Detecting TP53 mutations in diagnostic and archival liquid-based Pap samples from ovarian cancer patients using an ultra-sensitive ddPCR method

High-grade serous ovarian cancer (HGSOC) is the most common subtype of epithelial ovarian cancer and early detection is challenging. TP53 mutations are a hallmark of HGSOC and detection of these mutations in liquid-based Pap samples could provide a method for early diagnosis. Here we evaluate the use of IBSAFE, an ultra-sensitive droplet digital PCR (ddPCR) method, for detecting TP53 mutations in liquid-based Pap samples collected from fifteen women at the time of diagnosis (diagnostic samples) and/or up to seven years prior to diagnosis (archival samples). We analysed tumours for somatic TP53 mutations with next generation sequencing and were able to detect the corresponding mutations in diagnostic samples from six of eight women, while one patient harboured a germline mutation. We further detected a mutation in an archival sample obtained 20 months prior to the ovarian cancer diagnosis. The custom designed IBSAFE assays detected minor allele frequencies (MAFs) with very high assay sensitivity (MAF = 0.0068%) and were successful despite low DNA abundance (0.17–206.14 ng, median: 17.27 ng). These results provide support for further evaluation of archival liquid-based Pap samples for diagnostic purposes and demonstrate that ultra-sensitive ddPCR should be evaluated for ovarian cancer screening in high-risk groups or in the recurrent setting.

A promising development occurred in 2013, when Kinde et al. showed that somatic mutations in DNA shed from endometrial and ovarian cancers could be detected in standard liquid-based Pap test specimens by massively parallel sequencing 15 . While highly sensitive for endometrial cancer, the method was not able to detect more than 41% of ovarian cancers using a panel of 12 genes commonly mutated in these tumours. Subsequent studies from this research group in collaboration with others have attempted to increase the sensitivity for detection of ovarian cancer by introducing new procedures including lavage of the uterine cavity 16 , by combining Pap test and plasma sampling and by complementing the mutation assay with an assay for aneuploidy 17 . However, clinical sensitivity remains a challenge with this approach, and will require extensive modelling before application in clinical diagnostics [18][19][20] . Moreover, these approaches have only been applied in symptomatic patients, at the time of diagnosis, and have so far not been evaluated in pre-symptomatic women prior to the time of diagnosis.
Droplet Digital PCR (ddPCR) provides an alternative to sequencing-based methods, with the advantages of increased sensitivity, rapid turnover time and ease of use 21 . Analysis of circulating tumour DNA using ddPCR has shown great potential for prognostication and monitoring of treatment response in several tumour types including gynaecological cancers [22][23][24] .
In this study, we analysed liquid-based archival Pap samples (archival samples) from fifteen women collected approximately two to seven years before they were diagnosed with HGSOC, and from nine of these women also liquid-based diagnostic Pap samples (diagnostic samples) collected at the time of the HGSOC diagnosis. Mutations in TP53 were identified by next generation sequencing (NGS) of tumour tissue obtained at surgery. We used the ultra-sensitive ddPCR IBSAFE technology for mutation detection in Pap samples and used a commercially available approach from Bio-Rad as a control where applicable. The analysis of liquid Pap samples from pre-symptomatic women subsequently diagnosed with HGSOC has, to our knowledge, not been previously reported.

Results patient cohort.
A total of 20 archival samples were obtained from 15 patients from cohorts 1 and 2 ( Fig. 1A-C).
Fresh frozen tumour tissue was available from 11 patients, while FFPE tumour tissue was available from four patients. Corresponding blood was available from 14 patients, while nine patients from cohort 1 had diagnostic samples. Patient characteristics are provided in Table 1. The time from the collection of archival samples to the time of HGSOC diagnosis ranged from 20 to 95 months; two patients had more than one archival sample and the remaining patients had a single archival sample. DNA concentrations in diagnostic and archival samples varied between 5.2-55.2 ng/µl (median 16.3 ng/µl) and 0 (too low)-19.4 ng/µl (median 2.755 ng/µl) respectively but were not linked to disease stage. DNA concentrations in archival samples were lower than in diagnostic samples (P = 0.02). www.nature.com/scientificreports www.nature.com/scientificreports/ tumour sequencing and analysis. Paired tumour and blood samples were analysed for TP53 mutations using the INVIEW Oncopanel All-in-one from GATC (Supplementary Table S1). One patient lacked a corresponding blood sample; the tumour sequence was therefore analysed using a normal control constructed from five patients (patients 2, 3, 4, 6 and 9) within cohort 1 and two patients from cohort 2 (patients 11 and 14). At least one mutation was identified for each patient using GEMINI and a hard filter minor allele frequency (MAF) cut-off of 5%. Ten missense mutations, two nonsense mutations, and three frameshift deletions were identified. Mutations were dispersed across the TP53 gene, with no overlap between patients. Patient 4 displayed two mutations in positions adjacent to each other, which was handled as a single mutation in the downstream analysis ( Table 2). All mutations but one were previously recorded in COSMIC (Catalogue Of Somatic Mutations In Cancer); however five of the patients scored neutral or NA in the Functional Analysis through Hidden Markov Models (FATHMM) 25 scoring by GEMINI (GEnome MINIng). MAFs in the tumours ranged between 8-85% with a median of 64% (Table 3).
Mutation screening in diagnostic and archival samples using iBSAfe and Bio-Rad ddpcR. We analysed both diagnostic and archival samples using IBSAFE, an ultra-sensitive ddPCR-based method. We were also able to analyse three diagnostic samples (from patients 3, 8 and 9) with the Bio-Rad-based ddPCR assay to serve as an additional control for the ddPCR approach. Using IBSAFE we were able to detect a tumour MAF comparable to that of the NGS approach (Table 3, Fig. 2A). The Bio-Rad assay performed equally well in the tumour setting. The measured MAFs of the IBSAFE method ranged between 8-86% with a median of 63%.
The IBSAFE method was able to detect tumour-identified mutations in seven of nine diagnostic samples, however one mutation was determined to be a germline event (see below); hence, true somatic mutations were detected in six of eight diagnostic samples. Furthermore, IBSAFE detected the tumour-identified mutation in diagnostic samples from two of three patients with stage II disease (patients 5 and 8). The Bio-Rad assay failed to detect any mutations in the diagnostic sample from patient 3 but was able to detect mutations in the diagnostic samples from patients 8 and 9 (Table 3, Fig. 2B). The calculated MAF in the diagnostic sample from patient 8 (0.0089%) was below the theoretical limit of detection of the Bio-Rad assay 21 .
Despite the age and the low abundance of DNA in the archival samples, the IBSAFE method was successful in all the individual assays, measuring concentrations comparable to the QUBIT assay (Tables 1 and 3, Fig. 2C). Furthermore, we were able to detect a tumour corresponding mutation in the archival samples from patients 1 and 13 (sample 13C). Patient 13 was diagnosed with a stage IIIB tumour and had a total of three archival samples (13A, B and C) collected 46, 32 and 20 months prior to diagnosis. A TP53 mutation with a MAF of 0.042% was detected in sample 13 C obtained 20 months prior to diagnosis, with a total of 31.75 ng of DNA tested. The mutation was however not detectable in the earlier samples (13 A and B).
The MAF of 46% in the sample from patient 1 led us to suspect a germline TP53 mutation, which was confirmed using a normal tissue sample (endometrial biopsy), from the same patient (data not shown). www.nature.com/scientificreports www.nature.com/scientificreports/ No mutations were detected in the remaining archival samples (Table 3, Fig. 2C).

Discussion
In this study we evaluated the ability of IBSAFE, an ultra-sensitive ddPCR method, to detect known TP53 mutations in Pap samples from patients with HGSOC collected at the time of diagnosis and approximately two to seven years prior to diagnosis.
We were able to detect somatic TP53 mutations in diagnostic samples from six of nine women using IBSAFE. TP53 mutations were not detectable in two of the diagnostic samples, most likely due to the MAF being below the limit of detection of the IBSAFE method. The MAF estimates suggest that there is a huge variance in detectable somatic DNA mutations ranging from 0.0068% to 7.9%.
Importantly, IBSAFE was able to detect mutations with an in-sample limit of detection of 1 in 50,000. This sensitivity is much higher than that of conventional NGS [18][19][20]26 . Moreover, the MAF estimates of IBSAFE were higher   www.nature.com/scientificreports www.nature.com/scientificreports/ than those of the commercial approach for the diagnostic samples, despite exhibiting similar MAFs in the tumour setting, possibly indicating higher in-sample sensitivity. This is probably due to the experimental set-up in which the Bio-Rad assay is optimised for 100 ng DNA inputs. In reality, DNA inputs from liquid biopsies, such as plasma and Pap samples, are limited and vary greatly between patients. Therefore, a method that relies less on the amount of DNA input will allow for an improved limit of detection in low abundance DNA samples. Of note, IBSAFE was able to perform in samples with as little as 0.17 ng of DNA input. However, due to the small number of samples tested no significance comparisons between the two ddPCR methods could be performed. Furthermore, a direct comparison of MAFs is complicated due to differences between the methodologies. Such a comparison would require a larger study containing also healthy controls and is beyond the scope of this proof-of-principle study.
Notably, we were able to detect a tumour corresponding mutation in an archival sample collected 20 months before diagnosis from a patient subsequently diagnosed with a stage IIIB tumour, while we did not detect any mutations in the two other samples collected at time points earlier than 20 months prior to diagnosis. Although the archival samples in our study were obtained within the suggested median time frame proposed by Labidi-Galy et al. (2017) and IBSAFE assays were successful in all the archival samples, we did not detect any TP53 mutation in the remaining samples from the other women also obtained more than 20 months prior to diagnosis. We therefore speculate that the time-frame to detect HGSOC-derived mutations in Pap samples may be narrower. One explanation might be that although TP53 precursor lesions have been suggested to be present approximately seven years prior to overt HGSOC, these precursors might not shed cells or DNA until later in the tumorigenic process. Importantly however, a 20-month window for early detection may confer a better prognosis for the patient, as suggested by the diagnosis in this case being stage IIIB HGSOC. Unfortunately, the limited number of patients in the present study precluded the possibility of exploring this further, and a larger study, with multiple archival samples per patient, collected at time-points closer to the time of diagnosis would be required to address this.
Although sequencing costs have been reduced during the last decade and population-wide screens for rare disorders have been suggested, deep sequencing is still not feasible for population-wide screens of genetically complex somatic diseases like cancer 27,28 .
Digital droplet PCR offers a clinically feasible platform, with ease of use, fast turnover and high sensitivity 21 . Although the assay sensitivity is high, one shortcoming of ddPCR is the limited number of mutations that can be detected in a single low abundance DNA sample. For ddPCR to be applicable in a clinical setting, attention should be given to collecting as much DNA as possible from liquid biopsies such as plasma or Pap samples. One such approach was illustrated by Wang et al., who evaluated a Tao brush for sampling in close proximity to the tumour 17 , which was found to improve the limit of detection. Other methods, such as a lavage of the uterine cavity have also been reported to improve detection of ovarian cancer 16 . However, for a diagnostic test to be www.nature.com/scientificreports www.nature.com/scientificreports/ successfully implemented in a screening setting, it is crucial that the sampling procedure imposes minimal stress to the patient.
Another limitation of conventional ddPCR is the ability to detect only a single mutation per assay. However, the emergence of multiplexing of ddPCR assays may allow for several mutations to be screened simultaneously 29 . This has recently proved efficient when genotyping KRAS mutations in non-small cell lung cancer 30 , and could possibly increase the utility of TP53 screening even further, as mutations span the entire TP53 gene 31 .
Although the size of the cohort limited the power of the present study, we were able to analyse both diagnostic and archival (pre-diagnostic) samples successfully despite low amounts of DNA. We were able to detect true somatic tumour corresponding mutations in diagnostic samples from six of eight patients. Of note, detection of low abundance TP53 mutant DNA in Pap samples from two patients with stage IIA HGSOC emphasises the potential of the method for early detection. These findings may also support the use of such a sensitive method in the recurrent setting, where patients can be monitored for treatment response on a regular basis based on a known TP53 mutation. Furthermore, we successfully detected a tumour-derived mutation in an archival sample collected 20 months prior to diagnosis from a non-symptomatic woman, and the IBSAFE ddPCR assay performed well in all archival samples. To our knowledge, this is the first time an ovarian cancer-derived mutation has successfully been identified in a pre-diagnostic Pap sample from a non-symptomatic woman.
Although the present study is based on a small number of patients, we believe that an ultra-sensitive ddPCR method should be evaluated in a larger cohort of patients with a greater number of serial pre-diagnostic samples collected prior to the time of diagnosis to provide a better resolution of the diagnostic potential of TP53 testing in Pap samples for the pre-symptomatic detection of HGSOC.

patients.
We identified 79 women with HGSOC from an on-going study prospectively recruiting women with a suspected adnexal tumour at the Department of Obstetrics and Gynaecology in Lund at Skåne University Hospital in the southern Swedish healthcare region. Among these, liquid-based archival Pap samples collected before the ovarian cancer diagnosis were available from nine women from the cervical cancer screening program within the region of Malmö, Sweden at the Department of Medical Microbiology, Lund University Hospital, Sweden (cohort 1). Further, we identified an additional six women from whom matched liquid-based archival Pap samples and tumour tissue were available (cohort 2). All tumours were classified according to WHO 2014 32 and staged according to the International Federation of Gynaecology and Obstetrics (FIGO) criteria 33 . The study was approved by the Ethics Committee at Lund University (Sweden) and was performed in accordance with the Declaration of Helsinki. The patients provided written informed consent.

Sequencing of tumours.
Tumour DNA was extracted using the AllPrep DNA/RNA Mini kit and blood DNA was extracted using the QIAmp DNA Blood Midi kit, both from Qiagen (Sollentuna, Sweden). Formalin-fixed paraffin embedded DNA was extracted using the QIAamp DNA FFPE Tissue Kit from Qiagen.
Paired tumour/blood samples were sequenced using the INVIEW Oncopanel All-in-one (Supplementary  Table S1) and analysed using the in-house analysis pipeline at GATC (Ebersberg, Germany). Samples were aligned against the human reference genome hg19 (chromosomes only, UCSC) with Burrows-Wheeler Aligner (version 0.7.15). Local realignment was carried out using the Genome Analysis Tool Kit (GATK, version 3.7), and duplicate reads were removed using Picard (version 1.131). For one patient lacking paired normal DNA, a pooled reference genome was constructed from seven patients with available blood samples. One patient had two tumour samples, one from each ovary. Bam files were analysed using the Bcbio-nextgen pipeline (version 1.1.0) for paired tumour samples 34 , with Mutect2 from GATK 35 , Freebayes 36 , VarDict 37 and Varscan2 38 as mutation callers. Mutations were annotated using the Variant Effect Predictor 39 and classified using the GEMINI framework 40 to filter out possible germline mutations.
DnA extraction from liquid-based pap samples. Diagnostic samples were collected using a ThinPrep (Hologic Inc., Sollentuna, Sweden) brush at time of diagnosis and were kept in DNAgard (Sigma Stockholm, Sweden). Archival samples were collected using the BD SurePath liquid-based Pap test (Becton Dickson, Stockholm, Sweden) until 2014, after which ThinPrep was used. Following pathology review, residual materials were transferred to new tubes and centrifuged. Residual liquid was removed and the cell pellets were stored at −80 °C. Each cell pellet was resolved in 420 µl Specimen Transport Medium buffer (STM) (Qiagen, Hilden, Germany) and used for extraction of DNA with the QIAmp DNA Mini kit (Qiagen), using the manufacturer's instructions but with two additional washes with each wash buffer. DNA was quantified using Qubit HS DNA kit (Thermo Fisher, Göteborg, Sweden).
Droplet Digital pcR of pap samples. All chemicals, primers and equipment were purchased from Bio-Rad (Solna, Sweden) and used according to the manufacturer's instructions, unless otherwise stated. Primers for mutations were designed using Bio-Rad's online tool (San Diego, CA, USA) (Supplementary Table S2).
Droplet digital PCR of diagnostic samples was performed using a QX100 ™ Droplet Digital PCR system (Bio-Rad) according to the manufacturer's instructions. Briefly, a 22 µl PCR reaction was prepared for each sample using 100 ng of diagnostic sample DNA in a final concentration of 1x ddPCR Supermix with no dUTP, primers (450 nM), probes (250 nM), and restriction enzyme (HaeIII or MseI (5U), Thermo Fisher). Subsequently droplets were generated and transferred to a 96-well PCR plate (VWR, Spånga, Sweden). The plate was heat-sealed with pierceable foil (VWR), and amplification performed using a C1000 Touch deep-well thermal cycler (Bio-Rad). The cycling conditions were as follows: an initial denaturation cycle of 10 min at 95 °C, followed by 40 cycles of denaturation for 30 s at 94 °C, annealing for 60 s at 55 °C (ramping rate set to 2 °C/s), and a final incubation for www.nature.com/scientificreports www.nature.com/scientificreports/ 10 min at 98 °C, ending at 4 °C. The plate was transferred to the QX100 droplet reader and analysed using the automated settings of the QuantaSoft analysis software (Bio-Rad, Version 1.4.0.99). Patient-matched tumour DNA was used as a positive control and Human Genomic Female DNA (Promega, Nacka, Sweden) was used as a negative control.
Detection of TP53 mutations in archival samples using iBSAfe. Diagnostic and archival samples were analysed for their respective tumour TP53 mutation using IBSAFE (SAGA Diagnostics, Lund, Sweden). IBSAFE utilises ddPCR droplets together with a proprietary methodology that allows for ultra-sensitive detection of mutations to a lower limit of detection of ~0.001% MAF. 120 ng of diagnostic sample DNA and varying amounts of archival sample DNA (0.17-206.14 ng) were analysed using IBSAFE by SAGA Diagnostics. IBSAFE reactions were performed in duplicate or quadruplicate. Patient specific tumour DNA (positive control) as well as normal Human Genomic DNA (Promega) (negative control) samples were included in every run to confirm assay performance: for all IBSAFE assays, zero false positive signals were present in the negative control analyses of at least 80,000 normal haploid human genome copies.
Statistics. Statistical tests were performed in R (version 3.3.3) using a two-sided Mann-Whitney U test with a significance threshold of 0.05.

Data availability
All data, materials and results are kept at the Division of Oncology and Pathology, Department of Clinical Sciences, Skåne University Hospital, Lund University, Lund, Sweden and can be made available upon reasonable request to the corresponding author.