G. Oppenheimer Center for Neurobiology of Stress and Resilience at UCLA

Methods and Resources

How the PAIN Repository Differs from Other Brain Imaging Repository Initiatives

  • PAIN aims to identify brain signatures of common chronic pain disorders, characterize underlying mechanisms and identify biomarkers related to chronic pain and chronic pain treatment.
  • PAIN is multimodal. It contains three types of MRI scanning data for each subject, high quality structural information, diffusion tensor imaging (DTI) measurements, and resting state functional imaging. Unlike site specific task related data, these three imaging modalities are obtainable with almost any high resolution scanner and are relatively context independent, provided standardized acquisition protocols.
  • Because of the known limitation of open access data bases without standardization, PAIN includes two imaging databases:
    • The PAIN Standardized Repository is the primary PAIN Repository. It contains scans that have been collected using validated protocols (developed with other sites and different scanners) to allow for combined multisite analysis. It also contains a standardized set of clinical metadata for each subject that covers demographics, diagnoses, pain severity, comorbidity, cognitive and affective measures. Scans are assessed for quality and compatibility with the repository standards and each scan has a quality control record for use in analysis.
    • The PAIN Archive Repository is an open repository in which contributors can deposit any structural, DTI or resting state scans of pain patients or healthy controls which do not conform to the guidelines for scans in the supervised repository. Scans in the archive have minimal clinical information and can be very diverse in terms of scanning procedures. The Archive is seen as a resource to deposit and access data from already completed studies, older datasets or those with minimal subject data available. Interactions with existing open access repositories are planned.

 

The PAIN Repository and the LONI Pipeline

Extensive phenotyping of patients with chronic pain conditions, using clinical, behavioral, biological (genetic and multiomics data) and multimodal brain imaging approaches results in the generation of massive data sets of unprecedented complexity.

Responding to this challenge, the PAIN Repository uses a collaborative website (painrepository.org) to coordinate data accrual, report quality assessments and discuss analysis and publication plans. Secure network protocols allow for easy data transfer to and from the repository. In close collaboration with the USC Laboratory of Neuroimaging (LONI), the standardized PAIN Repository will also provide efficient analytical tools to extract information from these data sets, including ways to link processing steps into comprehensive, end-to-end workflows. The LONI pipeline provides an open access extensive framework for interoperability of such resources by utilizing a graphical user interface to utilize processing modules and linking them into complex scientific workflows (Figure). The Pipeline environment (pipeline.loni.usc.edu) is a distributed infrastructure model for mediating communications between different data resources, software tools and web-services. The Pipeline software architecture design is domain and hardware independent, which makes the environment useful in different computational disciplines and on diverse hardware structures.

What Types of Analyses Will PAIN Enable Members to Perform?

Members can take advantage of the PAIN repository by downloading raw data sets and performing their own analyses, or they can utilize the publically accessible LONI workflows to analyze their own or combined data sets.

 

Identifying Brain Signatures from Multimodal Brain Imaging Data Sets

Brain imaging endophenotypes (brain “signatures”) derived from multimodal imaging studies and large scale network analyses provide a powerful and biologically relevant substrate to better examine the correlations between large scale biological (including genetic and multi omics data), behavioral and clinical data sets with the brain in chronic pain conditions.

 

Correlating Clinical and Behavior al Data with Brain Signatures

In order to determine the functional implications of brain signatures, such as their relationship to pain severity and duration, it is necessary to correlate behavioral and clinical data sets with brain endophenotypes. For example, a strong correlation of alterations in saliency network properties with disease duration may suggest that the brain changes are secondary to nociceptive input to the brain.

Correlations between certain aspects of brain signatures with pain severity would suggest a role of the brain alterations in central pain amplification. Correlations between alterations of cortical modulation networks and cognitive measures (attention, prediction error, hyper vigilance) would suggest a role of the observed brain signature alterations in these psychological constructs.

As the initial data sets will be cross sectional, longitudinal data sets will be required to definitively answer questions regarding underlying mechanisms and causality.

 

Large Scale Connectivity and Complex Network Analyses

Large scale connectivity and complex network analyses support the concept that symptoms or functional impairments are associated with disruption or abnormal integration of spatially distributed regions that comprise a large-scale network (i.e., abnormal topological organization of structural and functional brain networks).

Using graph theoretical methods brain networks can be characterized by metrics that describe local topological neighborhoods of individual nodes, global network communication and signaling, and local and global measures of centrality that permit quantification of each element to the network’s structural integrity and information flow. These network metrics have been useful in probing central changes in several clinical syndromes. Ultimately, compared to regional activity or structure, measures of network properties may prove to be more sensitive central biomarkers, endophenotypes and predictors of outcome.

 

Multivariate Pattern Analyses

Multivariate pattern analyses are applied to identify patterns in the brain that makes it possible to discriminate subjects with persistent pain from healthy control subjects, and possibly patients with different types of chronic pain (e.g. neuropathic pain from persistent pain as well as predict treatment responders from non-responders and functionally link alterations in brain architecture (morphometric measures, volume, shape, cortical thickness) and resting state networks with measurable clinical and biological parameters enabling identification of distinct brain endophenotypes which characterize sub-populations of patients with different pathophysiology and treatment responses.

Ultimately, the goal of PAIN is to provide Pain Repository members with large scale data sets which can be used to develop accurate and sensitive classification algorithm based on biological markers (genetics, brain morphometry, resting state characteristics, network metrics, sympathetic and vagal measures) to inform treatment options for persistent pain disorders. Current algorithms include sparse PLS for regression, as well as discriminate analysis (sPLS, sPLS-DA) and random forest classification.