Clinical implications of plasma circulating tumor DNA in gynecologic cancer patients

In gynecologic cancer patients, therapy matched to ctDNA alterations (N = 33 patients) was independently associated with improved overall survival (hazard ratio: 0.34, P = 0.007) compared to unmatched therapy (N = 28 patients) in multivariate analysis. Tissue and ctDNA genomic results showed high concordance unaffected by temporal or spatial factors. ctDNA may be an important tool to individualize cancer therapy in patients with gynecologic cancer.

Molecular characterization of cancers is important in dictating prognostic factors and directing therapy. Next-generation sequencing of plasma circulating tumor DNA (ctDNA) offers less invasive, more convenient collection, and a more real-time representation of a tumor and its molecular heterogeneity than tissue. However, little is known about the clinical implications of ctDNA assessment in gynecologic cancer. We describe the molecular landscape identified on ctDNA, ctDNA concordance with tissuebased analysis, and factors associated with overall survival (OS) in gynecologic cancer patients with ctDNA analysis. We reviewed clinicopathologic and genomic information for 105 consecutive gynecologic cancer patients with ctDNA analysis, including 78 with tissue-based sequencing, enrolled in the Profile-Related Evidence Determining Individualized Cancer Therapy (NCT02478931) trial at the University of California San Diego Moores Cancer Center starting July 2014. Tumors included ovarian (47.6%), uterine (35.2%), cervical (12.4%), vulvovaginal (2.9%), and unknown gynecologic primary (1.9%). Most ovarian and uterine cancers (86%) were high grade. 34% (N = 17) of ovarian cancers had BRCA alterations, and 22% (N = 11) were platinum sensitive. Patients received median 2 (range 0-13) lines of therapy prior to ctDNA collection. Most (75.2%) had at least one characterized alteration on ctDNA analysis, and the majority had unique genomic profiles on ctDNA. Most common alterations were TP53 (N = 59, 56.2% of patients), PIK3CA (N = 26, 24.8%), KRAS (N = 14, 13.3%), BRAF (N = 10, 9.5%), ERBB2 (N = 8, 7.6%), and MYC (N = 8, 7.6%). Higher ctDNA maximum mutation allele frequency was associated with worse OS [hazard ratio (HR): 1.91, P = 0.03], while therapy matched to ctDNA alterations (N = 33 patients) was independently associated with improved OS (HR: 0.34, P = 0.007) compared to unmatched therapy (N = 28 patients) in multivariate analysis. Tissue and ctDNA genomic results showed high concordance unaffected by temporal or spatial factors.
This study provides evidence for the utility of ctDNA in determining outcome and individualizing cancer therapy in patients with gynecologic cancer.

Background
In 2019, approximately 109 000 women were diagnosed with gynecologic cancer, and approximately 33 100 women died from their disease [1]. The major gynecologic cancers are ovarian, uterine, and cervical cancer, with rarer cases of vulvar and vaginal cancer. Gynecologic cancers are treated with surgery, radiation therapy, chemotherapy, or a combination of these modalities based on stage, histologic risk factors, and other tumor or patient-specific risk factors. Despite attempts to advance therapy, recurrences and treatment failures are common, particularly for those patients diagnosed with advanced stage disease.
It is now clear that cancers are driven by specific genomic abnormalities, many of which can be targeted using existing therapies [2]. Schwaederle et al. [3] examined 570 phase II single-agent studies (N = 32 149 patients) and found that patients who received a personalized, biomarker-directed therapy had significantly improved outcomes and fewer deaths related to treatment toxicity. Molecular signatures have recently predicted therapeutic response in patients with gynecologic malignancies. Breast cancer susceptibility gene (BRCA) 1/2 mutated and homologous recombination-deficient epithelial ovarian cancers have demonstrated dramatic responses to PARP inhibitors, HER2-positive uterine serous cancers benefit from the incorporation of trastuzumab, and mismatch repair deficient or microsatellite unstable high tumors benefit from immune checkpoint inhibition with guidelines recommending MSI testing in ovarian, cervical, and vulvar cancers [4][5][6]. Additionally, the FDA recently approved larotrectinib as its second tumor-agnostic approval based on tumor molecular genetics; of note, most patients in the study had cancer types where the frequency of TRK-fusion was less than one percent, highlighting the importance of considering molecular tumor analysis broadly [7,8].
Molecular analysis of tumor tissue is increasingly being incorporated in the development of treatment regimens for patients with solid malignancies. However, patients are often incompletely tested, or undergenotyped, for guideline-recommended targetable mutations. For tumor tissue-based testing, adequate samples are not always available or accessible for analysis. Even when available, tissue obtained at the time of primary surgical resection or biopsy may not reflect the current tumor molecular makeup or adequately capture tumor heterogeneity. Plasma-derived circulating tumor DNA (ctDNA) offers a more convenient, less invasive, and real-time option to analyze tumor for potentially actionable mutations. In an effort to better understand the value of ctDNA in the management of gynecologic cancer, we describe the molecular landscape identified on ctDNA analysis, determine concordance between ctDNA and tissuebased analysis, and identify factors associated with survival in this cohort of gynecologic cancer patients.

Study patients
We reviewed the clinicopathologic and genomic information for 105 consecutive gynecologic cancer patients with ctDNA analysis who were enrolled in the Profile-Related Evidence Determining Individualized Cancer Therapy (PREDICT, NCT02478931) trial at the University of California San Diego (UCSD) Moores Cancer Center starting July 2014. All investigations followed UCSD Internal Review Board guidelines, and consent was obtained for investigational therapies or procedures [9]. The study methodologies conformed to the standards set by the Declaration of Helsinki.

Circulating tumor DNA sequencing
All blood samples for ctDNA were evaluated at Guardant Health, Inc (Redwood City, CA, USA), a Clinical Laboratory Improvement Amendments (CLIA)certified and College of American Pathologists (CAP)accredited clinical laboratory. The assay sequences cancer-associated somatic mutations in ctDNA. The panel initially included 54 genes in 2015, and it has been expanded to include 73 genes (Table S1) [10].
To assess concordance between plasma ctDNA and solid tissue biopsies, we compared frequencies of alterations in the subset of patients who had both ctDNA and tissue sequencing. All tissue DNA analysis was performed by Foundation Medicine, Inc (Cambridge, MA, USA), a CLIA-licensed and CAP-accredited clinical laboratory. All tissue samples were collected between July 2011 and July 2018, either at primary surgery (N = 31) or at recurrence (N = 47). The assay analyzed up to 324 genes [10,11].

Outcome definitions and statistical method
Patients were described by primary disease site, histology, smoking status, body mass index (BMI), ethnicity, and number of lines of chemotherapy prior to ctDNA analysis. Categorical variables and continuous variables were compared with Fisher's exact tests and Mann-Whitney U-tests, respectively. We included only characterized genetic alterations, excluding variants of unknown significance (VUS) and synonymous alterations. Number of ctDNA alterations and percentage of ctDNA were reported. If more than one ctDNA sample was available, we used the first sample collected. Each sample was categorized as actionable by UCSD PREDICT criteria or OncoKB criteria [12,13]. Each patient's primary oncologist dictated which therapy a patient received. Rates of matching were reported by ctDNA and by tissue biopsy alone. Analysis was performed on patients who were prospectively or retrospectively matched. If patients were not matched, reasons for not matching were reported. Frequency and type [single nucleotide variant (SNV), amplification, or deletion, which included frameshift mutations, deletions, and insertion/deletions] of genetic alterations were then reported in the cohort of all gynecologic cancer patients and in the ovarian, uterine, and cervical/vaginal/vulvar cancer cohorts. If patients had multiple genetic alterations of the same type in the same gene, it was counted once. However, if, for example, patients had a PIK3CA SNV and PIK3CA amplification, each was counted once. All data were abstracted from patients' medical records by two independent investigators.
Genomic alteration concordance between ctDNA and tissue was determined using concordance rate and Kappa value with standard error (SE) for the three most commonly altered genes. Kappa value can range from 0 (rate of agreement expected by chance alone) to 1 (perfect agreement). Patients were stratified by time interval from tissue biopsy to ctDNA blood draw (≤ 6 months vs > 6 months) and tissue biopsy site (primary tumor vs metastatic site). Fisher's exact test was used to compare concordance rates.
Overall survival (OS) was determined from date of blood draw for first ctDNA to date of death or last follow-up. Patients still alive at last follow-up were censored on that date. Univariate analysis was performed to calculate hazard ratios (HR) for age, BMI, site of primary tumor, TP53 alteration, PIK3CA alteration, median maximum ctDNA mutation allele frequency (MAF), number of characterized alterations, and number of lines of chemotherapy prior to first ctDNA collection; in general, cohorts were divided at the median of each value. All variables with P < 0.10 were included in the multivariate analysis. A second survival analysis was calculated in all patients who received matched treatment to ctDNA or unmatched treatment by either ctDNA or tissue, excluding patients who did not receive treatment or who received treatment matched only by tissue-based molecular testing. This survival was determined from date of first matched treatment or date of unmatched treatment to date of death or last follow-up. We added number of lines of chemotherapy prior to treatment to account for those patients who received matched treatment in greater than one subsequent line after ctDNA analysis.
In the 105 ctDNA samples, there were 217 unique genomic alterations. Gynecologic cancer patients most commonly had TP53 (N = 59, 56.2%), PIK3CA  Overall, 79 patients (75.2%) had ctDNA alterations that were potentially targetable by UCSD PREDICT criteria, while 55 women (52.4%) had ctDNA alterations that were actionable by OncoKB criteria [12,13].  (Table 4). Concordance was not significantly correlated with location of tissue biopsy (primary vs metastatic site) or time interval between blood draw and tissue biopsy (Table 4). If no ctDNA detected or no actionable alteration detected on ctDNA, this was coded as the primary reason for not matching by ctDNA. However, 12 of these patients were matched by alterations on tissue-based molecular profiling, for a total of 24 patients matching by tissue molecular profiling alone.

Discussion
Gynecologic malignancies diagnosed at an advanced stage often recur and are difficult to treat. Although prognosis varies by primary disease site, recurrent gynecologic cancer is generally incurable and treatment options exhibit modest efficacy with accompanying toxicity. The molecular characterization of gynecologic malignancies has emerged as an area of active interest; however, the utility of ctDNA to guide treatment in gynecologic cancer and its correlation with clinical data have been limited [14,15]. We found that 75.2% of gynecologic cancer patients had ≥ 1 genomic alteration on ctDNA assessment (Table 1). TP53 alterations were seen in over 50% of patients, and PIK3CA alterations were seen in nearly 25% of patients. These numbers are similar to a recent publication of 2579 'pan-gynecologic cancer' patients (1087 breast cancers, 579 ovarian cancers, 548 endometrial cancers, 308 cervical cancers, and 57 uterine carcinosarcomas) in The Cancer Genome Atlas; they reported TP53 and PIK3CA alteration rates of 44% and 32%, respectively [14]. However, their cohort had higher rates of PTEN (20% vs 5.7%) and ARID1A (14% vs 5.7%) alterations than our cohort.
We considered 100% of the characterized alterations to be targetable by FDA-approved agents or therapies in development, while 71% of characterized alterations were considered targetable by OncoKB criteria [12,13]. This discrepancy is likely explained by UCSD  PREDICT criteria defining TP53 as targetable using antiangiogenic agents based on prior data, while OncoKB has not defined this relationship; multiple studies have now demonstrated that TP53 is a marker for increased VEGF expression and improved response to antiangiogenic agents [16][17][18][19][20]. This percentage is higher than that described in a prior report of 211 gynecologic cancer patients, where 48% had at least one actionable alteration or a recent study of 78 highgrade serous ovarian cancer ctDNA samples that    showed 58% had at least one actionable [21]. Regardless, these observations suggest that many gynecologic cancer patients may be candidates for matched treatment [9,[21][22][23][24][25][26]. Furthermore, these data indicate discerning druggable alterations can be achieved through ctDNA analysis, which is less invasive, more convenient, and may afford more contemporaneous samples than tissue biopsy. Similar to reports of ctDNA examination in other cancer patients, our gynecologic cohort mostly had unique genomic portfolios in ctDNA, emphasizing the opportunity for individualized therapy [27][28][29][30].
The overall concordance rate of genomic alterations between tissue and ctDNA was 75.6-88.5% for TP53, PIK3CA, and KRAS. Concordance rates were not significantly related to location of biopsy (primary vs metastatic site) or time interval between blood draw and tissue biopsy (Table 4). These concordance rates provide some reassurance for reliability of ctDNA in place of tissue biopsy; however, tissue biopsy may add more actionable targets than ctDNA alone, as tissuebased NGS panels often comprise a much larger targeted set of genes.
Similar to prior studies in a variety of nongynecologic cancers, we found that higher percentage of ctDNA was correlated with worse survival [27,[30][31][32][33]. ctDNA has recently been associated with increased risk of recurrence in colorectal cancer and poorer outcomes in advanced non-small-cell lung carcinoma, breast cancer, and ovarian cancer [34][35][36][37]. In 44 patients with ovarian or uterine serous cancers who completed frontline therapy, Pereira et al. showed that ctDNA can be used as a biomarker to predict disease persistence and recurrence and was associated with OS [38]. Somewhat surprisingly, in our cohort, younger age was significantly associated with poorer OS. This may be due to selection bias, as our younger patients may have been more likely to be referred to our Table 4. Overall concordance between ctDNA and tissue-based DNA by tissue biopsy site (primary or metastatic) and time interval between blood draw and tissue biopsy (N = 78). Kappa value can range from 0 (rate of agreement expected by chance alone) to 1 (perfect agreement), with a higher kappa value correlating to a better concordance. precision medicine program or offered ctDNA despite poorer performance status or more advanced malignancies. Alternatively, being a tertiary care center, it is conceivable that there is a referral bias for young patients with more aggressive disease. Matched therapy has previously been shown to have great promise in cancer therapy [3,24,28,29,39,40]. We demonstrate that matched therapy by ctDNA was associated with significant improvement in OS in univariate and multivariate analysis, with 20.0-month median OS in the matched cohort compared to 5.3 months in the patients who received unmatched treatment. This demonstrates that ctDNA may be used to direct therapy to improve OS in gynecologic cancer. However, because patients were treated with heterogeneous matched and unmatched treatments (Table S3), further study is warranted to definitively conclude that matched therapy to ctDNA improves survival in gynecologic cancer.
This study has some important limitations. For simplicity, we considered only each patient's first ctDNA sample in our analyses. It is possible that patients had subsequent ctDNA analyses that were used for matched therapy; however, these matches were not used in our analysis. Additionally, our cohort is relatively small with a mix of ovarian, uterine, and cervical cancers, and all patients were treated at a single institution; however, in univariate analysis, disease site was not associated with OS, so it was unlikely to be a confounder. Furthermore, the vast majority of our tumors were high grade, and therefore, the impact of grade on our findings could not be elucidated. Similarly, only 22% (N = 11) of the 40 ovarian cancers were platinum sensitive, reflecting this heavily pretreated population. Thirty-four percent (N = 17) of ovarian cancer patients had BRCA alterations detected on germline or somatic tissue or ctDNA testing, while only 4% (N = 2) had BRCA alterations detected on ctDNA (Table 1). Of note, Guardant does not report germline alterations on ctDNA, which is an important limitation of this test, and explains the much of the discrepancy in these numbers. Also, ctDNA may have missed some somatic alterations captured by tissue biopsy due to relatively lower disease burden, including three patients with maximal mean allele frequency < 2.0%. Of note, the BRCA-positive patient with the highest maximum allele frequency (48.1%) did have her BRCA alteration detected on ctDNA. We advocate for further study into the important issue of concordance of BRCA alterations on ctDNA with tissue biopsy and recommend combining ctDNA with germline and/or tissue testing to definitively rule out BRCA alterations, especially given its significant clinical relevance for predicted benefit from PARP inhibitors [4]. The small numbers of patients in each of these subsets rendered it difficult to assess these important variables, which should be evaluated in follow-up studies of larger numbers of patients. Also, not all patients with ctDNA testing also had tissue-based testing, potentially limiting our concordance analyses. Finally, although we found that patients matched to ctDNA had improved OS, we did not examine progression-free survival or response rates due to heterogeneous patient follow-up; it is therefore conceivable that OS could be confounded by subsequent treatment after matched therapy. Despite these limitations, our findings are important in informing the utility of ctDNA in the treatment of gynecologic cancers.
Importantly, financial barriers did not impact access to targeted therapy in this patient cohort, despite diverse socioeconomic backgrounds. This likely reflects implementation of a medication acquisition team as part of our precision medicine program, as well as a robust portfolio of clinical trials and the availability of clinical trial coordinators [28]. As shown in other cancer types, ctDNA has great potential as an important biomarker to predict response to immunotherapy, guide need for adjuvant therapy in the postoperative setting, monitor response to therapy, and predict resistance or recurrences months prior to imaging [31,[41][42][43]. Further study and validation are required for these exciting potential future uses.

Conclusions
Efforts to improve oncologic outcomes and treatment options for patients with gynecologic cancers remain a clinical priority. This study suggests that ctDNA assessment may have both prognostic and therapeutic implications, informing individualized cancer therapy in a cohort of patients with gynecologic cancer. We found that higher ctDNA maximum MAF was associated with worse OS, while therapy matched to ctDNA genomic alterations was independently associated with improved OS compared to unmatched therapy. Tissue and ctDNA genomic results showed high concordance unaffected by temporal or spatial factors. Additional studies are warranted to better define the utility of ctDNA assessment in the management of gynecologic cancer.

Supporting information
Additional supporting information may be found online in the Supporting Information section at the end of the article. Fig. S1. Kaplan Meier survival curves for OS. Table S1. 54-to 73-gene panels (Guardant, Inc.) Table S2. Alterations in gynecologic patients undergoing ctDNA testing (N = 105 patients)*. Table S3. Patient-level DNA alterations and therapy in patients who received treatment after ctDNA (N = 85 patients).