In vivo PDX CRISPR/Cas9 screens reveal mutual therapeutic targets to overcome heterogeneous acquired chemo-resistance

Resistance towards cancer treatment represents a major clinical obstacle, preventing cure of cancer patients. To gain mechanistic insights, we developed a model for acquired resistance to chemotherapy by treating mice carrying patient derived xenografts (PDX) of acute lymphoblastic leukemia with widely-used cytotoxic drugs for 18 consecutive weeks. In two distinct PDX samples, tumors initially responded to treatment, until stable disease and eventually tumor re-growth evolved under therapy, at highly similar kinetics between replicate mice. Notably, replicate tumors developed different mutations in TP53 and individual sets of chromosomal alterations, suggesting independent parallel clonal evolution rather than selection, driven by a combination of stochastic and deterministic processes. Transcriptome and proteome showed shared dysregulations between replicate tumors providing putative targets to overcome resistance. In vivo CRISPR/Cas9 dropout screens in PDX revealed broad dependency on BCL2, BRIP1 and COPS2. Accordingly, venetoclax re-sensitized derivative tumors towards chemotherapy, despite genomic heterogeneity, demonstrating direct translatability of the approach. Hence, despite the presence of multiple resistance-associated genomic alterations, effective rescue treatment for polychemotherapy-resistant tumors can be identified using functional testing in preclinical models.


Establishment and serial transplantation of transgenic patient derived xenograft (PDX) models
Eight to 16 weeks old male and female NOD.Cg-Prkdc scid Il2rg tm1Wjl /SzJ (NSG) mice (The Jackson Laboratory, Bar Harbour, ME, USA) were kept in individually ventilated cages (IVCs). The animal rooms were fully air-conditioned with a temperature of 20-24°C and 45-65% humidity according to Annex A of the European Convention 2007/526 EC. The maximum stocking density of the cages corresponds to Annex III of the 2010/63 EU. The cages were constantly filled with structural enrichment and the animals had unlimited access to food and water. The animals were inspected daily and scored weekly. Signs of monitoring were a body condition score (BCS), posture and movement, behavior, condition of care, eyes and breathing. Humane endpoints were a BCS of <1, marked lameness or inability to move. Animals with apathetic behavior and no reaction to external stimuli, as well as swollen eyes or swollen peritoneal area and gasping also lead to a termination. In therapy trials, loss of body weight beyond 15% for more than two days or beyond 20% relative to day of treatment start resulted in termination. Establishing serially transplantable ALL PDX models in NSG mice from patient material, re-isolating PDX cells from murine spleen and bone marrow (BM), PDX cell culture, lentiviral transduction, enrichment of transgenic cells and in vivo bioluminescence imaging (BLI) were performed as described previously (1)(2)(3)(4)(5). In brief, up to 1 x 10 7 fresh or thawed transgenic PDX ALL cells were injected into the tail vein of mice. To monitor engraftment, tumor growth and treatment response, tumor burden of mice was regularly assessed by IVIS Lumina II (Caliper) with Living Image version 4.4 software (PerkinElmer) starting at day 14 after transplantation and additionally, 50µl peripheral blood (PB) was repetitively collected from the tail vein starting at day 28 after transplantation. PB was analyzed by flow cytometry after staining for murine CD45 and human CD38 (2). At advanced leukemic disease, at defined time-points or at signs of clinical disease (rough fur, hunchback, reduced motility, paralysis) mice were sacrificed by exposure to CO2 or cervical dislocation and ALL PDX cells were reisolated from murine BM and/or spleen for further analyses.

In vivo treatments
Mice were treated systemically with vincristine (VCR) i.v. once per week and/or with cyclophosphamide (Cyclo) i.p. once per week and/or with venetoclax (ABT-199, SelleckChem) 100 mg/kg p.o. five times per week. In all combination treatment experiments, VCR was applied two days prior to Cyclo; In combination treatment experiments of all three drugs, venetoclax was applied on days 1-5, VCR on day 3 and Cyclo on day 5 of each treatment week. VCR and Cyclo were diluted in sterile PBS for efficient application, ABT-199 was dissolved in 1% Carboxymethyl cellulose and 5% DMSO. Animals of the control groups were treated with solvent (PBS) by the same route of administration. Drug concentrations were calculated from clinically relevant concentrations converting the human doses to mouse equivalent doses based on body surface area and differences in metabolism as described (2,6,7).
Drug dosages of VCR and Cyclo were further optimized for each PDX sample individually to ensure tumor reduction by at least 2 orders of magnitude: parental ALL-199 and mice in the multiplex or pretreatment experiment received 0.15 mg/kg VCR and 70 mg/kg Cyclo; ALL-50 received 0.25 mg/kg VCR and 70 mg/kg Cyclo; ALL-265 received 0.3 mg/kg VCR and 70 mg/kg Cyclo. In the ALL-199 "low dose" treatment course 0.1 mg/kg VCR and 50 mg/kg Cyclo was used, in the short term treatment scheme, mice bearing ALL-199 or ALL-265 received 0.5 mg/kg VCR and 100 mg/kg Cyclo and in the "high dose" treatment scheme, mice received 0.6 mg/kg VCR and 100 mg/kg Cyclo. Mice were closely monitored to prevent signs of treatment toxicity and sacrificed as soon as clinical signs of sickness became apparent (weight loss, rough fur, reduced motility, hunchback).
Criteria to define sensitive, persisting and resistant cells were as follows: Sensitive: imaging values in the "descending phase" in repetitive imaging, 2-3 weeks after start of treatment. Persisting: imaging values "stable" in repetitive imaging at or below 1% relative to start of treatment. Resistant: imaging values in the "ascending phase" in repetitive imaging; values rise by at least 10x compared to tumor burden at the persisting stage, despite continuous treatment.
Drug holiday experiments: resistant ALL-199 derivatives were injected into mice (n=1 per derivative) and grown without treatment. At BLI signals ≥ 1x10 10 p/s, PDX cells were re-isolated from murine BM and up to 5x10 6 cells were re-transplanted into next recipient mice. The procedure was repeated twice, resulting in approximately 6 month of drug holiday. In the fourth passage, cells were injected into two mice per derivative.
One mouse was left untreated while the other mouse received chemotherapy, starting at a BLI signal ≥ 1x10 9 p/s. Treatment response was analyzed by repetitive BLI.
Pretreatment experiments: mice were treated with chemotherapy or with PBS as a control for six weeks, before donor ALL-199 PDX cells were injected. ALL-199 cells were either injected in low numbers (n=5 mice per group) to analyze tumor growth by repetitive BLI and compare growth between pre-treated and untreated mice, or in high cell numbers to achieve a high tumor burden (n=5 mice per group) and initiate treatment with combination chemotherapy; treatment response was monitored by repetitive BLI.

Multiplexing PDX
For the multiplex in vivo experiment, individual PDX samples were selected to i) be suitable for transduction and in vitro cultivation; ii) cover a range of different ALL subtypes (Ph-like, hyperdiploid, TCF3-TBX1, KMT2A-AFF1, DUX4 fusion); iii) cover different stages of disease (initial diagnosis, first and second relapse); and iv) have a similar passaging time in vivo. PDX cells were marked by one or more fluorochromes (eGFP, mCherry, mtagBFP, iRFP or combinations of these) as described (1,8).
Transduced cells were kept in vitro for 3 days, followed by enrichment by FACS. A mix of all samples was injected into two mice to expand the transgenic cell populations. At high leukemic burden, cells were re-isolated from mice and individual populations were enriched by FACS. Fluorochrome expression profiles of each sample were recorded by flow cytometry and samples were mixed again. Resistant PDX ALL-199 cells, which were known to proliferate slower than untreated PDX ALL-199, were injected in a larger proportion (10x) compared to untreated PDX ALL-199. For ALL-50 and ALL-707, fewer cells had to be injected as less cells could be isolated from donor mice. Individual samples were injected in the ratio: 1x ALL-199U, 10x ALL-199R, 0.07x ALL-50, 1x ALL-265, 1x ALL-502, 0,75x ALL-707. Fluorochrome profiles of the mix were recorded as well and cells were injected into groups of mice (n=12). Engraftment was analyzed by repetitive BLI. At BLI signal ≥ 1x10 9 p/s, mice were randomly assigned to one of three groups (n=4 per group). One group was sacrificed, PDX cells were re-isolated from murine bone marrow and composition of the PDX population was analyzed by flow cytometry. One group received polychemotherapy for three weeks, one group received PBS as control. At the end of the experiment, PDX cells were re-isolated from murine bone marrow and fluorochrome expression analyzed to identify individual PDX populations. Relative proportion of the individual PDX sample within the mixture was determined for each PDX sample in each group. Proportion after treatment and in the PBS treated group was normalized to start of treatment.

Determination of doubling time of PDX in vivo
In vivo doubling time was calculated by non-linear fitting of an exponential growth curve to data points of repetitive in vivo imaging using GraphPad prism. Doubling time was based on two different time points ((x1|y1) and (x2|y2)) and calculated as follows: Doubling time = ln(2)/k with k = (ln(y2)-ln(y1))/(x2-x1). genes assigned to these genomic regions were extracted using Ensembl Biomart with genome assembly GRCh37.p13 (10).

Targeted sequencing of recurrently mutated genes
To obtain a high sequencing depth of identified TP53 mutations in ALL-199, an established sequencing panel of 68 recurrently mutated genes was utilized.
Experimental procedure and data analysis was performed as previously described (11).
Transcriptome sequencing with prime-seq and differential gene expression analysis Murine bone marrow containing PDX-ALL was thawed and 2000 human cells were enriched by FACS based on GFP expression. Cells were sorted into 100 µL RLT Plus Buffer (Qiagen, Düren, Germany) supplemented with 1 % β-Mercaptoethanol.
Samples were flash frozen and stored at -80 °C until further processing. RNA sequencing was performed using prime-seq (12), a bulk version of the single cell RNAseq method mcSCRB-seq (13). In brief, samples were digested with Proteinase K and nucleic acids were isolated using SPRI Beads and DNAseI digestion was performed on beads. Reverse transcription of the isolated RNA was done using barcoded oligo-dT primers and a template switching oligo. Exonuclease I digestion was performed to remove excess primers. cDNA was amplified using Kapa HiFi HotStart polymerase and quality was assessed using capillary gel electrophoresis. Library preparation was carried out in triplicates with 0.8 ng cDNA input each using the Nextera XT Kit (Illumina, San Diego, USA). Fragments of 300-900 bp size were selected by agarose gel electrophoresis and library sequencing was performed at LaFuGA (LMU, Munich) with Illumina HiSeq1500 (Illumina, San Diego, USA) with paired-end reads. Sequencing was designed to cover barcode and UMI sequences with 28 bp in the first read and cDNA fragments with 50 bp in the second read and aiming for at least 10 Mio reads per sample. Raw fastq files were processed using the zUMIs pipeline (14). Reads were mapped to a concatenated genome of human and mouse (hg38, mm10) and Ensembl gene models (GRCh38 v.84,GRCm38.85) were used for quantification of gene expression levels. Samples were analyzed in two different batches according to the same protocol with minor modifications. The first batch was carried out using individual amplification of cDNAs, the second batch using pooled amplification for less bias and improved quality. To correct the batch effect resulting from the use of different primeseq sequencing time points, an empirical Bayesian method or the methodologies implemented in the limma package of R were applied as described previously (15,16).
Normalization of read counts was performed using the voom function (17). Differential gene expression was calculated using the DESeq2 and limma packages in R following recommended workflows (16,17). Differential gene expression was calculated between resistant and untreated samples and significantly upregulated transcripts in resistant samples were defined with the following cut-offs: p < 0.001 and log2 foldchange >1. To test the association between BCL2 and overall survival, we used the publicly available microarray data set consisting of 449 AML patients treated in different trials of the Haemato-Oncology Foundation for Adults in the Netherlands (HOVON). The data were a part of a publicly available cohort analyzed by Affymetrix (GSE14468) (19,20).

Protein expression analysis
Proteome analysis was performed as described previously (21). Match between run feature was enabled for identification of peptide across runs based on mass accuracy and normalized retention times. After filtering of artefacts, such as reverse hits and contaminants, Perseus software was used to analyze Maxquant output tables (24).

Gene set enrichment analysis
Gene set enrichment analysis was performed using GSEA Desktop Application

CRISPR/Cas9 dropout screens in ALL PDX samples
CRISPR-Cas9 dropout screens were performed using a protocol that we had optimized to be performed in vivo in PDX models (27,28). Cloning of humanized Streptococcus pyogenes (hSp) Cas9 into lentiviral vectors was previously described (5).
Fluorochrome marker was changed to mCherry using restriction cloning and cutting sites for NotI and SalI. Cloning of sgRNA libraries was performed as previously described (28), except that a H2Kk-mtagBFP marker was used to allow both magnetic enrichment and FACS enrichment. The CLUE pipeline was used to generate a library of 1196 sgRNAs, including sgRNAs targeting candidate genes, non-targeting negative controls and sgRNAs targeting essential genes as positive controls (28)(29)(30) with the sgRNA library, aiming for a maximal transduction efficiency of 30% to achieve single integrations of sgRNAs. Cells were cultured for 3 days in vitro before transduction efficiency was analyzed by flow cytometry and transduced cells were magnetically enriched using the H2Kk-marker. Cas9-and sgRNA-library doublepositive cells were injected into groups of mice, aiming for at least 100x coverage per sgRNA within the homed population, i.e. at least 2.4x10 6 cells were injected assuming a homing frequency of 5% with a library size of 1196 sgRNAs. One fraction was used as input control. For the screen using D5, n=11 mice were injected with 2.4x10 6 cells per mouse. For the screen using D7, n=11 mice were injected with 6x10 6 cells per mouse. At BLI signal ≥ 1x10 9 P/s, mice were randomly distributed and either sacrificed (treatment start) or treated with polychemotherapy (treated) or solvent (PBS). After three weeks, cells were re-isolated from murine bone marrow and samples were subjected to NGS as previously described (28). Significant dropouts were identified using the model-based analysis of genome-wide CRISPR/Cas9 knockout (MAGeCK)(31) using the MAGeCK pipeline based on the Galaxy server (32). To analyze the evenness of sgRNA read counts in the library plasmid pool, fraction of sgRNAs was plotted against fraction of total read counts and GINI Index was calculated (33).

Barcoding of PDX cells
Barcodes were cloned and PDX cells transduced as previously described (8). For barcode analysis, gDNA was isolated from the pelleted, flash-frozen ALL PDX cells using the DNeasy Blood and Tissue Kit (Qiagen). When the total cell number exceeded 5*10 6 cells, the sample was split into two columns and combined after elution. Based on the fraction of human cells as measured by Flow Cytometry, the total number of human genomes per 5 µl of gDNA was calculated and gDNA was concentrated using a SPRI bead clean-up to achieve more than 1000 genomes per sample to ensure comprehensive barcode detection. The Library amplification protocol is based on SiMSen-Seq (34,35) consisting of a barcoding PCR and an adapter PCR; the first PCR adds unique molecular identifiers (UMI) in a limited number of cycles to ensure uniqueness of the UMI's, the second PCR adds the necessary sequencing adapters and sample indices needed for Illumina sequencing. To ensure recovery of the full barcode complexity in the gDNA, the reaction was performed in triplicates. Due to the low on-target ratios in some samples the barcoding PCR was performed for 6 cycles.
Based on the results of a PCR cycling test, samples were amplified for 20 -28 cycles to a final concentration between 1 and 5 ng/µL. Each Replicate was indexed with a unique i5 index and an i7 index that was shared between all replicates of a sample.
Library quantification was performed using picogreen (Thermo Fisher) and all samples were pooled equimolar. The final library was quantified and correct amplicon size was confirmed using capillary gel electrophoresis (Agilent, Bioanalyzer DNA HS). Barcode sequencing was performed on an Illumina HiSeq 1500 instrument for 110 cycles covering the barcode and UMI sequences. Raw fastq files were demultiplexed using deML (https://github.com/grenaud/deML). Barcode assignment and clustering of barcodes to account for sequencing errors and polymerase errors was performed using bartender (36). Clusters with a distance of less than or equal to 4 were collapsed.
All further processing and visualisation was performed in R version 3.6. To account for differences in sequencing depth counts were transformed to cpm. Replicates were combined by taking the average cpm value per barcode.

Statistics
All statistical analyses were performed using GraphPad Prism 7.05 or R 3.6.1.
Statistical testing between two groups was performed using unpaired t-test.
Comparison between more than two groups was calculated with one-way ANOVA followed by Tukey´s multiple comparison test. Levels of significance were defined as follows: p > 0.05: n.s., p ≤ 0.05: *, p ≤ 0.01: **, p ≤ 0.001: ***. Correlation of flow cytometry and bioluminescence in vivo imaging signal intensity was calculated using non-linear regression.