Towards useful cancer biomarkers to improve care for cancer

By Tobias Sjöblom, Uppsala University, Department of Immunology, Genetics and Pathology: Experimental and Clinical Oncology Henrik Rönnberg, Swedish Unviersity of Agricultural Science, the Faculty of Veterinary Medicine and Animal Science

Early discovery followed by efficient surgical resection of a local tumour translates into high long-term survival for people with cancer. Furthermore, the effectiveness of cancer care would greatly increase if there were simple tests that determine if a tumour will recur or respond to a specific treatment. Blood biomarkers for early detection, as well as prediction of drug response or disease recurrence, have therefore been intense areas of research for the past decades. However, progress has been limited as evidenced by the low rate of regulatory approval of such tests.

In this workshop, we will discuss measures to enhance and accelerate biomarker discovery and validation processes to improve effective transition into clinical use. A breakthrough in this field would be of great interest to many actors: oncologists, researchers, regulators, funders, the pharmaceutical industry, etc. A greater range of proven biomarkers for all types of cancer, if quickly and robustly discovered, validated and put to clinical use, will transform the lives of millions of patients by greatly improving cancer detection, treatment selection and knowledge in prognosis decisive for follow-up and decisions on adjuvant treatment. Much more effective ‘discovery to application’ processes would make better use of scarce funds, precious clinical samples, focus regulatory attention, target scientific effort, enhance academic-industry linkages, and hasten product to market business cycles.   

Questions to be addressed in the workshop are:

Which forms of cancer are most in need of diagnostic biomarkers versus prognostic versus predictive biomarkers?

How should we improve the design of academic and industrial biomarker discovery programmes to better address the criteria for regulatory approval?   

How can we most effectively build a sound, best-practice platform for biomarker development?  

What are the benefits of comparative oncology for accelerated biomarker discovery and application?

How can comparative models (non-rodent mammals) enhance biomarker discovery and validation? 

What is a biomarker and how is its performance assessed?

A biomarker is a characteristic in the body or bodily products that can be measured objectively. Tumour biomarkers can be diagnostic, prognostic and/or predictive.

Diagnostic markers

Population screening programmes are to discover early stage tumours in seemingly healthy individuals. Furthermore, better tumour markers can help to identify cases with unspecific symptoms that may be tumour-related, this will enable more advanced diagnostics to be focused on a lower number of cases. As there would then be a greater chance of making a correct cancer diagnosis, this would be of great ethical and economic value.

Effective early discovery programmes currently only exist for cancers of the breast, cervix, and colon but rely on labour-intense procedures. Academic and corporate biomarker discovery efforts have therefore sought effective diagnostic biomarkers based on blood sampling. Despite major investment from public and private research funding bodies, measurable success in terms of such biomarkers for early discovery regulatory approved by the Food and Drug Administration, FDA, in the US remains scant.

Prognostic markers

Prognostic markers are helpful for deciding if, and when, cancer treatment should be started, considering the marker reported risk of tumour progression or recurrence. If tumour markers could help in prognosticating a treatment outcome at an early stage of treatment planning, the probable result would again be positive effects in terms of both quality of life and health economy. Finally, if a marker proves to have the capacity to detect microscopic disease before relapse is clinically evident, rescue therapy might be commenced earlier and, at least in theory, have a better chance of success. Today there are very few useful tumour markers that meet these criteria.

Predictive markers

Predictive markers report whether a proposed treatment will be beneficial or not for the specific patient. For example, activating mutations in Ras genes confer resistance to epidermal growth factor receptor (EGFR) antagonist treatment in colorectal cancer. Many tyrosine kinases contribute to tumour formation when mutated, and a large class of drugs has been developed to inhibit them. Thus, assessment of the tyrosine kinase status may predict the efficacy of inhibitors. Examples of such mutated tyrosine kinases are Bcr-Abl in chronic myelogenous leukaemia and c-kit in human and canine mastocytosis. Promising advances in liquid biopsy technologies based on detection of mutated tumour DNA in blood and other bodily fluids have recently been made in this area.

How is the performance of a biomarker measured?

Ultimately, the performance criterion of a biomarker test is its accuracy: to what extent are subjects with the condition diagnosed positively and subjects with condition absent diagnosed negatively. The percentage of subjects with condition present that are diagnosed positively is referred to as sensitivity, and the percentage of subjects with condition absent that are diagnosed negatively is referred to as specificity.[1] The sensitivity and specificity pair is the primary measure of diagnostic accuracy [2]. Typically, biomarker-based diagnostic tests produce a positive/negative test result by comparing the biomarker measurement to a cutoff value. The choice of cutoff value will highly influence the sensitivity and specificity pair, and by altering the cutoff, many distinct pairs of sensitivity and specificity are obtained. A Receiver Operating Characteristic (ROC) plot is obtained by plotting all sensitivity and specificity pairs [3]. When the cutoff can be continuously altered, the ROC-points form a ROC-curve. A convenient way of expressing the performance of a biomarker in a single number is to compute the Area Under the ROC-Curve (AUC).[4] The AUC-value can be interpreted as the average sensitivity over the range of specificities. For example, the diagnostic prostate cancer biomarker Beckman-Coulter PHI (BC-phi) has AUC-value 0.71 [5]. From a statistics point of view, ROC-analysis has several desirable properties: it is not affected by normalizations of the measurement data, and the probability distribution of the measurement data has no influence on the ROC-points per se, yielding an intrinsic robustness. As ROC-analysis also addresses the core issue of sensitivity and specificity, it is justified as the method of choice of regulatory authorities.

Why have past and current discovery efforts failed to produce clinically useful blood biomarkers?

There are literally hundreds of research papers discussing the reasons for failure in cancer biomarker discovery. Some well-known shortcomings include too small cohorts being used for biomarker discovery and validation, cases and controls being drawn from different populations, variations induced by pre-analytical sample handling, and too little emphasis on statistical aspects of study design[6]. However, the literature describing how such studies should be designed to maximize the likelihood of success is scant. The field appears to lack a common approach, as studies are regularly underpowered, a wide variety of statistical approaches are applied to the datasets, and the statistical methods used in the discovery part of the vast majority of biomarker discovery studies do not concern the core biomarker properties sensitivity and specificity[7], but rather differences between means of groups. Since the purpose of the biomarker development effort is success in a prospective clinical study, the null-hypothesis of the statistical hypothesis testing during biomarker development must be chosen so that the biomarker is assessed as effective for its intended use. For example, the standard for effectiveness applied in the approval of the improved PSA test BC-phi [8] was superiority relative to an existing biomarker-based diagnostic test. Based on these known issues and observing sound statistical principles, would it be desirable to develop a consensus framework for biomarker discovery and early validation to improve the likelihood of successful translation into clinical practice?

How can the regulatory requirements guide biomarker discovery?

Regulatory (FDA) approval of a biomarker-based diagnostic test is based on three top-level criteria: effectiveness, safety, and benefit-risk. Effectiveness is evaluated through the sensitivity and specificity pair as determined through a pivotal clinical study, juxtaposed with the sensitivity and specificity pairs of existing diagnostic tests. Safety is determined through an analysis of the consequences of erroneous test results under the specified intended use and principles of operation of the diagnostic test, as well as analysis of laboratory test results of interference (effects of potential interfering substances), precision and reproducibility (consistency under different lots, runs and users), robustness (tests under potential failure modes), guard banding (accepted ranges for each reagent and process step), specimen stability (transport and storage over a five-year period), intermediate product stability (storage conditions for extracted intermediate products/sera), reagent stability (test kit shelf life), and any other notion that the regulator deems is useful for the assessment of the safety of the biomarker-based diagnostic test. Benefit-risk is evaluated through an analysis of the benefits of accurate test results under the specified intended use, versus the risks of the biomarker-based diagnostic test. An approval order may include post-approval requirements and restrictions.

How can the bench to bedside translation of biomarkers be accellerated?

The clinical trials required for clinical validation and FDA approval of a cancer biomarker may require 10 years or more to conclude, depending on tumor type and the type of biomarker, Given the significant financial risks involved with the clinical validation of new biomarkers, combined with the limited gain from sales of a diagnostic as compared to drugs, is there  a need for governments and funders to absorb some of the risk? Early phase risk reduction could be in the form of collecting and providing the retrospective samples required for the early validation and clinical assessment. The collection efforts need to be ongoing in several independent populations, sample patients with cancer at diagnosis and longitudinally, and be linked to population based studies where samples before cancer diagnosis are available[9]. Late phase support could target screening centres to enable prospective evaluation of new tests.

How can veterinary science and comparative oncology help biomarker development?

The domestic dog has become increasingly useful as a comparatively spontaneous cancer model to study genetic and environmental risk factors as well as easing the transition between rodent and human clinical trials for cancer drug development[10] [11]. The many similarities between various cancer types affecting humans and dogs and the spontaneous development of these cancers in immune competent canine individuals living in a shared environment with us suggest a common aetiology. The shorter lifespan of dogs and the shorter time to relapse after cancer treatment allows data regarding efficacy, short and long-term toxicity and side-effects of novel cancer drugs to be generated in years rather than decades as in human clinical trials.

However, certain limitations need to be overcome to make full use of the dog model. Slightly different classification systems for common cancers limits translation of data and clinical outcome from dog to human. The canine genome and annotation, especially of immune gene families, could be improved to allow a more careful and correct comparison with the human genome. Tumours such as mammary neoplasia (breast cancer), malignant lymphoma (Non Hodgkin Lymphoma) and osteosarcoma do all deserve attention and are found to be similar in tumour morphology, biology and response to treatment where improved models would benefit human studies.

Treatment of metastatic tumours in dogs is generally not successful today and disease prevention and early diagnostics will significantly reduce animal suffering, improve animal welfare and reduce costs for pet owners.  For biomarker research, many analyses will be done on already collected tissue and blood for routine diagnostics in the clinical setting and thus reduce discomfort for living animals.

The potential benefits in using dog spontaneous tumor models in biomarker research are many;

• Trying out new diagnostic techniques and refine methodology before using in human oncology where availability of samples many times are a hurdle for technique testing.

• Rodent models will yield less material and are not performed in spontaneous tumor models and translation into human clinical setting has not always been successful, both in therapeutic studies as well as in biomarker research.

• Informing on how comorbidities may reduce sensitivity specificity of one or several biomarkers in marker panels are of interest to know early, as comparison between tumor samples and blood donors seldom reflects how the markers are used in the clinical setting and early promising results usually are significantly revised, when used in more mixed patient populations. Again, benefiting from the greater sample pool often available in the dog models will spare unessecary or non-strategic use of limited patient material in humans.

• Many technical/diagnostic platforms used in medical research are already validated in dogs and with the knowledge of the dog genome and proteome, validation efforts still needed are many times easy and cost effective.

Together with studying spontaneous tumours instead of induced tumours in experimental animals, this is well in line with the Replace/Refine/Reduce (3R) principles. Although, biomarkers in tumour tissue have mostly been investigated in dogs, recent promising findings on serum thymidine kinase 1 (TK1) either alone or in combination with CRP and also circulating micro RNAs have been suggested as potential biomarkers for canine neoplastic disease. With the necessary tools and resources acquired/supported, the domestic dog is now a fully-fledged model that is ideally suited to biomarker research and to informing human oncology in a parallel comparative approach.

References

Baker SG., The central role of receiver operating characteristic (ROC) curves in evaluating tests for the early detection of cancer, J Natl Cancer Inst. 2003 Apr 2;95(7):511-5.

Baker SG., Improving the Biomarker Pipeline to Develop and Evaluate Cancer Screening Tests, J Natl Cancer Inst. 2009 Aug 19; 101(16): 1116–1119

Clinical Laboratory Standards Institute, CLSI, GP10 A Ed. 1 (1995/R2001), Assessment Of The Clinical Accuracy Of Laboratory Tests Using Receiver Operating Characteristic (roc) Plots, 2001

Clinical Laboratory Standards Institute, CLSI, EP12-A, User Protocol For Evaluation Of Qualitative Test Performance, 2002

Gordon I, Paoloni M, Mazcko C, Khanna C, The Comparative Oncology Trials Consortium: using spontaneously occurring cancers in dogs to inform the cancer drug development pathway, PLoS Med. 2009 Oct;6(10):e1000161.

Karlsson E.K, Lindblad-Toh, K., Leader of the pack: gene mapping in dogs and other model organisms, Nature Reviews Genetics volume 9, pages 713–725 (2008)

Paoloni M, Khanna C., Translation of new cancer treatments from pet dogs to humans, Nat Rev Cancer. 2008 Feb;8(2):147-56.

Pavlou MP, Diamandis EP, Blasutig IM., The long journey of cancer biomarkers from the bench to the clinic., Clin Chem. 2013 Jan;59(1):147-57

U.S. Food and Drug Administration, Statistical Guidance on Reporting Results from Studies Evaluating Diagnostic Tests, 2007

U.S. Food and Drug Administration, Pre-Market Approval of Access Hybryitech P2PSA on the Access Immunoassay Systems, Summary of Safety and Effectiveness, 2012