Final Report Summary - WINTHER (WINTHERapeutics: development of a systems biology method to predict efficacy of cancer drugs to optimize individualized therapeutic decision and improve clinical outcome for cancer patients.)
Executive Summary:
WINTHER project set out to improve methods for predicting the efficacy of drugs in solid tumours cancer patients in a ground-breaking new approach to targeted therapies. Today, cancer-targeted therapies are primarily aimed at tackling detected oncogenic mutations or other DNA aberrations. However, only 40% of patients benefit from this approach, because targetable DNA anomaly cannot be detected in all cases. In WINTHER, a non randomized
proof of concept study, the traditional DNA-based approach was complemented with RNAbased investigation, effectively enabling personalised therapy for those patients with any kind of advanced solid malignant tumour that does not display any targetable DNA anomaly. To this effect, WINTHER introduced – and this was a first in a clinical trial – the concept of dual biopsies from the tumour and the normal tissue with which it had been matched to see how
messenger RNA (mRNA) is expressed differently in the two. The assumption made was that drugs targeting those tumour genes that showed the greatest difference compared with the normal tissue would be most effective.
The WINTHER international trial (http://clinicaltrials.gov/show/NCT01856296) involving 6 cancer centers across 5 countries (France, Spain, Israel, Canada and USA) already enabled significant learning for the future although the trial is still on going at the end of the two years period of the European grant and the two US sites were not able to open due to regulatory approval delays.
The dual biopsy of the tumour tissue and normal matched tissue has proven to be acceptable by patients and feasible in main tumour types.
Dual tumour/normal biopsies were safe and amongst the 174 patients who have had dual tumour/normal biopsies to date, there were no significant complications.
We were able to provide a personalised treatment decision for all patients who had passed the stringent quality controls based on either genomic or transcriptomic information (n=100). The Clinical Management Committees (CMC) to discuss each patient therapeutic option with all principal investigators in web-conferences with the support of a web portal displaying all the clinical, genomic and transcriptomic information has proven to be one of the major
achievements of the trial as they were very efficient and highly informative.
We were able to develop interventional radiology and laboratory Standard Operating Procedures optimizing the processes of dual biopsies, extraction of DNA and RNA from fresh frozen material to enable optimal investigation of the RNA in particular. Those procedures will be made available for sharing with the scientific community.
The trial experienced both regulatory approval and drug acquisition challenges but was mainly delayed due to the technical learning curve linked to using new information for the first time in the clinic. At the end of the EU two years project, 217 patients were consented in the trial. 174 patients had undergone the dual biopsies of tumour and normal matched tissue,
100 treatment decisions were performed in CMC, 46 patients had started the treatment, 33 patients were waiting for progression to start WINTHER therapy, 5 patients were waiting for treatment decisions and 6 were waiting for biopsies to be performed. The WINTHER trial remains opened with the objective to treat and follow-up at least one hundred patients in total. At the time of the report, follow-up data is available for 32 patients, who started the treatment. The best response was stable disease for 15, partial response for 4, 20 patients were still alive. Progression free survival measured from the consent date was not different between treatment decision based on genomic aberration and treatment decision based on transcriptomic information, nevertheless data are too preliminary to draw any statistical conclusion.
Ariana pharma was able to develop a commercial tool in the form of a secure web-based application to provide, in the future, physicians with a report to support therapeutic decision based on the RNA when no actionable DNA aberrations are detected by next generation sequencing.
Project Context and Objectives:
1. Context
Discoveries in molecular biology and the development of targeted therapies for cancer are delineating an important new concept: personalized treatment through matching of a patients’ tumour with one or more molecularly targeted agents via the use of biomarkers. High rates of efficacy have been reported in some settings with this approach. Nevertheless, most metastatic cancers have a dismal prognosis, and many drugs increase survival by only a few weeks or months. There is an emerging realization that treating unselected patient populations is unlikely to yield more than incremental benefits because cancers of the same histologic type are comprised of many molecular subgroups. The next generation sequencing technology coupled with potent targeted agents is undeniably bringing major progress by showing high regression rates, albeit in only limited oncogene-driven subsets of patients with metastatic cancers. Only 40% of patients benefit from this approach, because a targetable DNA anomaly cannot be detected in all cases. Therefore, the current state of the art does not enable to offer personalized therapeutics for the vast majority of patients, who do not display such oncogene drivers.
2. WINTHER trial objectives
In this context, an international team of researchers led by Prof. Jean-Charles Soria planned an ambitious proof of concept clinical trial, WINTHER, designed to treat in a personalized manner all eligible patients included in the trial by using both cutting-edge genomic and transcriptomic technologies to navigate patients with advanced refractory cancer to a matched drug. Because WIN is a global organization, an international trial was conceived that would leverage expertise across five countries (Canada (McGill/ Segal Cancer Centre), France (Gustave-Roussy (IGR), Israel (Sheba Cancer Research Center), Spain (Vall d’Hebron Institute of Oncology, VHIO), and USA (UC San Diego Cancer Center and MD Anderson Cancer Center, MDACC). The design of the trial was built on the PREDICT genomic trial at MD Anderson Cancer Center with the objective of 200 patients treated (100 in Europe and 100 in America).
In the WINTHER approach, the traditional DNA-based investigations was complemented with RNA-based investigation, effectively enabling personalised therapy for those patients with any kind of advanced solid malignant tumour that does not display any targetable DNA anomaly.
To this effect, WINTHER introduced – and this was a first in a clinical trial – the concept of dual biopsies from the tumour and the normal tissue with which it had been matched to see how messenger RNA (mRNA) is expressed differently in the two. The assumption made was that drugs targeting those tumour genes, which showed the greatest difference compared with the normal tissue would be most effective. When a tumour biopsy is investigated by measuring levels of mRNA, it is not possible to distinguish between the ‘genetic background noise’ of the patient and the useful information related to tumour-intrinsic abnormalities. Thanks to the dual biopsy, WINTHER was able to filter out this background noise.
Each patient with metastatic disease therefore consented to a dual biopsy: from both the metastasis and from the matched “Normal” tissue of origin of the tumour. As an example: in the case of a non-small cell lung cancer patient with a liver metastasis, a biopsy of the liver metastasis would be obtained as well as a sample of bronchial mucosa as the “normal counterpart”.
The comparison of tumour with its normal histologic counterpart was performed to remain only with the meaningful information. By this approach, transcriptomic explorations achieved were specific to each patient (normal and tumour cells belong to the same patient) and organ specific (complexity and noise were reduced to a specific organ and cell types), which enhance the accuracy of the measurements by discarding much of the noise.
The DNA of the metastasis (tumour) and RNA of both tumour and normal were then extracted from frozen tissues in the local center under common standard operating procedures and sent to centralized laboratories for a holistic set of ‘omics investigations to be used for treatment decision.
o Genomic investigations – Arm A
The DNA from the tumoral tissues were sent to Foundation Medicine (FMI), Cambridge, MA for next generation sequencing (NGS) and issuance of the Foundation One report outlining the genomic aberrations identified and matched targeted drugs recommendations. Patients were navigated to arm A in case of an actionable genomic aberration detected.
o Transcriptomic investigations – Arm B
Simultaneously, the RNA of both tumour and normal tissues were sent to Gustave Roussy (IGR) to perform microarrays based gene expression profiling performed on both tumour tissue samples and matched normal tissue in order to determine the fold change of mRNAs in Tumour vs. Normal. The distance between expression of the genes in the tumour and expression of the genes in the matched normal tissue of the patient was measured. The assumption taken was that the higher the difference (“distance”) the higher those genes are relevant to be targeted thus enabling higher chances that the drug targeting those genes will be efficient.
The WINTHER algorithm was then applied on the transcriptomic information of the patient in order to establish a list of drugs and presumed score of efficacy that matches the transcriptomic profile of the patient. This algorithm was based on the integration of a comprehensive knowledgebase of gene_drug interactions generated by reviewing the literature and extracting information from the Cytotoxicity Database (CTD), which was in turn generated by literature review {http://ctdbase.org/add}. Basically, were identified all the genes shown to be deregulated or in relation with drugs, and genes classified into major target genes, genes of resistance and minor targets (as example HER2 is the main target of Trastuzumab, VEGFA can be considered the target of Bevacizumab, etc.).The therapeutics included in the database were: (i) all compounds available in the clinical trial portfolio of the cancer centres participating in the trial; (ii) registered molecularly targeted agents; (iii) standard chemotherapeutic agents.
Patients were navigated to arm B in case no actionable genomic aberrations were detected by NGS.
3. Development of an industry tool
One important objective of the project was the development by Ariana Pharma of a software and bio-computing tools based on the clinically validated WINTHER 1.5 algorithms and developed according to industry standards so that a wider dissemination could be achieved through their commercialization. The tools are aimed to support the decision process leading to the patient treatment allocation according to the transcriptomic profile of the patients. This implies to develop robust and reliable software products, to set up a process for their maintenance and evolution, in full consideration of the industry standards and answering the regulatory requirements in this domain. The Winther 1.5 algorithm and the connected knowledge base have been incorporated into a web-based clinical decision system, Onco KEM®, developed by Ariana Pharma in the course of the WINTHER project with the objective of helping oncologists to select the best treatment for their patients.
4. Ancillary biomarker investigations
A secondary objective of the project was to identify blood-based immune-associated markers that could be employed for prognosis and therapeutics efficacy. More specifically, the goal was to develop a knowledge base of blood-based immune parameters that would be relevant for cancer prognosis and therapeutic efficacy. Then, to rationally hypothesize, which immune-related parameters could be associated with therapeutic efficacy based on therapeutic regimens tested in WINTHER.
Project Results:
1. Preamble
Multiple new processes were needed for an international, with numerous stakeholders, personalized trial that incorporated for the first time two major attributes: (i) introduction of the dual biopsy from tumoral and matched normal tissues; (ii) use of transcriptomic information for therapeutic decision. To attenuate these potential challenges, the trial was restricted to highly experienced institutions and/or investigators. Even so, initiating WINTHER faced considerable hurdles, which differed by country.
(i) Clinical Management - Regulatory
The most daunting obstacles to initiation were regulatory, especially in the USA, with the recent mandate for FDA oversight of the laboratory-based omics technologies in prospective clinical trials. As of March 2014, all sites had IRB approval and all except those in the USA have initiated. Timeline from protocol submission to IRB approval was about 1, 1, 4, 6, 9, and 10 months in France, Spain, USA-San Diego, Israel, Canada and USA-MD Anderson Cancer Center, respectively. Timeline from concept to activation was approximately 19 months in France, 22 months in Spain and Israel, and 30 months in Canada. The differences in timeline to activation related almost entirely to national regulatory requirements. At 2+ years, neither USA site has activated. In the USA, the FDA had requested a 400 pages Investigational Device Exemption (IDE) document be prepared. The approval was eventually obtained but too late to be able to activate the USA sites given the time to standardize processes in a Clinical Laboratory Improvement Amendments (CLIA) certified environment. CLIA regulate laboratory testing and require clinical laboratories to be certificated by their state as well as the Center for Medicare and Medicaid Services (CMS) before they can accept human samples for diagnostic testing. This has resulted in a 100 patients shortfall in patient accrual.
An important learning and consequence for future precision medicine trials is that regulatory agencies need to be informed and brought along in the design of future breakthrough trials such as WINTHER to avoid any delay in opening of the trial.
(ii) Clinical Management - Acquisition of molecularly guided drugs: The whole point of a personalized medicine trial is to individualize the drug(s) chosen for each particular patient. This means that, while the strategy to choose therapy is consistent between patients, the actual drugs administered will differ, based on the complexity of human tumour genomics. Therefore, access to a wide variety of approved or experimental agents is necessary. From the trial standpoint, approaching the drug manufacturers was difficult because patients might require any of a number of drugs from different sponsors. This represented a major rate- limiting step. Careful selection of sites that had many clinical trials with targeted agents available helped secure drugs for some, but not enough patients. In addition, contributions from a number of pharmaceutical companies and non for profit organization defrayed some of the cost of drug acquisition.
2. Technical and scientific results
(i) Trial biological investigations results
The major paradigm change in WINTHER was the use of two matched biopsies: from the tumour and matched normal tissue strictly from the same patient and same histological origin. This is in contrast to actual standard of care, which is only based on investigation of tumoral biopsies.
The use of tumoral biopsies only for gene expression leads to very low analytical performance because information concerning tumours is mixed with information of genetic background variability between individuals. Gene expression profiling of tumour biopsies cohorts requires thousands of specimens in order to identify useful information, validated statistically.
At entry in the study, biopsies of tumour and their normal matched counterpart as well as blood samples were collected as follow:
a. Tumour and normal matched tissues were obtained according to common Standard Operating Procedures (SOPs), guided by radiology or echography. Preferably 18 gauges trucut biopsies were advised to be used for tumours and appropriate procedures for the normal histologic counterpart (endoscopies and fibroscopies). The normal tissue is the tissue, which is the histological origin of primary tumours (example: normal colic mucosa for colon cancer, normal bronchial mucosa for lung cancers, normal muscle for rhabdomyosarcomas, etc.). The option of blood as the normal counterpart could not be retained due to the fundamental differences in term of transcriptome.
b. Biopsies used for the biological investigations were fresh frozen and not Formalin Fixed Paraffin Embedded (FFPE) as commonly used in the clinic. The reason for using fresh frozen was that FFPE samples show very poor result for the investigations of the RNA.
c. A histology control was performed by each center. The percentage of tumour cell threshold for the subsequent omics analyses was set to 60 % to ensure sufficient quantity and quality of the RNA in particular. Originally, macro dissection of the sample was planned to enrich the material in tumour cells but failed in most cases.
d. Normal and tumour biopsies were used to extract nucleic acids at each cancer center. Very stringent quality controls were put in place following common Standard Operating Procedures.
e. DNA was sent to Foundation Medicine, Cambridge, MA for next generation sequencing. A panel of 400 cancer relevant genes was sequenced using Illumina technology according to SOP of Foundation Medicine (CLIA Certified). This technology allows detection of mutations, all known translocations, and investigates copy number variations.
f. RNA was sent and used by the Genomic Platform at Gustave Roussy (ISO9001 certified laboratory) for gene expression for non-US clinical sites (gene expression analyses for US sites was planned to be done by Ambry Genetics, CLIA certified). The technology used was direct comparison of tumour and normal RNAs.
g. In addition to tumour and normal biopsies, a 20 ml blood sample was taken to store plasma and serum for further ancillary investigations done at Ben Gurion University of the Negev, Dr. Angel Porgador Laboratory.
Results of the various analyses were available within 4 to 6 weeks.
A usual turnaround time per patient from biopsy procedure to shipment of nucleic acid to Foundation Medicine or Gustave Roussy was all centers together about 20 days on average. The whole procedure could be done at the fastest in one week if both tumour and normal biopsies were done the same day.
One of the secondary objectives of the study was the optimization of the use of biopsies, and increase knowledge in handling biopsies of tumour and normal tissues and optimization of histological preparation and extraction of DNA and RNA from strictly the same tumour or normal cells.
a. Histology control optimization. This is one of the key features for all the clinical trials using biopsies. Content of tumour cells is crucial to be determined and enriched.
b. DNA suitable for detection of mutations and/ or amplifications through structural genomic approaches
c. RNA suitable for determination of gene expression profiling
All these three points have been achieved at the end of the two first years of the trial. The study being multicenter, standardization of the handling of biopsies from tumour and normal biopsies has been a key objective for the success in handling biological investigations. As a result, weekly laboratory conferences with the laboratory teams of each center have been convened. These teleconferences were used to follow sample processing, discuss any particular case, which needs advice in handling processing. They have been used to optimize and update laboratory SOPs established at the beginning of the study according to the practice and success of the laboratories. Those protocols have been updated to the state of the art of successful biopsy process to biological procedures. SOPs have been developed and optimized for standard biopsy procedure, histo-pathological control, preparation for and extraction of nucleic acids (DNA and RNA), each step being carefully controlled for quality before sending material for omics analyses.
Major pitfalls delayed completion of the study in the 2 years’ time frame allowed by the EU grant. The first year of the trial in each of the center can be considered as a learning / training phase to master all procedures / biological steps encompassing selection of the patient to be registered in the study to passing all the quality control checks for sequencing / gene expression related results reports.
Critical failure points in the whole process have been identified:
a. No normal tissue.
b. No tumour tissue.
c. Percentage of tumour cells in the tumour tissue < 60 %.
d. Bad quality of DNA. DNA could not be sent to Foundation Medicine.
e. Bad quality of RNA. RNA could not be used for gene expression analyses.
f. Gene expression analysis quality control failed.
Not reaching the requested percentage of tumour cells in the tumour tissue to pass the RNA quality control to be able to perform the gene expression experiment was the main reason of failure.
Normal biopsies were accepted well by patients and passed histological quality controls at 87%
WINTHER has proven to be a pioneering study on multiple levels and in particular training on the biopsy procedures, processing of frozen biopsies and nucleic acid extraction in multiple sites with very diverse level of experience. From the start of the study to the end of the project, the failure rates decreased with sites’ improved performance. Failures related to biopsy procedures went from 39% to 15% at the end of the project. Nevertheless, the progress made to master those steps in all centers has been far longer than anticipated.
(ii) Treatment decision by the Clinical Management Committee of physicians (CMC)
Clinical Management Committee (CMC) meetings to discuss each patients’ therapeutic options with all principal investigators in web conferences, with the support of a web portal displaying all the clinical, genomic and transcriptomic information, has proven to be very efficient and highly informative:
Data interpretation and report analysis were provided to the investigators with a comprehensive list of oncogenic events (if identified) for patients in ARM A, and with a table of drugs together with their predictive score based on WINTHER method, ARM B, enabling therapeutic decision making at weekly web based teleconference with PIs from each center (CMC). Those web conferences CMC were convened every Mondays and all principal investigators from all cancer centres participating in WINTHER were able to discuss each of the patient ready to be reviewed with both the genomic and transcriptomic reports. All the CMC were recorded and transcribed for future reference.
As an academic support tool to the clinical study and the CMC, Ben Gurion University of the Negev (BGU) had created a web portal that captured all the necessary information to be discussed during the CMC in particular: (i) Clinical information about the patient; (ii) Foundation One report on genomics results for arm A, (ii) Transcriptomic based report scoring drugs efficacy for Arm B with available detail for each of the scored drugs and relevant genes over expressed in the tumour of the patient as compared to normal; (iii) therapeutic choice agreed upon by the CMC and rationale; (iv) treatment administered to patient and start date.
The portal offered an organized framework to review all these files, an interactive viewer to explore the Arm B report, and an interface for recording CMC decisions.
(iii) Bioinformatics results
a. WINTHER score
The score was derived from a dual personal expression pattern (Tumour and Normal) based on a knowledge base, which collected existing evidence about genes-drug associations. The knowledge base was prepared by filtering entries from the Comparative Toxicogenomics Database (CTD) (www.ctdbase.org) and manually curating data from the literature according to a list of 170 chemical that were available at the medical centers participating in the WINTHER trial at the onset of the trial. Interactions described in CTD were filtered to select entries that (1) involved human genes or proteins, and (2) associated efficacy with expression (either at the mRNA or the protein level), either directly (expression was found to predict efficacy) or indirectly (changes in expression were associated with drug exposure). After filtering for chemicals available at the WINTHER trial centers, 10,304 interactions, representing 91 chemicals and 5,742 genes, were extracted from the CTD. Subsequently, we conducted manual review of the literature, focusing on drugs that were not found in the CTD. The review was initiated by searching PUBMED (http://www.ncbi.nlm.nih.gov/pubmed/) and/or Google scholar (http://scholar.google.com). Lastly, we manually extracted gene expression data from Gene Expression Omnibus datasets (http://www.ncbi.nlm.nih.gov/gds) that were published as supplementary data in relevant articles. The result of the manual process provided 1921 additional interactions, representing 83 additional chemicals that interacted with 1395 new and different genes. The data was planned to be updated periodically. The last update took place in August 2013.
The score was designed to have higher values for drugs that were expected to be more effective based on existing knowledge. Simply put, the score was designed to quantitatively estimate the compatibility between a patient and a drug based on prior knowledge from the literature.
Development of the next generation algorithm SIMS
Based on the observations and lessons learnt from WINTHER trial, we concluded that the optimized version of WINTHER algorithm would need to integrate both DNA, RNA and miRNA information to inform therapeutic decisions. Potentially, the algorithm should also enable to direct to rationally selected combinations of targeted therapies to achieve a more important impact on patients outcome. WINTHER trial was focused on mono therapies but the scientific community agrees that to tackle more efficiently mechanisms of resistance only the combination of drugs based on a systems biology approach could be more successful.
In order to develop the next generation algorithm, we focused on the most prevalent cancer type in WINTHER for which the dual biopsies procedures were manageable: non-small cell lung cancer (NSCLC). Because the biological and outcome data on WINTHER trial was not yet available on a sufficient number of patients, we studied in silico data generated and published by the CHEMORES initiative (www.chemores.org) which was an EU funded (FP6) Integrated Project. Tissue samples from a cohort of 121 patients who underwent complete surgical resection at the Institut Mutualiste Montsouris (Paris, France) between 30 January 2002 and 26 June 2006 were analysed.
The Simplified Interventional Mapping System (SIMS) algorithm was developed. SIMS merges molecular information with knowledge about the impact of drugs on the hallmarks of cancer. SIMS describes 183 key genes grouped into 24 interventional nodes that targeted drugs can act on. The targeted drugs include immune-modulators (anti-PDL1, anti PD1 and anti CTL4) as well as small molecules (TKI).
-The first difference with WINTHER algorithm lies in the transition operated from a gene by gene based interaction with each specific drug to a systems biology knowledge base that links specific interventional therapeutic interventional point to a class of drugs.
-The second difference with WINTHER algorithm is the integration of all genomic information.
SIMS algorithm generates a simple 1 to 10 scoring system that integrates thousands of omics investigations of the genes grouped in the 24 interventional points that can be efficiently targeted and blocked by a drug or a class of drugs. The scoring enables to rank the activation of the interventional points and to identify the trends of co-activation. The activation status is investigated for each patient through genomic investigations including DNA aberrations, CNV as well as differential mRNA and miRNA in matched tumour and normal dual biopsies obtained from the same patient.
The reporting is simplified and can be easily used by principal investigators in CMC either for assessment of mono-therapies or combinations.
The same strategy developed on the subset of NSCLC patients may also be applicable to other deadly malignancies.
b. Hosting web portal Raw data processing and integration
A secured web portal was built and hosted at BGU. Each Principal Investigator and relevant participants were provided with the https link and assigned a username and password. The data portal presents 4 reports for each patient: A clinical report: electronic Clinical Research Form (eCRF), the Foundation One™ report for mutations/CNV based drug prioritization, the Arm B report, which is generated using the WINTHER 1.5 score and a knowledgebase developed for this purpose to rank 173 drugs according to their expected efficacy, and - when available – the therapeutic decision made at the CMC. The portal offers an organized framework to review all these files, an interactive viewer to explore the Arm B report, and an interface for recording CMC decisions. A pipeline was established for the processing of raw data, its integration, and its fusion with prior knowledge.
c. Development of a clinically applicable interactive report for expression-based drug prioritization ("the Arm B / BGU report"):
An interactive viewer and Simple Expression Based Intervention Prioritizer (SEBIP) viewer (was developed). SEBIP was developed to offer nested levels of evidence: without additional actions, the viewer offered a ranked list of intervention, with best scoring therapeutic agents in the top and the lowest scoring therapies at the bottom. For each drug, a succinct summary of the evidence behind the rank was presented (in the form of a score, as well as a count of genes, which expression was concordant or discordant with the literature). A more detailed report about the evidence could be opened by simply following a link, which offered a comprehensive report explaining why a given drug received a high (or low) score.
(iv) Study statistical results
The WINTHER trial started on March 1st, 2013. The 1st patient was registered on April 19th, 2013. Today, accrual is still continuing and a total of 243 patients have been included. The present report concerns the 221 patients included in WINTHER trial before March, 1st 2015. Four patients who withdrew their consent are excluded from the analyses.
Sixteen patients had one or several criteria that should have blocked their inclusion. The proportion of patients with inclusion criteria that did not match the trial specification decreased with time since start of the trial. These patients were not excluded from the trial. The initial diagnosis of the patients was extremely heterogeneous; the most frequent primary cancers were lung tumors (26%). No other type of cancer had been present in more than 20% of the patients. At inclusion most of the patients had metastases, the most frequent metastases sites were lung (52%) and liver (46%).
Between the initial cancer diagnosis and the entry in Winther, the patients had received an average of 4 treatment lines (median 3; minimum 1; maximum 16). The median delay between initial diagnosis and inclusion was 25 months (minimum: 3, maximum: 245). Tumor tissue biopsies were performed on average at day 7 after inclusion and normal tissue biopsies at day 8. The report from Foundation medicine was available after a median of 37 days after biopsy and the BGU report after a median of 33 days. Treatment decisions occurred after a median of 8 days after FM report and 15 days after BGU report. One hundred patients had treatment decisions performed in Clinical Management Committees (CMC). 46 patients have started the treatment decided during the CMC, 30 had the treatment decision based on genomic aberration (arm A) and 16 based on transcriptomic information (arm B).
Concerning patients included before March 2015, 6 are still waiting for biopsy, and 5 are waiting for treatment decision and 33 for treatment start. The trial is still accruing; during the 2 months between March and May 2015, 22 patients have been included. The database is continuously updated and end of trial statistical assessment will be available upon recruitment and patient follow-up completion. Follow-up data are available for 32 patients, who started the treatment decided during the CMC. The best response was stable disease for 15, partial response for 4; 20 patients are still alive. Progression free survival measured from the consent date was not different between arm A and arm B. Data are too preliminary to draw any statistical conclusion at this point in time.
(v) Results Serum-omics and biomarkers – General and Immune-related parameters associated with cancer therapeutic efficacy
The main objective was to identify blood-based immune-associated markers that could be employed for prognosis and therapeutics efficacy. More specifically, sub-objectives included: developing of a knowledge base of blood-based immune parameters that was relevant for cancer prognosis and therapeutic efficacy. Then, to rationally hypothesize which immune-related parameters could be associated with therapeutic efficacy based on therapeutic regimens tested in WINTHER. After the establishment of parameter’s list, to experimentally measure this defined parameter subset in the blood samples deposited in the Bio bank of the clinical trial. Finally, to analyze the raw data for defining specific blood-based immune-related markers that correlate with progression-free survival (PFS) following WINTHER algorithm-based treatment.
Knowledge-base for blood-based immune-associated markers that were connected to the chemotherapeutic drugs employed by WINTHER was established. We chose the best 16 parameters and tested them experimentally in sera samples taken from WINTHER-enrolled and treated patients (sera were taken before beginning of treatment). At this stage, PFS data were not yet available and follow up of therapeutic efficacy will be performed until the end of the trial. Therefore, correlation with clinical outcome will be only available at the end of the follow-up. Here, we report biological correlations that have been identified to date.
A total of forty sera samples, taken before treatment from forty patients that were enrolled to WINTHER, were sent from four different centers (IGR, VHIO, CSM and CSS). It is important to mention that tumours were of different origin, lung, breast, colon, sarcomas etc.). This enables to obtain an interesting insight across different types of tumours, in relationship with the host. To our knowledge, this is the first report putting in perspective rather the malignant disease than a specific type of tumour.
We studied 16 blood analytes (cytokines). Only 10 cytokines (MMP.9 MMP. 3, IL.8 MMP-1, MCP-1, IL.1Ra P.Selectin G.CSF HGF, Gro alpha) showed detected and variable distribution for the cancer samples that we had and were further characterized. First, a significant correlation was observed between serum concentration of the cytokines and differential expression of certain genes in tumour vs. normal tissues (as determined with Agilent microarrays), Genes with correlation (P value < 0.001) in expression levels to serum cytokines were computed. Out of the 10 cytokines under study, 9 cytokines showed significant correlation with the differential expression of certain genes in tumoral vs. normal tissues (3 to 34 genes) while 1 cytokine, HGF, didn’t show any correlation. Data obtained for HGF suggested possible post-transcriptional regulation induced by miRNAs
This observation, obtained on 40 patients indicated that tumour was well the origin of the secreted cytokines, and their measurements met the criteria of cancer disease specificity.
Preliminary Conclusion:
These results demonstrated the power of the immune-reactome approach to detect distinct patterns of expression of cytokines in sera of cancer patients: this approach has until now never been integrated in correlative studies, and in particular in better understanding and characterization of therapeutic efficacy of patients entering a clinical trial.
(vi) SME industry tool results
Onco KEM® has been designed by Ariana as a secure and scalable web-based application for the commercial implementation and deployment of the WINTHER 1.5 algorithm for data-driven personalized medicine. Onco KEM® can thus be defined as a predictive rule based system supporting the therapeutic strategy decision according to each individual molecular profile.
The WINTHER report on treatment ranking provided to the clinician has been developed with a professional design that highlights key information about recommended ranked therapies and uses linked tables for progressive disclosure of more detailed information explaining the obtained scores and the considered drug-gene interactions.
A first “proof of concept” (PoC) deployment of Onco KEM® for the clinics participating in the WINTHER clinical trial is planned in 2015 during the extension of the trial and is expected to generate valuable clinical feedback to help future enhancements to Onco KEM®.
Initial market analysis performed by Ariana showed that Onco KEM® has a large market potential, as better tailoring chemotherapies in oncology remains a real medical need. In the perspective of the deployment of Onco KEM®, Ariana has built on a strategic marketing plan to ensure its efficient market entry compliant with regulations and market pressure.
a. Prototype development
During the first year of the project, the prototype of the WINTHER bio-computing tool has been specified and implemented using a standard flexible three-tier architecture pattern, providing data, application and presentation tiers.
At the heart of the tool is the WINTHER algorithm, executed by a high performance predictive analytics engine (KNIME Server) running alongside the database storage and retrieval component that presents the results to the final user (medical oncologists). In this prototype, KNIME application server (Glassfish J2EE) and KNIME Web Portal serve as the application and presentation tiers respectively.
The prototype development had focused on implementation of the core functionality (scoring algorithm and report generation), using anonymised patient data collected during the WINTHER clinical trial. Specifically, two automated workflows had been developed to (1) process new patient (consisting of four main tasks: gene expression data file upload, efficacy scores computing, results storage and patient report generation) and (2) generate reports form archived results (retrieving archived results and generating reports).
This work had led to the demonstration of the functional prototype usage through the standard web portal provided by KNIME serves used as a front-end. A MySQL database schema had been implemented to support the core workflows consisting of tables representing registered patients, drug-gene interaction knowledge base, patient files to process, and calculation results.
b. Final product development
The final product is a scalable hosted application and consists of 6 virtual machines, providing a combination of load balancing, redundancy and fault tolerance.
The final Onco KEM® platform is designed to support the following generic workflow:
o An Onco KEM® treatment selection report, based on a defined algorithm (here the WINTHER 1.5 algorithm), is requested on behalf of a patient.
o Tissue samples are obtained following pre-defined SOPs and sent to the lab for profiling.
o Patient profiling data, in a pre-defined format, are uploaded to Onco KEM®.
o Onco KEM® generates the treatment selection report using the specified algorithm, and notifies the clinician of the pending report.
The WINTHER treatment selection report provides a three linked levels of information:
o Summary of treatment scores based on expression data, showing scores for top 10 ranked treatments.
o Score explanations for each treatment, showing genes contributing to the score, and brief sentences extracted from literature describing the drug-gene interactions.
o Optional references table for each treatment, showing whole paragraphs describing the drug-gene interactions. The paragraphs are extracted from the primary scientific literature using text mining tools, Onco KEM® Builder, developed specifically for the WINTHER project.
c. Marketing Plan for the integrated tool
To ensure an efficient entry of Onco KEM® on the market and its implementation in the routine clinical practice, Ariana has defined a strategic marketing plan built on four activities with complementary objectives:
o Demonstration of clinical and cost effectiveness of the solution
o Regulatory compliance
o Business model and Reimbursement strategy
o Communication and Dissemination Plan
This strategic plan will enable highlighting the benefits provided by Onco KEM® to the patients’ medical outcome as well as to the health care systems. The goal will be to overcome economic hurdles in securing coverage and adequate reimbursement of the solution, and to specifically target and convince the relevant audiences for Onco KEM®.
By predicting and better tailoring chemotherapies in oncology, clearly defined as a real medical need by cancer key opinion leaders, Onco KEM® does have a large market potential and will target cancer patients that can undergo a biopsy of the cancer and a neighbouring tissue and whose oncologist needs to choose the appropriate therapy, either prior or after surgery. Taking into account the number of new cancer patients per year in Europe and the United States, the expected initial number of patients that can benefit from for Onco KEM® ranking report is ~ 700 000 per year
Potential Impact:
(i) Improved efficacy of therapeutics and clinical outcome: provide to a broader spectrum of cancer patients, irrespective of tumour type, a rational selection of drugs that target cancer biological abnormalities. Indeed, today the selection of cancer-targeted therapies is based on molecular abnormalities (mutations / translocations). However, this benefits only to a limited subset of patients. For the remaining majority, therapeutic decision is still based according to decade-old protocols with limited efficacy. Using transcriptomic information based on the novel approach of the distance between tumour and normal tissues enables patients without genomic aberrations to be treated in a personalized manner too. This novel approach developed by WINTHER offers therefore an increased number of patients that will benefit from a rational scientific selection of therapies based on molecular portraits and specific biology of the patient and the tumour.
(ii) Reducing the cost of clinical care: by improving patient’s allocation to efficient treatments and thus minimizing costs associated with inappropriate and ineffective treatments as well as diminishing adverse events or downstream complications, the industrial product Onco KEM® is expected to have major positive impact towards reducing healthcare costs. Policy makers, regulators and payers are more and more recognizing the benefits of personalized medicine tools and services but are particularly seeking additional evidence of their clinical and economic values.
(iii) Improved efficacy of drug development: by streamlining clinical trials and reduce time by directing patients to clinical trials of drugs which actions have reasons to fit the patient’s individual molecular abnormalities and genomic molecular profiling (instead of sending patients arbitrarily thus a staggering 95% failure rate of current clinical trials).
List of Websites:
http://clinicaltrials.gov/show/NCT01856296
WINTHER project set out to improve methods for predicting the efficacy of drugs in solid tumours cancer patients in a ground-breaking new approach to targeted therapies. Today, cancer-targeted therapies are primarily aimed at tackling detected oncogenic mutations or other DNA aberrations. However, only 40% of patients benefit from this approach, because targetable DNA anomaly cannot be detected in all cases. In WINTHER, a non randomized
proof of concept study, the traditional DNA-based approach was complemented with RNAbased investigation, effectively enabling personalised therapy for those patients with any kind of advanced solid malignant tumour that does not display any targetable DNA anomaly. To this effect, WINTHER introduced – and this was a first in a clinical trial – the concept of dual biopsies from the tumour and the normal tissue with which it had been matched to see how
messenger RNA (mRNA) is expressed differently in the two. The assumption made was that drugs targeting those tumour genes that showed the greatest difference compared with the normal tissue would be most effective.
The WINTHER international trial (http://clinicaltrials.gov/show/NCT01856296) involving 6 cancer centers across 5 countries (France, Spain, Israel, Canada and USA) already enabled significant learning for the future although the trial is still on going at the end of the two years period of the European grant and the two US sites were not able to open due to regulatory approval delays.
The dual biopsy of the tumour tissue and normal matched tissue has proven to be acceptable by patients and feasible in main tumour types.
Dual tumour/normal biopsies were safe and amongst the 174 patients who have had dual tumour/normal biopsies to date, there were no significant complications.
We were able to provide a personalised treatment decision for all patients who had passed the stringent quality controls based on either genomic or transcriptomic information (n=100). The Clinical Management Committees (CMC) to discuss each patient therapeutic option with all principal investigators in web-conferences with the support of a web portal displaying all the clinical, genomic and transcriptomic information has proven to be one of the major
achievements of the trial as they were very efficient and highly informative.
We were able to develop interventional radiology and laboratory Standard Operating Procedures optimizing the processes of dual biopsies, extraction of DNA and RNA from fresh frozen material to enable optimal investigation of the RNA in particular. Those procedures will be made available for sharing with the scientific community.
The trial experienced both regulatory approval and drug acquisition challenges but was mainly delayed due to the technical learning curve linked to using new information for the first time in the clinic. At the end of the EU two years project, 217 patients were consented in the trial. 174 patients had undergone the dual biopsies of tumour and normal matched tissue,
100 treatment decisions were performed in CMC, 46 patients had started the treatment, 33 patients were waiting for progression to start WINTHER therapy, 5 patients were waiting for treatment decisions and 6 were waiting for biopsies to be performed. The WINTHER trial remains opened with the objective to treat and follow-up at least one hundred patients in total. At the time of the report, follow-up data is available for 32 patients, who started the treatment. The best response was stable disease for 15, partial response for 4, 20 patients were still alive. Progression free survival measured from the consent date was not different between treatment decision based on genomic aberration and treatment decision based on transcriptomic information, nevertheless data are too preliminary to draw any statistical conclusion.
Ariana pharma was able to develop a commercial tool in the form of a secure web-based application to provide, in the future, physicians with a report to support therapeutic decision based on the RNA when no actionable DNA aberrations are detected by next generation sequencing.
Project Context and Objectives:
1. Context
Discoveries in molecular biology and the development of targeted therapies for cancer are delineating an important new concept: personalized treatment through matching of a patients’ tumour with one or more molecularly targeted agents via the use of biomarkers. High rates of efficacy have been reported in some settings with this approach. Nevertheless, most metastatic cancers have a dismal prognosis, and many drugs increase survival by only a few weeks or months. There is an emerging realization that treating unselected patient populations is unlikely to yield more than incremental benefits because cancers of the same histologic type are comprised of many molecular subgroups. The next generation sequencing technology coupled with potent targeted agents is undeniably bringing major progress by showing high regression rates, albeit in only limited oncogene-driven subsets of patients with metastatic cancers. Only 40% of patients benefit from this approach, because a targetable DNA anomaly cannot be detected in all cases. Therefore, the current state of the art does not enable to offer personalized therapeutics for the vast majority of patients, who do not display such oncogene drivers.
2. WINTHER trial objectives
In this context, an international team of researchers led by Prof. Jean-Charles Soria planned an ambitious proof of concept clinical trial, WINTHER, designed to treat in a personalized manner all eligible patients included in the trial by using both cutting-edge genomic and transcriptomic technologies to navigate patients with advanced refractory cancer to a matched drug. Because WIN is a global organization, an international trial was conceived that would leverage expertise across five countries (Canada (McGill/ Segal Cancer Centre), France (Gustave-Roussy (IGR), Israel (Sheba Cancer Research Center), Spain (Vall d’Hebron Institute of Oncology, VHIO), and USA (UC San Diego Cancer Center and MD Anderson Cancer Center, MDACC). The design of the trial was built on the PREDICT genomic trial at MD Anderson Cancer Center with the objective of 200 patients treated (100 in Europe and 100 in America).
In the WINTHER approach, the traditional DNA-based investigations was complemented with RNA-based investigation, effectively enabling personalised therapy for those patients with any kind of advanced solid malignant tumour that does not display any targetable DNA anomaly.
To this effect, WINTHER introduced – and this was a first in a clinical trial – the concept of dual biopsies from the tumour and the normal tissue with which it had been matched to see how messenger RNA (mRNA) is expressed differently in the two. The assumption made was that drugs targeting those tumour genes, which showed the greatest difference compared with the normal tissue would be most effective. When a tumour biopsy is investigated by measuring levels of mRNA, it is not possible to distinguish between the ‘genetic background noise’ of the patient and the useful information related to tumour-intrinsic abnormalities. Thanks to the dual biopsy, WINTHER was able to filter out this background noise.
Each patient with metastatic disease therefore consented to a dual biopsy: from both the metastasis and from the matched “Normal” tissue of origin of the tumour. As an example: in the case of a non-small cell lung cancer patient with a liver metastasis, a biopsy of the liver metastasis would be obtained as well as a sample of bronchial mucosa as the “normal counterpart”.
The comparison of tumour with its normal histologic counterpart was performed to remain only with the meaningful information. By this approach, transcriptomic explorations achieved were specific to each patient (normal and tumour cells belong to the same patient) and organ specific (complexity and noise were reduced to a specific organ and cell types), which enhance the accuracy of the measurements by discarding much of the noise.
The DNA of the metastasis (tumour) and RNA of both tumour and normal were then extracted from frozen tissues in the local center under common standard operating procedures and sent to centralized laboratories for a holistic set of ‘omics investigations to be used for treatment decision.
o Genomic investigations – Arm A
The DNA from the tumoral tissues were sent to Foundation Medicine (FMI), Cambridge, MA for next generation sequencing (NGS) and issuance of the Foundation One report outlining the genomic aberrations identified and matched targeted drugs recommendations. Patients were navigated to arm A in case of an actionable genomic aberration detected.
o Transcriptomic investigations – Arm B
Simultaneously, the RNA of both tumour and normal tissues were sent to Gustave Roussy (IGR) to perform microarrays based gene expression profiling performed on both tumour tissue samples and matched normal tissue in order to determine the fold change of mRNAs in Tumour vs. Normal. The distance between expression of the genes in the tumour and expression of the genes in the matched normal tissue of the patient was measured. The assumption taken was that the higher the difference (“distance”) the higher those genes are relevant to be targeted thus enabling higher chances that the drug targeting those genes will be efficient.
The WINTHER algorithm was then applied on the transcriptomic information of the patient in order to establish a list of drugs and presumed score of efficacy that matches the transcriptomic profile of the patient. This algorithm was based on the integration of a comprehensive knowledgebase of gene_drug interactions generated by reviewing the literature and extracting information from the Cytotoxicity Database (CTD), which was in turn generated by literature review {http://ctdbase.org/add}. Basically, were identified all the genes shown to be deregulated or in relation with drugs, and genes classified into major target genes, genes of resistance and minor targets (as example HER2 is the main target of Trastuzumab, VEGFA can be considered the target of Bevacizumab, etc.).The therapeutics included in the database were: (i) all compounds available in the clinical trial portfolio of the cancer centres participating in the trial; (ii) registered molecularly targeted agents; (iii) standard chemotherapeutic agents.
Patients were navigated to arm B in case no actionable genomic aberrations were detected by NGS.
3. Development of an industry tool
One important objective of the project was the development by Ariana Pharma of a software and bio-computing tools based on the clinically validated WINTHER 1.5 algorithms and developed according to industry standards so that a wider dissemination could be achieved through their commercialization. The tools are aimed to support the decision process leading to the patient treatment allocation according to the transcriptomic profile of the patients. This implies to develop robust and reliable software products, to set up a process for their maintenance and evolution, in full consideration of the industry standards and answering the regulatory requirements in this domain. The Winther 1.5 algorithm and the connected knowledge base have been incorporated into a web-based clinical decision system, Onco KEM®, developed by Ariana Pharma in the course of the WINTHER project with the objective of helping oncologists to select the best treatment for their patients.
4. Ancillary biomarker investigations
A secondary objective of the project was to identify blood-based immune-associated markers that could be employed for prognosis and therapeutics efficacy. More specifically, the goal was to develop a knowledge base of blood-based immune parameters that would be relevant for cancer prognosis and therapeutic efficacy. Then, to rationally hypothesize, which immune-related parameters could be associated with therapeutic efficacy based on therapeutic regimens tested in WINTHER.
Project Results:
1. Preamble
Multiple new processes were needed for an international, with numerous stakeholders, personalized trial that incorporated for the first time two major attributes: (i) introduction of the dual biopsy from tumoral and matched normal tissues; (ii) use of transcriptomic information for therapeutic decision. To attenuate these potential challenges, the trial was restricted to highly experienced institutions and/or investigators. Even so, initiating WINTHER faced considerable hurdles, which differed by country.
(i) Clinical Management - Regulatory
The most daunting obstacles to initiation were regulatory, especially in the USA, with the recent mandate for FDA oversight of the laboratory-based omics technologies in prospective clinical trials. As of March 2014, all sites had IRB approval and all except those in the USA have initiated. Timeline from protocol submission to IRB approval was about 1, 1, 4, 6, 9, and 10 months in France, Spain, USA-San Diego, Israel, Canada and USA-MD Anderson Cancer Center, respectively. Timeline from concept to activation was approximately 19 months in France, 22 months in Spain and Israel, and 30 months in Canada. The differences in timeline to activation related almost entirely to national regulatory requirements. At 2+ years, neither USA site has activated. In the USA, the FDA had requested a 400 pages Investigational Device Exemption (IDE) document be prepared. The approval was eventually obtained but too late to be able to activate the USA sites given the time to standardize processes in a Clinical Laboratory Improvement Amendments (CLIA) certified environment. CLIA regulate laboratory testing and require clinical laboratories to be certificated by their state as well as the Center for Medicare and Medicaid Services (CMS) before they can accept human samples for diagnostic testing. This has resulted in a 100 patients shortfall in patient accrual.
An important learning and consequence for future precision medicine trials is that regulatory agencies need to be informed and brought along in the design of future breakthrough trials such as WINTHER to avoid any delay in opening of the trial.
(ii) Clinical Management - Acquisition of molecularly guided drugs: The whole point of a personalized medicine trial is to individualize the drug(s) chosen for each particular patient. This means that, while the strategy to choose therapy is consistent between patients, the actual drugs administered will differ, based on the complexity of human tumour genomics. Therefore, access to a wide variety of approved or experimental agents is necessary. From the trial standpoint, approaching the drug manufacturers was difficult because patients might require any of a number of drugs from different sponsors. This represented a major rate- limiting step. Careful selection of sites that had many clinical trials with targeted agents available helped secure drugs for some, but not enough patients. In addition, contributions from a number of pharmaceutical companies and non for profit organization defrayed some of the cost of drug acquisition.
2. Technical and scientific results
(i) Trial biological investigations results
The major paradigm change in WINTHER was the use of two matched biopsies: from the tumour and matched normal tissue strictly from the same patient and same histological origin. This is in contrast to actual standard of care, which is only based on investigation of tumoral biopsies.
The use of tumoral biopsies only for gene expression leads to very low analytical performance because information concerning tumours is mixed with information of genetic background variability between individuals. Gene expression profiling of tumour biopsies cohorts requires thousands of specimens in order to identify useful information, validated statistically.
At entry in the study, biopsies of tumour and their normal matched counterpart as well as blood samples were collected as follow:
a. Tumour and normal matched tissues were obtained according to common Standard Operating Procedures (SOPs), guided by radiology or echography. Preferably 18 gauges trucut biopsies were advised to be used for tumours and appropriate procedures for the normal histologic counterpart (endoscopies and fibroscopies). The normal tissue is the tissue, which is the histological origin of primary tumours (example: normal colic mucosa for colon cancer, normal bronchial mucosa for lung cancers, normal muscle for rhabdomyosarcomas, etc.). The option of blood as the normal counterpart could not be retained due to the fundamental differences in term of transcriptome.
b. Biopsies used for the biological investigations were fresh frozen and not Formalin Fixed Paraffin Embedded (FFPE) as commonly used in the clinic. The reason for using fresh frozen was that FFPE samples show very poor result for the investigations of the RNA.
c. A histology control was performed by each center. The percentage of tumour cell threshold for the subsequent omics analyses was set to 60 % to ensure sufficient quantity and quality of the RNA in particular. Originally, macro dissection of the sample was planned to enrich the material in tumour cells but failed in most cases.
d. Normal and tumour biopsies were used to extract nucleic acids at each cancer center. Very stringent quality controls were put in place following common Standard Operating Procedures.
e. DNA was sent to Foundation Medicine, Cambridge, MA for next generation sequencing. A panel of 400 cancer relevant genes was sequenced using Illumina technology according to SOP of Foundation Medicine (CLIA Certified). This technology allows detection of mutations, all known translocations, and investigates copy number variations.
f. RNA was sent and used by the Genomic Platform at Gustave Roussy (ISO9001 certified laboratory) for gene expression for non-US clinical sites (gene expression analyses for US sites was planned to be done by Ambry Genetics, CLIA certified). The technology used was direct comparison of tumour and normal RNAs.
g. In addition to tumour and normal biopsies, a 20 ml blood sample was taken to store plasma and serum for further ancillary investigations done at Ben Gurion University of the Negev, Dr. Angel Porgador Laboratory.
Results of the various analyses were available within 4 to 6 weeks.
A usual turnaround time per patient from biopsy procedure to shipment of nucleic acid to Foundation Medicine or Gustave Roussy was all centers together about 20 days on average. The whole procedure could be done at the fastest in one week if both tumour and normal biopsies were done the same day.
One of the secondary objectives of the study was the optimization of the use of biopsies, and increase knowledge in handling biopsies of tumour and normal tissues and optimization of histological preparation and extraction of DNA and RNA from strictly the same tumour or normal cells.
a. Histology control optimization. This is one of the key features for all the clinical trials using biopsies. Content of tumour cells is crucial to be determined and enriched.
b. DNA suitable for detection of mutations and/ or amplifications through structural genomic approaches
c. RNA suitable for determination of gene expression profiling
All these three points have been achieved at the end of the two first years of the trial. The study being multicenter, standardization of the handling of biopsies from tumour and normal biopsies has been a key objective for the success in handling biological investigations. As a result, weekly laboratory conferences with the laboratory teams of each center have been convened. These teleconferences were used to follow sample processing, discuss any particular case, which needs advice in handling processing. They have been used to optimize and update laboratory SOPs established at the beginning of the study according to the practice and success of the laboratories. Those protocols have been updated to the state of the art of successful biopsy process to biological procedures. SOPs have been developed and optimized for standard biopsy procedure, histo-pathological control, preparation for and extraction of nucleic acids (DNA and RNA), each step being carefully controlled for quality before sending material for omics analyses.
Major pitfalls delayed completion of the study in the 2 years’ time frame allowed by the EU grant. The first year of the trial in each of the center can be considered as a learning / training phase to master all procedures / biological steps encompassing selection of the patient to be registered in the study to passing all the quality control checks for sequencing / gene expression related results reports.
Critical failure points in the whole process have been identified:
a. No normal tissue.
b. No tumour tissue.
c. Percentage of tumour cells in the tumour tissue < 60 %.
d. Bad quality of DNA. DNA could not be sent to Foundation Medicine.
e. Bad quality of RNA. RNA could not be used for gene expression analyses.
f. Gene expression analysis quality control failed.
Not reaching the requested percentage of tumour cells in the tumour tissue to pass the RNA quality control to be able to perform the gene expression experiment was the main reason of failure.
Normal biopsies were accepted well by patients and passed histological quality controls at 87%
WINTHER has proven to be a pioneering study on multiple levels and in particular training on the biopsy procedures, processing of frozen biopsies and nucleic acid extraction in multiple sites with very diverse level of experience. From the start of the study to the end of the project, the failure rates decreased with sites’ improved performance. Failures related to biopsy procedures went from 39% to 15% at the end of the project. Nevertheless, the progress made to master those steps in all centers has been far longer than anticipated.
(ii) Treatment decision by the Clinical Management Committee of physicians (CMC)
Clinical Management Committee (CMC) meetings to discuss each patients’ therapeutic options with all principal investigators in web conferences, with the support of a web portal displaying all the clinical, genomic and transcriptomic information, has proven to be very efficient and highly informative:
Data interpretation and report analysis were provided to the investigators with a comprehensive list of oncogenic events (if identified) for patients in ARM A, and with a table of drugs together with their predictive score based on WINTHER method, ARM B, enabling therapeutic decision making at weekly web based teleconference with PIs from each center (CMC). Those web conferences CMC were convened every Mondays and all principal investigators from all cancer centres participating in WINTHER were able to discuss each of the patient ready to be reviewed with both the genomic and transcriptomic reports. All the CMC were recorded and transcribed for future reference.
As an academic support tool to the clinical study and the CMC, Ben Gurion University of the Negev (BGU) had created a web portal that captured all the necessary information to be discussed during the CMC in particular: (i) Clinical information about the patient; (ii) Foundation One report on genomics results for arm A, (ii) Transcriptomic based report scoring drugs efficacy for Arm B with available detail for each of the scored drugs and relevant genes over expressed in the tumour of the patient as compared to normal; (iii) therapeutic choice agreed upon by the CMC and rationale; (iv) treatment administered to patient and start date.
The portal offered an organized framework to review all these files, an interactive viewer to explore the Arm B report, and an interface for recording CMC decisions.
(iii) Bioinformatics results
a. WINTHER score
The score was derived from a dual personal expression pattern (Tumour and Normal) based on a knowledge base, which collected existing evidence about genes-drug associations. The knowledge base was prepared by filtering entries from the Comparative Toxicogenomics Database (CTD) (www.ctdbase.org) and manually curating data from the literature according to a list of 170 chemical that were available at the medical centers participating in the WINTHER trial at the onset of the trial. Interactions described in CTD were filtered to select entries that (1) involved human genes or proteins, and (2) associated efficacy with expression (either at the mRNA or the protein level), either directly (expression was found to predict efficacy) or indirectly (changes in expression were associated with drug exposure). After filtering for chemicals available at the WINTHER trial centers, 10,304 interactions, representing 91 chemicals and 5,742 genes, were extracted from the CTD. Subsequently, we conducted manual review of the literature, focusing on drugs that were not found in the CTD. The review was initiated by searching PUBMED (http://www.ncbi.nlm.nih.gov/pubmed/) and/or Google scholar (http://scholar.google.com). Lastly, we manually extracted gene expression data from Gene Expression Omnibus datasets (http://www.ncbi.nlm.nih.gov/gds) that were published as supplementary data in relevant articles. The result of the manual process provided 1921 additional interactions, representing 83 additional chemicals that interacted with 1395 new and different genes. The data was planned to be updated periodically. The last update took place in August 2013.
The score was designed to have higher values for drugs that were expected to be more effective based on existing knowledge. Simply put, the score was designed to quantitatively estimate the compatibility between a patient and a drug based on prior knowledge from the literature.
Development of the next generation algorithm SIMS
Based on the observations and lessons learnt from WINTHER trial, we concluded that the optimized version of WINTHER algorithm would need to integrate both DNA, RNA and miRNA information to inform therapeutic decisions. Potentially, the algorithm should also enable to direct to rationally selected combinations of targeted therapies to achieve a more important impact on patients outcome. WINTHER trial was focused on mono therapies but the scientific community agrees that to tackle more efficiently mechanisms of resistance only the combination of drugs based on a systems biology approach could be more successful.
In order to develop the next generation algorithm, we focused on the most prevalent cancer type in WINTHER for which the dual biopsies procedures were manageable: non-small cell lung cancer (NSCLC). Because the biological and outcome data on WINTHER trial was not yet available on a sufficient number of patients, we studied in silico data generated and published by the CHEMORES initiative (www.chemores.org) which was an EU funded (FP6) Integrated Project. Tissue samples from a cohort of 121 patients who underwent complete surgical resection at the Institut Mutualiste Montsouris (Paris, France) between 30 January 2002 and 26 June 2006 were analysed.
The Simplified Interventional Mapping System (SIMS) algorithm was developed. SIMS merges molecular information with knowledge about the impact of drugs on the hallmarks of cancer. SIMS describes 183 key genes grouped into 24 interventional nodes that targeted drugs can act on. The targeted drugs include immune-modulators (anti-PDL1, anti PD1 and anti CTL4) as well as small molecules (TKI).
-The first difference with WINTHER algorithm lies in the transition operated from a gene by gene based interaction with each specific drug to a systems biology knowledge base that links specific interventional therapeutic interventional point to a class of drugs.
-The second difference with WINTHER algorithm is the integration of all genomic information.
SIMS algorithm generates a simple 1 to 10 scoring system that integrates thousands of omics investigations of the genes grouped in the 24 interventional points that can be efficiently targeted and blocked by a drug or a class of drugs. The scoring enables to rank the activation of the interventional points and to identify the trends of co-activation. The activation status is investigated for each patient through genomic investigations including DNA aberrations, CNV as well as differential mRNA and miRNA in matched tumour and normal dual biopsies obtained from the same patient.
The reporting is simplified and can be easily used by principal investigators in CMC either for assessment of mono-therapies or combinations.
The same strategy developed on the subset of NSCLC patients may also be applicable to other deadly malignancies.
b. Hosting web portal Raw data processing and integration
A secured web portal was built and hosted at BGU. Each Principal Investigator and relevant participants were provided with the https link and assigned a username and password. The data portal presents 4 reports for each patient: A clinical report: electronic Clinical Research Form (eCRF), the Foundation One™ report for mutations/CNV based drug prioritization, the Arm B report, which is generated using the WINTHER 1.5 score and a knowledgebase developed for this purpose to rank 173 drugs according to their expected efficacy, and - when available – the therapeutic decision made at the CMC. The portal offers an organized framework to review all these files, an interactive viewer to explore the Arm B report, and an interface for recording CMC decisions. A pipeline was established for the processing of raw data, its integration, and its fusion with prior knowledge.
c. Development of a clinically applicable interactive report for expression-based drug prioritization ("the Arm B / BGU report"):
An interactive viewer and Simple Expression Based Intervention Prioritizer (SEBIP) viewer (was developed). SEBIP was developed to offer nested levels of evidence: without additional actions, the viewer offered a ranked list of intervention, with best scoring therapeutic agents in the top and the lowest scoring therapies at the bottom. For each drug, a succinct summary of the evidence behind the rank was presented (in the form of a score, as well as a count of genes, which expression was concordant or discordant with the literature). A more detailed report about the evidence could be opened by simply following a link, which offered a comprehensive report explaining why a given drug received a high (or low) score.
(iv) Study statistical results
The WINTHER trial started on March 1st, 2013. The 1st patient was registered on April 19th, 2013. Today, accrual is still continuing and a total of 243 patients have been included. The present report concerns the 221 patients included in WINTHER trial before March, 1st 2015. Four patients who withdrew their consent are excluded from the analyses.
Sixteen patients had one or several criteria that should have blocked their inclusion. The proportion of patients with inclusion criteria that did not match the trial specification decreased with time since start of the trial. These patients were not excluded from the trial. The initial diagnosis of the patients was extremely heterogeneous; the most frequent primary cancers were lung tumors (26%). No other type of cancer had been present in more than 20% of the patients. At inclusion most of the patients had metastases, the most frequent metastases sites were lung (52%) and liver (46%).
Between the initial cancer diagnosis and the entry in Winther, the patients had received an average of 4 treatment lines (median 3; minimum 1; maximum 16). The median delay between initial diagnosis and inclusion was 25 months (minimum: 3, maximum: 245). Tumor tissue biopsies were performed on average at day 7 after inclusion and normal tissue biopsies at day 8. The report from Foundation medicine was available after a median of 37 days after biopsy and the BGU report after a median of 33 days. Treatment decisions occurred after a median of 8 days after FM report and 15 days after BGU report. One hundred patients had treatment decisions performed in Clinical Management Committees (CMC). 46 patients have started the treatment decided during the CMC, 30 had the treatment decision based on genomic aberration (arm A) and 16 based on transcriptomic information (arm B).
Concerning patients included before March 2015, 6 are still waiting for biopsy, and 5 are waiting for treatment decision and 33 for treatment start. The trial is still accruing; during the 2 months between March and May 2015, 22 patients have been included. The database is continuously updated and end of trial statistical assessment will be available upon recruitment and patient follow-up completion. Follow-up data are available for 32 patients, who started the treatment decided during the CMC. The best response was stable disease for 15, partial response for 4; 20 patients are still alive. Progression free survival measured from the consent date was not different between arm A and arm B. Data are too preliminary to draw any statistical conclusion at this point in time.
(v) Results Serum-omics and biomarkers – General and Immune-related parameters associated with cancer therapeutic efficacy
The main objective was to identify blood-based immune-associated markers that could be employed for prognosis and therapeutics efficacy. More specifically, sub-objectives included: developing of a knowledge base of blood-based immune parameters that was relevant for cancer prognosis and therapeutic efficacy. Then, to rationally hypothesize which immune-related parameters could be associated with therapeutic efficacy based on therapeutic regimens tested in WINTHER. After the establishment of parameter’s list, to experimentally measure this defined parameter subset in the blood samples deposited in the Bio bank of the clinical trial. Finally, to analyze the raw data for defining specific blood-based immune-related markers that correlate with progression-free survival (PFS) following WINTHER algorithm-based treatment.
Knowledge-base for blood-based immune-associated markers that were connected to the chemotherapeutic drugs employed by WINTHER was established. We chose the best 16 parameters and tested them experimentally in sera samples taken from WINTHER-enrolled and treated patients (sera were taken before beginning of treatment). At this stage, PFS data were not yet available and follow up of therapeutic efficacy will be performed until the end of the trial. Therefore, correlation with clinical outcome will be only available at the end of the follow-up. Here, we report biological correlations that have been identified to date.
A total of forty sera samples, taken before treatment from forty patients that were enrolled to WINTHER, were sent from four different centers (IGR, VHIO, CSM and CSS). It is important to mention that tumours were of different origin, lung, breast, colon, sarcomas etc.). This enables to obtain an interesting insight across different types of tumours, in relationship with the host. To our knowledge, this is the first report putting in perspective rather the malignant disease than a specific type of tumour.
We studied 16 blood analytes (cytokines). Only 10 cytokines (MMP.9 MMP. 3, IL.8 MMP-1, MCP-1, IL.1Ra P.Selectin G.CSF HGF, Gro alpha) showed detected and variable distribution for the cancer samples that we had and were further characterized. First, a significant correlation was observed between serum concentration of the cytokines and differential expression of certain genes in tumour vs. normal tissues (as determined with Agilent microarrays), Genes with correlation (P value < 0.001) in expression levels to serum cytokines were computed. Out of the 10 cytokines under study, 9 cytokines showed significant correlation with the differential expression of certain genes in tumoral vs. normal tissues (3 to 34 genes) while 1 cytokine, HGF, didn’t show any correlation. Data obtained for HGF suggested possible post-transcriptional regulation induced by miRNAs
This observation, obtained on 40 patients indicated that tumour was well the origin of the secreted cytokines, and their measurements met the criteria of cancer disease specificity.
Preliminary Conclusion:
These results demonstrated the power of the immune-reactome approach to detect distinct patterns of expression of cytokines in sera of cancer patients: this approach has until now never been integrated in correlative studies, and in particular in better understanding and characterization of therapeutic efficacy of patients entering a clinical trial.
(vi) SME industry tool results
Onco KEM® has been designed by Ariana as a secure and scalable web-based application for the commercial implementation and deployment of the WINTHER 1.5 algorithm for data-driven personalized medicine. Onco KEM® can thus be defined as a predictive rule based system supporting the therapeutic strategy decision according to each individual molecular profile.
The WINTHER report on treatment ranking provided to the clinician has been developed with a professional design that highlights key information about recommended ranked therapies and uses linked tables for progressive disclosure of more detailed information explaining the obtained scores and the considered drug-gene interactions.
A first “proof of concept” (PoC) deployment of Onco KEM® for the clinics participating in the WINTHER clinical trial is planned in 2015 during the extension of the trial and is expected to generate valuable clinical feedback to help future enhancements to Onco KEM®.
Initial market analysis performed by Ariana showed that Onco KEM® has a large market potential, as better tailoring chemotherapies in oncology remains a real medical need. In the perspective of the deployment of Onco KEM®, Ariana has built on a strategic marketing plan to ensure its efficient market entry compliant with regulations and market pressure.
a. Prototype development
During the first year of the project, the prototype of the WINTHER bio-computing tool has been specified and implemented using a standard flexible three-tier architecture pattern, providing data, application and presentation tiers.
At the heart of the tool is the WINTHER algorithm, executed by a high performance predictive analytics engine (KNIME Server) running alongside the database storage and retrieval component that presents the results to the final user (medical oncologists). In this prototype, KNIME application server (Glassfish J2EE) and KNIME Web Portal serve as the application and presentation tiers respectively.
The prototype development had focused on implementation of the core functionality (scoring algorithm and report generation), using anonymised patient data collected during the WINTHER clinical trial. Specifically, two automated workflows had been developed to (1) process new patient (consisting of four main tasks: gene expression data file upload, efficacy scores computing, results storage and patient report generation) and (2) generate reports form archived results (retrieving archived results and generating reports).
This work had led to the demonstration of the functional prototype usage through the standard web portal provided by KNIME serves used as a front-end. A MySQL database schema had been implemented to support the core workflows consisting of tables representing registered patients, drug-gene interaction knowledge base, patient files to process, and calculation results.
b. Final product development
The final product is a scalable hosted application and consists of 6 virtual machines, providing a combination of load balancing, redundancy and fault tolerance.
The final Onco KEM® platform is designed to support the following generic workflow:
o An Onco KEM® treatment selection report, based on a defined algorithm (here the WINTHER 1.5 algorithm), is requested on behalf of a patient.
o Tissue samples are obtained following pre-defined SOPs and sent to the lab for profiling.
o Patient profiling data, in a pre-defined format, are uploaded to Onco KEM®.
o Onco KEM® generates the treatment selection report using the specified algorithm, and notifies the clinician of the pending report.
The WINTHER treatment selection report provides a three linked levels of information:
o Summary of treatment scores based on expression data, showing scores for top 10 ranked treatments.
o Score explanations for each treatment, showing genes contributing to the score, and brief sentences extracted from literature describing the drug-gene interactions.
o Optional references table for each treatment, showing whole paragraphs describing the drug-gene interactions. The paragraphs are extracted from the primary scientific literature using text mining tools, Onco KEM® Builder, developed specifically for the WINTHER project.
c. Marketing Plan for the integrated tool
To ensure an efficient entry of Onco KEM® on the market and its implementation in the routine clinical practice, Ariana has defined a strategic marketing plan built on four activities with complementary objectives:
o Demonstration of clinical and cost effectiveness of the solution
o Regulatory compliance
o Business model and Reimbursement strategy
o Communication and Dissemination Plan
This strategic plan will enable highlighting the benefits provided by Onco KEM® to the patients’ medical outcome as well as to the health care systems. The goal will be to overcome economic hurdles in securing coverage and adequate reimbursement of the solution, and to specifically target and convince the relevant audiences for Onco KEM®.
By predicting and better tailoring chemotherapies in oncology, clearly defined as a real medical need by cancer key opinion leaders, Onco KEM® does have a large market potential and will target cancer patients that can undergo a biopsy of the cancer and a neighbouring tissue and whose oncologist needs to choose the appropriate therapy, either prior or after surgery. Taking into account the number of new cancer patients per year in Europe and the United States, the expected initial number of patients that can benefit from for Onco KEM® ranking report is ~ 700 000 per year
Potential Impact:
(i) Improved efficacy of therapeutics and clinical outcome: provide to a broader spectrum of cancer patients, irrespective of tumour type, a rational selection of drugs that target cancer biological abnormalities. Indeed, today the selection of cancer-targeted therapies is based on molecular abnormalities (mutations / translocations). However, this benefits only to a limited subset of patients. For the remaining majority, therapeutic decision is still based according to decade-old protocols with limited efficacy. Using transcriptomic information based on the novel approach of the distance between tumour and normal tissues enables patients without genomic aberrations to be treated in a personalized manner too. This novel approach developed by WINTHER offers therefore an increased number of patients that will benefit from a rational scientific selection of therapies based on molecular portraits and specific biology of the patient and the tumour.
(ii) Reducing the cost of clinical care: by improving patient’s allocation to efficient treatments and thus minimizing costs associated with inappropriate and ineffective treatments as well as diminishing adverse events or downstream complications, the industrial product Onco KEM® is expected to have major positive impact towards reducing healthcare costs. Policy makers, regulators and payers are more and more recognizing the benefits of personalized medicine tools and services but are particularly seeking additional evidence of their clinical and economic values.
(iii) Improved efficacy of drug development: by streamlining clinical trials and reduce time by directing patients to clinical trials of drugs which actions have reasons to fit the patient’s individual molecular abnormalities and genomic molecular profiling (instead of sending patients arbitrarily thus a staggering 95% failure rate of current clinical trials).
List of Websites:
http://clinicaltrials.gov/show/NCT01856296