Parkinson's disease biomarkers: perspective from the NINDS Parkinson's Disease Biomarkers Program

Discovery

Discovery in biomarkers research generates new knowledge, drives innovation and provides the input for biomarkers pipeline development. Methods of discovery may include physiological/clinical assessments, genomic/proteomic/metabolomic/RNA methodologies and imaging approaches. Some of these approaches derive directly from basic research (Figure 1). Application of basic research in biomarker development capitalizes on the strengths of the NIH process. This is consequently an appropriate space for NINDS to play a role in filling the gaps in PD biomarkers research. NINDS established the PDBP based on the concept that biomarkers discovery and replication are needed prior to validation and application in clinical trials.

Replication & validation

Replication and validation are essential components of any scientific process, as biomarkers will only be useful if generalizable and reliable. However, validity and generalizability criteria are not well delineated in the biomarker field. Replication is not a simple or straightforward designation. Replication can mean consistent results obtained in the same laboratory with the same samples; results replicated in the same laboratory with different samples; or results replicated in a different setting with the same samples. In addition to replication, analytic and clinical validation are necessary for moving any biomarker forward. Analytic validation is generally applied to studies that establish that the performance characteristics of the test are acceptable in terms of its sensitivity, specificity, accuracy and precision, as applicable [13]. Clinical validation is defined as establishing that the candidate biomarker acceptably identifies, measures or predicts the clinical, biological, physical or state of interest [13]. Thus to meet the requirements for replicating and validating robust biomarkers, the PDBP is a milestone-driven program, in which go/no-go decision points are specified for all discovery and replication projects. In addition, standard blood-based laboratory tests are performed on all PDBP participants and lab results are captured in the PDBP DMR. Quality control of plasma, serum and cerebrospinal fluid (CSF) by the NINDS BioSEND biorepository includes hemoglobin assessment, while the RNA Integrity Number, 260/280 and 260/230 ratios are used to determine RNA and DNA quality, respectively. NIH also sponsored a funding announcement that supported the use of PDBP biosamples for discovery or replication of promising PD biomarkers defined by specificity, selectivity, accuracy and precision criteria. Study design required the identification of independent discovery and replication cohorts and the use of pooled samples for standardization across laboratories and platforms.

Qualification

Qualification is a term that applies largely to a regulatory process and designation. For instance, considerations for biomarker qualification at the FDA include: the context of use for drug development, the relationship of the biomarker to clinical outcomes and treatment, assay considerations (including variability), biological rationale for use, strength of association of the biomarker with the clinical state, reproducibility of data and strength of evidence. Note that qualification is not required for biomarker use in a clinical trial prior to submitting biomarker trial data to the FDA for review [14].

PDBP achievements

The overall philosophy for establishing the NINDS PDBP was to foster scientific breakthroughs and progress, as well as resources, and to facilitate integration of the existing and future projects into a dovetailed sequence from discovery to qualification for application of PD biomarkers in Phase II or III clinical trials. The PDBP concrete goals established in 2012, based on stakeholder and outside expert input, as noted above, were data management, clinical cohorts with standardized biospecimen collection and laboratory-based discovery science [7].

Data management

The PDBP DMR is the central hub for coordinating data-related activities integral to PDBP and includes seven modules that enable data entry, quality control, data access, data query, biosample access and account management. The DMR contains a common data dictionary used across multiple studies and built upon NINDS common and PD-disease specific data elements [15]. Several unique features associated with various modules of the DMR include ProFoRMS, use of a Global Unique Identifier (GUID) and the Query tool. The ProFoRMS module is an electronic data entry system, which also enables patient scheduling and quality assessment of data prior to uploading to the data repository. The GUID, is a software application that generates a unique identifier for each study participant. This software application uses a subject's personal identifiable information to create a one-way hash code made up of a prefix, an alphanumeric sequence and a check character. The DMR database contains only deidentified data, so the GUID is never directly associated with the personal identifiable information. The GUID enables linking of all data (genetic, imaging, clinical and biomarker) for an individual subject both within and across studies, and it is the identifier used to join subject level data in the Query tool. The Query tool enables a researcher to select data based on either a form, a data element, or through the use of a defined query. In the Query tool, selected data can be joined (3-way), processed at an individual level and further filtered based on data elements and permissible value ranges. Data can be downloaded either from data generated from queries or directly from studies listed in the data repository. To date, there are have been over 4.7 million records and 400,000 datasets accessed through the DMR query tool. These data have contributed to biomarker discovery [16–18] and disease modeling efforts [19].

Clinical cohort & biospecimen collection

A total of 1547 individuals have been enrolled in the PDBP studies to date (Table 3). Out of this, 825 individuals have PD, 44 individuals have Progressive Supranuclear Palsy (PSP), 28 individuals are diagnosed with Multiple System Atrophy (MSA), 36 individuals have atypical PD, 23 individuals with Essential Tremor and 525 are control subjects. Twenty-five percent of PDBP participants have a family history of PD and 184 participants were diagnosed with PD within 1 year of PDBP study enrollment. Clinical assessment tools overlap with those used in subject ascertainment in PPMI [8] and BioFIND [9]. The PDBP has collected samples from 218 individuals with at least a 2-year follow-up to date. Biospecimens obtained on all subjects include DNA, RNA, plasma, serum, whole blood and many subjects also have provided CSF. These samples were collected in a standardized fashion across sites, utilizing comparable protocols as used in the Alzheimer's Disease Neuroimaging Initiative [20], PPMI [8] and BioFIND [9]. Samples are processed and stored at the BioSEND repository at Indiana University using standard operating procedures. All specimens undergo uniform quality assessment at BioSEND upon receipt. Biosample requests are submitted through an online application form [21] and reviewed by the PD Biospecimen Resource Access Committee (BRAC) established by NINDS. The PD BRAC also reviews requests for biosamples from BioFIND, the Michael J Fox Foundation LRRK2 cohort and the HBS through this online system, thus enabling a researcher to request samples from multiple cohorts in a single application process. Once approved, biospecimens are distributed from BioSEND in a systematic manner [22].

PDBP clinical data are made available via a controlled process which requires the establishment of an account with the DMR. Requirements for account access include agreements to: not share the data with third parties; to not seek to identify any individual participating in the study; and acknowledge the PDBP in resulting publications. Once the PD BRAC has approved biosample access, the PDBP DMR Query tool enables an investigator to link clinical data to biosample availability through the GUID. PDBP biosamples selected by the investigator are processed through the DMR Order Manager. The DMR Order Manager provides a unique tool for linking biorepository and data management functions. To date, a total of 24 PDBP biosample requests have been received and 5228 biosamples have been distributed. A total of 967 DNA samples have been used for genotyping with the NeuroX chip [19,23]. NeuroX data are available through the DMR, once a DMR account holder has completed the PDBP Genomic Data Use Certificate. A total of 1308 DNA samples are currently being used for whole genome sequencing analysis and data from this analysis will be available through the DMR in 2017. Eight hundred and thirteen DNA samples have been distributed for targeted sequencing of validated PD risk alleles and data will be available through the DMR in 2018. One hundred and thirty-eight RNA samples have been distributed for RNA biomarker replication and data will be available in the DMR in 2018. One thousand five hundred and thirty-one plasma samples are currently being analyzed for urate and vitamin D levels. This data will be available in the DMR in 2017. A complete list of PD BRAC-approved biomarker discovery projects are provided in Supplementary Table 1.

Clinical & laboratory-based discovery science

A typical biomarker ‘pipeline’ is represented in Figure 1. Discovery science is an essential part of the biomarkers pipeline. While it is represented commonly as a uniform enterprise in most biomarkers research discussions, biomarker discovery research is complex, highly textured and not monolithic as often represented in biomarkers literature; the latter typically delves more deeply into clinical trial requirements and the US FDA qualification process. A challenge which warrants further evaluation is how the biomarker discovery approaches will interact with each other as well as with the other components of the biomarker qualification process.

Clinical biomarkers

We consider a clinical biomarker to be an objectively measured assessment that characterizes specific traits of individuals with PD. Examples include gait measurements, smell testing, cognition and neuropsychiatric assessments. Some of these candidate biomarkers overlap with clinical end points for other biomarker investigations. It is not entirely clear how to differentiate a clinical biomarker per se from well-validated clinical assessment tools. The reality is that several putative biomarkers are also clinical assessment tools. Well-validated clinical assessments tools that are considered likely to be useful as biomarkers are shown in Table 4.

Clinical biomarkers may be useful to measure disease activity and progression. Clinical measures, in particular, the Unified Parkinson's Disease Rating Scale (and the more recently updated Movement Disorder Society-Unified Parkinson's Disease Rating Scale) have long been used as outcome measures in clinical trials in PD [29]. There are a number of major limitations to using this scale for disease modification research in PD including: the scale objectively measures only motor symptoms and nonmotor symptoms are measured only by survey of the subject; the scale shows progression (worsening) of PD only very slowly such that for a typical trial of <3 years duration, insufficient worsening is seen in the placebo control patient group making it impossible to detect the existence of a modestly disease-modifying agent; the scale depends on trained raters who bring subjectivity to the assessment process; the scale is particularly susceptible to confounding by dopaminergic treatment effects. To partially address these limitations, the NIH NET-PD LS1 trial of creatine as a putative disease-modifying agent in PD used a new composite outcome measure called the global statistical test [38]. This consisted of several items in each of five clinically relevant domains (activities of daily living, cognitive function, ambulatory capacity, quality of life and global disability). Unfortunately, this study was terminated early when an interim analysis showed that continuing the trial was futile.

Convenient biomarkers may be useful in identifying asymptomatic individuals at risk for developing PD. Hyposmia is present in up to 90% of people with PD and occurs commonly years prior to the onset of motor symptoms. It also occurs in other neurodegenerative disorders and thus lacks specificity [30]. The most commonly used test to assess smell is the University of Pennsylvania Smell Identification Test [39]. This test has also been criticized from a practical standpoint for possible cultural biases, the time it takes to administer and its cost (approximately US$30/test). A recent positron emission tomography (PET) imaging study showed that odor identification is a more robust measure of forebrain cholinergic activity than odor memory/discrimination tests in PD, and odor identification tests would be the preferred olfactory method to screen for more severe cholinergic forebrain denervation in PD [40]. As it is lacking specificity, olfaction may be a more useful measure in a multimodal biomarker approach [19].

REM behavior disorder, when diagnosed by polysomnogram, is highly predictive for the development of a neurodegenerative synucleinopathy but lacks specificity for PD. In reviewing the criteria for prodromal PD, the task force assigned polysomnogram-proven REM behavior disorder a positive likelihood ratio of 130 [25]. Obtaining this measure in large populations is very difficult, however, which is why this was omitted in recent prediction models [19]. The same is true for constipation, which is present very frequently as a premotor sign in PD but lacks specificity for PD.

Impaired cognition is a major feature of advancing PD and measuring cognitive impairment may be a useful measure of disease progression or even trial end point. The Montreal Cognitive Assessment [41] was shown to be superior to other widely used clinical tools, such as the Mini-Mental State exam in that it lacks floor or ceiling effects [30]. Other neuropsychiatric tools are validated as progression measures: the Neuropsychiatric Inventory [42], Scales for Outcomes in PD-Cognition [43], the Beck Depression Scale and the Geriatric Depression Scale [44]. It is clear that clinical measures such as smell testing and sleep disorders, when added to other assessments such as genetic risk score, can be a useful component in predicting PD risk [19].

New technologies raise the prospect of collecting large amounts of ‘real-world’ data for assessing disease activity. Analysis of spiral drawing on a digitizing tablet offers the ability to objectively measure motor control in the ecologically relevant task of writing and discriminates early PD from controls. The use of wearable and smart-phone technology for monitoring treatment assessments has been an exciting development in PD research [45] and may prove useful for biomarker discovery. One PDBP project utilizes the APDM Mobility Lab, a sensor-based gait and balance assessment tool, to evaluate its performance as a putative state and rate biomarker platform. In a study of 135 PD subjects and 66 age-matched controls assessed at baseline, a set of gait and balance variables reported by this device correlated with PD severity measures and successfully differentiated PD subjects from controls [46]. In an ongoing study, longitudinal measures from this device are being evaluated over a period of two years in PD subjects and controls to determine if changes in parameters over time can be used as a rate biomarker for the disease.

Harmonization

We recommend continued harmonization of clinical and laboratory data and biospecimen collection. The PDBP DMR has been created and provides essential infrastructure to the project to allow harmonization to continue. The clinical measures are standardized via the use of shared assessments and clinical data elements across studies in the DMR. Additional harmonization efforts have been essential in PDBP in terms of standardized operating procedures for the collection and storage of biospecimens. The standard operating procedures can be found online [124]. The BioSEND Repository houses not only PDBP and BioFIND samples, but also the PPMI samples, which will facilitate across all PD-based biomarker initiatives standardized collection, handling and request for access practices as noted above [22]. Additional challenges with harmonization include the identification of the same biomarker with different technologies or methodologies. Again, standardized operating procedures can assist with this challenge as can the process of identifying the best and most widely reproducible methodology.

Replication

We also recommend continued replication. Replication of results, as noted above, is essential for moving biomarkers projects forward toward clinical trial and practice usefulness. Toward this goal, the PDBP has released a Funding Opportunity Announcement (FOA) in the past to allow for replication of discoveries using the PDBP samples and data collection; future opportunities are anticipated [125]. To date, seven replication projects have been funded under PDBP mechanisms, and others additionally have been approved via the PD BRAC. One of these projects has led to a publication in which PDBP has been used as a replication sample [19].

Publication of negative results

We recommend a systematic method for reporting and collating negative results. The phenomenon of bias toward positive results in the literature has been noted [126] and it arguably presents a particularly significant obstacle toward biomarkers replication/validation. Negative results, especially from replication/validation studies are essential to inform the biomarkers process. Some journals publish and encourage negative results (such as PLoS ONE [127] and eNeuro [128]), and we urge the use of these forums for dissemination of negative and of failure to replicate studies in these and other journals. Ideally, both positive and negative studies could be collated and managed in a searchable resource. A data management solution, which does not yet exist but can be envisioned, where findings could be surveyed, would be an extremely useful tool to biomarkers researchers.

Use of de novo cohorts for discovery efforts & longitudinal follow-up of well-characterized cohorts

There are two important cohorts that are necessary to identify biomarkers. First, creation of a cohort of newly diagnosed individuals for discovery of biomarkers would add to the current biomarker landscape. Second, we should continue to follow-up existing cohorts to determine their clinical outcomes.

The value of large, pragmatic trials is increasingly being recognized [2]. The NINDS PDBP was designed to reflect the real-life clinical situation in its recruitment, inclusion/exclusion criteria and assessment goals [10]. As a result, PDBP subjects are typically on symptomatic treatment. However, the importance of the use of de novo, or newly diagnosed, cohorts for discovery remains an unmet need. There are challenges to collecting and studying de novo PD patients that are both ethical as well as sample-size related. PPMI has successfully recruited and followed over 400 de novo subjects with clinical PD. However, to our knowledge, these samples and data are approved for validation and qualification use. Expanding the scope of this and other cohorts, such as the DeNoPa cohort [44] and other international PD cohorts [129] to support discovery biomarker research would be a boon to the field.

The original PDBP FOA was designed for three or more years of follow-up for each study. Considering the long clinical course and relatively slow rate of decline in PD, longitudinal follow-up will inform hypotheses tested and data collected on the PDBP cohort. Challenges in accomplishing this include the cost for long-term follow-up, especially for individuals where no intervention nor hypothesis testing is occurring. Strategies for overcoming this obstacle include novel approaches to collecting meaningful longitudinal data in a cost-effective manner such as the use of phone survey(s), wearable technology approaches and virtual care visits.

There are validated phone surveys for use in PD [130]. These could be utilized for periodic follow-up, alone or in combination with other tools. Wearables are particularly adaptable to the measure of tremor and movement and have the advantage of convenience and likely good compliance [131]. Telemedicine is used in clinical care, as well as in some research endeavors [132]. It could be possible to extend the use of telemedicine into a well-defined longitudinal follow-up collection of data on the PDBP and other cohorts toward the goal of enriching the outcome information of existing biospecimen collections.

The need for expanding the types of biospecimens was also discussed. As postmortem evaluation remains the gold standard in PD and other neurodegenerative diseases, and because several sites in the PDBP have an autopsy program (via the NINDS-funded Udall Centers of Excellence as well as at other clinical sites), the committee recommended dovetailing brain banking with the ongoing PDBP activities. For instance, the Neurobiobank, which is supported by the National Institute of Mental Health, NINDS and the Eunice Kennedy Shriver National Institute of Child Health and Human Development, brings together multiple stakeholders to facilitate research advancement through the collection and distribution of human postmortem brain tissue [133]. While this resource is currently available to PDBP researchers, a formalized integration could also be pursued. Common standards for brain collection, preparation and neuropathologic analysis should be developed, perhaps along the lines used by the National Alzheimer's Coordinating Center.

Additional biologicals to consider for banking include resources for the development of induced pluripotent stem cells. Although the potential of this cell resource for future drug discovery efforts is significant, the banking of fibroblasts and peripheral blood mononuclear cells should be prioritized in order to assure leveraging of resources. Some possible subtypes within the PDBP and other cohorts could include: those with extremes in clinical measures, patients with a change in clinical diagnosis, those who meet the Nalls criteria for risk [19] and those with genetic components as risk factors or causal mutations.

Additional future directions include biomarkers studies on cognitive impairment in Parkinsonism. Cognitive impairment, including dementia, is commonly seen in those with PD. Pathologically, PD with dementia (PDD) is most often associated with the presence of cortical Lewy bodies, as is the closely related dementia with Lewy bodies [134]. Clinically, however, the two disorders are not always distinct. Biologically, there is clear overlap and some controversy regarding biological differences between PDD and Dementia with Lewy Bodies (DLB). The PDBP is funding four new studies under a recently released FOA directed at furthering discovery science in PDD and DLB biomarkers, including building additional clinical cohorts with well-characterized subjects and standardized clinical specimens banked via BioSEND.

Early in the disease course, it can be difficult to differentiate PD from other conditions such as PSP and MSA. Even with the correct diagnosis, PD has wide variability in disease course, rate of progression and response to treatment, making an objective marker extremely useful [135]. Therefore, in response to this need, NINDS has also recently released an FOA (PAR 16–112) toward discovery of biomarkers differentiating PD from others forms of Parkinsonism and from ET. This FOA also seeks to recruit subjects in discovery projects toward a better understanding of ethnic, genetic and other subtypes and stratifications within the PD patient population.

Incorporation of larger datasets, such as those associated with whole genome sequencing will require a movement from a server-based data repository to a cloud-based solution. Storage and access of data in the cloud will also facilitate a new approach to data handling where funding support is needed for cloud-based assessments and collaborative types of studies. Data types generated through biomarker discovery platforms require the development of standard analysis pipelines to enable data to be readily shared across studies. The inclusion of information on the pipeline used (providence) in the data management resource will facilitate the replication of biomarker results.

Determining PD biomarkers from diverse data sources depends critically on the use of rigorous analytic techniques. First, it is imperative for methods to yield not only accurate biomarkers but also ones that are highly reliable, that is, likely to attain comparable accuracy in other samples and across discovery, replication/validation and qualification phases. To achieve such reliability, reproducible methods should be used, ranging from data processing to software to the choice of statistical methods. Second, several of the data modalities in the PDBP DMR involve large-scale high throughput measures such as GWAS, proteomic, metabolomics and various neuroimaging techniques. Statistical methods applied should be able to cope with high-dimensional data, addressing issues such as multiple testing adjustments for error control, multicollinearity, overfitting and variable selection. Third, the PDBP DMR and similar resources provide rare opportunities for access to extremely rich data, potentially enabling the combination of different data types, which may reveal distinct manifestations of PD pathology. The data should be fully leveraged, when appropriate, to integrate across different data types and to conduct multimodal analyses. Suitable tools to investigate PD biomarkers using large-scale multimodal data are beginning to emerge [136,137], but there is an important need to ensure that proper methods are applied and, in some cases, to develop new statistical techniques.