Whole genome sequencingGenomic DNA (gDNA) was extracted from both cell lines and human organoids by means of Maxwell® RSC Blood DNA Kit AS1400 (Promega, Madison, WI, USA).Starting from 500 ng of cell line-derived gDNA, Next Generation Sequencing (NGS) libraries were prepared in-house by means of Nextera DNA Flex Library Prep kit (Illumina Inc., San Diego, CA, USA), according to manufacturer’s protocol. The quality of libraries was checked with Qubit™ dsDNA Quantification Assay Kits (ThermoFisher Scientific, Waltham, MA USA), while DNA fragments’ size distribution was assessed using the 2100 Bioanalyzer with a High-Sensitivity DNA assay kit (Agilent Technologies, Santa Clara, CA, USA). Equal amounts of final DNA libraries were pooled and sequenced on NovaSeq 6000 (Illumina Inc., San Diego, CA, USA) as paired-end 150 bp reads. In the case of organoid samples, 2ug of gDNA have been used as starting material for TruSeq DNA PCR-Free Library Prep Kit (Illumina Inc., San Diego, CA, USA), in order to generate WGS data according to manufacturer’s protocol. DNA fragmentation step has been obtained by using M220 Focused-ultrasonicator (Covaris LLC, Woburn, MA, USA) with settings for 500 bp and 130ul AFA tubes. The quality of libraries was checked with Qubit™ dsDNA Quantification Assay Kits (ThermoFisher Scientific, Waltham, MA USA), while DNA fragments’ size distribution was assessed using the 2100 Bioanalyzer with a High-Sensitivity DNA assay kit (Agilent Technologies, Santa Clara, CA). Equal amounts of final DNA libraries were pooled and sequenced on NovaSeq 6000 (Illumina Inc., San Diego, CA, USA) as paired-end 150 bp reads.HRDetect pipelineWith the aim to reproduce the original results by Davies et al. (Davies, Glodzik et al. 2017), we develop a pipeline that is able to generate an HRDetect score starting from alignment data (Fig. 1). This workflow is composed of these steps: annotation of BAM files, variant calling, filtering and finally HRDetect method calling. Alignment files must be compliant with Cancer Genome Project pipeline that requires them to have ReadGroup identifier set (“RG:”) on every row along with several information in the BAM header:-read group (“RG” line)-genome assembly identifier (“AS” field in “SQ” line)-species (“SP” filed in “SQ” line)-sample name (“SM” filed in “RG” line).A raw BAM input file is annotated as requested by a custom script.The official Cancer Genome Project for WGS pipeline (also called “cgpwgs”) by Wellcome Sanger Institute was used for variational analysis; this workflow is implemented in a Docker image and a corresponding Dockerfile is hosted as “dockstore-cgpwgs” on GitHub website (https://github.com/cancerit/dockstore-cgpwgs). Some of the dependencies in the Dockerfile were broken, so it was then modified in order to fix them and have a running container. This step always requires two samples as input: one healthy and one tumoral. Custom scripts were used to filter variants retaining only the most confident ones. Raw variations results were converted to match the format requested by the “HRDetect_pipeline” function in “signature.tools.lib” R library 20 with a custom script.HRDirect pipelineStarting from the previous pipeline, we developed a new workflow able to be independent of normal or healthy data. To make limited changes to the HRDetect workflow, we generated a meta-normal sample using 14 healthy WGS from the Davies breast dataset. These data were merged and subsampled to achieve a median total depth of 40× (typical depth for WGS data), resulting in a new single BAM file. This will be used instead of the matched (and possibly missing) normal sample as input for the workflow. Four methods have been developed in order to enrich real somatic variants (Supplementary Fig. 3). The first two methods took into account only SNVs or indels whose VAF is in the range of 10–20% (strategy A) or 5–35% (strategy B) respectively. These values come from the observation of the distribution of VAFs for matched comparison (hereafter reference distribution), both in SNVs and indels (Supplementary Figs. 1 and 2).A third strategy was evaluated, in which variations are not filtered based on their value (i.e., range filter), but are sampled proportionally to the reference distribution, trying to reproduce this latter (strategy C): most represented VAF are more sampled. This approach had similar results to strategy B.Strategy D is not based on the reference distribution but on the sample data. Germline and somatic variations are mixed, but the formers are the vast majority; since the germline VAFs distribution is essentially bell-shaped and specular and the somatic distribution is not enriched for frequencies larger than 50%, this part of the real distribution can be considered as pure germline. Therefore, we can hypothesize that the germline shape is the mirror of the right part of the total distribution. The exceeding fraction is then the somatic one. This is the core of strategy D.Based on each strategies’ results, the HRDirect workflow implements an “ensemble” approach collecting all the classifications using a majority vote rule and outputting the final prediction (Fig. 2). Notably, the outcome of HRDirect is no more a number but a classification label.Ethics, organoid culture, and drug screeningTumor samples were obtained from patients treated at Niguarda Cancer Center, (Milano, Italy) and IRCCS (Candiolo, Turin, Italy). All patients signed a dedicated informed consent in accordance with guidelines of the ALFAOMEGA Master Observational Trial (NCT04120935, IFOM-CPO003/2018/PO002) and the PROFILING protocol (001-IRCC-00IIS-10, 6.0 version, dated 24th August 2015).The study was conducted in accordance with the Declaration of Helsinki and under the approval of the local Independent Ethical Committees of each participating center (for ALFAOMEGA study: Ethical Committee Niguarda Cancer Center Milano Area 3, decision n. 617-122018 dated 13/12/20218 and Ethical Committee IRCCS Candiolo, decision n. CE IRCCS 102/2021 dated 25/03/2021, for Profiling study: Ethical Committee IRCCS Candiolo, decision n. CE IRCCS 225/2015 dated 10/09/2015). PDOs #1, #2, #3, #4, #5, #11, and #12 were previously established and characterized as described in ref. 14,29,30.Organoids from CRC patients #7, #8, #9, #10, #13, #14, #15, #16, #17, #18, #20 were established directly from tissue biopsy obtained at the time of surgery, while organoids from patients #6, #19 were generated from patient-derived xenografts (PDXs) models following procedures described in full details in ref. 14.To generate PDXs, tumor specimens were subcutaneously implanted in 7-week-old NOD-SCID mice (Charles River Laboratory). All animal procedures were approved by the Ethical Committee of the Candiolo Cancer Institute and by the Italian Ministry of Health.Organoids from patients #1-#9 were tested with olaparib in a 7-day-long viability assay as described in ref. 14. Olaparib response data for patients 1 to 5 are retrieved from previous publication14, while patients 6 to 9 have been de novo screened.Hematoxylin & Eosin staining of patients’ tumor tissue and organoidsThe tumor specimens were fixed with 10% buffered formalin for 24 hours at room temperature. After rinsing with running water, the specimens were transferred to the Tissue Processor Donatello Series 2 Diapath (Diapath, Martinengo (BG), Italy) for dehydration. The blocks were embedded into paraffin in the Embedding Module Canova Diapath (Diapath, Martinengo (BG), Italy). The FFPE blocks were sectioned at 3 µm using manual rotary microtome Leica RM 2255 (Leica Biosystems, Frankfurt, Germany) and were stained with Hematoxylin & Eosin solution.Immunohistochemical staining of CRC organoidsThree micron-thick sections were cut from formalin-fixed paraffin-embedded cell blocks of PDOs and stained with antibodies raised against RAD51C (rabbit polyclonal antibody, E185, Life Technologies, Thernofisher) and ATM (rabbit monoclonal antibody Y170, Abcam). The protocol for RAD51C was optimized on the Leica BOND staining system (Leica Biosystems), whereas the protocol for ATM was optimized on an automated immunostainer Ventana Benchmark ULTRA (Ventana Roche). Positive and negative controls were included in each immunohistochemical run. For RAD51C, the pattern and intensity of membranous, cytoplasmic, and nuclear staining were recorded; for ATM, the presence or lack of nuclear staining was recorded.Western blot analysis on organoidsOrganoids were enzymatically dissociated using TrypLE Express Enzyme for 10 to 20 minutes at 37 °C to obtain single-cell suspensions and seeded at a density of 140.000 cells per well in 6-well plates precoated with basement membrane extract (BME; Cultrex BME Type 2; Amsbio) overlayed with 2 mL of growth media containing 2% BME. The treatment with drugs started on day 4 after seeding when formed growing organoids were visible. Organoids were treated in fresh 2 mL medium without BME with 1 μM ATRi ceralasertib for 24 hours and 2.5 mmol/L HU for 4 hours. Collection of the organoids was performed using a Cell dissociation Solution. Complete removal of BME was conducted following several washes with cold PBS. Organoids were subsequently lysed in using boiling SDS buffer [50 mmol/L Tris-HCl (pH 7.5), 150 mmol/L NaCl, and 1% SDS] to extract total cellular proteins, quantified by the BCA Protein Assay Reagent kit (ThermoFisher Scientific), and prepared using LDS and Reducing Agent (Invitrogen). Western blot analysis was performed with Enhanced Chemiluminescence System (GE Healthcare) and peroxidase-conjugated secondary antibodies (Amersham). The following primary antibodies were used for Western blotting: anti-phospho-RPA32 (Ser33; Bethyl Laboratories A300-246A; 1:1000), anti-RPA32 (Abcam AB252861; 1:2000), anti-RPA32 (S4/S8; Bethyl Laboratories A300-245A; 1:3000), anti-phospho-Histone H2AX (Ser139; Cell Signaling Technology, 80312S; 1:1000), anti-H2AX (Cell Signaling Technology, 7631S; 1:1000), anti-phospho-DNA-PK (Ser2056; Cell Signaling Technology, 68716S; 1:1000), anti-DNA-PK (Cell Signaling Technology, 12311S; 1:1000), anti-ATM (Cell Signaling Technology, 2873S; 1:1000); anti- phospho-ATM (Ser1981; Cell Signaling Technology, 4526 s; 1:1000) and anti-Vinculin (MERCK 05-386; 1:3000). Detection of the chemiluminescent signal was performed with ChemiDoc Imaging System (Bio-Rad).The following primary antibodies were used for Western blotting in supplementary Fig.4: anti-HSP90 (ABCAM, ab2928; 1:1000), anti-RAD51C (Santa Cruz Biotechnology SC-56214; 1:1000), anti-ATM (Cell Signaling Technology, 2873S; 1:1000).HRD score by commercial targeted assaysFirst, DNA extracted from 31 cell lines was subjected to targeted sequencing using the TruSightTM Oncology 500 HRD panel (TSO500-HRD; Illumina Inc., San Diego, CA, USA). The panel covers 533 genes for a total sequenced size of 1.94 Mb, allowing the detection of SNVs and small indels, and the assessment of MSI status (120 loci), TMB, and copy number for 59 genes. TruSight Oncology 500 HRD also includes probes specifically designed to assess genomic scars taking advantage of the Myriad Genetics Genomic Instability Score (GIS) algorithm to enable HRD evaluation. GIS score is an a-dimensional value ranging from 0 to 100, the sum of three independent scars: LOH, TAI, and LST (PMC6773427). Briefly, following the manufacturer’s protocol, 150 ng of genomic DNA was used as starter material to generate libraries, while 80 ng of post-fragmentation material was used for the subsequent steps. Final libraries were sequenced on the Novaseq 6000 instrument (Illumina, San Diego, California, USA) to reach a minimum of 500× read depth. Raw data were then processed on a local DRAGEN™ server v3 by the DRAGEN™ TruSight Oncology 500 v2 Analysis Software which incorporates a proprietary GIS algorithm powered by Myriad Genetics for HRD assessment. Details about DRAGEN™ pipelines were reported in the user-manual https://support.illumina.com/content/dam/illumina-support/documents/documentation/software_documentation/trusight/trusight-oncology-500/200019138_01_DRAGEN-trusight-oncology-500-analysis-software-v2_1-local-user-guide.pdf.HRD status was also evaluated with the CE-IVD AmoyDx HRD Focus Panel provided by AmoyDx (AmoyDx, Xiamen, China), according to the manufacturer’s instructions. Briefly, 120 ng of DNA was used for library preparation and then sequenced on an Illumina Novaseq 6000 instrument (Illumina, San Diego, California, USA). Raw data were analyzed using the AmoyDx NGS Data Analysis System-ANDAS Software to estimate a GSS. The GSS by AmoyDx returns a score between 0 and 100, based on the evaluation of the same genomic scars identified by the GIS (PMID: 36191839), encompassing a trademarked genomic region with the proprietary AmoyDX algorithm.
