-27632 in anti-CD28 antibody precoated 96-well plates. Where indicated, target cells were incubated with 50 µg ml−1 MHC class I blocking antibody (clone W6/32) for 30 min at 37 °C before incubation with T cells. Data from functional T cell assay was analyzed using Graphpad Prism v.9.
Exome and RNA sequencing
Tumor genomic DNA and RNA was extracted from formalin-fixed paraffin embedded tumor material using the AllPrep DNA/RNA kit (Qiagen). For patient ITO66, genomic DNA and RNA were isolated from tumor organoids. Genomic DNA of patient PBMCs was extracted using the DNeasy Blood & Tissue kit (Qiagen). Exome enrichment was performed using the SureSelect XT2 Human All Exon V6 kit (Agilent) and strand-specific libraries were generated using the TruSeq Stranded mRNA sample preparation kit (Illumina) according to the manufacturer’s instructions. Resulting libraries were sequenced on HiSeq 2500 or NovaSeq 6000 DNA analyzers (Illumina). Whole-exome and RNA sequencing was processed using bcbio-nextgen. Briefly, DNA reads were mapped against GRCh38 using Burrows–Wheeler aligner (BWA), duplicates were marked with Picard MarkDuplicates and low complexity regions were excluded. Somatic and germline mutations were identified using Mutect2 and HaplotypeCaller, respectively, followed by annotation by SnpSift. RNA reads were quality filtered and mapped with STAR or TopHat2, transcript-level expression was quantified by Salmon and gene fusions were determined by Arriba12,21.
Antigen library design
To design the model antigen library used to validate the screening system, protein sequences of genes encoding known human nonmutated cancer regression antigens, as well as selected immunodominant epitope-encoding genes of Epstein-Barr virus, cytomegalovirus and influenza, were collected from the Uniprot database (https://www.uniprot.org/) (Supplementary Table 1). Protein sequences were reverse-translated and codon-optimized, and resulting nucleotide sequences were segmented into 93 nucleotide (nt) minigenes with 45 nt overlap between neighboring minigenes. In addition, a set of previously characterized neoantigens was included, all encoded by 93 nt minigenes in which the mutant codon was flanked on either side by 45 nt of the relevant nonmutant gene sequence. Minigene sequences encoding the corresponding nonmutated peptides were included for each model neoantigen. A stop codon was added directly following each minigene sequence, and internal BbsI recognition sites were removed without altering the encoded peptide sequences. Each 93 nt sequence was duplicated for a total of 4,764 sequences, and a unique 12 nt barcode sequence was incorporated into each minigene sequence following the stop codon. The resulting sequences were flanked by sequences to enable PCR amplification and subcloning using BbsI (New England Biolabs) into a pMSCV retroviral vector that also encodes the puromycin N-acetyltransferase resistance gene and mCherry (pMSCV-puroR-mCherry).
To design NKIRTIL027 and NKIRTIL063 patient mutanome libraries, all single nucleotide variants (SNVs) and frameshifting indels with confirmed RNA expression within tumor cells were encoded as 93 nt minigenes. RNA sequencing data of tumor ITO34 was unavailable, and the library was designed without taking RNA expression of tumor variants into account. For SNVs, minigenes were designed that encoded peptides in which the mutant codon was flanked on either side by 45 nt of the relevant nonmutant gene sequence. In the case of frameshifting indels, or when SNVs resulted in loss of a stop codon, the newly formed open reading frame was segmented in 93 nt minigenes with 45 nt overlap between adjacent minigenes. Minigenes encoding corresponding WT sequences were included for all tumor variant minigenes. Minigenes encoding the MART126–35 and CDK4R24L epitopes were included in all libraries as internal controls. Internal BbsI recognition sites were removed without altering encoded peptide sequences, and minigenes were flanked by sequences for PCR amplification and subcloning as described above. For patient ITO66, the mutanome library was designed to encode tumor variants as 63 nt minigenes, and no corresponding WT minigenes were included. All minigene libraries were synthesized by Twist Biosciences.
Generation of a universal antigen expression vector
To establish a library expression system that enables the concurrent processing and presentation of minigene products through both the MHC class I and class II pathways, constructs were designed in which a TMG encoding two previously identified neoantigens recognized by either CD4+ or CD8+ TIL of patient NKIRTIL027 (LEMD2P>L (ref. 28) and TTC37A>V (unpublished data), respectively) was either fused or not fused to the signal sequence of CD74 (Supplementary Fig. 4)29. Codon-optimized constructs were synthesized (Twist Biosciences) and subcloned into the retroviral pMSCV-puroR-mCherry vector. NKIRTIL027 immortalized B cells were transduced with TMG constructs, selected to over 90% purity (by measuring mCherry expression) with 5 μg ml−1 puromycin and incubated with NKIRTIL027 CD4+ or CD8+ TIL at a ratio of 1:1 for 48 h in T cell medium with 30 U ml−1 IL-2. T cell activation was subsequently assessed by measuring IFNγ levels in the culture supernatant using the Cytometric Bead Array kit (BD Biosciences), following the manufacturer’s instructions.
Library cloning and transduction
Oligonucleotide libraries were amplified by 12 cycles of PCR using Phusion High-Fidelity DNA Polymerase (New England Biolabs) and primers Preamp Forward (5′-ACTGTCAGAAGACTGCAAGC-3′) and Preamp Reverse (5′-TGACAGCGAAGACCATAGTG-3′). For first proof-of-concept screening experiments using MHC class I-restricted TCRs, the amplified model antigen library was cloned by Golden Gate assembly using BbsI into the pMSCV-puroR-mCherry retroviral vector. For all other screens, amplified libraries were cloned into the pMSCV-puroR-mCherry vector modified to include the sorting sequence of CD74. Subcloned libraries were amplified using Endura electrocompetent cells (Lucigen) and library DNA was extracted using the PureLink HiPure Maxiprep kit (Invitrogen). During all cloning steps, a library representation of at least 100× was maintained.
Libraries were retrovirally transduced in duplicate into immortalized B cell lines, as described above. To ensure single retroviral integrations, B cells were transduced at an infection rate of less than 10%. One day after transduction, B cells were transferred to B cell medium in the presence of irradiated CD40L+ L cells. Transduction efficiency was assessed 3 days post-transduction by measuring mCherry expression by flow cytometry, followed by selection with 5 µg ml−1 puromycin for 2 days and expansion of the B cell cultures until used in screens.
Antigen discovery screens
For proof-of-concept screens using MHC class I-restricted TCRs, the antigen library encoding known cancer regression antigens was transduced into a previously immortalized HLA-A*02:01+ patient B cell line (OVC21)12. Library-expressing B cells were coincubated in duplicate with donor CD8+ T cells transduced with the CDK4R24L-specific TCR #53 or MART26–35-specific TCRs DMF4, DMF5 or 1D3 (all HLA-A*02:01-restricted) in T cell medium with 25 U ml−1 IL-2 at a T cell:B cell ratio of 5:1 and at a density of 2 × 106 total cells cm−2. Cultures were resuspended on day 1 and 2 of the experiment. For screens using patient-derived MHC class II-restricted neoantigen-specific TCRs, the model library was transduced into patient-matched immortalized B cells (patient NKIRTIL017), and library-expressing B cells were cocultured with donor CD4+ T cells transduced with either the MANSC1D>H– or SNORD73AR>W-specific TCR as described above. To simulate screening conditions using clonally diverse T cell populations, TCR-expressing T cells were mixed with donor-matched mock-transduced T cells at indicated ratios. Library coverage of at least 300× was maintained in all experiments. After 72 h of coincubation, cells were washed in PBS, and cell debris was removed by either Ficoll-Paque density gradient separation or using the Dead Cell Removal kit (Miltenyi Biotec). Isolated cells were subsequently resuspended in DirectPCR Lysis Reagent (Viagen) containing 500 µg ml−1 proteinase K and lysed by incubation at 55 °C for 60 min, 85 °C for 30 min and 94 °C for 5 min. Minigene sequences were then amplified by PCR using NEBNext Ultra II Q5 Master Mix (New England Biolabs), using the following primers:
Prep-I Forward (for screens with MHC class I-restricted TCRs):
5′-CAAGCAGAAGACGGCATACGATGGAGGAGAACCCTGGACCTACAAGC-3′
Prep-II Forward (for all other screens):
5′-CAAGCAGAAGACGGCATACGACCTGCGGATGAAGCTGCCCG-3′
Prep Reverse:
5′-AATGATACGGCGACCACCGAGATCTACACTCTTTCCCTACACGACGCTCTTCCGATCTNNNNNNNG ATCCGACTCGGTGCCACTTTTTCAAC-3′
The 7-nt stretch of N nucleotides indicates a unique barcode sequence used to enable the multiplexed preparation of sequencing libraries. Following PCR, samples were pooled equimolarly and run on a 1% agarose gel to separate minigene amplicons from potential primer dimers. Minigene amplicons were extracted from gel using the Monarch DNA Gel Extraction Kit (New England Biolabs) and deep sequenced on an Illumina HiSeq 2500 Sequencing system (single read 65 bp). Sequencing data were deposited in the National Center for Biotechnology Information (NCBI) Sequence Read Archive under accession code PRJNA884260 (ref. 30).
For patient neoantigen screens, mutanome libraries were transduced into autologous immortalized B cells, followed by selection with puromycin. The cytotoxic potential of expanded patient TIL was confirmed before neoantigen screens by measuring their capacity to degranulate. To this end, CD4+ and CD8+ TIL were polyclonally stimulated using CD3/CD28 Dynabeads in T cell medium in the presence of Golgistop (BD Biosciences) and an antibody against CD107 for 12 h. Following incubation, cells were stained with IR-Dye and an anti-CD4 antibody and analyzed by flow cytometry. Neoantigen screens were subsequently performed by incubating library-expressing B cells in duplicate with patient T cells at a T cell:B cell ratio of 5:1. Library-transduced B cells cultured in the absence of patient T cells served as a negative control. After 72 h of coincubation, cells were processed as described above.
Sequence analysis
Initial sequence quality profiles were quantified by FastQC and demultiplexed using fastq-multx (ea-utils) with one mismatch allowed. Vector sequences were trimmed from sequence reads using fastq-mcf (ea-utils) against the UniVec database, and samples were subsequently quality filtered using cutadapt. The unique 12 nt barcodes that were added to individual minigene sequences were extracted using seqkit and mapped using Bowtie2 with no multimatched hits allowed. For the ITO66 neoantigen screen, high-quality reads were mapped against the full minigene sequences of the patient library using BBMap with ambiguously mapped reads removed and only perfect mappings allowed. Per sample count tables were differentially compared and normalized using DESeq2. Minigenes with an average abundance below the fourth percentile and a coefficient of variation greater than one across the two internal replicates were removed from analyses. Statistical testing was performed using the DESeq2 Wald test and log-fold change cut-off of 0.25. Tumor variants were defined as screen hits when at least one of the duplicate mutant sequences, but neither of the corresponding WT-encoding minigenes, had an false discovery rate-corrected P value less than 0.2 and a log2 fold change of less than −0.5. All data analysis was performed using R and visualized using the ggplot2 package.
To validate patient neoantigens identified in screens, minigenes encoding the screen hits, as well as their WT counterparts, were synthesized as individual gBlocks (IDT), cloned into the CD74 signal sequence-modified pMSCV-puroR-mCherry vector and transduced into immortalized patient B cells. Following selection with puromycin, minigene-transduced B cells were cocultured with expanded patient CD4+ or CD8+ TIL for 48 h in T cell medium with 30 U ml−1 IL-2 and T cell activation was assessed by measuring IFNγ levels in the culture supernatant using the Cytometric Bead Array kit (BD Biosciences). Reactivity towards neoantigens was considered confirmed when T cells secreted at least twofold more IFNγ in response to the mutant sequence compared with the WT control sequence.
NKIRTIL063 TMG screen
The number of tumor variants of patient NKIRTIL063 selected for TMG screening was reasonably limited to 200 (of a total of 685 nonsynonymous expressed mutations). Mutations were selected by first including the 25 most clonal mutations (based on variant allele frequency), followed by including mutations with highest gene expression up to a total of 200 tumor variants. TMG constructs were designed to encode ten variant-encoding minigenes (93 nt each) in which the mutant codon was flanked by 45 nt of nonmutant gene sequence on either side. Codon-optimized sequences were synthesized (Twist Biosciences) and subcloned into the CD74-modified pMSCV-puroR-mCherry retroviral vector. NKIRTIL063 immortalized B cells were transduced with TMG constructs and selected to more than 80% purity with 5 μg ml−1 puromycin. Next, TMG-expressing B cells were cocultured with NKIRTIL063 CD4+ or CD8+ TIL (from the same expansion cultures as used for the antigen discovery screen) at a ratio of 1:1 for 48 h in T cell medium with 30 U m−1 lL-2, and activation of T cells was determined by measuring IFNγ levels in the culture supernatant using the Cytometric Bead Array kit (BD Biosciences). To validate that the observed reactivity to selected TMG constructs was mediated by neoantigens identified using our neoantigen discovery screen, modified versions of these TMGs were designed such that exclusively the minigene that encoded the identified neoantigen was reverted to its WT sequence. Reactivity of NKIRTIL063 CD4+ or CD8+ TIL to B cells transduced with these modified TMGs was subsequently assessed as above.
In silico selection of neoantigen vaccine targets
The computational tool Vaxrank31 was used to rank tumor mutations of patients NKIRTIL063, NKIRTIL027 and ITO66 for use in a putative personalized cancer vaccine. Patient ITO34 was omitted from this analysis because RNA expression data were unavailable. HLA typing of patients was performed using OptiType for HLA-A, -B and -C alleles. The set of somatic variant calls and aligned RNA reads were used as input, with parameters set to a peptide length of 25, an epitope length of 8–11 and utilization of the MHCFlurry prediction algorithm. In line with ongoing clinical trials of personalized neoantigen-based vaccines32, the 20 top ranking predicted neoantigens were considered for putative neoantigen vaccines (Supplementary Table 2).
Reporting summary
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.
