Patient samples and clinical annotations
In this exploratory analysis, 10 PDX were used, derived from patients with HRR-related breast (triple negative, ER+ , or HER2-positive breast cancer), ovarian (high-grade serous), or pancreatic cancer from patients carrying germline pathogenic variants in BRCA1, BRCA2, or PALB2 genes. One ER+ BC model was generated from a clinical progression to talazoparib (PDX474.7). PDX196 was generated from the pericardial effusion of an ovarian cancer patient on treatment with PARPi shortly before clinical progression, and PDX168 came from a platinum-refractory pancreatic cancer patient (Additional file 1: Supplementary Table S1). Clinicopathologic characteristics, including sex, race, histological subtype, treatment outcome, and history before and after biopsy for PDX generation, were collected.
Generation of PDX models and in vivo treatment experiments
Fresh tumor samples from patients were collected for implantation into nude mice following the European Union’s animal care directive (2010/63/EU) and were approved by the Ethical Committee of Animal Experimentation of the Vall d’Hebron Research Institute. To generate patient-derived xenografts (PDXs), surgical or biopsy specimens from primary tumors or metastatic lesions were immediately implanted in mice. Fragments of 30 to 60 mm3 were implanted into the lower flank of 6-week-old female athymic Rj:NMRI-Foxn1nu/nu (Janvier) mice. Animals were housed in air-filtered laminar flow cabinets with a 12-h light cycle and food and water ad libitum. Mice bearing breast cancer models were supplemented with 1.5 µM 17β-estradiol (Sigma-Aldrich) in drinking water. Upon growth of the engrafted tumors, models were perpetuated by serial transplantation. In each passage, flash-frozen and formalin-fixed, paraffin-embedded (FFPE) samples were taken for genotyping and histological analyses.
To evaluate the sensitivity to the drugs, tumor-bearing mice were equally distributed into treatment groups with tumors ranging 100 to 300 mm3. AZD5305 (saruparib) and olaparib were administered orally (p.o.) six times per week in water/HCl pH 3.5–4 at 1 mg/kg and 10% v/v DMSO/10% w/v Kleptose [HP-β-CD] at 100 mg/kg, respectively. AZD6738 was administered p.o. five times per week in 10% v/v DMSO, 40% v/v PEG300 at 25 mg/kg. Carboplatin was given intraperitoneally (i.p.) once a week in 0.85% physiologic saline at 37.5 mg/kg. Tumor growth was measured with caliper bi-weekly from first day of treatment. To generate PDX models with acquired resistance to olaparib, treatment was maintained for up to 150 days in olaparib-sensitive tumors until individual tumors regrew or maintained in observation for up to 450 days. In addition, pharmacodynamic experiments were conducted and collected after 12 days of dosing and 2 h after final doses. Tumor volume was calculated as V = 4π/3 × L × l2, “L” being the largest diameter and “l” the smallest. In all experiments, mouse weight was recorded twice weekly. Mice were euthanized using carbon dioxide overdose (100% at 5 PSI) for a minimum of 3 min in an euthanasia chamber as recommended by the Euthanasia Guidelines for Investigators. Mice were left undisturbed for an additional 5 min and death was confirmed by cervical dislocation. Euthanasia was performed according to humane endpoints, e.g., when tumors reached 1500 mm3, in accordance with institutional guidelines.
Evaluation of response to therapy in PDXs
The antitumor activity in therapy-resistant models was determined by comparing individual tumor volumes at 21 days to their respective baseline values: % tumor volume change = (V21days − Vinitial)/Vinitial × 100. For therapy sensitive PDXs, the best response was defined as the minimum value of % tumor volume change sustained for at least 10 days. To classify the overall response of each PDX, we modified the RECIST (mRECIST) criteria, to be based on the mean % tumor volume change: complete response (CR), best response < − 95%; partial response (PR), − 95% < best response < − 30%; stable disease (SD), − 30% < best response < + 20%, progressive disease (PD), best response > + 20%.
Response to therapy was evaluated by measuring different preclinical readouts: preclinical complete response rate (pCRR), preclinical overall response rate (pORR), and preclinical benefit rate (pCBR), calculated over the mean of individual tumors from olaparib-sensitive PDX models showing at least one CR, CR + PR, and CR + PR + SD upon PARPi treatment, respectively. More specifically, pCRR was defined as the fraction of PDXs with at least one individual tumor reaching a CR for at least 10 days. pORR was defined as the fraction of PDXs with at least one individual tumor reaching a CR or a PR for at least 10 days, and pCBR as the fraction of PDXs with at least one individual tumor reaching a CR or a PR for at least 10 days or a SD for a minimum of 80 days. Preclinical progression-free survival (pPFS) was defined as the time to disease progression or death from any cause. Preclinical time to progression (pTTP) was defined as the number of days until initially sensitive tumors (CR + PR + SD) regrew.
Targeted sequencing
All laboratory methods were performed using the manufacturer’s protocols. Genomic DNA was isolated from fresh-frozen PDX tissue using the DNeasy Blood & Tissue Kit (Qiagen). All samples were quantified using the Qubit® dsDNA HS Assay Kit (catalog #Q32851) and Qubit 2.0 fluorometer (Thermo Fisher Scientific, Waltham, MA, USA). Hybridization, capture, and sequencing of exonic and intronic regions (< ± 10 bp) of interest were performed using the DNA NGS-based gene panel Hereditary Plus OncoKitDx (Healthincode) in a MiSeq device (Illumina). Bioinformatic analysis of SNVs, insertion/deletions, CNVs, and Alus was performed using Data Genomics software (Imegen; v19.1). Reversion mutations were verified by manual inspection of alignments in IGV. Variants were described and classified according to HGVS (http://www.hgvs.org) and ACMG/GAMP (2015), with reference hg19 (GRCh37). No variants were detected in regulatory or intronic zones > ± 10 bp.
Exome sequencing
All laboratory methods were performed using the manufacturer’s protocols. Genomic DNA was isolated from fresh-frozen PDX tissue using the DNeasy Blood & Tissue Kit (Qiagen). All samples were quantified using the Qubit® dsDNA BR Assay Kit (Invitrogen; #Q32853) and by Qubit Flex Fluorometer (Invitrogen), DNA purity was determined using a NanoDrop Eight (Thermo Scientific), and DNA integrity was measured using a 4200 Tapestation (Agilent). Exome libraries were constructed using the Illumina DNA Prep with Exome 2.0 Plus Enrichment (Illumina, #20077596). Paired-end sequencing with a read length of 150 bp was performed using Illumina NovaSeq 6000 with approximately 10 Gbp per sample for ~ 200-fold average sequence depth. Library sizes and quantification were determined by 4200 Tapestation (Agilent) and pooled libraries were subsequently pooled equimolar. Each library was loaded onto one lane of an S4 v1.5 flow cell (300 cycles) (Illumina, #20028312). Sequencing data was demultiplexed, passed through a bcl-to-fastq conversion program (bcl2fastq v2.20.0.422). Fastq files were analyzed using pipeline software bcbio-nextgen v1.2.9 (https://doi.org/https://doi.org/10.5281/zenodo.3564938). Reads were aligned to the human hg38 and mouse mm10 reference using bwa mem v0.7.17, and sequencing duplicates for each UMI were collapsed into a single consensus read using fgbio v1.4.0. All software were run using best practice parameters established within the bcbio workflow or in-house. Mouse-derived sequences were removed using Disambiguate [30]. Variant calling was performed using VarDict v1.8.2 [31], down to a variant allele frequency (VAF) of 1% (before filtering and curation) and variant effects annotated by snpEff 5.0 [32]. Filtering of non-cancer variants (i.e., common polymorphisms) was performed as per VarDict best practice.
Additionally, the following filters were applied using the NGS Report App in SolveBio (https://www.solvebio.com/): preset filter: Tissue; Hide Variant Depth (VD) below: 3; Hide Total Depth below: 50; Additional SolveBio filters: “type” does not equal “synonymous_variant” and “dkfzbias” does not equal “strand.” Copy number analysis was performed using Seq2C v1.3 [31]. The change in the normalized Log2 values was used to determine potential copy number changes. Chromosome Y was excluded and only deletions with log2ratio < 0 or whole gene amplifications with log2ratio > 0 were kept.
Structural variants were reported by manta v1.6.0 and filtered for having annotation_parsed.detail__exact = ON_PRIORITY_LIST, split_read_support_1 > 15, split_read_support_2 > 15.
RNA sequencing
RNA was extracted from 15 to 30 mg of fresh-frozen tumor from PDX samples by using the RNeasy Mini kit (Qiagen). RNA concentration was determined by Qubit Flex Fluorometer (Invitrogen), RNA purity was determined using a NanoDrop Eight (Thermo Scientific), and RNA integrity was measured using a 4200 Tapestation (Agilent). Libraries were prepared using NEBNext Ultra II Directional RNA Library Prep Kit for Illumina (New England BioLabs, E7760L) or NEBNext® Ultra™ II RNA Library Prep Kit (New England BioLabs, E7770L) as per manufacturer’s guidelines. Ribosomal RNA (rRNA) was removed using the NEBNext® Poly(A) mRNA Magnetic Isolation Module (New England BioLabs, E7490L) or using the NEBNext® rRNA Depletion Kit v2 (Human/Mouse/Rat) (New England BioLabs, E7400X). Paired-end sequencing with a read length of 150 bp was performed using Illumina NovaSeq 6000. Library sizes and quantification were determined by 4200 Tapestation (Agilent) and libraries were subsequently pooled equimolar. Each library was loaded onto one lane of an S4 v1.5 flow cell (300 cycles) (Illumina, #20028312).
The RNAseq pipeline implemented in bcbio-nextgen (version 1.2.9) was used for quality control and gene expression quantification. Reads were aligned to the UCSC build GRCh38 Homo sapiens genome, augmented with transcript information from Ensembl release 86 using STAR’s 2-pass mapping mode (version 2.6.1d), and to mouse mm10 genome. Alignments were evaluated for evenness of coverage, rRNA content, genomic context of alignments, and complexity using a combination of FastQC, Qualimap, and custom tools. Transcripts per million (TPM) measurements per isoform were generated by alignment-based quantification using Salmon (version 1.6.0) and used to estimate abundance of genes [33]. The aggregated gene counts were used for differential gene expression analyses with DESeq2 [34]. Log2 transformation was used for data analysis.
BRCA1 isoforms analysis
RNA was extracted using the RNeasy Mini Kit (Qiagen) with an additional step of DNase digestion using the RNase-Free DNase Set (Qiagen). A total of 200 ng of RNA were retrotranscribed to yield cDNA using the PrimeScript RT reagent kit (Takara), combining random and oligo-dT primers. Quantification of the BRCA1 Δ11q isoform transcript levels was performed by reverse transcription-quantitative PCR (RT-qPCR) using TaqMan probes targeting exon 11 cryptic donor and exon 12 acceptor junction (custom assay). Additionally, a TaqMan probe expanding exon 23–24 (Hs01556193_m1) junction was used to measure BRCA1 global expression. The geometric means of the expression values for both RPLP0 (Hs99999902_m1) and GAPDH (Hs02758991_g1) housekeeping genes were used to normalize the expression. For each sample, qPCR assays were performed in triplicate using 1 µl of the 20 × TaqMan gene expression assay (TaqMan probes), 10 µl of 2 × TaqMan Fast Advanced Master Mix (Applied Biosystems), 8 µl of water, and 1 µl of the cDNA (20 ng) previously generated. Samples were run for 40 cycles on a QuantStudio 6 Flex PCR System (Applied Biosystems) with the following thermal cycler conditions: 2 min at 50 °C, 20 s at 95 °C, 3 s at 95 °C, and 30 s at 60 °C. Triplicates were individually analyzed, and the corresponding mean values were considered. Data obtained in the form of quantification cycle (Ct) was normalized using the average values of the two reference genes (ΔCt) [35]. Splicing fraction of the Δ11q isoform was calculated as 2−ΔCt(Δ11q)/(2−ΔCt(23–24)) × 100 [36]. Samples presenting BRCA1 Δ11q splicing fraction > 1.5 × the splicing fraction at baseline were classified as hypomorphic BRCA1 due to Δ11q isoform overexpression.
PARylation assay
A frozen tumor specimen was homogenized in ice-cold 1 × radioimmunoprecipitation assay (RIPA) buffer (Tris–HCl pH 8.0 10 mM, EDTA 1 mM, Triton-X-100 0.1%, SDS 0.1%, SDC 0.1%, and NaCl 140 mM) supplemented with 1X protease inhibitor cocktail (cOmplete, Roche), NaF 10 mM, Na2VO4 200 mM, and PMSF 5 mM. Protein concentration was calculated using DC™ Protein Assay (Bio-Rad). PARylation was then determined by western blot. Briefly, a total of 20 µg of protein was used on 8% and 12% SDS-PAGE acrylamide gels at 100 V and transferred to nitrocellulose membrane for 1.5 h at 100 V. Membranes were blocked for 1 h in 5% milk in Tris-buffered saline (TBS) with 0.1% Tween20 (T-TBS) and then hybridized using the corresponding primary antibodies in 5% bovine serum albumin (BSA, Sigma-Aldrich, #A9647) with T-TBS overnight. The rabbit anti-poly/mono(ADP-ribose) (E6F6A) (Cell Signaling #83,732, 1:1000) and human GAPDH (Abcam, ab128915) antibodies were used. Membranes were incubated for 1 h with mouse and rabbit horseradish peroxidase (HRP)-conjugated secondary antibodies (GE Healthcare) in 5% milk in T-TBS. Proteins were detected with Immobilon Western Chemiluminescent HRP substrate (Millipore). Immunoblots were captured at the chemiluminescence imager Amersham Imager 600 (GE Healthcare). Images were captured with FUJIFILM LASS-4000 camera system.
Immunofluorescence and biomarker scoring
The following primary antibodies were used for immunofluorescence (IF): rabbit anti-RAD51 (Abcam ab133534, 1:1000), mouse anti-geminin (NovoCastra NCL-L, 1:60), rabbit anti-geminin (ProteinTech 10,802–1-AP, 1:400), mouse anti-BRCA1 (Santa Cruz Biotechnology sc-6954, 1:50), mouse anti-γ-H2AX (Millipore #05–636, 1:200), rabbit anti-phospho RPA32/RPA2 (S4/S8) (pRPA, Bethyl Laboratories A300-245A, 1:500), and rabbit anti-53BP1 (Cell Signaling 4937, 1:100). Goat anti-rabbit Alexa fluor 568, goat anti-mouse Alexa fluor 488, donkey anti-mouse Alexa fluor 568, and goat anti-rabbit Alexa fluor 488 (all from Invitrogen; 1:500) were used as secondary antibodies. The IF staining was performed on FFPE PDX tumors as described in Castroviejo-Bermejo et al [37].
RAD51, BRCA1, γ-H2AX, and pRPA scores were quantified as the percentage of geminin-positive cells with 5 or more nuclear foci. Geminin is a master regulator of cell-cycle progression that ensures the timely onset of DNA replication and was used as counterstaining to mark for S/G2-cell cycle phase [38]. Samples with low γ-H2AX score (< 25% of positive cells) or with < 40 geminin-positive cells were not evaluated due to insufficient endogenous DNA damage or insufficient tumor cells in the S/G2-phase of the cell cycle, respectively. Progressing tumors with high BRCA1 score (> 1.5 × the percentage of BRCA1 score of the same PDX at baseline) were classified as harboring hypomorphic BRCA1. Recruitment of 53BP1 to DNA damage was evaluated by the qualitative assessment of geminin-positive cells with 5 or more 53BP1 nuclear foci. One hundred geminin-positive cells from at least three representative areas of each sample were analyzed for each biomarker. Genomic instability was scored as the percentage of tumor cells with at least one micronucleus based on DAPI staining (micronuclei score). One hundred cells from at least three representative areas of each sample were analyzed for the micronuclei score. All scorings were performed blindly onto life images using a 60 × -immersion oil lens. At least two biological replicates per PDX model were analyzed.
Statistical analysis
Data was analyzed with GraphPad Prism version 8.2.1 (GraphPad Software) and R software. Shapiro–Wilk test was used to assess normality of data distributions. If the null hypothesis of normal distribution was not rejected, statistical tests were performed using unpaired two-tailed t-test (for two groups comparison of RAD51 score). Otherwise, the non-parametric pairwise Wilcoxon test corrected for multiple testing (false discovery rate correction) was used (for two groups comparison of treatments as percentage of change from baseline) [39]. Bars represent the mean of at least three technical replicates. For preclinical readouts of response (pCRR, pORR, and pCBR), two binomial generalized linear mixed-effects models were performed, both with PDX models as random factor [40]. First, preclinical readouts of response to AZD5305 were analyzed using the response that the same PDX showed upon treatment with olaparib as covariate. The aim of these models was to test whether AZD5305 response rates correlated with olaparib response. Second, a deeper analysis was performed only with those models that showed at least one individual tumor sensitive to olaparib. These models included the preclinical readouts of response as dependent variable and the treatment (olaparib and AZD5305) as fixed factor. In the case of preclinical readouts assessing progression (pPFS and pTTP), a mixed-effects cox regression model was performed [41], all of them in R software [42]. Progression rate of tumors treated with AZD5305 and olaparib was compared using a chi-square test. To calculate the association between RAD51 score and pTTP to both PARPi in PDXs and between RAD51 and BRCA1 scores of progressing tumors, a linear regression model was fitted to estimate the R2 with 95% confidence intervals (CI). To analyze biomarker (RAD51, BRCA1, γ-H2AX, pRPA, and micronuclei) scores according to response to AZD5305, a linear mixed-effects model was used.
Add Comment