From prospective biobanking to precision medicine: BIO-RAIDs – an EU study protocol in cervical cancer

FIGO (International Federation of Gynecology and Obstetrics) classification for CC was revised in 2009. The use of MRI imaging was encouraged towards improving staging accuracy (92 %) [4‐7]. While screening has decreased the incidence of CC in Western EU countries, its treatment has not changed over the last 15 years, the most recent improvement being the association of radio-chemotherapy [8, 9]. Addition of Bevacizumab [10] or Gemcitabine [11] to chemotherapy did show small improvements, but these treatments are costly, have side effects and are not available in many countries in need.

In early stage disease (≤IB1) and in a subset of patients of stage IB2-IIA, a surgical approach may cure the patient [12‐15]. For stages IB2 to III, concurrent chemo-radiation with a platinum-based reagent is the recommended standard of therapy [9]. Unfortunately, advanced stage (III-IV) disease remains a significant public health problem [16]. In on-going clinical trials, the impact of additional chemotherapy either as neo-adjuvant (INTERLACE trial [17] or adjuvant (OUTBACK trial: ASCO meeting 2012) strategies, are currently being assessed.

Molecularly targeted agents in CC are still being tested in clinical trials. CC patients may have variable outcomes and specific genetic mutations have already been shown to impact clinical response to standard therapy by us [18] and others [19‐21]. The Cetuxicol trial, a phase 2 trial sponsored by Institut Curie, showed that the addition of Cetuximab over a 6 week period, did not improve DFS. Of interest was the finding that PI3K pathway tumor mutations in the Cetuximab treatment arm led to a worse outcome (De la Rochefordiere 2015). We are presently lacking prognostic and predictive biomarkers for CC treatment and there is a growing need for the development of these to follow up the course of the disease. Due to the multiplicity of potential genetic alterations, retrospective molecular assessments in small patient populations are mostly inconclusive. For these reasons, we initiated BIO-RAIDs, a prospective study with extensive biobanking which aims to identify predictive biomarkers for treatment response of cervical cancers in both Western and Eastern European countries. To our knowledge, BIO-RAIDs is the first large prospective trial of this type in the field of CC. At the medical/scientific level, BIO-RAIDs will be crucial in setting the ground for future precision medicine studies by identifying a set of stratification criteria for cervical carcinomas as well as other cancers with similar molecular alterations.

BIO-RAIDs is a prospective European study, involving oncology centers from France, Germany, Serbia, the Netherlands, Romania and Moldova. 700 patients with previously untreated, advanced stage CC (stage IB2-IV) will be enrolled. The recommendations for standards of treatment by stage of disease as well as the timing of biopsies are shown in Fig. 2. The primary objective of the study is to assess dominant mutations and activation of signaling pathways in cervical cancers predictive of response to standard treatment. Secondary objectives of the study are: 1. The determination of PFS at 18 months as a function of dominant genetic or protein alterations; 2. The descriptive analysis of standard treatments applied in participating European countries; 3. A descriptive analysis of grade 3 and 4 side effects; 4. A descriptive analysis of the frequency of molecular alterations according to geographic location.

Approval for the BIO-RAIDs study has been obtained from ethics committees in all participating countries (Comité pour la protection des personnes (CPP) Ile de France in France, The Serbian Medical Society Belgrade in Serbia, The Protocol Toetsingscommissie of the Antoni van Leeuwenhoek (PTC) in the Netherlands, The Medizinische Hochschule Hannover Ethik-kommission in Germany, The National Ethics Committee of Health Ministry of Republic of Moldova, Academy of Medical Sciences National Ethics Committee for medicines and medical devices in Romania). BIO-RAIDs is conducted in accordance with the guidelines of the Declaration of Helsinki, and the principles of Good Clinical Practice as defined by the International Conference on Harmonization (ICH-E6, 17/07/96) as well as specific laws and regulations of the countries were the study is performed. Centers participating in BIO-RAIDs study are listed in the Table 1.

Eligible patients with stage IB2-IV disease, scheduled for primary surgery, chemo-radiation or primary chemotherapy are invited to participate in this study. Table 2 details the inclusion and exclusion criteria. Documented informed consent is obtained for all patients. Patient data is anonymized and recorded in an electronic Case Report Form (eCRF) (Quanticsoft). The eCRF generates patient specific unique barcode numbers for each sample. At study entry, baseline demographic characteristics, medical history and findings of staging such as complete physical and gynecological examination, abdominal and pelvic CT, pelvic MRI (+/- optional PET-CT) are recorded. A central review of MRI imaging is planned to be performed in Serbia.

Tumor and serum samples are collected at defined time points as shown in Table 3 and stored at −80 °C (mutational data) or at room T° in case of preparation of fixed paraffin embedded sections (IHC). Standard operating procedures have been established by the RAIDs consortium for biopsy handling as well as for blood and sera collections. All samples are centralized in the RAIDs biobank located in Morahollum (SeQomics, Hungary).

Early clinical follow up may vary according to the chosen treatment strategy and a function of the FIGO stage of the disease. Detailed clinical and imaging evaluations are carried out at the end of treatment and at 6, 12 and 18 months. All clinical, sampling and imaging data are registered in the eCRF.

Standardization for External Beam Radio-chemotherapy (EBRT) and Brachytherapy (BT) delineation in participating centers was improved through online workshops (ODW) according to the GEC-ESTRO recommendations [23] and the EMBRACE protocol (An International study on MRI-guided Brachytherapy in locally advanced CC: EMBRACE. https://​www.​embracestudy.​dk/​). Centers with deviations from the treatment protocol are being offered an on-site training period in a reference center. Contour evaluation methodology: Intraobserver variability between 3 contouring periods: 1. Quantitative differences: DICE index. 2. Qualitative assessment: target Volumes, areas contoured. Organs at Risk A: Optimal: > 0.81; B: Suboptimal: < = 0.81 (Breunig et al. IJROB 2012) A: Optimal: > 0.81 Target Vol. B: Average: 0.65 – 0.81 C: Suboptimal: < 0.65 (Dimopoulos et al. Radioth Oncol2009; Petersen et al. IJROB2007) Initial analyses demonstrated interobserver variability for baseline contouring. Quantitative analyses were performed between centers and years of experience. Qualitative analyses compared group contours with reference contours. ANOVA was applied for analysis based on DICE, the significance of: institution; organ at risk (OAR) versus target volume (TV); organ; participant’s years of experience (grouped in 2 levels: Residents vs. Specialists). As an example of results of contouring workshop: most centers have an average DICE index for each volume between 0.65 and 0.81. This loosely falls within the average (B) category. If the participants improve in the guideline and final contouring sessions, these centers are initially prepared to participate within the RAIDs study, and only would need to complete a dummy run to validate the dosimetry as well. Half of the RAIDs institutions have suboptimal (C) DICE indexes for the GTV-node volume. The qualitative analyses showed some conceptual errors: 1. delineating vessels or clinical nodal target volumes instead of the actual macroscopic lymph nodes. 2. these results are partly due to the clinical case itself: 1. paraaortic lymph node which on the computed tomography image set for contouring seemed enlarged, although the clinical information explicitly stated that no paraaortic lymph nodes were pathological. 2. suspicious lymph node in the left groin, which within the live sessions was deemed as inflammatory by an experienced radiotherapist.

Acute or delayed toxicities of treatment will be documented in the eCRF.

We try to assess which are the dominant mutations and which signaling pathways are activated in cervical cancers based on potentially deleterious genomic or proteomic alterations (COSMIC database). Furthermore we want to ascertain which abnormalities will be predictive of response to standard treatment and outcome. Analysis of a large number of patients is necessary because if a specific driver mutation is present in only 7 % of patients, 700 patients need to be evaluated to detect this mutation in 50 patients. No statistical hypothesis applies, but appropriate statistical methods will be used to analyze the results.

Complete response (CR) is assessed by MRI at the end of treatment (latest at 6 months) based on RECIST criteria. In the case of surgery, complete response will be defined as a pathological complete response. Progression-free survival (PFS) is defined as the time from diagnosis to the date of the first progression or death. If patients are alive and free of progression, they will be censored and their PFS will be defined as the time from diagnosis to the date of last known follow-up visit. Overall survival (S) will be defined as the time from diagnosis to the date of death or last follow-up. Survival rates will be estimated using the Kaplan-Meier method. PFS and OS will be compared to molecular phenotypes or clinical factors using the log-rank test.

Univariate and multivariate analyses will be performed to evaluate the association between each set of biomarkers (gain or loss of function) and clinical outcome. If the outcome measure is CR, a logistic regression is used. If the outcome measure is PFS or OS, a Cox regression model is preferred using 95 % confidence intervals. Multivariate analyses will take into account correlations between the different factors allowing to define a genomic or proteomic “response signature”, capable to predict the different outcome measures as compared to the presently available criteria. The level of significance is fixed at 5 %. Corrections for multiple testing to correct for the occurrence of false positives will be by the Benjamini-Hochberg method. Stratification according to geographical location will take into account the heterogeneity of standard treatment. Analyses will be performed using R° software by the Biostatistics Department of Institut Curie.

The bioinformatics’ platform at Institut Curie together with SeQomics (Hungary) will ensure reliable downstream bioinformatics analysis of patient samples. The tumors will be characterized by a list of features such as: mutations, structural variants, protein expression +/- phosphorylation, and presence of an immune signature by IHC among others. In an unsupervised analysis, an exhaustive exploration of the molecular profiles, without an a priori model will be carried out to characterize the dominant molecular abnormalities. This approach will involve a principal component analysis, an independent component analysis as well as clustering. This exploratory analysis is meant to detect possible biases due to technological aspects or sample handling (batch effect). The supervised analysis will attempt to identify and validate biomarkers using machine learning techniques such as LASSO, ridge regression, elastic net or SVM. The prediction of the influence of specific molecular abnormalities on patient outcome needs to be validated by their impact on the major endpoints which have been defined above: 1° Complete Response (CR). 2° Progression-free survival (PFS); 3° Overall survival (OS). Finally, results of both the unsupervised and supervised analyses will be compared to published classifications.

The biomarkers identified in the second step will be integrated with well-known clinical (FIGO stage, node involvement etc.) and histological prognostic factors in a multivariate model as defined in the first step. Our objectives are to study correlations and prioritize markers for their distinctive ability to predict complete response, progression free survival and overall survival.