In silico analyses of RFWD3 expression were carried out using The Cancer Genome Atlas (TCGA) dataset of 489 HGSOC cases (TCGA-OV) described in (5). Median follow up was 30 months (range 0-179 months). In this cohort, overall survival was defined as the interval from initial surgical resection to the last known contact or death. Disease free survival was defined as the interval from initial surgical resection to disease progression, date of recurrence, or date of last known contact if the patient was alive and had not recurred. Analysis was performed using the cBioportal online tool (www.cbioportal.org) (23) and Graphpad Prism v9.4.1.

Human HGSOC cell lines ES2, 59M, COV318 and PEO1 were obtained from in-house liquid nitrogen frozen stocks. Cell lines were authenticated by short tandem repeat profiling, and regularly mycoplasma tested. Cells were grown in RPMI-1640 (Thermo Fisher, Loughborough, UK) supplemented with 10% fetal bovine serum (Thermo Fisher), 1% penicillin streptomycin (Thermo Fisher), and 2mM L-glutamine (Thermo Fisher).

Cultured cells were treated with lysis buffer on ice, centrifuged at 15,000g, and supernatant collected. Protein content was quantified using the bicinchoninic acid assay. Lysates were loaded on 7.5% acrylamide gels, and subjected to SDS-PAGE to separate proteins. Proteins were transferred to PVDF membrane, incubated with blocking buffer (LI-COR, Cambridge, UK) for 1h, and incubated overnight with primary antibody diluted in blocking buffer. Primary antibodies targeting RFWD3 (Abcam, Cambridge, UK; ab138030; 1:500 dilution), tubulin (Abcam; ab7291; 1:5000 dilution), PARP (Cell Signaling Technology, Danvers, MA, USA; 9542; 1:1000 dilution) and FANCD2 (Abcam; ab108928; 1:1000 dilution) were used. Membranes were incubated with anti-mouse IRDye 680LT (LI-COR 926-68020) and anti-rabbit IRDye 800CW (LI-COR 926-68023), and imaged using the Odyssey Infrared Imaging System (LI-COR Biosciences). Quantification of band intensity was performed using ImageJ v1.53 (24).

Cytotoxicity and proliferation were measured using the SRB assay. 250-3000 cells were plated per well in 96-well plates, treated with 1:4 or 1:2 dilution series of carboplatin (Sigma-Aldrich, Gillingham, UK), mitomycin C (Selleckchem, Houston, TX, USA) paclitaxel (Selleckchem), or olaparib (Selleckchem) and incubated for 5 days at 37°C. These were fixed using 25% trichloroacetic acid (Sigma-Aldrich), and stained with 0.4% w/v sulforhodamine B dye (Sigma-Aldrich). Dye was dissolved by addition of Tris pH 10.5, and optical density at 540nM was measured using a BP800 Microplate Reader (Biohit). Absolute IC₅₀ values were interpolated from concentration-response curves using Graphpad Prism v9.4.1.

Transfection of cell lines with siRNA was performed using Lipofectamine 3000 (Thermo Fisher) according to manufacturer’s instructions. RFWD3 targeting siRNA CCAUUUGAGGUGAACCGUAtt or negative control siRNA (1027281, Qiagen, Manchester, UK) was used. Transfection was performed twice, with cells allowed to recover for 24h between transfections.

sgRNA sequences targeting two distinct sequences in the RFWD3 gene (CACCGCTCTCAGGGTTCGGGCATAA and CACCGGGCTCTCAGCATTACGCTGT; Integrated DNA technologies (IDT), Ilkeston, UK) were cloned into pSpCas9(BB)-2A-Puro (PX459) V2.0 plasmid backbones (25). Cells were transfected with the resulting plasmids using Lipofectamine 3000 (Thermofisher) according to manufacturer’s instructions. Selection of transfected cells was carried out for 48h using 2.5µg/mL puromycin, and individual cells were subcloned to generate monoclonal cell lines. Monoclonal cell lines were evaluated for gene editing via Western blot and Sanger sequencing. CRISPR/cas9 mediated editing was carried out using the same methodologies as described for RFWD3, and sgRNA sequences targeting two positions in the FANCD2 gene (CACCGCATCCTCAATGTAAGACTCC and CACCGGATAGGAAGGGTGTCTCCTC; IDT).

DNA extractions were carried out using an Allprep DNA/RNA kit (Qiagen) following manufacturer’s instructions. DNA was amplified for sequencing via PCR using AmpliTaq Gold 360 PCR Master Mix (Thermo Fisher) and following manufacturers’ protocols. PCR cycling conditions were as follows; 95˚C for 5 minutes, followed by 40 cycles (95˚C for 30 seconds, 55˚C for 30 seconds, and 72˚C for 30 seconds), and 72˚C for 5 minutes. Forward primer sequence GTTCTGTAGCCCTTTTGATTGTATCAT, reverse primer sequence TTTCTGTATGGAGAACTGCTGTGG (IDT). PCR product was analysed for heterozygosity by gel electrophoresis and Sanger sequenced.

Cells were seeded in 10cm dishes, and treated with 10µM bromodeoxyuridine (BrdU; Abcam) for 30min at 37°C. 2x10⁶ cells were fixed in 70% (v/v) ethanol. These were digested to nuclei using 1mg/mL pepsin (Sigma-Aldrich), treated with 2M hydrochloric acid, and blocked using 0.5% BSA (Sigma-Aldrich) solution. Staining was with 1:100 anti-BrdU antibody (Invitrogen; 14-5071-82), followed by anti-mouse AF488 secondary antibody (A11029, Invitrogen). Nuclei were then treated with RNase A (Thermofisher) and 50µg/mL propidium iodide (PI; Thermofisher), and analysed using the LSRFortessa (Becton Dickinson, Franklin Lakes, USA). Cell cycle phase was assigned based on PI and BrdU staining gates in FlowJo v10.8.1.

A total of 3000 cells were seeded per well, and were fixed and stained as described above using the SRB assay. Plates were fixed at D0, 6h, 24h, 48h, and 72h time points. 6 wells were seeded per condition, and the mean absorbance was used to plot growth curves. Doubling times were calculated from growth curves plotted in Graphpad Prism v9.4.1.

Cell migration was analysed using the Incucyte S3 live imaging system (Essen, Royston, UK). 96 well Imagelock plates (Essen) were coated with 50µg/mL collagen I (prepared in-house according to (26)) or 10µg/mL laminin (Thermo Fisher). 1000 cells were seeded per well, and imaged every 15 min. Image sequences were analysed using the mTrackJ plugin on ImageJ v1.53. Individual cells were tracked for 10h, and mean migration velocity per track was calculated using mTrackJ (27). 45 cells were tracked per biological replicate.

Microarrays of HGSOC patient tumours are described in (28). Briefly, from ovarian cancer cases treated at the Edinburgh Cancer Centre prior to 2007, 365 high grade serous ovarian cancer cases were identified by expert pathological review. All were treated with first line platinum containing chemotherapy after primary debulking surgery. Samples of treatment naïve tumour were taken at primary surgery, fixed in formalin and embedded in paraffin. Sections were cut for immunohistochemistry (IHC) staining. Median follow up was 13.5 years (range 0.2-22.7 years). Overall survival (OS) in this cohort was defined as the interval from pathologically confirmed diagnosis to patient death. Progression free survival (PFS) was defined as the interval from pathologically confirmed diagnosis to disease progression or recurrence, respectively. Ethical approval was obtained from South East Scotland Human Annotated Bioresource (Lothian NRS Bioresource Ethics Committee reference 20/ES/0061-SR705 and SR1518). This study was carried out in accordance with the principles of the Declaration of Helsinki.

Slides were dewaxed and rehydrated using the Leica Autostainer XL (Leica, Newcastle, UK), and treated with 3% hydrogen peroxide solution (Sigma-Aldrich). Antigen retrieval was performed by heating slides in a pressure cooker containing Tris-EDTA buffer for 12 min. Slides were permeabilised using 0.5% Triton-X100. Blocking was with 5% goat serum for 1h. Antibody staining was with anti-RFWD3 antibody (Proteintech, Manchester, UK) (19893-1-AP) diluted 1:400 overnight at 4°C. Slides were then incubated with peroxidase conjugated secondary antibody (A0545 Sigma-Aldrich) and 3, 3’-diaminobenzidine (DAB) solution (Agilent, Wokingham, UK) for 10 min. Slides were then dehydrated and counterstained with haematoxylin on the Leica Autostainer XL before mounting. Slides were imaged in the brightfield channel using the Nanozoomer XR (Hammamatsu, Welwyn Garden City, UK).

IHC staining of the tumour microarray (TMA) was analysed using QuPath software version 0.3.2 (https://qupath.github.io/) (29). Cells were automatically detected, and tumour areas annotated. Thresholds for nuclear DAB staining were set at optical density (OD) 0.1, 0.3 and 0.5 to determine negative, weak, moderate and strong staining in individual cells. Histoscores for each tumour core were generated based on the proportion of cells classified within each of these categories automatically within QuPath. Evaluable tumour cores were available for 231/365 cases. Where available, triplicate (54/231) and duplicate (84/231) cores for each case were analysed, and mean values were used for statistical analysis. Duplicate and triplicate cores demonstrated good concordance between histoscores (median coefficient of variation 12.9%).

All graphs were plotted using Graphpad Prism v9.4.1. The Cox proportional hazards test for performing multivariate analysis of survival data, and testing of proportionality of the variables, was performed using the survival package in R v4.2.2. All other statistics were performed in Graphpad Prism v9.4.1. Tests used are described in individual figure legends.

The TCGA-OV dataset is a large and well characterised publicly available dataset consisting of 489 cases of HGSOC, with associated mRNA expression and clinical data (5). A subset of 316 cases also have coding exon sequences. This dataset was interrogated to determine whether alterations in RFWD3 sequence or expression occur in HGSOC. Summary clinicopathological information for this cohort is given in Supplementary Tables S1 and S2 .

No instances of mutations in the RFWD3 gene were observed in the dataset. However, low mRNA expression, defined as a Z-score of <-2 compared to tumours in the dataset with diploid RFWD3, occurred in 9% of cases (28/489). While two of these coincided with copy number deletions, the reason for low expression in the others is unknown. In contrast, high expression, defined as a Z-score of >2, occurred in just 1.4% (7/489) of cases. Low expression of RFWD3 therefore occurs in a proportion of HGSOCs ( Supplementary Figure S1 ).

The subset of low RFWD3 expressing cases were further investigated to determine whether they were associated with differences in clinical outcomes and tumour phenotype. These, and cases with mutations in other bona fide FA genes (FANCA, FANCB, FANCC, FANCD2, FANCE, FANCF, FANCG, FANCL, FANCM, FANCI, UBE2T, SLX4, ERCC4, MAD2L2, BRIP1, BRCA2, PALB2, RAD51C, RAD51, BRCA1, XRCC2), were compared with a control group of cases with neither an FA mutation nor low expression of RFWD3. There were no differences in stage or residual disease between the groups. While there was an apparent difference in patient age between the groups, no significant differences compared to the control group were observed after adjusting for multiple testing (Bonferonni corrected p=0.076) ( Supplementary Table S1 ). Both the cases with low RFWD3 expression and those with mutations in other FA genes were found to have significantly higher non-synonymous tumour mutational burden (control VS RFWD3 low expression p=0.025; control VS FA mutated p<0.0001) compared with the control group ( Figure 1 ). This suggests that low RFWD3 expression can influence tumour genotype in a similar way to mutation of FA genes including BRCA1 and BRCA2.

Differences in overall survival (OS) and disease-free survival (DFS) were also assessed. While both DFS and OS were significantly improved for the FA mutated cases compared to the control group (median OS 58 VS 43 months, Bonferonni-adjusted p=0.009; median DFS 19 VS 15 months, Bonferonni-adjusted p=0.0006), there was no difference in DFS (Bonferonni-adjusted p>0.999) or OS (Bonferonni-adjusted p>0.999) between the control and RFWD3 low-expressing groups ( Supplementary Figure S2 ).

RFWD3 expression was investigated across three HGSOC cell lines with varying carboplatin sensitivities, 59M, ES2 and COV318 (mean IC₅₀ values 10.2 µM, 5.4µM, 2.6µM respectively) ( Figures 2A-C ). Although multiple bands were observed in Western blot analysis, these were all reduced in intensity by siRNA knockdown ( Supplementary Figure S3 ), and likely represent different isoforms or posttranslational modifications of RFWD3. While similar RFWD3 expression was observed in two of the cell lines, the COV318 cells showed minimal expression levels ( Figure 2B ). This was significantly reduced compared with the other cell lines (p=0.0001, p=0.002 for 59M and ES2 respectively) ( Figure 2C ). Interestingly, reduced expression occurred in the most chemosensitive cell line of the three ( Figure 2A ).

Therefore, a potential link between RFWD3 expression and chemosensitivity was investigated. As the COV318, ES2 and 59M cell lines have different genetic backgrounds, and therefore a combination of multiple factors is likely responsible for differences in chemosensitivity, isogenic cell lines were used to determine whether modulation of RFWD3 could contribute to chemosensitivity. Transient knockdown of RFWD3 was performed in the 59M and ES2 cell lines using siRNA ( Supplementary Figure S3 ). Endogenous expression levels of RFWD3 in the COV318 cell line were too low to obtain a measurable expression difference upon siRNA knockdown. Reduced RFWD3 expression decreased the IC₅₀ of carboplatin by approximately 2-fold in both 59M (mean IC₅₀ 3.9µM VS 2.0µM, p=0.036) and ES2 (mean IC₅₀ 2.7µM VS 1.0µM, p=0.016) cell lines ( Figures 2D-I ). Therefore, low RFWD3 expression appeared to be associated with increased carboplatin sensitivity.

To study the effects of differential RFWD3 expression in a stable model, CRISPR/cas9 mediated gene editing was used to disrupt the RFWD3 gene in the ES2 cell line. Two sgRNAs were designed and used to delete a DNA section of 197bp across exons 5 and 6 of the RFWD3 gene. Clones with homozygous edits were obtained ( Supplementary Figure S4 ), and the edits were confirmed by sequencing ( Supplementary Figures S5 - S7 ). However, edited clones still retained expression of RFWD3 protein ( Supplementary Figure S8 ), although this was reduced by approximately 50% compared with empty vector transfected controls. It was therefore reasoned that the mutations may be hypomorphic, resulting in reduced protein expression. These cell lines are hereafter referred to as RFWD3^Δ/Δ cells. Interestingly, we were able to produce knockout cell lines using the PEO1 ovarian cancer cell line, for another FA pathway protein, FANCD2 ( Supplementary Figure S9 ) using the same methodology. While multiple other groups have attempted to generate knockout models of RFWD3, these have also been unsuccessful and have resulted in similarly reduced protein levels (15, 18, 30). Therefore, RFWD3 is speculated to be an essential gene for cell survival.

As the siRNA data showed that changes in RFWD3 expression can modify response to platinum, the response of RFWD3^Δ/Δ cells to a range of chemotherapeutic drugs was assessed, in order to identify potential strategies to best treat tumours with differing expression of RFWD3. Consistent with our previous results, the RFWD3^Δ/Δ cells were hypersensitive to carboplatin, demonstrating an approximately 10-fold increase in sensitivity compared with controls (mean IC₅₀ 2.9µM control VS 0.3µM RFWD3 ^Δ/Δ clone 3, p=0.0004; mean IC₅₀ 2.9µM control VS 0.8µM RFWD3 ^Δ/Δ clone 44, p=0.003) ( Figures 3A, B ). RFWD3^Δ/Δ cells showed similarly enhanced sensitivity to mitomycin C (mean IC₅₀ 25.9nM control VS 1.9nM RFWD3 ^Δ/Δ clone 3, p<0.0001; mean IC₅₀ 25.9nM control VS 5.2nM RFWD3 ^Δ/Δ clone 44, p=0.0003) ( Figures 3C, D ); both of these drugs introduce ICLs to DNA. This validated the ability of the RFWD3^Δ/Δ cells to show phenotypic changes consistent with FA deficiency.

Apoptosis in response to carboplatin treatment was assessed via Western blot analysis of caspase-cleaved PARP (cPARP). The proportion of cPARP expression increased for all cell lines upon treatment with carboplatin ( Figures 3E, F ). At 100µM carboplatin treatment, there was a significant increase in the proportion of cPARP in the RFWD3^Δ/Δ cell lines compared with controls (mean percentage of cPARP 47% control VS 81% RFWD3 ^Δ/Δ clone 3, p<0.0001; mean percentage of cPARP 47% control VS 89% RFWD3 ^Δ/Δ clone 44, p<0.0001) ( Figure 3F ). This indicates that RFWD3 deficiency can lead to increased apoptosis upon carboplatin treatment.

Cell response to treatment with paclitaxel, another commonly utilized chemotherapeutic in HGSOC treatment, was also assessed. No consistent differences in paclitaxel sensitivity of control and RFWD3^Δ/Δ cells were observed ( Supplementary Figure S10 ). Therefore, ICL inducing agents appear to be the most effective treatment for tumour cells with low expression of RFWD3. To assess whether there was a differential response to PARP inhibitors, the response of the RFWD3 Δ/Δ clone 3 was compared with the control clones, however only a non-significant small change was observed ( Supplementary Figure S11 ).

RFWD3 has been previously reported to participate in non-canonical signalling pathways distinct from its role in ICL repair in other cancer types (21, 22). Therefore, the effect of RFWD3 deficiency on cancer relevant phenotypes was assessed using the RFWD3^Δ/Δ cell lines.

Proportions of cells in different phases of the cell cycle were assessed using flow cytometry ( Supplementary Figure S12 ). There was no difference in the proportions of cells in G0/G1 or S-phase of the cell cycle ( Figures 4A, B ). However, both RFWD3^Δ/Δ clones showed an accumulation of cells in G2/M phase compared with controls (mean percentage of cells 10.7% control VS 15.3% RFWD3 ^Δ/Δ clone 3, p=0.013; mean percentage of cells 10.7% control VS 18.6% RFWD3 ^Δ/Δ clone 44, p=0.0001) ( Figure 4C ). This indicates that RFWD3 can modify cell cycle progression, either directly or indirectly through accumulation of DNA breaks due to ICL repair loss, and deficiency of RFWD3 results in G2/M checkpoint arrest.

Cell growth was measured over 72h ( Supplementary Figure S13 ). Doubling times of RFWD3^Δ/Δ cells were significantly increased compared with controls (mean doubling time 16.1h control VS 20.8h RFWD3 ^Δ/Δ clone 3, p=0.028; mean doubling time 16.1h control VS 29.8h RFWD3 ^Δ/Δ clone 44, p<0.0001) ( Figure 4D ), demonstrating that deficiency of RFWD3 results in impairment of tumour cell proliferation. This again may be either a direct effect of RFWD3 loss, or an indirect effect resulting from accumulation of DNA damage.

Motility of cells in 2D was measured on collagen I and laminin substrates ( Supplementary Figure S14 ), as these are major components of the ovarian extracellular matrix. Wild type and empty vector control ES2 cells were capable of migration on both substrates. In comparison, RFWD3^Δ/Δ cells showed significantly impaired migration velocities both on collagen I (mean velocity 1.28µm/min control VS 0.94µm/min RFWD3 ^Δ/Δ clone 3, p=0.001; mean velocity 1.28µm/min control VS 0.72µm/min RFWD3 ^Δ/Δ clone 44, p<0.0001) and laminin (mean velocity 0.99µm/min control VS 0.66µm/min RFWD3 ^Δ/Δ clone 3, p=0.014; mean velocity 0.99µm/min control VS 0.60µm/min RFWD3 ^Δ/Δ clone 44, p=0.005) ( Figures 4E, F ), indicating that RFWD3 can influence cell migration.

To explore potential relationships between RFWD3 expression patterns at the protein level and outcomes in HGSOC, tumour microarrays were immunostained for RFWD3. Microarrays were generated consisting of 231 evaluable tumour cores sampled from a cohort of chemonaïve HGSOC patients at primary surgery. Summary clinicopathological information for this cohort is given in Supplementary Tables S3 and S4 .

RFWD3 staining predominantly occurred in cell nuclei, and intensity was highly varied across the cohort ( Figures 5A-C ). Histoscores, describing the overall proportion of strong, moderate, weak and negative staining, were generated for each core based on the nuclear staining intensity in tumour cells. These were normally distributed ( Figure 5D ), therefore the cases were split into “high” “mid” and “low” expression groups based on quartiles, with histoscores greater than Q3 assigned as “high” expression, less than Q1 assigned as “low” expression, and remaining cases assigned as “mid” level expression. Differences in clinical characteristics for the RFWD3 expression groups were assessed. No significant associations were found between RFWD3 expression and patient age, tumour stage, residual disease, or BRCA1/2 mutation status ( Supplementary Table S3 ).

Response to first line therapy was assessed for the tumours. Radiological response data was available for 83 of the tumours analysed for RFWD3 expression. A total of 42 of these were treated with platinum and taxanes, while 41 were treated with platinum without taxanes. As the number of cases with radiological response data available was low, quartile group data could not be analysed due to insufficient group numbers to give meaningful results. Therefore, response was assessed by comparing RFWD3 histoscores for each outcome for the different treatment types.

Significantly lower RFWD3 expression was observed for those patients with a complete response (CR) compared to those who showed progressive disease (PD) for the patients treated with platinum only (median histoscores 105 CR VS 178 PD, p=0.046) ( Figure 5E ). This supports a role for RFWD3 expression in tumour platinum response. In contrast, for those patients treated with platinum and taxane based therapy, there was no significant difference between those with a complete response to chemotherapy and those with progressive disease ( Figure 5F ).

The survival data was also analysed for this patient cohort, to determine whether the differences in therapeutic outcomes observed would translate into survival differences for patients with altered RFWD3 expression. Multivariate analyses of OS and PFS were carried out using the Cox proportional hazards model ( Table 1 ). Age and residual disease were accounted for in the models, and the models were stratified by stage as this did not fit the proportional hazards assumption. Counter to expectations, the RFWD3 high-expression group had an improved PFS compared to the mid-expression group (hazard ratio 0.46-0.99, p=0.045). There was no significant difference in survival for the low-expression group. This finding however did not translate to a change in OS, as there were no significant changes for any of the RFWD3 expression groups.