Enabling Retrospective Biomarker Studies: Resolving the Conflict between Short and Long Term Goals

In the field of clinical biomarker research, it is common to need to explore new hypotheses after the conclusion of a clinical study (i.e. retrospective studies).  However, if the proper samples are not available, even the best ideas are no more than fantasies.  While this statement might seem trivial, it is surprising to discover that many bio / pharmaceutical companies struggle to implement the proper strategic and tactical steps to enable retrospective biomarker studies. 

In my experience, the most common strategic issue facing bio / pharmaceutical companies in this area is in resolving the conflict between short term and long term corporate goals.  Specifically, in the context of the conduct of clinical studies, the need to meet recruitment quotas and deadlines often clashes with the proposal to acquire supplementary samples that, at the time, have only theoretical value (i.e. potential use in retrospective studies).  Indeed, there is a general consensus among the teams responsible for running clinical trials (i.e. clinical operation) that adding sample collection procedures can complicate approval of protocols by the Institutional Review Boards (responsible for clinical study protocol approval on behalf of the institution and their patients) and can impede patient recruitment.  Therefore, unless there is a strong concrete justification for collecting certain samples, additional sample collections tend to be excluded from clinical protocols.  The solution to this apparent conflict resides in a strong corporate policy in support of biomarker research in general and retrospective biomarker research in particular.  Without the assurance that the logistical constrains imposed by sample acquisition for biomarker research will be fully acknowledged as a factor affecting the conduct of clinical studies, clinical operation will favor the bottom line (i.e. completion of studies in the shortest possible time).

Beyond a biomarker-friendly corporate attitude, the scientists and clinicians responsible for biomarker research need to have a sound understanding of the logistical impact of additional sample collection on clinical studies.  Biomarker researchers need to be able to negotiate intelligently with their colleagues in clinical operation.  Reciprocally, clinical operation staff needs to be with the scientific questions explored by the biomarker researchers.  Therefore, cross-training of biomarker researchers and clinical operation staff is one of the key aspects of a successful clinical biomarker research program.

In some cases, clinical samples that were collected for one purpose (e.g. pharmacokinetics) can be repurposed for biomarker research.  However, if the proper informed consent was not put in place at the time of sample collection, using these samples for retrospective biomarker studies is not acceptable.  Indeed, current ethical and legal standards mandate that all individuals enrolled in a clinical study be fully informed about the use of the biological samples collected in course of the study.  The issue of drafting informed consent forms that adequately inform the patients about future biomarker research can be quite tricky.  While it is impossible to describe all potential future use of clinical samples for biomarker research, it is important to define the overall intent and the limit of this research.  Also, it is often desirable to draft the informed consent form with the option for the patient of opting in (or out) of future biomarker research. 

Finally, assuming that clinical samples exist and are properly consented, efficient retrospective biomarker research requires a solid sample management system.  Beyond the physical inventory of samples, such as system ideally needs to seamlessly integrate anonymized patient medical information, clinical study specific information, consent status (whether patient opted in or out of biomarker research), and prior data obtained from these samples.  Hence, an efficient patient sample management is as much about inventory management as it is about information management.

Thierry Sornasse for Integrated Biomarker Strategy 

Magnetic Resonance Spectroscopy to Monitor Tumor Metabolism: Not a Biomarker Yet But a Promising Concept

In the October 12^th issue of PLos One, Alessia Lodi and Sabrina Ronen from UCSF published the results of their work on the use of Magnetic Resonance Spectroscopy (MRS) to monitor the metabolic activity of tumor cells (link).  Although this work was entirely conducted in vitro on cell lines, the concept presented in this paper offer a glimpse at a possible new approach to monitor drug effect early during treatment.  Indeed, there is ample evidence that anti-cancer drugs alter the metabolic profile of cancer cells before producing detectable effects on tumor size (detectable by CT scan or MRI) or even overall metabolic activity (detectable by FDG-PET).

The utility of MRS to monitor early metabolic changes induced by drugs in cancer cells has been demonstrated before.  However, these earlier studies focused on single metabolites which limited their observations to the specific drug – cell combination studied.  In this work, the authors expanded on earlier work MRS use for the monitoring of cancer cell metabolism by used an unbiased 1H MRS-based metabolomics approach to investigate the overall metabolic consequences of treatment with the phosphoinositide 3-kinase inhibitor LY294002 and the heat shock protein 90 inhibitor 17AAG in prostate and breast cancer cell lines.

Obviously, translating this concept to human patients, in which complexity will be several orders of magnitude greater, will not be easy but one can speculate that as MRS technology further progresses, tracking multiple metabolites in vivo will become trivial. 


Thierry Sornasse for Integrated Biomarker Strategy

An Old Diagnostic Gets the Boot: PSA Testing in Healthy Men is No Longer Recommended

This week, the United States Preventive Services Task Force is due to release its draft recommendation on the use of the Prostate-Specific Antigen (PSA) test in healthy men of all ages.  The PSA test has been a standard tool in urology to assist in the diagnosis of prostate cancer.

Essentially, this recommendation states that the PSA test in healthy men has no clinical benefit, does not save lives, and actually may lead to unnecessary follow up tests and procedures that can have deleterious effects on the patient’s health (see The New York Times article).

These conclusions are based on the results of five well-controlled clinical studies which confirm the general empirical consensus about the PSA test: its lack of specificity and sensitivity result in unacceptable numbers of false positive and false negative tests, respectively.  In particular, false positive tests are particularly troublesome since a positive test will usually lead to a biopsy and treatment that can lead to impotence and/or incontinence.  While those risks of complications are somewhat acceptable for actual prostate cancer patients, they are unacceptable for individual who have misdiagnosed.

This recommendation by the United States Preventive Services Task Force is already producing strong reactions from prostate cancer survivors and advocacy groups.  The idea of shelving the PSA test is unacceptable to those who feel that this diagnostic saved their live.  The truth is that neither the PSA test nor other currently available tests are particularly useful in detecting prostate cancer.  Hence, there is an urgent need to develop, clinically validate, and deploy effective tools for the early detection of prostate cancer in apparently healthy men.  Ironically, the dominance of the PSA test on the market has probably a substantial obstacle to the development of new diagnostic in this field.  Indeed, the protectionism from a segment of the diagnostic industry with financial interest in PSA testing, as well as the difficulty to change medical practices among physicians have probably contributed to the lack of alternative prostate cancer diagnostics.  One can speculate that the new recommendation about PSA testing will open a breach for innovative tools that will actually save lives.


Thierry Sornasse for Integrated Biomarker Strategy

A New RNA Biomarker for Huntington’s Disease: Going Beyond Nerve Pathology (bis)

In the early online issue of the Proceedings of the National Academy of Sciences of October 3^rd (link), Hu and colleagues reports on a new biomarker of disease activity for Huntington’s disease (HD) based on the differential expression level of the transcript for H2AFY gene in peripheral blood mononuclear cells (PBMC).  HD is an autosomal recessive genetic disorder in which nerve cells in certain parts of the brain waste away, or degenerate.

Similarly to the recent work on ALS biomarkers published in PLoS One this month (see earlier post: New Potential ALS Multiprotein Biomarker: Going Beyond Nerve Pathology), Hu et al. hypothesized that the key pathobiology affecting neurons in HD would be detectable in other cell types than neurons.  Indeed, the huntingtin protein, which has been shown to be at the center of HD pathobiology, is expressed in most tissues, including PBMC.

Using a standard transcriptomics approach, the authors surveyed the entire genome for differential RNA expression between the PBMC of HD patients, healthy controls, and other neurological disorders (Parkinson’s disease, Alzheimer’s disease, corticobasal degeneration, essential tremor, progressive supranuclear palsy, and multiple system atrophy).  Using stringent statistical criteria and pathobiological knowledge, the team selected the transcriptional modulator H2A histone family member Y (H2AFY) as the most relevant biomarker for HD.  This initial discovery was confirmed by two independent studies.  First, a cross-sectional case controlled study of an additional 36 HD patients, 9 carriers of the HD mutation with no clinical symptoms (the HD mutation has 100% penetrance and therefore all carriers will eventually develop the disease), 50 healthy controls, and one individual with spinocerebellar ataxia.  Second, a longitudinal case-control study where 25 HD patients and 21 healthy controls were followed for at least 2 years (37 subjects were followed for 3 years).

In order to link the transcriptional difference observed in PBMC of HD patients to the pathobiology of the disease, the authors analyzed the expression of the H2AFY-encoded protein MacroH2A1 in the frontal cortex of postmortem brains obtained from 12 HD patients.  While the expression of MacroH2A1 was clearly elevated in the brain of patients with grade 2 or 3 disease, this trend was not maintained in grade 4 patients.  This was most likely due to the fact that MacroH2A1 is expressed at high level in neurons and that this stage of the disease is characterized by a substantial loss of these cells.  Finally, the authors assessed the translational value of the H2AFY / MacroH2A1 biomarker in a mouse model of HD (knock-in of exon 1 fragment of the human huntingtin gene).  There again, the progression of the disease was associated with an elevation of the MacroH2A1 protein in relevant brain substructures and treatment with the experimental HDAC inhibitor sodium phenylbutyrate resulted in a decrease in the biomarker signal.

Altogether, if these observations are further confirmed, the availability of a disease progression and a disease modification biomarker for HD should constitute a major advance in the field.  Indeed, the development of drugs for the treatment of HD has been hampered by the lack of sensitivity and precision of standard clinical end points.  Similarly to other neurodegenerative diseases such Alzheimer’s and Parkinson’s disease, clinical progression in HD is slow, erratic, and relatively unpredictable at the individual level.

Beyond the direct impact of this work, the approach used by Hu and colleagues seems to signal a new trend in biomarker research: instead of limiting the scope of biomarker research to the specific anatomical compartment primarily affected by the disease, which in the case of the central nervous system is essentially inaccessible, the field may significantly benefit from considering accessible peripheral tissues which may display secondary pathobiology similar to that affecting the primary tissues.  Indeed, a similar approach was used by Nardo and colleagues to identify a potential new protein biomarker for Amyotrophic Lateral Sclerosis (PloS One October 5^th; see earlier post: New Potential ALS Multiprotein Biomarker: Going Beyond Nerve Pathology)


Thierry Sornasse for Integrated Biomarker Strategy

New Potential ALS Multiprotein Biomarker: Going Beyond Nerve Pathology

In the October 5^th issue of PLos One, Nardo and colleagues (link) present the results of their work on a new peripheral blood cell-bases biomarker for the diagnosis and monitoring of Amyotrophic Lateral Sclerosis (ALS: a disease of the nerve cells in the brain and spinal cord that control voluntary muscle movement).

Based on the assumption that ALS pathobiology is no restricted to the nervous system, the authors conducted a classic proteomics analysis (2D-DIGE) of pooled peripheral blood mononuclear cells (PBMC) collected from healthy controls and patients suffering from ALS (grouped into two disease severity cohorts based on the ALS functional rating scale revised [ALSFRS-R]).  The first set of 71 candidate biomarkers was first refined to a 14-protein biomarker panel by validation against healthy controls.  This subset was further refined to a 5-protein ALS-specific biomarker panel (table 1) by validation against other neurological disease controls that may clinically resemble ALS.  

Table 1

Out of this 5-protein panel, the combination of IRAK4 and CypA was the most associated with ALS versus other neurological disorders, yielding a discriminatory power of 91% at the appropriate cut-off value (Receiver Operator Curve AUC = 0.905).

From the original 14-protein biomarker panel, the authors also derived a 3-protein ALS severity biomarker panel (table 2) by comparing patient samples from moderate disease (ALSFRS-R > 24) to samples from patients with severe disease (ALSFRS ≤ 24).  Out of this 3-protein panel, ERp57 was the most associated with disease severity with 89% discriminatory power at the appropriate cut-off level (Receiver Operator Curve AUC = 0.893).

Table 2

Finally, the authors investigated the translational value of the 14-protein biomarker panel by analyzing those proteins in the PBMC and the spinal cord from a rat model of ALS (G93A SOD1-transgenic rats).

By showing that disease biomarkers for a neurological disease can be identified in easily obtainable PBMC, this work represents an important step in the evolution of biomarker research.  Instead of limiting the scope of biomarker research to the specific anatomical compartment primarily affected by the disease, which in the case of the central nervous system is essentially inaccessible, the field may significantly benefit from considering accessible peripheral tissues which may display secondary pathobiology similar to that affecting the primary tissues. 


Thierry Sornasse for Integrated Biomarker Strategy

The Parkinson Progression Marker Initiative: An Emerging Success Story

In an earlier post Biomarker Qualification Consortia: The ADNI Success Story, I discussed the value of biomarker qualification consortia by highlighting the success of the NIH sponsored Alzheimer’s Disease Neuroimaging Initiative (ADNI).  Here, I would like to raise awareness to another biomarker qualification consortium in the field of neurodegenerative medicine: the Michael J. Fox Foundation sponsored Parkinson Progression Marker Initiative (PPMI).

PPMI is a consortium of academic, industrial, and non-profit organizations dedicated to the assessment and qualification of biomarkers of Parkinson’s disease (PD) through the longitudinal monitoring of early PD patients.  In contrast to ADNI, PPMI main sponsor is the Michael J. Fox Foundation: a non-profit organization dedicated to the advancement of PD treatment through scientific research and public awareness.  PPMI has assembled a total of 21 sites in the US and Europe which will recruit and follow 400 early PD patients and 200 age-matched controls volunteers over a 5-year period.  Similarly to ADNI, PPMI has defined a set of standardized protocols for the assessment of clinical (motor assessment, neuropsychiatric, olfaction) end points and imaging (DATScan, MRI, DTI), biochemical (alpha-synuclein, DJ-1, urate), and genetic biomarkers.  In keeping with the non-competitive spirit of the consortium, all data collected by PPMI will be made available to the scientific community through a centralized data repository.  PPMI also intends to facilitate access the biosamples collected from the study participants.

Although PPMI has only been active since June 2010 and not all sites have been active since the study start, the study has already enrolled 50% of the control participants (goal: 200 individuals) and 31% of the PD participants (goal: 400 individuals).

As it is common for studies of this magnitude, PPMI has encountered a few bumps in the road.  Although DATScan was approved by the FDA for US use on January 14^th 2011, this critical imaging biomarker technology has not been available since February 2011 (status may have changed since this information was released).

Thierry Sornasse for Integrated Biomarker Strategy

A Clinically Qualified Biomarker of Response to anti-CD20 Therapy in Rheumatoid Arthritis

In the September 21^st issue of Science Translational Medicine (link), scientists at Genentech reveal their findings about a new, clinically qualified biomarker of non-response to antibody therapy to CD20 in rheumatoid arthritis (RA).  B cell depleting therapy using the anti-CD20 mAb rituximab in RA (link) is reserved for patients who have failed standard disease-modifying antirheumatic drugs (specifically methotrexate) and/or with inadequate response to anti-TNF antibody therapy.  Considering the cost, complexity, and relative risk associates with anti-CD20 therapy and considering that about 50% of RA patients do not respond to rituximab, there is a strong impetus to target this therapy to patients who are most likely to respond favorably.

Hypothesizing that RA patients with high frequency of antibody producing plasma B cells are less likely to respond to rituximab (plasma cells do not express CD20), the team at Genentech surveyed a set of B cell and plasma cell specific RNA transcripts in blood samples from a subgroup of patients who had been treated with rituximab (REFLEX study).  Using the American College of Rheumatology 50% improvement criteria (ACR50), they identified a clear association between failure to meet ACR50 and elevated levels of RNA for the immunoglobulin J chain (IgJ) at baseline.  They confirmed this observation using blood samples from patients enrolled in two additional independent rituximab studies (DANCER and SERENE), and one study of ocrelizumab (a second generation anti-CD20 mAb) in RA (SCRIPT).  When all four trials were combined, the ACR50 response rate in the active arms was 28% for the IgJ^lo group (n = 471) and 12% for the IgJ^hi group (n = 122) (Odd ratio: 2.7; 95% confidence interval: 1.5 to 5.3).  The predictive power of the IgJ RNA level was further refined by combining this parameter with the RNA levels for the B cell specific splice variant of Fc Receptor-like 5 (FCRL5).  Together, elevated levels of IgJ RNA and low levels of FCRL5 at baseline (biomarker positive: IgJ^lo / FCRL5^hi) were strongly associated with low probability of positive response to anti-CD20 mAbs therapy (figure 1).  Indeed, in the combined 4 clinical studies, 28% of biomarker-negative patients responded to treatment while only 9% of biomarker positive patients responded under the same conditions (Odd ratio: 3.6; 95% confidence interval: 1.8 to 8.4).  Of note, this combination biomarker does not appear to be an indicator of more severe diseases since it was not associated with different response rate in the placebo groups from those clinical studies.

Fig. 1

Beyond representing a major advance in the area of treatment decision in RA patients, this work represents a remarkable example of the power of well-planned, well-executed prospective retrospective studies for the discovery and clinical qualification of novel biomarkers

Thierry Sornasse for Integrated Biomarker Strategy

New Biomarker to Guide Antibiotic Prescription Decisions: Procalcitonin as Barometer of Infection

In the early online issue of September 22^nd of BMC Medicine (link; provisional paper), Philip Schuetz, Werner Albrich, and Beat Mueller review the present and the future promises of procalcitonin (PCT) as a potential generalized biomarker of infection and potential guide to antibiotic prescription in clinical settings.  As the authors point out, the field currently lacks reliable biomarkers of bacterial infection that can be assessed rapidly from easily accessible samples, resulting in suboptimal management of antibiotics administration.  Therefore, beyond the direct benefit of expediting the diagnosis of bacterial infection, PCT could be used to develop an antibiotic prescription algorithm that would potentially optimize antibiotics usage by eliminating their use in circumstances where they are not needed (fig. 1)

While strong evidences from randomized clinical trials support the use of PCT to guide the prescription of antibiotics for the treatment of lower respiratory tract infections (upper respiratory tract infection, pneumonia, COPD exacerbation, and acute bronchitis), and severe sepsis, more work needs to be done to establish PCT as a clinically relevant tool in the management of infections such as bacteremia, abdominal infection, neutropenia, and postoperative fever.

Thierry Sornasse for Integrated Biomarker Strategy

Catching Metabolic Pathways in the Act: Navigating the Heavy Water World

A press release on September 16^th on Market Watch about KineMed caught my attention (link).  KineMed, based in Emeryville CA, has developed new proteomics and metabolomics tools that enable the monitoring of metabolic flux through complex biological pathways by exploiting the power of deuterated water (or heavy water: ²H₂O) labeling.  By monitoring the kinetic of predictable mass shift of molecules of interest by mass spectrometry, the scientists at KineMed have been able to ascertain complex dynamic processes such as blood clotting, complement cascade activation, epidermal turnover in psoriasis patients, anterograde neuronal transport in ALS and PD patients, and DNA turnover rate in leukemia and breast cancer (see a video presentation by Marc K. Hellerstein, M.D., Ph.D.; co-founder of KineMed)

Because of its non-radioactive nature and ease of deployment (deuterated water is simply administered as a glass of water), this technique offers the prospect of identifying new biomarkers related to disease processes, drug mechanism of action, and drug toxicities.  It is important to remember though that this technique does not allow for in situ metabolism monitoring and thus still requires sample collection.  Therefore, the usual limitations associated with the collection of biosamples do apply to this new technique.

Thierry Sornasse for Integrated Biomarker Strategy

From Biomarker to Companion Diagnostic: of Analytical and Clinical Validation, Regulatory Affairs, and Intellectual Property

In the August 24^th early online issue of Drug Discovery Today (reference), Michael Nohaile from Novartis Pharma AG discusses the key factors required to translate a promising biomarkers into an effective companion diagnostic (CDx).  Based on a pragmatic staging scheme of drug – CDx co-development (figure 1), the author dissects the complex cross-functional interactions between of analytical and clinical validation, regulatory affairs, and intellectual property management.

Fig.1

On the analytical validation front, the author stresses the importance of timely assay platform selection, the need for proper consideration of pre-analytical parameters (see my earlier post: Biomarker Research: The Pre-analytical Puzzle), and the critical issue of the synchronization of the assay validation process to meet clinical development milestones.  Failure to complete assay validation before the initiation of pivotal clinical will require the conduct of complex and expensive bridging studies to satisfy the regulatory requirement for CDx.

On the clinical validation front, the author discusses the issue of adequate sample ascertainment rate from clinical studies in the context of prospective-retrospective (predefined analysis of samples from a completed study) CDx clinical validation strategies, and the issue of the statistical power for purely prospective CDx clinical validation studies.  In particular, serious consideration should be given to the decision of including or excluding marker negative patients in such studies.  On the one hand, inclusion of marker-negative patients is required to determine the positive and negative predictive value of the candidate CDx.  On the other hand, beyond being less expensive and potentially faster, studies that exclude marker-negative patients may also present an ethical advantage in cases where the potential treatment benefit is expected to be negligible in marker-negative individuals.

From a regulatory affairs perspective, the fact that CDx are regulated by the Center for Devices and Radiological Health (CDRH) implies that specific regulatory expertise is required for the successful prosecution of CDx (see also my earlier post about recent FDA guidance: Companion In Vitro Diagnostics (IVD) Development: some clarity at last).  In particular, the fact that the risk / benefit analysis for CDx is entirely tied to the risk / benefit profile of the associated drug implies a close collaboration between the drug reviewing authorities (CDER/ CBER) and the device reviewing authorities (CDRH).

Finally, from an intellectual property, the author discusses the issue of the timing of patent filing and the more global issue of patentability of biomarkers.

Thierry Sornasse for Integrated Biomarker Strategy

System Biology-Based Prognostic Biomarkers of Clinical Complications in Acutely Injured Patients

In the September 13^th issue of PLoS One (link), John D. Storey and colleagues report on inflammation-related gene expression signatures associated with differential clinical outcome in acute trauma patients.  Specifically, the authors analyzed the expression of inflammation-related genes in 168 blunt-force trauma patients over a 28-day period.  The genes and gene pathways that clustered differently between patients’ clinical outcome subgroups (based on Marshall multiple organ failure clinical score) were assembled into predictive modules of clinical outcomes.  Of particularly interest, the down-regulation of MHC II expression within 48 hours of trauma and up-regulation of p38-MAPK within 100 hours of trauma were particularly robust independent predictors of negative clinical outcome in this patient sample.

Considering that up to 60% of late trauma mortality is caused by infections, sepsis, and multiple organ failure multiple organ, the management of these inflammation-related complications remains a major unmet medical need.  In particular, the ability to predict the individual patient clinical trajectory early during trauma treatment remains a significant challenge for the medical community.  Therefore, the prospect of using gene expression as a prognostic biomarker to manage the care of trauma patients is of particular significance.

Thierry Sornasse for Integrated Biomarker Strategy

Brain Imaging Biomarker of Pain: I see how you feel

Our perception of biomarkers tends to be limited to the realm of measures that provide information about disease and drug activity.  In fact, biomarkers can provide a means to assess additional biological processes relevant to patient well being such as anxiety and pain.  In paper published in the September 13^th issue of PLoS One (link), a team of the Department of Anesthesia, Stanford University describes a new functional MRI-based (fMRI) biomarker for the identification of pain.  Because the sensation of pain can be subjective and can occur in the absence of detectable injury, the standard for assessing pain is based on patient self report.  While this traditional measure is readily assessable, it does not differentiate between the sensory and the psychological components of pain perception.  In addition, patient self reported pain assessment is impossible in individuals who are not able to communicate.  Therefore, development of an objective biomarker of pain is of great interest to the medical community. 

The team at Stanford performed a pilot study involving 24 individuals who were monitored by fMRI while being subjected to painful and non-painful thermal stimuli.  Using the results from the first 8 volunteers, the team used Support Vector Machine learning to develop a predictive model that then validated on the remaining 16 individual volunteers.  In this setting, the model accurately identified the type of stimulus with 81 % accuracy.

While the size of this study is not sufficient to draw definitive conclusions, it is tempting to speculate that the future of pain management in patients who are unable to communicate may improve dramatically

Thierry Sornasse for Integrated Biomarker Strategy

Prognosis of conversion from MCI to AD: of verbal memory, brain volume, and CSF biomarkers

In the September 2011 issue of the Archives of General Psychiatry (reference), Dr. Goldberg and colleagues report the results of the first study that examined the respective predictive values of cognitive measures, brain imaging, and cerebrospinal fluid (CSF) biomarkers in determining the risk of conversion from Mild Cognitive Impairment (MCI) to Alzheimer’s disease (AD).

In contrast with the multiple recent publications derived from the Alzheimer’s Disease Neuroimaging Initiative about biomarkers in AD (ADNI; see earlier post), this work identified measures of delayed verbal memory (Logical Memory delayed recall and Auditory Verbal Learning Test delayed recall) as the most reliable predictors of progression from MCI to AD.  While brain volume assessed by MRI (Left middle temporal lobe thickness) was identified as an additional predictive factor, the levels of Ab42 and Tau in the CSF did not add significant predictive value to their model (systematic stepwise logistic regression).

In commentary provided to Medscape (link), the lead author urged caution in interpreting this finding by stating that “Biomarkers unarguably work. However, cognitive markers, which are less expensive and less invasive, also work and provide strong complementary information”.

In my mind, the question is not so much whether cognitive assessment tools work better than CSF biomarkers but more about the applicability of these findings to the general practice of medicine.  Indeed, while CSF biomarkers are objective measures, the results of even the best cognitive tests are partially subjective: the skills of the person administering the test can have an influence on the results.  Therefore, one can wonder if, in the hands of the average neurologist or neuropsychiatrist, the verbal memory testing would perform as well and would outperform the objective measure provided by CSF biomarkers.

Thierry Sornasse for Integrated Biomarker Strategy

FDA Pharmacogenomic Biomarkers in Drug Labels

The list of pharmacogenomic biomarkers included the labels of FDA approved drugs has grown substantially over the last 10 years.  The most recent update from the FDA (Table of Pharmacogenomic Biomarkers in Drug Labels; 08/25/2011) lists 109 pharmacogenomic biomarkers included in the labels of 97 drugs (the labels of some drugs such as Imatinib and Warfarin include more than one pharmacogenomic biomarkers).

From a regulatory perspective, these biomarkers can be included in different sections of the drug labels (e.g. box warning, contraindication, clinical pharmacology), informing the prescribing physicians and the patients about identification of responders / non-responders, avoiding adverse events, and optimizing drug dosage.  The label information about pharmacogenomic biomarker can describe:

• Drug exposure and clinical response variability
• Risk for adverse events
• Genotype-specific dosing
• Mechanisms of drug action
• Polymorphic drug target and disposition genes

Functionally, the majority of the pharmacogenomic biomarkers currently included in the label of approved drug fall into the category of safety and efficacy markers related to drug exposure due to altered drug metabolism.  Indeed, 60 of the 109 pharmacogenomic biomarkers belong to the liver cytochrome P450 enzymes (CYP) which play a critical role in drug metabolism.  Other functional variants of enzymes involved in drug metabolism such as dihydropyrimidine dehydrogenase (DPD) and thiopurine S-methyltransferase (TPMT) also fall into this category. 

Although still representing a minority of cases, the number of drug efficacy pharmacogenomic biomarkers included in cancer drug labels has been growing (i.e. response biomarkers, predictive biomarkers).  In general, these biomarkers are designed to assist in the prescription decision by testing for the presence of the drug target. 

Examples:

• Imatinib: C-kit, BCR-Abl, PDGFR
• Trastuzumab: Her2/neu
• Vemurafenib: BRAF
• Tositumomab: CD20

As the field of biomarker development in support of drug development evolves, it is expected that this list of pharmacogenomic biomarkers included in drug labels will grow substantially, making the promise of personalized medicine a reality.

Thierry Sornasse for Integrated Biomarker Strategy

Pairing GWAS with in-depth metabolomics: assigning functions to genetic variants

In the September 1^st issue of Nature (reference), scientists from the Helmholtz Zentrum Munchen Institute in Munich, Germany, the Wellcome Trust/Sanger Centre, King’s College, and Metabolon, Inc. present the most comprehensive Genome Wide Association Study (GWAS) aimed at identifying relationship between individual genetic variations and specific metabolic pathways.  Using ultra-high performance LC-MS and GC-MS, the levels of over 250 metabolites, representing over 60 metabolic pathways, were analyzed in serum samples from volunteers enrolled in the German KORA F4 study (n= 1768) and in the British TwinsUK study (n= 1052).  From these measures, over 37,000 metabolic traits (concentrations or ratios of metabolite pairs) were derived and their association with about 600,000 SNPs was assessed.  The team identified 37 independent genetic loci with genome-wide significant associations with metabolic traits, 23 of which represented novel associations.  Moreover, among these 37 genetic loci, 15 overlapped with known disease-associated genetic loci, shedding new light on possible new pathobiological mechanisms of diseases such as diabetes, kidney failure, venous thromboembolism, and coronary artery disease. 

This remarkable work represents a major evolution in the field of GWAS by providing a means to place genetic information within a functional biological context.  Indeed, despite identifying thousands of disease risk loci, most GWAS are cataloging exercises offering little or no information about the biological processes potentially associated with the identified genetic variants.  

Thierry Sornasse for Integrated Biomarker Strategy

Biomarker of depression: Doctor I feel blue or maybe just over-methylated

In the August 30^th issue of PLoS One, Fuchikami and colleagues report their findings about a new biomarker of severe depression based on Brain-Derived Neurotrophic Factor (BDNF) gene methylation profiles (reference).  Briefly, the authors analyzed the methylation profile of the BDNF gene in blood samples collected from 20 clinically diagnosed severe depression patients and 18 healthy human volunteers (see figure 1).  Their analysis covered 81 CpG units upstream of exon 1 (CpG I) and 28 CpG units upstream of exon 4 (CpG IV) of the BDNF gene.  Differential methylation status in CpG I appeared markedly different between patients and controls, with an overall trend for hypo-methylation in patients with major depression.  The biological implication of this methylation profile is currently unknown. 

Fig.1

Considering the small sample size used in this study, these findings should be viewed as an initial screening for potential biomarker candidates which will require substantially more work to be confirmed.  First, because the individuals enrolled in this first study were exclusively of Japanese origin, the relevance of BDNG gene methylation status as a biomarker of depression remains to be established in a more ethnically diverse population.  Second, as I have mentioned in an earlier post (link), the reductionist sample selection process used in this study probably yielded over-optimistic statistical association values that may not translate well to the more complex real-world.  Indeed, the diagnosis of major depression is almost never made as a simple binary determination of “healthy” vs. “depressed”.  Rather, the diagnosis of depression is a process of eliminating other conditions that manifest themselves with similar symptoms.  Hence, analysis of the BDNF gene methylation profile in clinically related psychiatric conditions should constitute an important follow up to this initial study.  Finally, assuming that these biomarker candidates are confirmed, it would be particularly interesting to determine whether current treatments for depression affect the methylation profile of the BDNF gene.

Of note, it seems that the field of biomarker discovery in the area of depression is picking up speed lately. This paper comes one day after the announcement by Lundbeck Canada of a $2.7 million donation in support of biomarker discovery in the area of major depression and bipolar disorder (announcement), and a few weeks after the cover story of Ridge Diagnostics’ depression blood test in the August issue of Psychiatric Time (see earlier post). 

Thierry Sornasse for Integrated Biomarker Strategy

Genome Wide Association Studies: Beyond the Sample Size Barrier

Genome Wide Association (GWA) studies (GWAS) are hypothesis-free experiments aimed at identifying possible associations between subtle genetic variations and disease risk and/or disease state (see overview in Nature Review Drug Discovery 7, 221).  Over the last few years, the number of GWAS has exploded thanks to the shrinking cost and improved performance of all-genome analysis tools.  Despite their power to decipher the genetic susceptibility of many diseases, GWAS suffer from a major limitation: the sample size required to identify credible associations.  Because GWAS are hypothesis-free, all possible genetic variations are tested for possible association with the phenotype of interest, requiring much more stringent thresholds for statistical significance: depending on the penetrance of the genetic variation the type I error threshold (aka alpha) is usually set between 10^-7 and 10^-5.  Thus, sample sizes in GWAS tend to be well above a 1,000 cases and often exceed the 10,000 mark.  When you are considering a relatively rare disease or condition, these numbers can become a limiting factor.  Furthermore, these large sample sizes only permit testing of direct association hypothesis and not more complex hypotheses such as interactions between variants; the latter would require even larger sample size.  So, if sample size constitutes an inherent limit in GWAS, how can this field progress beyond this barrier?

A recent paper published by Hicks and colleagues in Cancer Informatics (Cancer Inform 2011; 10: 285-204) offers a possible solution.  Focusing on breast cancer, the authors combined GWAS information with gene expression data to determine the combined contribution of multiple genetic variants acting within genes and putative biological.  In addition, thanks to this approach, the authors were able to identify novel genes and biological pathways that could not be identified using traditional GWAS.

Thierry Sornasse for Integrated Biomarker Strategy

First approved companion diagnostics for a lung cancer drug: XALKORI (crizotinib) and ALK FISH Test

On August 26^th 2011, Pfizer announced approval by the FDA of XALKORI (crizotinib) – an ALK-specific kinase inhibitor ‑  for the treatment of patients with ALK-positive, locally advanced or metastatic non-small cell lung cancer (NSCLC).  XALKORI was developed and approved in parallel with a molecular companion diagnostic, developed by Abbot Molecular, aimed at detecting the rearrangement of the ALK gene on the 2p23 chromosome by Fluorescent In Situ Hybridization (FISH). 

The parallel approval of this new drug with the companion diagnostic ALK FISH marks the first example of personalized therapy for lung cancer and reinforces the growing trend of personalized medicine in oncology.  Indeed, earlier this month, the FDA approved Zelboraf (vemurafenib) and companion diagnostic for BRAF-mutation positive metastatic melanoma (See: A biomarker finds its drug), adding to the list of cancer treatments that depend on companion diagnostics (See: Personalized cancer medicine review).

Thierry Sornasse for Integrated Biomarker Strategy

Personalized cancer medicine review: predictive biomarkers

In the advanced online issue of Nature Review Clinical Oncology of August 23^rd, Nicholas La Thangue and David Kerr published a detailed analysis of the state and the future of personalized cancer medicine (reference).  In this review, the authors offer a thorough analysis of the history, significance, and evolution of current predictive biomarkers for drug response in cancer (see table 1).

Table 1

First, a note about terminology. The authors use a biomarker nomenclature popular in the oncology field where the term “predictive” biomarker describes two types of disease biomarkers (see previous post for details): 

1. Trait biomarkers: a stable predictor of disease risk or response to treatment (e.g. genotype, liver CYP expression)
2. State biomarkers: evolving predictor of a disease stage (e.g. most medical diagnostics).

Second, beyond their analysis of predictive biomarkers, the authors also point to a potential unintentional negative consequence of personalized cancer medicine.  While novel targeted treatments of cancer offer the prospect of greater efficiency and lower risk to the patients, these new personalized treatment tend to displace older, more affordable “untargeted” drugs, potentially reducing access to treatment for the patients.  Ideally, the use of these older broad spectrum “untargeted” treatments could be optimized through the identification of novel predictive biomarkers of response to treatment.  However, considering the limited incentives associated with such efforts, it is unlikely that the industry will actively pursue this option.

Finally, this review highlights the growing need to revisit some of the initial assumptions associated with the development of predictive biomarkers and established companion diagnostics such as HER2 testing.  As I wrote in a previous post (link), the apparent direct connection between these early biomarkers (i.e. presence of target) and the biological process of interest (i.e. response to treatment) resulted in a development process that put little emphasis on biological qualification.  Now, it has become clear that this apparent connectivity is much more complex than initially assumed, forcing the cancer biomarker community to dedicate substantial efforts to clarify the biological significance of these early cancer biomarkers.

Thierry Sornasse for Integrated Biomarker Strategy

Biomarker Studies: samples, hypothesis, and statistics

In a recent post on BiomarkerBlog, David Mosedale highlights a common problem with the design of biomarker discovery studies: reductionist clinical samples selection.  While it is tempting to initially explore for potential new biomarkers in highly contrasted clinical samples (e.g. healthy vs. diseased, benign early cancer vs. advanced metastatic cancer), this approach is almost guaranteed to yield over-optimistic results that do not translate easily to the real, complex world.  As a solution to this common problem, the author proposes that the design of the initial biomarker discovery study should reflect more accurately the intended application of the biomarker by including a spectrum of cases representative of the true complexity of the target patient population.  While I fully agree with David’s point, I would like to suggest an alternative view on this issue.

I would argue that the root-cause of the disconnection between biomarker discovery and their translation to medical use is the application of the right statistics to the wrong questions (i.e. statistical hypothesis).  Based on this premise, is there a fundamental issue with biomarker exploration using highly contrasted clinical samples?  I would argue that this approach can be useful as long as it is recognized for what it is: an initial screening step designed to test the minimalist hypothesis of whether a distinguishing factor (or factors) can be detected under artificially contrasted conditions.  Thus, the strength of the statistical association between the distinguishing factor and the selected sample phenotypes only reflects the pre-defined sample choice, not the true nature of the factor’s statistical association in the real-world population.  Hence, the use of this approach should be limited to the selection of potential biomarker candidates intended to be studied in a representative clinical sample.

Another case of inappropriate hypothesis definition is often encountered in the so-called validation of candidate biomarkers where a subset of the clinical sample used for discovery is used to determine the predictive value of the candidate biomarker using techniques such as Receiver Operator Curve analysis.  Here again, the strength of the statistical predictive value (Positive and Negative Predictive Values) derived from this approach is skewed by the initial sample selection, offering limited information about the predictive value of candidate biomarkers in the real-world.

So what is the solution to this somewhat frustrating trend in biomarker research?  I would argue that biomarker scientists should learn to ask the right questions to statisticians, and that statisticians should learn to challenge biomarker scientists about the actual hypothesis they wish to test.

Thierry Sornasse for Integrated Biomarker Strategy