Circulating Cell-Free Tumor DNA Analysis of 50 Genes by Next

Published OnlineFirst January 12, 2016; DOI: 10.1158/1078-0432.CCR-15-2470

Personalized Medicine and Imaging Clinical Cancer Research Circulating Cell-Free Tumor DNA Analysis of 50 Genes by Next-Generation Sequencing in the Prospective MOSCATO Trial Cecile Jovelet1, Ecaterina Ileana1,2, Marie-Cecile Le Deley3,4,5, Nelly Motte1, Silvia Rosellini3, Alfredo Romero1, Celine Lefebvre6, Marion Pedrero6,Noemie Pata-Merci7, Nathalie Droin7, Marc Deloger8, Christophe Massard2, Antoine Hollebecque2, Charles Ferte9, Amelie Boichard1, Sophie Postel-Vinay2,6, Maud Ngo-Camus2, Thierry De Baere10, Philippe Vielh11, Jean-Yves Scoazec1,5,11, Gilles Vassal12, Alexander Eggermont5,9, Fabrice Andre5,6,9, Jean-Charles Soria2,5,6, and Ludovic Lacroix1,6,11,13

Abstract

Purpose: Liquid biopsies based on circulating cell-free DNA 48.4%–61.6%] at the patient level. Among the 220 patients with (cfDNA) analysis are described as surrogate samples for molecular mutations in tDNA, the sensitivity of cfDNA analysis was signif- analysis. We evaluated the concordance between tumor DNA icantly linked to the number of metastatic sites, albumin level, (tDNA) and cfDNA analysis on a large cohort of patients with tumor type, and number of lines of treatment. A sensitivity advanced or metastatic solid tumor, eligible for phase I trial and prediction score could be derived from clinical parameters. Sen- with good performance status, enrolled in MOSCATO 01 trial sitivity is 83% in patients with a high score (8). In addition, we (clinical trial NCT01566019). analyzed cfDNA for 51 patients without available tissue sample. Experimental Design: Blood samples were collected at inclu- Mutations were detected for 22 patients, including 19 oncogenic sion and cfDNA was extracted from plasma for 334 patients. variants and 8 actionable mutations. Hotspot mutations were screened using next-generation sequenc- Conclusions: Detection of somatic mutations in cfDNA is ing for 50 cancer genes. feasible for prescreening phase I candidates with a satisfactory Results: Among the 283 patients with tDNA–cfDNA pairs, speciﬁcity; overall sensitivity can be improved by a sensitivity 121 had mutation in both, 99 in tumor only, 5 in cfDNA only, score allowing to select patients for whom cfDNA constitutes a and for 58 patients no mutation was detected, leading to a reliable noninvasive surrogate to screen mutations. Clin Cancer Res; 55.0% estimated sensitivity [95% conﬁdence interval (CI), 1–9. 2016 AACR.

1Laboratoire de Recherche Translationnelle et Centre de Ressources- Introduction Biologiques, AMMICA, INSERM US23/CNRS UMS3655, Gustave Roussy,Villejuif, France. 2Departement d'Innovation Therapeutique et Personalized medicine in oncology, which has seen a rapid d'Essais Precoces, Gustave Roussy,Villejuif, France. 3Departement de development in the last decade due to advances in cancer genomic Biostatistiques et Epidemiologie, Gustave Roussy, Villejuif, France. 4 5 research and technology, consists of proposing a treatment tai- INSERM U1018, Gustave Roussy, Villejuif, France. Facultede fi Medecine, Universite Paris Sud, Le Kremlin-Bicetre,^ Orsay, France. lored to the individual patient's molecular pro le (1). This 6INSERM U981, Gustave Roussy, Villejuif, France. 7Plateforme de assumes that the identified genomic mutations are driver altera- Genomique Fonctionnelle, AMMICA, INSERM US23/CNRS UMS3655, tions, and that matched targeted drugs are available (2). With the 8 Gustave Roussy, Villejuif, France. Plateforme de bioinformatique, use of high-throughput molecular technologies, the feasibility of AMMICA, INSERM US23/CNRS UMS3655, Gustave Roussy, Villejuif, France. 9Departement d'Oncologie Medicale,Gustave Roussy,Villejuif, such approach has been demonstrated in several pilot trials (3, 4) France. 10Departement de Radiologie, Gustave Roussy, Villejuif, as well as in large scale studies (5). 11 France. Departement de Biologie et Pathologie Medicales, Gustave To characterize the genomic alterations, tissue from primary Roussy,Villejuif, France. 12Direction de la Recherche Clinique, Gustave Roussy,Villejuif, France. 13Faculte de Pharmacie, Universite Paris Sud, tumor or metastasis is generally used for molecular analysis. In Chatenay-Malabry,^ France. the absence of suitable tissue samples from surgical or endo- Note: Supplementary data for this article are available at Clinical Cancer scopic resections, targeted biopsy is required. However, biopsy Research Online (http://clincancerres.aacrjournals.org/). techniques present ethical and logistic challenges, due to their invasiveness, morbidity, cost, variable accessibility of the tumor C. Jovelet and E. Ileana contributed equally and are the co-first authors of this article. sites, as well as potential sampling bias due to intratumor genetic heterogeneity (6,7). In contrast, tumor-derived nucleic J.-C. Soria and L. Lacroix share senior authorship acids in plasma (also known as circulating cell-free tumor DNA Corresponding Author: Cecile Jovelet, Laboratoire de Recherche Translation- - cfDNA), only requires a blood sample, can be collected easily nelle, 114 rue Edouard Vaillant, Villejuif 94800, France. Phone: 331-4211-4364; and repeatedly, and may be a noninvasive alternative to tissue Fax: 331-4211-4364; E-mail: [email protected] biopsy (8). doi: 10.1158/1078-0432.CCR-15-2470 cfDNA is composed of small circulating DNA fragments shed 2016 American Association for Cancer Research. into the plasma following apoptosis or necrosis of tumor cells, or

www.aacrjournals.org OF1

Jovelet et al.

(ECOG ¼ 0 or 1) and a primary tumor or metastasis accessible to Translational Relevance biopsy. The study was approved by the local Institutional review Molecular screening using next-generation sequencing board and all patients provided written informed consent for (NGS) of tumor-derived nucleic acids in plasma or circulating genetic analysis of their tumor and plasma samples prior to cell-free DNA (cfDNA) only requires a blood sample, and participation in this study. CT scan or ultrasound-guided biopsies appears as an attractive noninvasive alternative to DNA anal- were performed in metastatic or primary tumor sites to carry out a ysis extracted from tissue sample (tDNA). cfDNA-based anal- comprehensive molecular characterization. The percentage of ysis might also overcome the issue of sampling bias. This tumor cells out of all nucleated cells was assessed on a 3-mm article confirms the feasibility of NGS mutations screening section after a hematoxylin and eosin staining by a senior pathol- based on cfDNA samples on a large cohort of phase I clinical ogist. Samples with at least 10% of tumor cells were considered for trial. Although cfDNA specificity was very high, sensitivity further tDNA analysis through NGS; at least 30% of tumor cells appeared limiting for using this surrogate analysis in some within the sample was required for CGH analysis. Patients with patients. We consequently developed a sensitivity prediction less than 10% of tumor cells on the tissue sample underwent only score, which allowed appropriately selecting patients for cfDNA analysis. A weekly molecular tumor board reviewed all the cfDNA analysis. The analysis of cfDNA using NGS represents NGS and CGH results and determined the most relevant targeted an attractive noninvasive option for patients in whom a tumor therapy available through early clinical trials according to the molecular profile is required. molecular profile. Treatment efficacy was evaluated by Response Evaluation Criteria in solid tumors (RECIST) version 1.1 (18). Peripheral blood samples were collected just before the tissue biopsy. through spontaneous release from cancer tissue, and possibly fi circulating tumor cells (CTC; refs. 9, 10). The detection and Extraction and quanti cation of cfDNA – clinical use of cfDNA have long remained challenging due to the Blood samples (3 10 mL) were collected in EDTA tubes (BD mixture between tumor circulating free DNA and normal circu- Vacutainer, Beckton Dickinson and Company) and centrifuged lating free DNA, the low levels of cfDNA, and the difficulty to for 10 minutes at 1,000 g within 4 hours of the blood draw. The accurately quantify the number of DNA fragments. However, supernatant containing the plasma was further centrifuged at recent technological advances now enable the detection, quanti- 14,000 g for 10 minutes at room temperature and stored at fication, and analysis of cfDNA (11), thereby opening up possi- 80 C until analysis. Initially, the plasma were selected on the bilities for clinical use. Added to the minimal invasiveness of basis of their availability, and then, consecutively. m blood sampling, the attractiveness of liquid biopsy also consists in DNA was extracted from 500 L of plasma using the QIAamp avoiding cost and delay linked to the tissue biopsy. DSP virus kit (Qiagen), according to the manufacturer's instruc- m The potential applications of cfDNA are currently under exten- tions, and resuspended in 40 L of elution buffer. A real-time sive investigations, and include diagnosis, patient selection for quantitative PCR TaqManassay targeting GAPDH was used to targeted therapies, prognostic assessment, monitoring of treat- measure plasma DNA concentration. ment response and resistance, as well as screening for acquired tDNA was extracted from the fresh frozen biopsy sample using fi mutations at the time of resistance (10, 12). Recent studies have the AllPrep DNA/RNA Mini Kit (Qiagen) and quanti ed with shown a correlation between cfDNA levels and treatment Qubit 2.0 (Life technologies). response in breast cancer (13), the ability of cfDNA to identify fi mutations associated with acquired resistance (14), and its poten- Identi cation of somatic genomic mutations by NGS tial for diagnosis and patient selection of targeted therapy (15). Targeted sequencing libraries were generated using the Ion The ongoing MOSCATO 01 (Molecular Screening for Cancer AmpliSeq Library kit 2.0 according to the manufacturer's instruc- Treatment Optimization) trial (clinical trial NCT01566019) is a tions (Life Technologies). Tumor and plasma samples were ana- prospective molecular screening program of patients eligible for lyzed independently with Cancer Hotspot Panel v2(CHP2) tar- phase I trials, which uses high-throughput technologies [next- geting 50 cancer genes covered by 207 amplicons (Life Technol- generation sequencing (NGS) and comparative genome hybrid- ogies). For this study, any other genomic alterations found with ization (CGH)] to match patients' tumor molecular profile with other biopsy analysis (additional NGS panel, CGH, FISH, etc.) targeted therapy and evaluates the clinical benefit of such performed in the MOSCATO trial were not considered. fi approach (16, 17). The primers used for ampli cation were partially digested by In the current study, we evaluated whether detection of muta- FuPa enzyme. The digested product was then ligated with adapters fi fi tions in cfDNA is a reliable alternative to tissue biopsies for and barcodes, then ampli ed and puri ed using Agencourt beads. molecular screening, by determining the performance of cfDNA The quantity of the libraries was assessed using the Qubit 2.0 analysis to detect mutations as compared to tumor tissue (tDNA). Fluorometer. An equal amount of each library was pooled and amplified with the Ion OneTouch 2 system by the emulsion PCR with the Ion PGM Template OT2 200 Kit (Life Technologies). The Patients and Methods enrichment was then performed with the Ion One Touch ES Patients and sample collection (Enrichment System). The enriched Ion Spheres were loaded into Patients referred to the Drug Development Department a 316v.2 Ion Sequencing Chip. Sequencing was made using Ion (DITEP) at the Gustave Roussy Cancer Center (Villejuif, France) Personal Genome Machine (PGM, Life Technologies) using real- were eligible for inclusion in the MOSCATO 01 trial if they had an time measurement of the hydrogen ions produced during DNA advanced or metastatic solid tumor progressive after at least replication. The sequencing data were analyzed with the Torrent one line of standard therapy, with a good performance status Suite Variant Caller 4.2 software and reported somatic variants

OF2 Clin Cancer Res; 2016 Clinical Cancer Research

cfDNA Next-Generation Sequencing in MOSCATO Trial

were compared with the reference genome hg19. The variants Calibration and discriminative ability of the final model were were called if >5 reads supported the variant and/or total base evaluated using Hosmer–Lemeshow test (21) and AUC (22). We depth >50 and/or variant allele frequency >1% was observed. All set the P value threshold at 0.05 in the multivariable analysis. The the variants identified were visually controlled on .bam files using final model was validated using the 10-fold cross-validation Alamut v2.4.2 software (Interactive Biosoftware). All the germline approach. All estimates are given with their 95% confidence variants found in 1000 Genomes Project or ESP (Exome Sequenc- intervals (95% CI) and tests are two-sided. ing Project database) with frequency >0.1% were removed. All All analyses were performed with SAS v9.3 (SAS Institute). somatic mutations were annotated, sorted, and interpreted by an expert molecular biologist according to available databases (COS- MIC, TCGA) and medical literature. Analyses of tDNA and cfDNA Results were performed and interpreted independently. Patient characteristics For validation of our analysis method, we used a Quantitative From November 2011 to October 2014, 669 heavily pre- DNA Reference Standard control (Horizon Diagnostics) harbor- treated patients with metastatic or advanced solid tumors were ing 16 mutations in 9 different genes. Ten nanograms of DNA enrolled in the MOSCATO 01 trial (Fig. 1). Among the 600 Reference Standard were spiked in 1 mL of control plasma (pool patients for whom a biopsy was performed, a total of 334 of healthy patients). Two independent analyses were performed (56%) patients had a cfDNA analysis. All plasma samples have with CHP2 panel using the cfDNA analysis protocol. All the been processed within 4 hours after the blood sampling, mutations detected in the plasma were analyzed down to variant ensuring a good homogeneity of all the preanalytic and analytic with 1% allele frequency. Nevertheless, the 1% allele frequency steps in our cohort. Evaluable NGS results were obtained by threshold could not be considered for all the samples in our both tDNA and cfDNA analysis in 283 patients (85%, main analysis, as the assay sensitivity depends on the coverage depth analysis set). The first 178 patients were selected on the basis of which is variable between samples and genomic positions. the plasma availability and knowing mutational status in MiSeq resequencing analysis (Illumina), using CHP2 library tumor, whereas from June 2014, all consecutive patients amplicons, was performed. enrolled in MOSCATO 01 trial (N ¼ 105) had a cfDNA analysis. For 51 additional patients without tissue samples available for Statistical analysis molecular testing (due to the low content in tumor cells in All patients with available tDNA and cfDNA results were biopsy samples), only cfDNA analysis was performed. Those included in the main analysis. We estimated sensitivity and patients were not included in statistical analysis. specificity of cfDNA analysis compared with tDNA considered Patientanddiseasebaselinecharacteristics(N ¼ 283) are as the reference test. The analysis was performed both at the gene summarized in Table 1. Patients had a median age at biopsy of level, with possibly multiple mutated genes per patient, and at the 58 years (range: 18–79). Thoracic, gastrointestinal, breast, head patient level. Kappa agreement coefficient was estimated at the and neck, gynecologic and urologic cancers were the most gene level in the subset of patients who were consecutively frequent tumor locations, with 19%, 18%, 15%, 14%, 9%, included. Among patients with mutations found in tumor, the and 7%, respectively. 95% of the patients were metastatic and cfDNA result was classified as fully concordant if all mutations were heavily pretreated with a median number of 3 previous identified in tumor were also found in cfDNA, and partially lines of treatment (range: 0–11). The median cfDNA concen- concordant if only part of the mutations found in tumor were tration found in plasma was 18.1 ng/mL (range: <0.1–727.0 identified in cfDNA. At the patient level, the sensitivity of cfDNA ng/mL). As illustrated in Fig. 2, cfDNA concentration was was computed as the probability of a positive concordant cfDNA significantly higher in patients with a gastrointestinal tumor result (fully or partially concordant) in patients for whom muta- than in patients with other tumor sites (median: 30.4 vs. 16.2, tions were found in tumor. P ¼ 0.02). We then modeled this probability of a positive concordant cfDNA result in patients whose mutations were found in tumor Mutation analysis tissue, and built a sensitivity prediction score using a logistic Overall, 283 patients had tumor and plasma samples collected regression. All variables associated with a P <0.15 in univariable at the same timepoint which were analyzed independently using model were then evaluated in multivariable analysis using a the same CHP2 panel. Mean depth for the cfDNA NGS analysis backward procedure. We first considered only patient character- was x492. A total of 347 mutations were identified in tDNA istics at study entry (age, gender, ECOG performance status, and 195 mutations were identified in cfDNA. In tDNA, at least primary tumor site, number of metastatic sites, LDH level, albu- 1 mutation was detected in 220 patients (78%) (1 mutation in min level, and number of prior lines of treatment), to build a 129 patients, 2–4 in 91 patients). Mutations are detailed in clinical score, available before tumor biopsy and DNA extraction. Supplementary Table S1. We also evaluated the diagnostic value of the RMH score, which is The most frequently mutated genes in tumor (N > 3) were TP53 a score developed by the team from Royal Marsden Hospital for (47.3% of patients), KRAS (17.3%), PIK3CA (13.1%), EGFR prognosis purpose in the setting of oncology phase I trial score (5.3%), CDKN2A (4.2%), APC (3.9%), PTEN (3.9%), SMAD4 (19). We then evaluated the additional value of cfDNA concen- (3.5%), AKT1 (2.5%), STK11 (2.5%), CTNNB1 (2.1%), FBXW7 tration adjusted on the clinical score. Continuous variables (age, (2.1%), KIT (1.8%), ERBB2 (1.4%), RB1 (1.4%), and BRAF LDH and albumin levels, number of metastatic sites and prior (1.4%). In cfDNA, at least 1 mutation was detected in 134 patients lines of treatment, RMH score, cfDNA concentration) were (47%; median ¼ 1 mutation per patient, range 1–4 mutations). entered in the model assuming a log linear relationship, after The most frequent mutations (n > 3) found in cfDNA were in TP53 checking the shape of the relation in the model. Models were (23.3%), KRAS (11.0%), PIK3CA (6.0%), APC (3.2%), PTEN compared using Akaike Information Criterion (AIC; ref. 20). (3.2%), EGFR (2.8%), ERRB2 (2.5%), SMAD4 (2.1%), CTNNB1

www.aacrjournals.org Clin Cancer Res; 2016 OF3

Jovelet et al.

Figure 1. Participant ﬂow diagram.

(1.8%), CDKN2A (1.4%), AKT1 (1.4%), KIT (1.4%), and RB1 Concordance between tDNA and cfDNA (1.4%) genes. No correlation was observed between allele fre- Among the 347 mutations identified in tDNA from 283 quency observed in tumor tissue and in plasma. patients, 173 mutations were identified in both tDNA and The cfDNA analysis of the 51 patients without tDNA analysis cfDNA, and 174 mutations in tDNA, but not in cfDNA. Results revealed at least one mutation in 22 patients (43.1%). Among the for the main genes are illustrated in Fig. 3A. Moreover, 22 29 mutations found, 19 were pathogenic variants and included 8 mutations were found in cfDNA, but not in tDNA, providing variants of therapeutic interest. Several of these cfDNA analyses additional information. The sensitivity of using NGS for 50 have been performed in parallel to tDNA analysis of patients targeted hotspot genes analysis in cfDNA compared with tDNA recently enrolled in MOSCATO 01 trial and the results were was 49.9% (95% CI, 44.6%–55.1%) and the specificity was discussed in dedicated molecular tumor board. Also, cfDNA 99.8% (95% CI, 99.8%–99.9%). Sensitivity in the five most analysis could be performed in the same conditions (turn-around frequent mutated genes (TP53, KRAS, PIK3CA, EGFR,andAPC, time and cost) as tDNA analysis. representing 75.9% of all cfDNA mutations, was 47%, 59%, To confirm mutations identified in cfDNA with the PGM-Ion 43%, 53%, and 46%, respectively. The Kappa coefficient cal- Torrent analysis, MiSeq resequencing analysis (Illumina) was culated on the 105 consecutive patients was 0.60 (95% CI, performed using the CHP2 library amplicons on a set of 83 0.50–0.69). patients. The mean depth for the cfDNA analysis with MiSeq At the patient level, NGS analysis allowed us to observe: (i) 58 was x748 and 91 mutations were identified. The MiSeq anal- patients without any mutation found in tumor or in cfDNA; (ii) 5 ysis allowed the detection of 16 additional mutations which patients with mutations found in cfDNA but not in tumor; (iii) were not identified in cfDNA analysis by PGM-Ion Torrent, 220 patients with mutations found in tumor and plasma. Among with 11 of them (69%) present in the tDNA analysis. Con- these latter, the mutations found in plasma were fully concordant versely, PGM-Ion Torrent analysis identified 6 mutations in for 87 patients (39.5%) and partially concordant for 34 patients cfDNA, which were also found in the tDNA, but which could (15.5%), whereas the mutations found in tumor were not found not be detected with MiSeq sequencing of cfDNA. Overall, in the cfDNA for the 99 other patients (45%), leading to an MiSeq and PGM sequencing provided concordant results for estimated sensitivity of 55.0% (95% CI, 48.4–61.6%) at the the cfDNA analysis. patient level (Fig. 3B).

OF4 Clin Cancer Res; 2016 Clinical Cancer Research

cfDNA Next-Generation Sequencing in MOSCATO Trial

Table 1. Patient and tumor characteristics goodness of fit (Hosmer and Lemeshow test, P ¼ 0.34). Using Frequency (%) the 10-fold cross validation, the estimated prediction error N ¼ Total 283 was 0.205. Age at study entry The relationship between the derived sensitivity prediction Median (range) 58 (18–79) Gender score and the probability of a positive cfDNA result is illustrated Male 149 (52.7%) in Fig. 4E. The estimated sensitivity of cfDNA was 28% when Female 134 (47.3%) focusing on the patients with a low score (first third of the Primary tumor site distribution), which mainly corresponded to patients with less Breast 41 (14.5%) advanced disease and/or patients with head and neck or urologic Gastrointestinal 50 (17.7%) cancer. It is noticeable that cfDNA was positive in only 1 of the 22 Gynecologic 26 (9.2%) fi Head and Neck 39 (13.8%) patients with the lowest score values (the rst decile which Thoracic 54 (19.1%) includes 21 patients with head and neck cancer). In contrast, Urologic 21 (7.4%) sensitivity of cfDNA is 83% in patients with a high score value (last Othera 52 (18.4%) third of the distribution), reaching 94% (17/18) in the highest ECOG Performance Status (64 MD) decile that is mainly represented by patients with advanced breast 0 100 (45.7%) or gastrointestinal cancer. 1 112 (51.1%) 2 7 (3.2%) The strong relationship between LDH level and cfDNA posi- Number of previous lines of therapy tivity observed in univariable analysis vanished in multivariable Median (Range) 3 (0–11) analysis (OR ¼ 1.33, 95% CI, 0.98–1.81, P ¼ 0.07), because LDH Number of metastatic sites level was significantly correlated with the other variables included 0 15 (5.3%) in the model (Spearman correlation coefficient between LDH and 1 101 (35.7%) clinical score ¼ 0.36, P < 0.0001). 2 87 (30.7%) fi 3 51 (18%) The RMH score was signi cantly associated with cfDNA pos- >3 29 (10.2%) itivity. However, the performance of the multivariable model LDH level (UI/L; 7 MD) described herein above was better than that of the model based Median (range) 220 (14–6,980) only on RMH score (AIC, 254.2 for the final clinical multivariable Albumin level (g/L) (3 MD) model vs. 264.5 for the model based only on RMH score, when Median (range) 36 (22–47) RMH score (10 MD)b 0 102 (37.4%) 1 106 (38.8%) 2 53 (19.4%) 3 12 (4.4%) cfDNA Concentration (ng/mL; 2 MD) Median (range) 18.1 (<1–727) Number of mutated genes/patient 0 63 (22.3%) 1 129 (45.6%) >1 91 (32.2%) Abbreviation: MD, missing data. aOther tumor sites: prostate (N ¼ 18), cancer of unknown primary (N ¼ 9), testis (N ¼ 5), thyroid (N ¼ 4), bone (N ¼ 4), skin (N ¼ 3), penis (N ¼ 2), peritoneum (N ¼ 2), connective tissue and soft tissue (N ¼ 3), lip (N ¼ 1), scalp and neck (N ¼ 1). bRMH Score: Prognostic score developed by the team from Royal Marsden Hospital in the setting of oncology phase I trials (ref Arkenau JCO 2009), based on LDH level, albumin level and number of sites of metastases, weighted equally: LDH normal (0) versus LDH > upper limit of normal (1), albumin 35 g/L (0) versus albumin < 35 g/L (1), and sites of metastases 2 (0) versus sites of metastases > 2 (1). The prognostic score for each individual was computed as the sum of the three components.

Table 2 details the modeling of cfDNA positivity among the 220 patients with mutations found in the tumor, including 121 patients with fully or partially concordant cfDNA result. The sensitivity of cfDNA test significantly increased with the number of metastatic sites (Fig. 4A, P ¼ 0.0006 in multivariable model), a decreased albumin level (Fig. 4B, P ¼0.0007), and the number of prior treatment lines (Fig. 4C, P ¼ 0.047); it was also significantly associated with the primary tumor site (Fig. 4D, P ¼ 0.002). The sensitivity was higher in breast and gastrointestinal cancers and lower in patients with head and neck or urologic cancer. The final multivariable analysis clinical model, based on these four clin- Figure 2. ical characteristics, had a good predictive power as indicated by cfDNA plasma concentration according to primary tumor site. Concentration the AUC equal to 0.78 (Supplementary Fig.S1), as well as a good values are reported in ng/mL of plasma; bar corresponds to the median.

www.aacrjournals.org Clin Cancer Res; 2016 OF5

Jovelet et al.

good surrogate of tissue biopsy and the one for whom the cfDNA NGS analysis will not be contributive. Overall, the specificity was above 95%, but the sensitivity of cfDNA analysis was estimated at only 55% (121/220) in the patients with at least one mutation found in tDNA. The observed discordance could be due to several reasons: poor mutation detection in cfDNA due to low cfDNA concentrations in the bloodstream (23); presence of the mutation in allele frequencies below the detection limit of our method; true disease heterogeneity (24); and poor DNA quality (25), true genetic differences between cfDNA (released mainly from necrotic cells) and tDNA. Ongoing technological advances are likely to improve the cfDNA analysis method and further increase the concordance between tDNA and cfDNA. For low cfDNA levels, on first hand, we chose to use the limited quantity of plasma available through the prospective collection set up. Carrying out cfDNA from larger plasma volume could increase the levels of cfDNA for molecular analysis for some cases. For several samples, even using four times the volume of plasma may lead to better results; indeed, we have tested 45 additional cases using 2 mL of plasma and observed 49% of concordance (data not shown). On the other hand, using alternative sequencing method as MiSeq with better coverage did not increase the number of mutations detected. Nevertheless, our results should be also considered in a context of novel and more sensitive techniques. For example, methodologies based on dig- ital PCR are making cfDNA analyses possible even in cases where only low levels of cfDNA are present, with a threshold of detection of 0.01% or lower (11). In addition, important progress has been made using sequence-specific assays that target predefined mutations and detect extremely rare alleles (26). False negative results (mutation found in tDNA but not in cfDNA) may be further reduced by a reestimation of the background frequency noise and the minimum allele frequency threshold. Another approach Figure 3. could be to prioritize screening of limited numbers of mutations Concordance between tumor biopsies (tDNA) and plasma (cfDNA) molecular – analysis. Concordance between tDNA and cfDNA for the 283 patients with with highly sensitive techniques instead of large tumor type paired samples; at gene level for the 16 most frequently mutated genes agnostic gene panel. However, whatever the threshold of detec- (among 50; A) and at the patient level (B; Tþ mutation in tumor biopsy, Pþ tion, a sufficient DNA concentration or DNA copy number is mutation in plasma sample; T no mutation in tumor biopsy and no mutation needed, but was not always observed in our cohort, with DNA in plasma samples). concentration ranging from <0.1 ng/mL to 727 ng/mL. Even if our cohort of phase I patients could be considered as both models include the 212 patients with no missing data for biased by low representation of some tumor types (e.g., melano- RMH score). ma), our results could be compared with several prior studies The positivity of cfDNA result was significantly higher with investigating the concordance of cfDNA and tDNA in different increasing cfDNA concentration (P ¼ 0.003 in univariable tumor types with varying results. Rothe and colleagues found a analysis and 0.045 in multivariable analysis). However, the sensitivity of cfDNA estimated at 82% in 11 patients with mutated added value of cfDNA concentration to the multivariable metastatic breast cancer (27). Another study found a sensitivity clinical model was marginal: the AUC increased from 0.78 rate of 58% in 50 mutated metastatic lung cancer patients (28). to 0.79. The estimated sensitivity of cfDNA in patients with breast or lung cancer in our series (74% and 55%, respectively) is consistent with the published results. In a recent study based on 39 patients with Discussion different types of tumor harboring mutations, Frenel and collea- In this study, we show that use of NGS for cfDNA analysis is gues estimated sensitivity of cfDNA at 59% (23/39), similar to our feasible in routine clinical settings for molecular characterization overall result (55%; ref. 29). in patients with locally advanced or metastatic solid tumors, Lebofsky and colleagues compared the cfDNA and tDNA using eligible for phase I trials and displaying a good performance NGS technology in 34 patients with 18 different tumor types, status. To our knowledge, this study reports the largest patient enrolled in the phase II SHIVA trial. Among the 18 patients with cohort used to evaluate the feasibility of NGS gene panel screening paired samples and mutations found in metastasis biopsies, using plasma samples matched with analysis of tumor samples cfDNA analysis was concordant in 17 cases, leading to a very issued for various tumor types obtained from 283 patients. This is high sensitivity rate (97% at the gene level, 98% at the patient also the first study reporting a sensitivity prediction score enabling level; ref. 15). Several factors may explain the apparent discrep- better selection of patients for whom liquid biopsy could be a ancy between our result and the SHIVA trial results, in addition to

OF6 Clin Cancer Res; 2016 Clinical Cancer Research

cfDNA Next-Generation Sequencing in MOSCATO Trial

Table 2. Modeling of the cfDNA positivity in the 220 patients with mutations found in the tumor Multivariable analysis clinical Univariable analysis modela OR (95% CI) P OR (95% CI) P Age at study entry 1.01 (0.99–1.04) 0.24 Gender 0.10 Female 1 (ref) Male 0.63 (0.37–1.09) ECOG performance status 0.83 0 1 (ref) 1 0.87 (0.47–1.60) 2 1.32 (0.23–7.62) Primary tumor site 0.0004 0.002 Breast 1 (ref) 1 (ref) Gastrointestinal 0.83 (0.29–2.39) 1.42 (0.42–4.78) Gynecologic 0.51 (0.15–1.70) 0.54 (0.14–2.08) Head and Neck 0.07 (0.02–0.27) 0.15 (0.03–0.64) Thoracic 0.42 (0.15–1.20) 0.66 (0.19–2.25) Urologic 0.16 (0.04–0.62) 0.14 (0.03–0.62) Other 0.53 (0.18–1.57) 0.82 (0.24–2.76) Number of metastatic sites 1.79 (1.35–2.36) <0.0001 1.79 (1.28–2.49) 0.0006 LDH level (OR associated with a 100 UI/L increase) 1.77 (1.26–2.50) 0.001 Albumin level (OR associated with a 10 g/L increase) 0.39 (0.20–0.77) 0.006 0.26 (0.12–0.57) 0.0007 RMH score <0.0001 0 1 (ref) 1 3.19 (1.66–6.12) 2 7.00 (2.99–16.4) 3 16.0 (1.90–135) Number of prior lines of treatment 1.26 (1.08–1.46) 0.003 1.21 (1.002–1.46) 0.0474 Concentration of cfDNA (OR associated with a 10 ng/mL increase) 1.14 (1.05–1.24) 0.003 aThe clinical model described in the table includes the four clinical variables found significantly associated with the cfDNA positivity (P < 0.05 in multivariable analysis): primary tumor site, number of metastatic sites, albumin level and number of prior lines of treatment. Overall 217 patients are included in this model due to missing value for albumin level in 3 patients. the play of chance in this small sample size series: patient ogenic variants of therapeutic interest. This data highlights the characteristics were different, with, in their series compared to feasibility and the added value of cfDNA analysis for patients for ours, more breast and GI cancers, less head and neck or urologic whom tissue biopsy is not available. Nevertheless, due to rela- cancer, more patients with a number of metastatic sites greater tively high rate of false negative result, it is important to identify than 2. In addition, most plasma samples (31/34) were collected patients for whom biopsy contribution is essential. after biopsy, often several months after the initial biopsy, when We showed that the sensitivity of cfDNA analysis was signif- the disease could be more advanced. icantly associated with the primary tumor site and increased with Part of the disparity in sensitivity rates of cfDNA reported by the number of metastatic sites, the number of prior lines of different studies could also be explained by the different proce- treatment and a decrease of albumin level. For a practical use in dures that are used in laboratories at both preanalytic (blood selecting patients for whom cfDNA could be used as a surrogate collection, processing, storage, baseline of patients) and analytic molecular screening maker, we have built, for the first time, a phases (DNA extraction, quantification, and analysis; refs. 11, 24, sensitivity prediction score. Patients with a high score are more 30, 31). To introduce the liquid biopsy in the clinical manage- likely to have a contributive result from cfDNA analysis. Also, our ment of cancer patients, it is important that these parameters be findings suggest that for those patients, cfDNA is a valuable standardized for consensus analysis and reporting (32). surrogate material for molecular screening by NGS analysis, Most of the published studies had a lower number of patients whereas tumor biopsy analysis will be recommended in patients tested and the method of patient selection is not described in with a low sensitivity prediction score. detail. Moreover, the performance status and the disease's char- Therefore, molecular characterization based on cfDNA anal- acteristics are generally poorly reported in most of the published ysiscouldbeperformedinthesameconditionsastDNA studies. Those points limit the possibility of comparison with our analysis (turn-around time and cost). cfDNA-based analysis cohort of patients having mainly performance status and a low could be used for diagnosis and treatment decisions, even for RMH score at the time of sample collection that appear to be patients for whom tumor biopsy is not feasible, which repre- crucial for paired cfDNA and tDNA molecular status concordance. sent a significant proportion of patients with metastatic cancers, In the current study, we identified mutations only in the cfDNA or when the quality of the histologic sample is too poor for analysis and not in tDNA in few patients. This point was consis- molecular analysis. But further studies on prospective valida- tent with previous data (26–29). This suggests that cfDNA analysis tion cohorts are required to determine whether an optimized using NGS could be also used as a complementary analysis to the version of the assay may be applicable in routine clinical tDNA to better evaluate the tumor heterogeneity (7). In addition, practice as a liquid biopsy. we analyzed the cfDNA of 51 patients without tDNA because of Nevertheless, there are still improvements possible to our low tumor cellularity (less than 10% of tumor cell) and we found study, beside preanalytic points discussed previously. We evalu- one or more mutations in 22 patients (43%), including 8 path- ated only the cfDNA at baseline compared with the tDNA

www.aacrjournals.org Clin Cancer Res; 2016 OF7

Jovelet et al.

Figure 4. cfDNA result in patients with mutations found in the tumor, according to albumin level (A), number of metastatic site (B), number of previous lines of therapy (C), primary tumor site (D), and sensitivity prediction score (E). D, H&N, head and neck; Urol., urologic; Thor., thoracic; G.I., gastrointestinal; Gyne., Gynecologic.

regardless of the tumor types and progression rate before the patients for whom it will likely be recommended to have analysis inclusion in the MOSCATO trial. Further cfDNA kinetics analysis performed on tumor tissue. (i.e., analysis at multiple time-point samples) and correlation of the cfDNA results with the clinical outcome is warranted. Disclosure of Potential Conflicts of Interest fl Despite its inconveniences, tumor biopsy remains the most No potential con icts of interest were disclosed. efficient diagnostic tool at present. However, several major advan- tages of cfDNA justify further investigation and development as a Authors' Contributions tool to detect tumor-specific genomic alterations in the circula- Conception and design: C. Jovelet, E. Ileana, M.-C. Le Deley, C. Ferte, S. Postel- tion, beyond what is currently possible with tumor biopsy. First, Vinay, T.D. Baere, F. Andre, J.-C. Soria, L. Lacroix Development of methodology: C. Jovelet, E. Ileana, M.-C. Le Deley, N. Motte, the possibility of cfDNA analysis even with low levels of cfDNA S. Rosellini, C. Ferte, T.D. Baere, L. Lacroix may lead to early detection of cancer (11). Second, even when Acquisition of data (provided animals, acquired and managed patients, biopsy is technically feasible, the minimally invasive, more con- provided facilities, etc.): C. Jovelet, E. Ileana, A. Romero, N. Pata-Merci, venient cfDNA quantification would be more acceptable to N. Droin, A. Hollebecque, C. Ferte, A. Boichard, S. Postel-Vinay, M. Ngo-Camus, patients than biopsy. Third, cfDNA analysis facilitates monitoring T.D. Baere, J.-Y. Scoazec, J.-C. Soria, L. Lacroix of molecular alterations associated with tumor response or Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): C. Jovelet, E. Ileana, M.-C. Le Deley, S. Rosellini, acquired resistance, while sparing the patient from repeated C. Lefebvre, M. Pedrero, M. Deloger, C. Ferte, T.D. Baere, J.-C. Soria, L. Lacroix biopsies and their potential harms. Writing, review, and/or revision of the manuscript: C. Jovelet, E. Ileana, M.-C. In conclusion, our results on a consistent cohort of patients Le Deley, S. Rosellini, M. Deloger, C. Massard, A. Hollebecque, C. Ferte, S. Postel- suggest that targeted NGS for the detection of somatic mutations Vinay, T.D. Baere, P. Vielh, J.-Y. Scoazec, A. Eggermont, F. Andre, J.-C. Soria, in cfDNA could be applicable in clinical for patient eligible for L. Lacroix phase I clinical trials and bearing advanced or metastatic solid Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): C. Jovelet, E. Ileana, M.-C. Le Deley, N. Motte, tumors. We have established a score that allows us to select A. Romero, C. Massard, A. Hollebecque, C. Ferte, J.-Y. Scoazec, G. Vassal, patients for whom the analysis of cfDNA using NGS offers an L. Lacroix attractive noninvasive option to screen mutations and to selected Study supervision: C. Ferte, J.-C. Soria, L. Lacroix

OF8 Clin Cancer Res; 2016 Clinical Cancer Research

cfDNA Next-Generation Sequencing in MOSCATO Trial

Acknowledgments MOSCATO 01 trial. The authors especially acknowledge Yuki Takahashi for her The authors acknowledge the clinical team, including Yohann Loriot, Ras- editing support. tislav Bahleda, Anas Gazzah, Andrea Varga, and Eric Angevin for their help with enrollment of patients in the MOSCATO trial, and Aljosa Celebic for Data Grant Support Management. The authors acknowledge laboratory and bioinformatic teams, This work was supported by grant INCa-DGOS-INSERM 6043. including Melanie Laporte, Isabelle Miran, Ludovic Bigot, Stephanie Coulon, The costs of publication of this article were defrayed in part by the payment of Marie Breckler, Catherine Richon, Aurelie Honore, Magali Kernaleguen, Glaw- page charges. This article must therefore be hereby marked advertisement in dys Faucher, Zsoﬁa Balogh, Jonathan Sabio, Lionel Fougeat, Marie Xiberras, accordance with 18 U.S.C. Section 1734 solely to indicate this fact. Leslie Girard, Lucie Herard, Catherine Lapage, Guillaume Meurice, Romy Chen- Min-Tao, and Yannick Boursin for their support in the preanalytic processing of Received October 14, 2015; revised December 7, 2015; accepted December 7, samples storage and for tumor sample analysis of the patients included in the 2015; published OnlineFirst January 12, 2016.

References 1. Tursz T, Andre F, Lazar V, Lacroix L, Soria JC. Implications of personalized 17. Massard C, Hollebecque A, Rosselini S, Ferté C, Lacroix L, Auger N, et al. medicine-perspective from a cancer center. Nat Rev Clin Oncol 2011; Enriching phase I trials with molecular alterations: Interim analysis of 8:177–83. 708 patients enrolled in the MOSCATO 01 trial. Ann Oncol 2015; 2. Arnedos M, Vielh P, Soria JC, Andre F. The genetic complexity of common 26:ii3–5. cancers and the promise of personalized medicine: Is there any hope? J 18. Eisenhauer EA, Therasse P, Bogaerts J, Schwartz LH, Sargent D, Ford R, et al. Pathol 2014;232:274–82. New response evaluation criteria in solid tumours: revised RECIST guide- 3. Von Hoff DD, Stephenson JJ, Rosen P, Loesch DM, Borad MJ, Anthony S, line (version 1.1). Eur J Cancer 2009;45:228–47. et al. Pilot study using molecular profiling of patients' tumors to find 19. Arkenau H-T, Barriuso J, Olmos D, Ang JE, de Bono J, Judson I, et al. potential targets and select treatments for their refractory cancers. J Clin- Prospective validation of a prognostic score to improve patient selection for Oncol 2010;28:4877–83. oncology phase I trials. J Clin Oncol 2009;27:2692–6. 4. Wagle N, Berger MF, Davis MJ, Blumenstiel B, De Felice M, Pochanard P, 20. Akaike H. "A new look at the statistical model identification." IEEE Trans et al. High-throughput detection of actionable genomic alterations in Autom Control 1974;19:716–23. clinical tumor samples by targeted, massively parallel sequencing. Cancer 21. Hosmer DW, Lemeshow S. Applied logistic regression. New York, NY: Discov 2012;2:82–93. Wiley. 2000. p147–56. 5. Andre F, Bachelot T, Commo F, Campone M, Arnedos M, Dieras V, et al. 22. Hanley JA, McNeil BJ. The meaning and use of the area under a receiver Comparative genomic hybridisation array and DNA sequencing to direct operating characteristic (ROC) curve. Radiology 1982;143:29–36. treatment of metastatic breast cancer: a multicentre, prospective trial 23. Bettegowda C, Sausen M, Leary RJ, Kinde I, Wang Y, Agrawal N, et al. (SAFIR01/UNICANCER). Lancet Oncol 2014;15:267–74. Detection of circulating tumor DNA in early- and late-stage human 6. Gerlinger M, Rowan AJ, Horswell S, Larkin J, Endesfelder D, Gronroos E, malignancies. Sci Transl Med 2014;6:224–49. et al. Intratumorheterogeneity and branched evolution revealed by multi- 24. Perkins G, Yap TA, Pope L, Cassidy AM, Dukes JP, Riisnaes R, et al. Multi- region sequencing. N Engl J Med 2012;366:883–92. purpose utility of circulating plasma DNA testing in patients with advanced 7. Burrell RA, McGranahan N, Bartek J, Swanton C. The causes and con- cancers. PLoS One 2012;7:e47020–9. sequences of genetic heterogeneity in cancer evolution. Nature 2013;501: 25. Elshimali YI, Khaddour H, Sarkissyan M, Wu Y, Vadgama JV. The clinical 338–45. utilization of circulating cell free DNA (ccfDNA) in blood of cancer 8. Bidard F, Weigelt B, Reis-filho JS. Going with the flow: from circulating patients. Int J Mol Sci 2013;14:18925–58. tumor cells to DNA. Sci Transl Med 2013;5:1–7. 26. Forshew T, Murtaza M, Parkinson C, Gale D, Tsui DWY, Kaper F, et al. 9. Schwarzenbach H, Hoon DSB, Pantel K. Cell-free nucleic acids as bio- Noninvasive identification and monitoring of cancer mutations by tar- markers in cancer patients. Nat Rev Cancer 2011;11:426–37. geted deep sequencing of plasma DNA. Sci Transl Med 2012;4:136–68. 10. Crowley E, Di Nicolantonio F, Loupakis F, Bardelli A. Liquid biopsy: 27. Rothé F, Laes J, Lambrechts D, Smeets D, Vincent D, Maetens M, et al. monitoring cancer-genetics in the blood. Nat Rev Clin Oncol 2013;10: Plasma circulating tumor DNA as an alternative to metastatic biopsies for 472–84. mutational analysis in breast cancer. Ann Oncol 2014;10:1959–65. 11. Diaz LA, Bardelli A. Liquid biopsies: genotyping circulating tumor DNA. J 28. Couraud S, VacaPaniagua F, Villar S, Oliver J, Schuster T, Blanche H, et al. Clin Oncol 2014;32:579–86. Non-invasive diagnosis of actionable mutations by deep sequencing of 12. Aparicio S, Caldas C. The implications of clonal genome evolution for circulating-free DNA in non-small cell lung cancer: findings from BioCAST/ cancer medicine. N Engl J Med 2013;368:842–51. IFCT-1002. Clin Cancer Res 2014;20:4613–24. 13. Dawson S-J, Tsui DWY, Murtaza M, Biggs H, Rueda OM, Chin S-F, et al. 29. Frenel J-S, Carreira S, Goodall J, Roda Perez D, Perez Lopez R, Tunariu N, Analysis of circulating tumor DNA to monitor metastatic breast cancer. N et al. Serial next generation sequencing of circulating cell free DNA Engl J Med 2013;368:1199–209. evaluating tumour clone response to molecularly targeted drug adminis- 14. Murtaza M, Dawson S-J, Tsui DWY, Gale D, Forshew T, Piskorz AM, et al. tration. Clin Cancer Res 2015;21:4586–96. Non-invasive analysis of acquired resistance to cancer therapy by sequenc- 30. Oxnard GR, Paweletz CP, Kuang Y, Mach SL, O'Connell A, Messineo MM, ing of plasma DNA. Nature 2013;497:108–12. et al. Noninvasive detection of response and resistance in EGFR-mutant 15. Lebofsky R, Decraene C, Bernard V, Kamal M, Blin A, Leroy Q, et al. lung cancer using quantitative next-generation genotyping of cell-free Circulating tumor DNA as a non-invasive substitute to metastasis biopsy plasma DNA. Clin Cancer Res 2014;20:1698–705. for tumor genotyping and personalized medicine in a prospective trial 31. Luke JJ, Oxnard GR, Paweletz CP, Camidge DR, Heymach JV, Solit DB, et al. across all tumor types. Mol Oncol 2015;9:783–90. Realizing the potential of plasma genotyping in an age of genotype- 16. Ferte C, Massard C, Ileana E, Hollebecque A, Lacroix L, Ammari S, et al.et al. directed therapies. J Natl Cancer Inst 2014;106:1–5. Molecular screening for cancer treatment optimization (MOSCATO 01): a 32. Martinez P, McGranahan N, Birkbak NJ, Gerlinger M, Swanton C. Compu- prospective molecular triage trial; Interim analysis of 420 patients. Cancer tational optimisation of targeted DNA sequencing for cancer detection. Sci Res 2014;74:CT240. Rep 2013;3:3309–18.

www.aacrjournals.org Clin Cancer Res; 2016 OF9

Circulating Cell-Free Tumor DNA Analysis of 50 Genes by Next-Generation Sequencing in the Prospective MOSCATO Trial

Cécile Jovelet, Ecaterina Ileana, Marie-Cécile Le Deley, et al.

Clin Cancer Res Published OnlineFirst January 12, 2016.

Updated version Access the most recent version of this article at: doi:10.1158/1078-0432.CCR-15-2470

Supplementary Access the most recent supplemental material at: Material http://clincancerres.aacrjournals.org/content/suppl/2016/01/12/1078-0432.CCR-15-2470.DC1

E-mail alerts Sign up to receive free email-alerts related to this article or journal.

Reprints and To order reprints of this article or to subscribe to the journal, contact the AACR Publications Subscriptions Department at [email protected].

Permissions To request permission to re-use all or part of this article, use this link http://clincancerres.aacrjournals.org/content/early/2016/03/21/1078-0432.CCR-15-2470. Click on "Request Permissions" which will take you to the Copyright Clearance Center's (CCC) Rightslink site.