Early Detection of High Grade Ovarian Cancer Using Uterine Lavage EHUD Study and Duplex Sequencing

Overview

In Phase I the sponsor will systematically test conditions for lavage filtration that increase tumor cell fraction without reducing tumor mutation yield. The Sponsor will also transition all lavages to luteal phase timing, when endometrial shedding is least. In Phase II the Sponsor will examine our data in context of clinical characteristics, particularly age, to develop a multivariate model that determines optimal mutant allele frequency (MAF) diagnostic threshold by patient. Furthermore, the sponsor will explore a highly innovative idea, entailing empirically determining each individual's background mutation load, agnostic of the aging or mutagenic exposures responsible, and using this as a personalized calibrator to determine optimal MAF diagnostic threshold.

APPROACH PHASE I The University of Washington has demonstrated diagnostic proof of principle; however, further optimization and validation is needed for industrial scale deployment. In Phase I the UtL filtration method will be optimized (Aim 1), analytical performance on the commercial workflow validated (Aim 2) and a sample collection milestone met (Aim 3). Aim 1. Determine utility of uterine lavage filtration to maximize enrichment for tumor derived DNA. Preliminary testing with two samples indicated that filtration of UtLs to remove clusters of endometrial cells could increase sensitivity for tumor mutations several folds. This data set will be expanded by analysing the filtration effect in 10 UtLs from patients with HGSC with known TP53 mutations. Each UtL will be divided in half: the first half will be filtered and the second will remain unfiltered. DNA will be extracted from the filtered, filtrate, and unfiltered fractions and analysed by TP53 DS (total of 30 samples). For each UtL, the DNA yield in each of the fractions and the MAF of the tumor mutation will be compared. It is anticipated that the filtered fraction will contain less DNA then the unfiltered fraction but the tumor mutation will be present at a higher frequency. If sensitivity is improved (i.e. more is gained by enrichment than is lost by reduction from less available DNA), filtration will be used on all future samples collected. Aim 2. Validation of optimized assay: accuracy, precision, limit of detection, and reproducibility. TwinStrand has developed a streamlined DS workflow using improved adapters, ligation chemistry and a high throughput 96 well plate-based format amenable to liquid handling robots and compatible with CLIA laboratory standards. TwinStrand has also developed an optimized cloud-based analysis pipeline that supports automated parallel processing of multiple samples. The optimized process for the detection of TP53 mutations in UtL samples will be validated. To assess accuracy, technical precision, and lower limit of detection, DNA samples from two individuals that differ in genotype at \>5 SNP sites in or near TP53 will be mixed, in ratios from 1:100 1:5,000, two UtL. The mixtures will be sequenced at \~10,000x molecular depth in two independent experiments. SNP allele fraction (AF) will be compared at different dilutions to determine accuracy (expected AF vs. observed AF), precision (AF variation among replicates), and lowest limit of detection achievable. To test assay reproducibility in its intended clinical use, sequencing will be repeated on the 30 samples used Aim 1, which will encompass a wide range of tumor MAFs. The Sponsor will calculate the coefficient of variation among replicates. Based on pilot studies The Sponsor anticipate excellent reproducibility (CV\<5%). Aim 3. Sample collection. Collection of samples proposed in Phase II will be started under appropriate IRB/Ethic Committee approval at each participating institution. After collection, samples are shipped to the Institute for Cancer Research at the Medical University of Vienna who will perform UtL filtration and DNA extraction. Paired DNA from Pap smears and peripheral leukocytes will also be extracted. Isolated DNAs will be assigned a de identification number and shipped to TwinStrand for Duplex Sequencing. Phase I Milestones: 1. Confirm utility of UtL filtration on tumor mutation enrichment. 2. Quantify streamlined assay's accuracy, precision, lower limit of detection, and reproducibility on UtLs. 3. Start samples collection for Phase II. Phase I Products: 1. DS workflow ready for application to UtL DNA at commercial scale in Phase II and Phase III. 2. Sample bank for Phase II APPROACH PHASE II Phase II will validate the use of TP53 DS on UtLs for early ovarian cancer detection in expanded case control patient cohorts encompassing both average and high risk populations. Predefined parameters that may affect the sensitivity and specificity of TP53 mutation frequency will be examine, and statistically modeled. Clinical sensitivity and specificity will be maximized by building personalized diagnostic threshold statistical models using multivariate clinical characteristics, as well as each individual's unique background mutation load based on leukocyte sequencing. In a subset of patients, the superior performance of UtL over Pap smear sampling will be confirmed. In addition, a proof of concept study will be conducted in a sub-set of included patients to study a set of 96 methylation markers relevant to HGSC in UtLs and corresponding tumor tissue/STICs. AIM 1. GENERAL POPULATION OVARIAN CANCER SCREENING This aim will develop a biomarker for HGSC detection in an average risk population. The Sponsor will conduct a case control study that integrates TP53 DS mutational data with clinic pathological information in order to identify women with HGSC with maximum sensitivity and specificity. More than 98% of HGSCs carry mutations in TP53, which means that ultra-deep DS can be cost effectively focused on just a small genomic region. The product of this aim will be a 200 patient biomarker data set that demonstrates the cost effective, commercially robust performance of this critically needed ovarian cancer diagnostic. Aim 1A: To assess the test performance of TP53 DS on UtLs for HGSC detection in average risk patients. Methods: UtL samples from 200 subjects from average risk population will be analyzed via DS: 100 subjects with HGSC (cases) and 100 subjects with lesions that were ultimately found to be benign after resection, i.e., without cancer (controls). In all patients UtL will be collected prior surgical intervention for an ovarian mass. Lavage will be carried out during the luteal phase of the menstrual cycle if pre-menopausal. Surgical specimens will be pathologically assessed, and a patient will be counted a HGSC positive or a control depending on the histological results. For cancer patients, the primary tumor will be sequenced via conventional methods by the Medical University of Vienna for a panel of genes that includes TP53, BRCA1 and BRCA2. For all de-identified patient samples, collected clinical information will include: age, smoking history, prior chemotherapy exposure, parity, age-of-menopause, age-of-menarche, history of oral contraceptive use, cancer family history, and seven-gene mutation status of the primary tumor. Cases and controls will be aged matched and as wide an age range as possible will be included to assess the effect of age on sensitivity and specificity. Statistical Analysis: A detailed mutation profile for each sample will be generated from the duplex sequencing output files including: mutant allele frequency (MAF) for all mutated positions, mutation spectrum, predicted pathogenicity to protein function, relationship to known hotspots, and overall mutation load (number of mutant nucleotides divided by the total number of nucleotides sequenced). Following this, samples will be unblinded for cases-controls status. TP53 MAF from DNA collected by UtLs will be used as predictor for differentiating between average risk patients (AIM I) with and without HGSC by logistic regression modelling. Age, smoking history, prior chemotherapy exposure, parity, age-of-menopause, age-of-menarche, history of oral contraceptive use, cancer family history and germline mutation status will be considered as potential confounders. Model prediction will be assessed by cross-validation. An analogous analysis will be applied to the group of high risk patients (AIM II) where presence and absence of STIC is defining the outcome variable. Cut off values for mutant allele frequency will be suggested in both cases and specificity and sensitivity will be estimated including appropriate confidence intervals. Aim 1B: Improving diagnostic performance with personalized calibration by background mutation load. The unprecedented sensitivity of DS led us then others to the novel discovery that cancer like mutations accumulate at very low levels with age in multiple human tissues. In a pilot study the sponsor discovered that the average BB mutation load is somewhat higher in UtLs than other tissues, potentially because of DNA contribution from endometrial tissue, which replicates extensively prior to menopause. While this did not compromise specificity in the pilot study, the sponsor recognize that BB could pose an obstacle with some very early tumors where MAF signal is low, or with very elderly women where background is high. The Sponsor expect that lavage filtration and luteal phase collection in premenopausal women will reduce BB signal, but as an added measure the sponsor will examine whether performance can be further improved by normalizing for an individual's background mutation load. The sponsor hypothesize that the level of TP53 mutations in circulating leukocytes can serve as an empirically measured personal calibrator that captures, not only known factors that increase BB, such as age, but also unknown mutagenic exposures or other factors occurring during a person's life. Although DNA from leukocytes probably only contributes a minimal amount to the total pool of BB mutations in UtLs, the sponsor posit that they may serve as a proverbial "canary in a coal mine" that will be proportionally representative of BB mutations elsewhere in the body. The BB mutation load in a lavage, itself, cannot be directly measured in a real world setting when the mutation(s) contributed by a tumor are unknown. The plausibility of this concept was shown by our initial study of TP53 mutations in peritoneal fluid, which included a subset of matching blood samples and indicated a strong association between BB and age. Methods: DNA from the peripheral blood mononuclear cells (PBMC) component - collected immediately prior to a surgery - of randomly chosen 50 from 100 HGSC cases and 50 from 100 controls from Aim 1A will be extracted and subjected to TP53 DS at \~10,000x molecular depth. TP53 mutations from PBMCs will be subtracted from TP53 mutations found in the UtL. Statistical Analysis: The Sponsor will examine the association between TP53 mutation frequencies in UtL and leukocytes and will determine whether false negatives in Aim 1A correspond to cases with increased BB in leukocytes. In addition, the sponsor will analyze TP53 mutation frequencies in leukocytes as a predictor of case control status, again using the leave out 10% procedure. This innovative calibration method will be compared with the simple ROC metrics achieved with the univariate model using a fixed mutation fraction as well as the adjusted multivariate model developed based on other patient characteristics such as age. As a scientific question unrelated to the present aims, the sponsor are eager to see whether leukocyte TP53 mutation load will serve as an independent predictor of patient age, overall health or history of mutagenic exposures. Aim 1C. Comparison of diagnostic performance of uterine lavage vs. Pap smear collected DNA. The first report of using NGS for ovarian cancer detection from a trans vaginal liquid biopsy relied on Pap smear collection and SafeSeqS as the sequencing technology. The cited sensitivity of 41% provided an important proof of principle, but also ample room for improvement. The sponsor have definitively demonstrated the superior accuracy of DS over SafeSeqS like methods and ongoing studies in our lab indicate limited sensitivity of Pap smears compared to UtL. However, a direct comparison of the two collection methods on the same patients remains to be performed. Formally demonstrating the degree of superiority of UtL over Pap smears will help with commercial market entry (the sponsor note that a related NCI SBIR has been awarded to PapGene Inc, a company founded by the above-mentioned study's authors). The superiority of UtL is expected based on the fact that the lavages reach the fallopian tubes and ovarian surfaces whereas Pap smears rely on disseminated cancer cells reaching the cervical canal. Nevertheless, in the event that TP53 DS on Pap smears does not drastically underperform UtL, the sponsor would further explore this complementary collection method. Because Pap smears are routinely performed by primary care providers, the ability to use these could help accelerate market adoption. However, current evidence points to the probable inferiority of Pap smears. Methods: The sponsor will carry out TP53 DS as above but using DNA from Pap smears collected prior to surgery of randomly selected, age matched subset of 25 cancer cases and 25 controls from Aim 1A. Statistical Analysis: Data on TP53 MAF from DNA collected by conventional pap smear tests from a subsample of average risk patients will be considered as additional predictor in models developed within aim 1A and 1B.The sponsor will investigate if any additional power to discriminate cases from controls can be gained by combining information from Pap smears and UtLs. AIM 2. HIGH RISK POPULATION OVARIAN CANCER SCREENING This aim will take a similar approach to that of Aim 1 but will focus on women at high risk for ovarian cancer due to hereditary breast and ovarian cancer (HBOC) mutations. HBOC women will have been identified by a strong family history of breast or ovarian cancer and found to carry mutations in cancer susceptibility genes, usually BRCA1 and BRCA2, which confer a lifetime HGSC risk of 35%, 46% and 13%, 23%, respectively. Standard of care is risk reducing salpingo oophorectomy (RRSO) at an early age, typically after completion of child bearing. Although this approach does decrease mortality, it is imperfect: approximately 10% of ovarian cancers in this population develop before age 40. In the Vienna case series, cancers have been seen at as young an age as 20, which is well before a prophylactic surgery would normally have been undertaken. Thus, even with standard of care, HBOC women in their reproductive years are at a significantly elevated risk of ovarian cancer relative to other women their need for an effective screening tool is even greater. At present, affected women must make the decision to either accept the increased risk of a highly lethal cancer or elect to forgot (or prematurely end) childbearing. This is an excruciatingly challenging decision faced by hundreds of thousands of women in the US alone. In certain ethnic groups, particularly Ashkenazi Jews, the risk of being a carrier is as high as 1 in 40. Yet the majority of BRCA carriers, even those who do not have surgery, will never develop HGSC. There is an urgent need for early detection diagnostic tools in women at high risk of HGCS in order to save lives and preserve the choice of fertility. The product of this aim will be a 115 patient biomarker data set that demonstrates commercially robust performance of such an assay. Aim 2A. To assess the test performance of TP53 DS on UtLs for HGSC detection in high risk patients. The sponsor will approach this study per the general methods of Aim 1A. One significant difference is that UtL will be collected from HBOC women undergoing RRSO, rather than from women with known ovarian masses. After RRSO, the ovaries and fallopian tubes undergo a meticulous sectioning protocol known as SEE-FIM to carefully examine the tissues, particularly the fimbriae, for STICs and occult cancers. Among this high risk population, such lesions are found in roughly 5% of resections. The cancer group will encompass women where STICs or occult tumors were found during SEE-FIM the control group will include women with no lesions. This study design is innovative in that it focuses, not only on high risk women, but specifically on the ability of our assay to detect very early stage, potentially even microscopic, cancers when they remain surgically curable. Methods: UtL samples from up to 115 women with hereditary high risk ovarian cancer syndromes will be analyzed via DS: 15 with STICs or occult cancer (cases) and up to 100 cancer free (controls). Case and control will be matched by age. UtLs will be performed immediately prior to RRSO. Surgical specimens will be pathologically assessed by a centrally standardized SEE-FIM protocol. After RRSO, the ovaries and fallopian tubes undergo a meticulous sectioning protocol known as SEE-FIM to carefully examine the tissues, particularly the fimbriae, for STICs and occult cancers. The sponsor expects to have identified 5 to 15 women with STIC or occult cancers in this cohort, 100 age matched women will have no detected lesions. The smaller number of cancer cases than controls reflects the low percentage of lesion positive surgeries and the sample number the sponsor project realistically being able to achieve. Although a larger number would be optimal, such samples are extremely limited. The cancer group will encompass women where STICs or occult HGSC were found during SEE-FIM the control group will include women with no lesions Statistical analysis: This will be as per Aim 1A. Aim 2B. Diagnostic threshold calibration in high risk women by background mutation load. This sub-aim will use the same protocol and statistical methods of aim 1B and apply them to the high- r i s k population of Aim 2. Methods:Leukocyte DNA from all available cancer cases and a randomly chosen 30-woman subset of the control group will be evaluated. It is well-established that the risk of malignancy in HBOC women is elevated relative to others by virtue of defects in DNA repair that increase the probability of oncogenic mutations arising. As such, the sponsor anticipate that the non-cancer-derived background mutation load measured in UtLs as well as in peripheral blood may be proportionally elevated as well. Adjusting for the high background mutation load measured in blood may be particularly important to increase specificity in this group. Statistical analysis: This will be as per Aim 1B. AIM III To define a methylation signature for HGSC detection in UtLs. Over the last 15 years, the value of DNA methylation analysis for the detection of epigenetic, tumour-specific changes has been demonstrated, particularly in cancer diagnostics. Studies of DNA-methylation in ovarian cancer (OC) focused on detecting DNA fragments shed by OC cells into the bloodstream (i.e. cell-free DNA - cfDNA) suggest that DNA methylation patterns in cfDNA have the potential to detect a proportion of OCs up to two years in advance of diagnosis. This study also clearly shows the limitation of this approach because only 50% of all patients who finally developed high-grade serous ovarian cancer (HGSC) within two years could be detected. The underlying problem is that due to the low concentration of cfDNA in the blood stream a very weak signal needs to be detected. With the aim of potentially increasing the sensitivity of HGSC detection the sponsor will in cooperation with Prof. Andreas Weinhäusel, Austrian Institute of Technology (AIT), Competence Unit Molecular Diagnostics, perform a proof of concept study in a sub-set of patients included in this project. Prof. Andreas Weinhäusel, identified 96 methylation markers relevant to HGSC. Only a very small amount of DNA is required for DNA methylation. It is therefore advisable to use the already available DNA from the lavages and the corresponding tumor tissue, which has been extracted in the course of this study, for the further development of the specificity and sensitivity of the test procedure. The product of AIM3 will be a 140 patient methylation markers data set that demonstrates high sensitive performance of such an assay. Methods: Genomic DNA from UtLs and corresponding tumor tissue from 30 HGSC patients and 10 STICs/occult cancers will get enriched for methylated DNA with the aid of methylation sensitive restriction enzymes. This DNA will be analyzed by performing a highthrough put-qPCR in which 5 x 96 marker x 96 DNA will be analyzed in parallel. The most informative markers will be combined to a methylation signature. To proof the specificity of the signature DNA from UtLs from 60 controls will be analyzed (30 average risk and 30 high risk cases). Assessment of DNA methylation pattern will be carried out on the pre-resection lavages and on up to 10 STIC lesions/occult cancers. When local pathology results identify STICs or occult cancer cells (FFPE tissue), laser microdissection and DNA isolation will be performed. Subsequently applying standard NGS TP53 sequencing will be performed as described in AIM 1 and AIM 2. In this pilot sub-study, high-through put-qPCR will be applied to remaining DNA material for assessment of methylation status. Statistical analysis: Aiming to evaluate the potential for defining a DNA-Methylation signature for simple PCR testing, dCT-PCR values from 96-plexed high-throughput MSREqPCR analysis will be analysed by bioinformatics \& biostatistics to conduct 1) class comparison between the relevant clinical subtypes and classes for defining significantly differentially methylated genes; 2) multivariate class prediction analysis using different feature selection approaches based on single-marker p-values. Different classification algorithms (e.g. k-nearest neighbor, support vector machines, linear discrimination analyses etc.) will be applied, using 10-fold cross validation and/or leave-one-out cross validation for selecting best candidate methylation markers - and algorithms; in addition, the top-AUC values of single candidate methylation-marker's will be defined in ROC analysis and considered for marker selection. A subset of 48 candidate markers will be selected for confirmation in a separate sample set. The methylation data derived thereof will be analyzed using class comparison and class prediction analysis in a similar manner as the first set of data derived from 480plexed analysis. Best performing single markers and combinations of markers from multivariate analysis will be defined. Classification results and methylation data will be compared with p53 mutation test-results to evaluate potential of methylation-based diagnostic classification, as a sole and in combination with p53 mutation analysis. This will be conducted applying different class-prediction approaches integrating both p53 and methylation data in multivariate models.