Personalized gene therapy to change the tumor environment and improve CAR T cell treatment in solid tumors

Human tumour and matched normal tissue samples (n = 54) from patients with neuroblastoma were analysed. The cohort consisted of paediatric and adolescent patients (age range: 1–17 years). Information on participant sex or gender was not available in the anonymised dataset. Sex and gender were not considered as variables in the study design, as there is currently no evidence that these factors influence the occurrence of CRISPR target sites or affect gene-editing efficacy or safety, and were therefore not relevant to the scientific objectives of this analysis. WGS was performed for patient tumour and matched normal tissue, and expression data were generated from RNA-sequencing (RNA-seq). For comparative in vitro analyses, unpublished data from several neuroblastoma cell lines (n = 14) were used. WES and RNA-seq were performed to determine mutational and transcriptional profiles. Extrachromosomal DNA (ecDNA) and structural variants (including deletions, insertions, inversions, tandem duplications and translocations) data were obtained from 20 neuroblastoma primary tumours with matched controls corresponding to the published WGS data of the discovery cohort³⁸. We used the processed SV calls and ecDNA reconstructions provided by Rodríguez-Fos et al. and deposited with that study³⁸; no additional SV calling or ecDNA reconstruction was performed. For downstream PAM analysis, ecDNA coordinates were lifted over to GRCh38 to match the reference genome used in this study⁶⁶. Additionally, publicly available sequencing data from neuroblastoma cell lines (n = 48) were included from the Cancer Dependency Map Portal (DepMap 23Q4, 22Q2, 20Q4)^3,67–71. These datasets comprise mutational profiles generated through high-coverage WES, read-depth analysis and RNA sequencing. Gene dependency scores, gene-expression data, and copy-number variations were also obtained from DepMap. Somatic variants were identified using MuTect2 (version) to call variants between tumour-normal pairs for patients and against the reference genome GRCh38.p14 (release 44) for DepMap and NB cell lines. For comprehensive gene annotation, the reference genome GRCh38.p14 (release 44) was used across all data sets.

CancerPAM (Python) was designed to process the called variants (csv-files) based on their genome builds (e.g. hg38) and identify somatic variants that produce novel NGG PAMs. The 5′ and 3′ genomic sequences surrounding the somatic variants were retrieved through an API and the UCSC Genome Browser. This analysis was conducted to assess whether novel C were adjacent to existing C or novel G to existing G. The output included information on the somatic variant, the potential gRNA sequence, the novel PAM and whether the novel PAM was located on the plus or minus strand of the genome. Novel PAMs were analysed for CRISPR efficiency and specificity using CRISPOR³⁷. Access to the platform was facilitated through an API, which provided the relevant efficiency and specificity values. CancerPAM was designed using Snakemake to automate and streamline the analysis pipeline⁷².

To identify tumour-specific CRISPR-Cas9 target sites in structural variants (SVs) involving extrachromosomal DNA (ecDNA), we developed the bkp-PAM analysis pipeline. This approach builds on previously published neuroblastoma WGS data and breakpoint annotations (Rodriguez-Fos et al.)³⁸ and extends them with systematic PAM and gRNA discovery at ecDNA/SV-associated junctions.

For each ecDNA/SV breakpoint, we extracted 60 bp windows flanking the junction in the GRCh38 reference genome (−30 to +30 relative to the breakpoint, where the breakpoint lies between positions −1 and +1). Both flanking sequences were strand-corrected according to the reported orientation of the rearrangement breakpoints. For tumour-specific-breakpoint-generated novel PAM site detection, we first assessed whether the breakpoint itself generated a novel CRISPR-Cas9 PAM sequence. Bases at positions −1 and +1 were inspected for the presence of "GG" (NGG PAM motif) or "CC" (CCN motif, reverse complement PAM). To distinguish genuine breakpoint-derived PAMs from pre-existing sites, the flanking sequences were aligned to the wild-type reference genome. A PAM was considered novel if the required G or C was absent in the unaltered genomic context but introduced by juxtaposition at the breakpoint. For NGG PAMs, the corresponding gRNA sequence was defined as bases −22 to −3 upstream of the PAM; for CCN PAMs, the gRNA was defined as the reverse complement of bases +3 to +22 downstream of the PAM. In addition to de novo PAM creation, we searched for gRNAs spanning the breakpoint that would bind uniquely due to sequence divergence from the reference. Candidate NGG or CCN motifs were identified within ±30 bp around each breakpoint (NGG starting between +6 and +16; CCN starting between −18 and −8). To ensure breakpoint specificity, we required that candidate gRNAs include at least four nucleotides complementary to both sides of the junction. For each breakpoint, the pipeline generated an annotated table with the number and type of novel CRISPR target sites, their genomic coordinates, PAM motif, associated gRNA sequence, and classification as either (a) novel PAM created at the breakpoint or (b) novel gRNA.

In this study, four CRISPOR-derived scores were utilized alongside gene dependency, expression and copy number data. Two scores, MIT and CFD, assess on-target specificity by estimating the guide RNA's genomic uniqueness and considering mismatches with the target DNA. These scores, ranging from 0 to 100, account for the number, position and distribution of mismatches in a sequence-dependent manner, with higher values indicating reduced off-target effects. Notably, the CFD score correlates more strongly with the total off-target cleavage fraction of a guide than the MIT score^5,35,73. To predict the on-target efficiency the Doench and Moreno-Mateos scores were chosen. The higher the efficiency score, the more likely cleavage occurs at this position. The Moreno-Mateos score is based on CRISPRscan and predicts gRNA activity by analysing molecular features such as guanine enrichment, adenine depletion, nucleotide truncation and 5′ mismatches, effectively capturing the sequence determinants influencing CRISPR/Cas9 activity in vivo³⁴. The Doench score, developed from a large-scale CRISPR study, evaluates sequence traits linked to high or low guide RNA activity, creating a scoring algorithm based on desirable nucleotide patterns. These scores also range from 0 to 100⁵. Beside that gene dependency data was used to estimate how essential a gene is for cell survival and the lethality of its knock-out. Highly negative dependency scores indicate critical gene functions, with values below −0.5 signifying depletion in most cell lines and scores below −1 representing strong lethality, corresponding to the median of all essential genes. A value of 0 reflects non-essential genes³. To calculate a single score for each novel PAM, they were first ranked for each feature individually. PAMs lacking data for a given feature were assigned the lowest rank for that feature. In the case of ties in feature values, the corresponding PAMs shared the same rank. Subsequently a final score for each novel PAM site was determined by summing the weighted ranks of all features. The Doench and Moreno scores (which assess cutting efficiency but not knock-in efficiency) were averaged. In contrast, the CFD score was assigned a weighting factor of 2, to increase specificity and hence safety. The final score was calculated as follows: 2* rank CFD + rank MIT + 0.5*(rank Doench + rank Moreno) + rank Dependency + rank expression + rank copy number. Scores were then normalized within each dataset to a range of 0–100, where a score of 100 represents the most promising target and a score of 0 the least promising. Finally, PAMs were ranked from highest to lowest score in the output table. In the case of ties the corresponding PAMs shared the same rank.

Sequences for all homology-directed repair templates (HDRTs) and gRNAs used are provided in Supplementary Table 3. Human cytokine coding sequences were obtained from the NCBI genome viewer (https://www.ncbi.nlm.nih.gov/gdv/↗). Q8 tag and sPA were used as published^74,75. DNA fragments were acquired from Twist Bioscience after codon optimisation. We used homology-based seamless cloning NEBuilder HiFi DNA Assembly, NEB) to insert DNA fragments into a high-copy-number pUC-based plasmid vector, containing a ColE1 origin and Ampicillin resistance (AmpR) before transformation of ultracompetent XL10-Gold (Agilent) Escherichia coli cells. Transformed cells were selected on LB agar plates containing ampicillin. Positive clones were identified by colony PCR and their plasmid DNA was isolated using the ZymoPURE Plasmid Miniprep Kit (ZYMO RESEARCH) before Sanger sequencing (LGC Genomics) was conducted for comprehensive analysis. Linear dsDNA HDRTs were produced by PCR amplification using KAPA HiFi DNA Polymerase (Roche) in a 400 μl reaction. Following amplification, the HDRTs were purified using AMPure XP Beads (Beckman Coulter) and their correct size confirmed by gel electrophoresis. For vector expression testing the Effectene transfection reagent (Qiagen) was used in HEK293T cells at a dose of 1 µg DNA per 1 × 10⁶ cells. Custom genetic constructs and oligonucleotides generated in this study are available from the corresponding author upon reasonable request.

CRISPR cutting and knock-in experiments were conducted using the Lonza 4D-Nucleofector (Lonza, Basel, Switzerland)⁷⁶. Ribonucleoprotein (RNP) complexes were assembled from synthesized single-guide RNAs (gRNAs) (Synthego, Supplementary Table 3) and Alt-R™ S.p. Cas9 Nuclease V3 (IDT) with or without HDRTs and polyethylene glycol (PGA). Electroporation was performed using the X-unit of the 4D-Nucleofector with 16 well Nucleocuvette strips applying pulse programs DN110 for SK-N-BE2c and FF104 for SK-N-AS cells. For CRISPR-Cas knock-in experiments, the RNP complex was assembled immediately before nucleofection. Per reaction, a gRNA to Cas9 protein molar ratio of 2:1 was used (80 pmol gRNA (2.5 µg) and 40 pmol Cas9 enzyme (6.67 μg) per 1 × 10⁶ cells) with 50 μg polyethylene glycol (PGA) were mixed with sterile water. The complex was incubated at room temperature for 15 min. HDRT dsDNA (2.0 μg per 1 × 10⁶ tumour cells and 1.0 µg per 1 × 10⁶ T cells) was then added. Tumour cells were trypsinized, counted, washed twice (200 × g, 5 min, room temperature) and resuspended in SF buffer containing Supplement 1 (1:4.5 dilution). T cells were resuspended in supplement containing buffer P3. For each reaction, 20 μL of the cell suspension was added to the corresponding RNP-HDRT mixture and electroporation was carried out using Lonza 4D-Nucleofector 16-well strips. Following nucleofection, cells were recovered in antibiotic-free RPMI medium supplemented with 10% FBS, incubated at 37 °C with 5% CO₂ and transferred to appropriate cell culture plates. For T cell knock-ins Alt-R HDR Enhancer V2 (IDT) at 1 µM final concentration was added. CRISPR knock-in experiments targeting tumour cells did not include an Alt-R HDR Enhancer V2 (IDT). After 24 h, culture conditions were adjusted as necessary, including the removal of HDR Enhancer (IDT) when applicable. For CRISPR editing efficiency assessment, custom PCR assays were designed using the GeneGlobe tool (Qiagen) to amplify the cut sites. Out/Out PCR reactions were performed using the QIAprep & CRISPR Kit (Qiagen) with AllTaq Master Mix. Sanger sequencing was conducted and the resulting chromatograms were analysed using the ICE (Synthego) and EditR (http://baseeditr.com/↗) tools to quantify indel formation.

To assess the efficiency and stability of genetic modifications in transgenic cell lines, digital PCR (dPCR) was performed using the QIAcuity digital PCR system (QIAGEN). Primers and hydrolysis probes were designed based on thermodynamic and structural parameters optimized for the QIAcuity digital PCR system. GC content (40–60%), primer melting temperature (55–65 °C), and probe Tm (1–3 °C higher than primers) were adjusted to avoid secondary structures and cross-dimerization. Specificity was confirmed by NCBI BLAST. All primer and probe sequences are available in Supplementary Table . Primers were synthesized by Eurofins Genomics and hydrolysis probes by Metabion. Copy number variation (CNV) analysis was conducted using a duplex probe-based approach to quantify the integrated transgenes relative to a reference gene. Genomic DNA was extracted from stable transgenic and wild-type control cell lines. The dPCR reaction mix contained the QIAcuity Probe PCR Master Mix (QIAGEN), gene-specific primers and probes and fragmented template DNA. The total reaction volume was 12 µl for a 96-well nanoplate format and 40 µl for a 24-well format. The final reaction composition included: 1× QIAcuity Probe PCR Master Mix, 800 nM forward and reverse primers for the target transgene, 400 nM for the reference gene (AFF3) primers, 400 nM target-specific hydrolysis probe (FAM-labelled) and 200 nM AFF3 probe (HEX-labelled). To optimize template accessibility, 0.05 U/µl XbaI (New England Biolabs) was added. Samples were pipetted into the QIAcuity Nanoplate (QIAGEN) and sealed with a QIAcuity Nanoplate Seal (QIAGEN). The dPCR thermal cycling protocol was performed on the QIAcuity One digital PCR system (QIAGEN) with the following conditions for In/In transgene-specific assays: Enzyme activation at 95 °C for 2 min, denaturation at 95 °C for 15 s, primer annealing and elongation at 58 °C for 30 s. This cycle was repeated for 40 cycles, followed by an imaging step for fluorescent signal detection. For Out/In site-specific knock-in assays annealing at 5 8 °C for 30 s was followed by an additional elongation step for 1 min at 72 °C and cycle count was increased to 55 cycles followed by a 2-min extension at 72 °C to enhance signal resolution. Following amplification, the QIAcuity software (QIAGEN) analysed partitioned fluorescence signals to determine absolute DNA copy numbers per microliter using Poisson statistics. Each sample was processed as a duplex reaction, normalizing transgene copy number to the AFF3 reference to account for variations in DNA input and to calculate cell counts. The software generated graphical representations of positive and negative partitions, scatterplots and quantitative tables. Negative control wells (no-template controls) were included to validate specificity and fluorescence thresholding was manually reviewed to ensure correct partition classification. Final CNV values were expressed as copies per 100 cells. 4

CAR T cells were generated from PBMCs isolated from the blood of healthy donors using a density gradient centrifugation method¹⁸. Blood was diluted 1:1 with phosphate-buffered saline (PBS) and carefully layered over Ficoll-Paque (Sigma-Aldrich) in a centrifuge tube. Following centrifugation at 300 × g for 20 min without brake, the mononuclear cell layer was collected. Red blood cells were lysed using a haemolysis buffer, and the remaining cells were washed with PBS and counted. T cells were isolated from PBMCs the CD3⁺ Pan-T Cell Isolation Kit (Miltenyi), followed by magnetic-activated cell sorting (MACS) as per the manufacturer's instructions. Isolated T cells were seeded at a concentration of 1 × 10⁶ in a 24-well plate and activated with anti-CD3/CD28 beads at a 1:1 cell-to-bead ratio. Depending on the CAR construct, T cells were transduced with lentiviral vectors (SIN epHIV7) propagated in 293T cells on Day 1 post-activation at a multiplicity of infection of 1. non-transduced T cells served as negative controls. Transduced T cells were cultured in T cell medium (RPMI + 10% FBS + 1% GlutaMAX (Gibco) supplemented with 0.5 ng/ml interleukin-15 (IL-15, Miltenyi)) and 5 ng/ml interleukin-7 (IL-7, Miltenyi), with medium and cytokine replenishment every 2–3 days. Twelve days post-transduction, transduction efficiency was assessed by immunostaining for epidermal growth factor receptor truncated (EGFRt) and analysed by flow cytometry. EGFRt-positive cells were enriched using MACS. Cells were stained with a PE-labelled anti-EGFRt antibody, followed by incubation with magnetic anti-PE beads and subsequently separated magnetically. The enriched T cells were then cryopreserved for future use. Prior to experimental use for in vitro experiments, cryopreserved CAR T cells were thawed and subjected to an expansion protocol. For in vivo experiments, CAR T cells were enriched using MACS, expanded according to the rapid expansion protocol, and cryopreserved until use. This involved co-culturing the CAR T cells with freshly thawed, irradiated (80 Gy) PBMCs and irradiated (35 Gy) EBV-transformed lymphoblastoid feeder cells (TM-LCL), in the presence of OKT3 (Miltenyi) CD3 activating antibody complex, IL-7 and IL-15. The culture medium was refreshed every 2–3 days with the addition of 0.5 ng/ml IL-15 and 5 ng/ml IL-7 for in vitro and NOG xenograft experiments and 10 ng/ml IL-15 and 10 ng/ml IL-7 for humanized NOG (HuNOG) experiments. Experiments were conducted 12–15 days after the initiation of the second stimulation.

To assess CAR T cell infiltration into 3D neuroblastoma tumours, bioprinted tumour models were generated³⁹. Bioprinting was performed by Cellbricks GmbH (Berlin, Germany) using a standardized hydrogel-based bioink optimized for neuroblastoma culture, allowing the creation of standardized cylindrical tumour constructs embedding SK-N-AS and SK-N-BE2c neuroblastoma cells (±cytokine knock-in) with precise size and volume (Ø 3 mm × 1 mm; 7.07 mm³), enabling controlled cell distribution and tumour architecture. L1CAM-targeting CAR T cells were then co-cultured with the 3D tumours for 12 h and infiltration efficiency was quantified by flow cytometry, using the T cell-to-tumour cell ratio as a readout. To further model physiological barriers to T cell migration, a transendothelial migration and tumour infiltration assay was developed by incorporating a Human Umbilical Vein Endothelial Cells (HUVEC) monolayer within a Boyden transwell insert, positioned atop the bioprinted tumour constructs. This system simulated vascular endothelium, requiring CAR T cells to migrate through an endothelial barrier before reaching the tumour mass. Vybrant™ DiO Cell-Labelling Solution (Thermo Fisher, Cat# V22886) was used to confirm HUVEC layer formation before T cell addition. After 4 h of CAR T cell migration, the insert was removed, and CAR T cell infiltration into the tumour mass was assessed 8 h later by flow cytometry.

Mice were housed under specific pathogen-free conditions in individually ventilated cages with a 12 h light/dark cycle, at an ambient temperature of 23 ± 1 °C and relative humidity of 45–65%, with ad libitum access to food and water. The animal welfare was checked twice daily. Body weights, tumour volume and general health conditions were recorded throughout the whole study.

All animal experiments were conducted in female NOD.Cg-Prkdc^scid Il2rγ^tm1Sug/JicTac mice (CIEA NOG; Taconic Biosciences, Inc.) aged 6–8 weeks at the start of the experiment. A total of 24 mice were used for humanization and 48 mice for non-humanized xenograft CAR T-cell experiments. Only female mice were used to minimize variability in tumour growth kinetics and immune reconstitution related to sex hormones; therefore, sex was not included as an experimental variable in the analyses. For transplantation 5 × 10⁶ tumour cells were mixed with Matrigel (1:1) and transplanted subcutaneously in a final volume of 100 µl in the right flank. After engraftment (palpable tumour), tumour size was measured at least twice a week with a digital caliper. Tumour volume was calculated with the formula V = (L × W²)/2. CAR T cell transplantation was performed at a tumour volume of at least 50 mm³. For in vivo experiments, 10 × 10⁶ CAR T cells were injected into the tail vein (i.v.) in 200 µl PBS by slow injection. For BLI mice were anaesthetised with Isoflurane (Baxter, San Juan, Puerto Rico) and received intraperitoneally 150 mg/kg D-Luciferin (Biosynth, Staad, Switzerland) dissolved in PBS. BLI was performed with the NightOwl II LB983 in vivo imaging system. The IndiGO 2.0.5.0 software was used for initial analysis, colour-coding of the signal intensity and quantification. BLI was performed d1, d4, d7, d11 and d14-26 for SK-N-AS and d1, d4, d7, d11, d17, d24 and d28 for SK-N-BE2c transplanted animals. All mice treated in this study were sacrificed upon reaching a tumour volume of 1.5 cm³, which represents the maximal tumour size permitted by the institutional animal ethics approval (LaGeSo, Reg E0023-23), or earlier if other humane endpoints were met. All animals were euthanized before exceeding this limit. Blood samples were collected after retrobulbar venous plexus puncture in MiniCollect® tubes containing Lithium Heparin for serum samples and processed according to the manufacturer's instruction. Serum samples were stored at −80 °C. For tumour tissue collection, mice were sacrificed. Tumours were removed, their weights determined and subsequently divided into two pieces, one of which was formalin-fixed and paraffin-embedded and one as a snap-frozen sample.

All experiments were performed in accordance with relevant ethical regulations for research involving human participants and animals. Collection and reuse of human tumour and matched normal tissue samples (n = 54) were conducted under written informed consent and approved by the Ethics Committee of Charité - Universitätsmedizin Berlin (protocol EA2/055/17), in accordance with the Declaration of Helsinki and applicable data protection laws. The generation of CAR T cells was conducted under the ethical approval EA2/262/20 from Charité - Universitätsmedizin Berlin. Human umbilical cord blood and buffy coat samples used for generation of CD34⁺ haematopoietic stem cells were obtained under informed consent and processed at EPO Berlin-Buch in compliance with institutional and national regulations. No participant compensation was provided for sample donation. All animal procedures (humanized and non-humanized xenograft NOG models) were approved by the Landesamt für Gesundheit und Soziales (LaGeSo), Berlin, Germany (approval Reg E0023-23) and performed in compliance with German animal-welfare legislation. Mice were housed under standardized conditions with veterinary oversight, and humane endpoints were pre-defined and observed.

Further information on research design is available in the linked to this article. Nature Portfolio Reporting Summary