Summary
Pathology consensus review for clinical trials and disease classification has historically been performed by manual light microscopy with sequential section review by study pathologists, or multi-headed microscope review. Limitations of this approach include high intra- and inter-reader variability, costs, and delays for slide mailing and consensus reviews. To improve this, the Nephrotic Syndrome Study Network (NEPTUNE) is systematically applying digital pathology review in a multicenter study using renal biopsy whole slide imaging (WSI) for observation-based data collection. Study pathology materials are acquired, scanned, uploaded, and stored in a web-based information system that is accessed through a web-browser interface. Quality control includes metadata and image quality review. Initially, digital slides are annotated, with each glomerulus identified, given a unique number, and maintained in all levels until the glomerulus disappears or sections end. The software allows viewing and annotation of multiple slide sections concurrently. Analysis utilizes “descriptors” for patterns of injury, rather than diagnoses, in renal parenchymal compartments. This multidimensional representation via WSI, allows more accurate glomerular counting and identification of all lesions in each glomerulus, with data available in a searchable database. The use of WSI brings about efficiency critical to pathology review in a clinical trial setting, including independent review by multiple pathologists, improved intraobserver and interobserver reproducibility, efficiencies and risk reduction in slide circulation and mailing, centralized management of data integrity and slide images for current or future studies, and web-based consensus meetings. The overall effect is improved incorporation of pathology review in a budget neutral approach.
Introduction
In the early 19th century, Richard Bright developed a system for pathologic classification of kidney diseases based on macroscopic abnormalities (1,2). The 20th century brought the use of light microscopy (LM) and correlation of histology and cytomorphology with clinical phenotype, consenting the classification of renal disorders to be organized in three major categories: degenerative diseases (nephrosis), inflammatory diseases (nephritis), and chronic nephrosclerosis (3,4). In the mid-20th century, evaluation of kidney percutaneous needle biopsies by histology (5,6), immunofluorescence (IF) (7,8) and electron microscopy (EM) (9) have allowed advances in understanding of renal disease and their subcategorization based on histopathologic and clinical features, outcome, and only in some instances, pathogenetic mechanisms. Clinically useful classification systems have been developed and validated, to varying degrees, by collaborative consensus studies using standard morphologic evaluation. These include classification systems for lupus GN (10), IgA nephropathy (11,12), FSGS (13) pauci-immune crescentic GN (14), diabetic glomerulosclerosis (15), and renal transplant rejection (16), among others. This approach has been used for the last 50 years; however, histologic analysis by LM carries acknowledged limitations, including high intra- and inter-reader variability. Reproducibility of scoring is an important factor in classification formulation processes, which may modulate in the integration of morphologic parameters in certain classification systems as exemplified in the process that resulted in the Oxford Classification of IgA Nephropathy (11,12). The development of molecular-based classification systems, which may address disease heterogeneity, better define prognosis, and guide therapeutic intervention, requires a systematic, rigorously validated, morphologic evaluation.
In the 21st century, the advent of whole slide imaging (WSI), a technology that enables the creation of high-resolution images from traditional glass microscope slides, represents a substantial advance in the morphologic analysis of renal biopsies (see the subsequent section on general principles of WSI) (Table 1).
Definitions of terms commonly used in digital pathology and this manuscript
The Nephrotic Syndrome Study Network (NEPTUNE) is the first multicenter study to systematically apply digital pathology review to morphologic analysis of nephrotic syndrome (NS) in the setting of system biology. NEPTUNE is a North American multicenter collaborative consortium, part of the Rare Disease Clinical Research Network, supported by the National Institutes of Health Office of Rare Diseases Research and the National Institute of Diabetes and Digestive and Kidney Diseases, the NephCure foundation, and the University of Michigan, whose goal is to facilitate translational research in patients with NS (https://rarediseasesnetwork.epi.usf.edu/NEPTUNE) (Table 2) (17). The NEPTUNE cohort study targets enrollment of 450 adults and children with NS (including minimal change disease [MCD], FSGS, or membranous nephropathy [MN]) for detailed clinical, morphologic, and molecular phenotyping. More than 400 parameters are captured for each patient over at least 30 months. In addition, single nucleotide polymorphism and exon analysis, and genome-wide gene expression profiling of renal biopsy compartments for mRNA and miRNAs will be performed, and patient-specific molecular profiles will be linked to phenotypic, genetic, and morphologic data for comprehensive analyses using systems biology approaches as part of the core study aims. To fulfill the requirements of the NEPTUNE study design the pathology committee took an innovative approach by (1) creating the NEPTUNE digital pathology protocol and using WSI of renal biopsies for morphologic and morphometric assessment rather than conventional LM, and (2) applying observation-based data collection, called “descriptors,” in place of consensus diagnosis.
The NEPTUNE consortium enrolling centers
General Principles of WSI
Conventional microscopy isolates the field of view of the observer to the microscopic image, which is projected directly from the specimen, and is limited only by the quality of the illumination and optics of the microscope. WSI is the reduction of a microscope slide to a digital image that is viewed on a computer screen rather than the immersive environment of the binocular microscope (terminology currently in use is listed in Table 1). The image is displayed at the nominal “magnification” at the desired level on the computer screen. Metadata, including embedded images of entire slides, the slide label, the image details, and the instrument used to generate the image, are included in the WSI. Some instruments offer Z-stacking (the capacity to “focus” the image at different planes). Manufacturers utilize proprietary WSI formats, which may include user-definable features pertaining to image compression; however, most software allows the manipulation of WSI from different hardware vendors. Like digital radiology, a Digital Imaging and Communications in Medicine standard promises interoperability between systems (18). Ultimately, the success of a WSI solution depends on three factors: the quality of the image storage/image serving hardware, the quality of the network connecting data collection and storage, and the user viewing environment. Inadequate hardware substantially degrades system function and results in user dissatisfaction.
Instrumentation
The current instrumentation for WSI is based on one of two approaches: (1) montage imaging, in which a two-dimensional camera captures images at high magnification, which are then “stitched” into a digital image; and (2) a scanning approach by which a line scan, or stripe of image, is collected and “stitched” into a digital image. The instrument optics for these approaches is fundamentally identical, as is the resultant image pyramid. Issues such as focus, tissue finding, and image compression are managed by software.
Visualization
Another aspect of WSI is visualization, done on a computer screen rather than the immersive environment of the binocular microscope. Desktop computers and monitors are most commonly used for review of the digital images. Although high-resolution, color-calibrated screens similar to those used in digital radiology are available, they are rarely utilized due to cost and space needed for dedicated workstations/cockpits. Laptops are generally suboptimal because of screen size/resolution limitations; however, the development of high-resolution displays for laptops and other wireless devices is seeing a rapid evolution of applications for viewing images on non-desktop computing platforms, although still with limited software controls.
Software
The core of WSI is the software, used to control image acquisition, manage the image database, and generate a user-interface. Image acquisition software is tightly integrated with the instrumentation, with a goal of “set and forget” functionality, offering limited options for individualizing in the generation of WSI. Data management, in the form of servers where the storage, distribution, and annotated information of WSI is coordinated, remains subject to undergoing evolution. Some WSI manufacturers do not offer server-based file/information management software, whereas others offer research-focused platforms that incorporate flexible image analysis tools and simple to manipulate data-export features. Other more clinically focused platforms are based on servers with laboratory information systems or laboratory information systems–integration features and attempt to offer “case management” software that mimics current slide management methods (virtual slide trays). These data management platforms operate on servers that are best managed within centralized information technology environments of the institution where they exist. Substantial information technology security challenges exist to protect both data and patient confidentially.
At the desktop level, the user interacts with the software with either a browser-based interface, which promotes diversity of supported platforms, or via a client software package, which offers an increased number of user-defined features for image viewing.
Technical and Operational Issues of the NEPTUNE Pathology Imaging Protocol
The NEPTUNE protocol (summarized in Figure 1) requires participants to have a renal biopsy at the time of enrollment, which is performed at US and international facilities. Whereas the application of digital pathology systems to NEPTUNE patients enables acquisition and distribution of renal biopsy imaging in an efficient manner, adhesion to relevant human subject regulations and Good Laboratory Practice guidelines is strictly maintained (19,20). Compliance with Good Laboratory Practice and Health Insurance Portability and Accountability Act regulations, acquisition of pathology material, scanning, quality control, and organization of the web-based information managing system organization (Figure 2) are discussed in the Supplemental Appendix.
NEPTUNE digital pathology protocol. Scanning and uploading of pathology material can occur at the local enrolling centers (local scanning) or centrally (central scanning). For local scanning, glass slides are scanned into whole slide images and locally uploaded to the web-based pathology information management system by a password-protected access together with the electron microscopy and immunofluorescence digital images and PDF of the pathology report. For central scanning, all pathology material is mailed to the image management coordinating site at the National Institutes of Health, where glass slides are scanned into whole slide images and uploaded into the web-based pathology information management system together with the rest of the pathology material. Once cases pass quality control processes they are ready for triage, annotation and analysis. Analysis of whole slide images is performed with a two-step process: first the pathologist annotates all glomeruli in all levels available (multi-level representation of glomeruli). Reader pathologists score each case by using the NEPTUNE Pathology Scoring System. The entire process utilizes digital pathology rather than conventional light microscopy. NEPTUNE, Nephrotic Syndrome Study Network.
The NEPTUNE Digital Pathology Scoring System
The NEPTUNE pathology committee has created a defined language scoring system, the NEPTUNE Pathology Scoring System (NPSS). This approach was utilized to improve intraobserver and interobserver reproducibility, and to capture the complex morphologic features of the heterogeneous diseases presenting with NS. The NPSS provides a system for comprehensive morphologic analysis of the biopsy digital images by utilizing “descriptors” for patterns of injury in the renal parenchymal compartments, rather than assigning a conventional diagnosis to each biopsy, as described below in the discussion on scoring. Each descriptor has been defined by specific criteria through consensus by the pathology committee members. The NPSS analysis is based on a two-step process: (1) annotation of glomeruli, and (2) scoring of glomerular, vascular, and tubulointerstitial morphologic features.
Web-based pathology information management system organization. Access to the web-based information managing system organization is password protected. The system is organized into three categories of folders: the major study categories (MCD/FSGS, MGN, and others), the cases ready for triage, and the cases that fail quality control review (A). Within each of these major folders, there are individual patient folders labeled with the study PID (B). Each patient folder contains whole slide images labeled with the PID, slide level, and stain (C–E), digital images of electron microscopy (F), and of immunofluorescence (G) in addition to the PDF of the original pathology report (not shown). MCD, minimal change disease; MGN, membranous GN; PID, patient identification code.
Annotation
Annotation is performed on slide images using only one section per slide (Figures 3–5). Each glomerulus is first identified in a single section, given a unique number (G1, G2,….G12, etc), followed through all levels, and given the same number in each level until there are no further sections or the glomerulus disappears (Figure 3). The software enables four slide images to be viewed simultaneously, allowing the pathologist to identify and follow glomeruli more easily and facilitating multi-level glomerular reconstruction (Figure 4). The ability to view up to four images that move synchronously on the same screen enables more efficient analysis, with little time lost waiting for images to upload or having to reposition the images. The WSI-based annotation allows for accurate count of glomerular number, and enables the scoring pathologist to identify all descriptors associated with each glomerulus.
Simultaneous visualization of multiple virtual slides. The software allows simultaneous visualization of multiple sections, which facilitates annotation and scoring of glomerular lesions. The glomeruli are identified by the blue annotations. Annotation is generally initiated in the section with most glomeruli, as is evident in A. As the annotation process proceeds to other biopsy levels, new glomeruli can appear or disappear. The number of glomerulus decreases moving through the biopsy levels in A–D. However, a new glomerulus does appear in C.
Multi-level representation of glomeruli. Using simultaneous section visualization and retrieval by unique glomerular identifiers, annotated glomeruli can be easily followed through the various levels. This allows determination of all lesions in a multi level representation of each glomerulus, rather than assessing single level representations. Level 1 (hematoxylin and eosin stain) shows two glomeruli annotated as glomerulus 1 and 2. Glomerulus 1 is histologically unremarkable and remains such throughout all levels here represented (level 2, silver methenamine; level 6, hematoxylin and eosin; level 7, periodic acid–Schiff; level 12, hematoxylin and eosin). Glomerulus 2 appears entirely sclerotic on level 1; however, in all other levels, it would be counted as a segmentally sclerotic glomerulus with an area of solidification of the tuft and adhesion to the Bowman’s capsule. No hyalinosis or foam cells are detected in any of the represented levels.
The NEPTUNE Pathology Scoring System: Examples of descriptors. (A) Multi-level representation of a glomerulus annotated as glomerulus 3 throughout various levels. Although on level 1 there are no lesions of segmental sclerosis or collapse and this glomerulus would be counted as nonsclerotic if viewed on one level only, the same glomerulus reveals two distinct lesions of segmental sclerosis with hyalinosis on levels 8 and 9 (periodic acid–Schiff and silver stain, respectively). There is one segmental lesion containing hyalinosis on level 11 (hematoxylin and eosin) and the glomerulus has no segmental lesions again on level 12 (hematoxylin and eosin). The histologic profiling of this glomerulus includes the following descriptors: segmental sclerosis with hyalinosis cannot determine location ×2, and hyalinosis cannot determine location. (B) This glomerulus stained with silver is characterized by three descriptors: segmental sclerosing tip lesion, hyalinosis away from the vascular pole ×2, and mid-glomerular hyalinosis. (C) Three descriptors are used to profile this glomerulus (silver stain): segmental proliferative collapse, segmental podocyte hyperplasia, and segmental podocyte hypertrophy. (E) Descriptors are applied to this case of membranous glomerulopathy with segmental sclerosis. Both features of sclerosis and membranous are scored (silver stain). (F) A case of membranous glomerulopathy is illustrated (silver stain on the left and trichrome on the right). Descriptors used to profile this glomerulus include the following: diffuse spikes/hopes and global podocyte hypertrophy. (D and G) Descriptors are also applied to electron microscopy analysis. D shows that there is 100% effacement; condensation of the actin-based cytoskeleton and microvillous transformation are also scored. Primary processes are scored as presented. G shows subepithelial electron dense deposits. NEPTUNE, Nephrotic Syndrome Study Network.
Scoring
The NPSS assesses all renal parenchymal compartments and includes descriptors for LM, IF, and EM specimens. For histologic evaluation, each glomerulus is assessed individually in all annotated sections, and all descriptors for each glomerulus are recorded (Figures 4 and 5). Descriptors refer to lesions or abnormalities in each glomerulus, without giving a specific diagnosis. For example, glomerulus 6 in a case of FSGS may have no abnormalities at level 1, a perihilar segment of sclerosis at levels 3 and 5, no abnormalities at level 7, and an adhesion to Bowman’s capsule at the tip area on level 9. The use of descriptors allows a more accurate data collection regarding the lesions in each glomerulus as a whole. In addition, other morphologic, genomic, proteomic, and clinical data can be correlated with these described glomerular lesions for each case to better determine their significance for prognosis and therapeutic response, possibly resulting in novel classification systems.
A scoring sheet is used to record all glomerular descriptors for each glomerulus and their locations (descriptor mapping) using the slide image ID. With this process, not only is it possible to create a permanent record of morphologic data but retrieval of any specific description is also facilitated for consensus and/or reconciliation, if needed. The renal parenchyma is also assessed for percentage of acute and chronic injury of tubules (acute tubular injury and tubular atrophy) and interstitium (interstitial inflammation and fibrosis), and for the degree of damage in arteries and arterioles using a semi-quantitative system, because these are not the main focus of the NEPTUNE protocol. Although morphometric analysis is not a primary goal of the NEPTUNE pathology morphologic scoring system, the digital images are available for detailed morphometric analysis of these renal compartments for ancillary studies to address specific research questions. Similarly, EM digital images are evaluated for well defined descriptors of glomerular and podocyte damage, also recorded on the score sheet. Scoring of IF usually is based on the pathology report findings, although images may be included in the case folder. Once the case scoring sheet is completed, the morphologic data are transferred into the NEPTUNE database.
Advantages and Applications of the NEPTUNE Digital Pathology Imaging Protocol
The purpose of the NEPTUNE Digital Pathology Review Protocol is to enable a more robust means of consensus review, leveraging the capacity to view and annotate multiple slide sections concurrently and to have descriptive data available and easily searchable. Advantages and disadvantages are listed in Table 3. Reviewers are all presented the same material, especially when annotation of specific structures precedes scoring; however, the review is independent of time constraints of the complexities of group histopathology review sessions. This approach enables detailed evaluation of clinical samples by multiple pathologists for clinical trials on a scale not previously feasible.
Advantages and limitations of digital pathology
The financial aspect of utilizing digital pathology over conventional LM should also be considered. In digital pathology studies involving multiple investigators at disparate locations, the upfront costs of WSI preparation are an added financial burden. At the current time, a scanner and related hardware required for WSI costs approximately $200,000 USD, including installation and training. Depending on the software required, costs can greatly increase. Recurring software license fees and fees for maintenance and support for both the hardware and software are estimated to cost between 15% and 20% of total cost of purchase of the system annually. Supporting personnel can also add approximately an additional 15%–20%. However, this initial cost may be mitigated by the absence of mailing and reductions in technical and study coordinator effort. The ability to review digital images from remote sites concurrently enables webinar-based consensus meetings, resulting in cost containment, timely interpretations, and efficient use of professional time. In addition, there are no costs for slide re-cuts or potential loss of data if there is no remaining tissue to replace lost or damaged slides. The initial investment in creating a permanent reference set of images and annotations is compensated by the possibility to use, simultaneously or at different times, the same cases for more than one study. If the costs of hardware and pathologist time are excluded from the evaluation, estimates of financial cost to implement WSI as a consensus review approach are cost neutral compared with consensus panel review, in which the pathologists meet in a central location to review the material. The costs of “round-robin” review, where slides are shipped to study pathologists in sequence, are less. (Table 4) However, the process is markedly inefficient, and liability issues concerning loss of multiple clinical cases at one time, have seen this approach fall out of favor. With WSI-based review, the costs are associated with scanning time, data management, and, to a lesser extent, shipping the slides to a central location for scanning. With round-robin review there is substantial cost in shipping the slides to each study pathologist in sequence. Consensus review incurs travel cost, as well as the personnel cost of the data coordinator. Ultimately, the cost-benefit equation has to be evaluated within the context of the evaluation and the nature of the data being collected.
Panel review methodologies: Cost and issues
The application of WSI is an enabling application, which front-ends the cost of assembly of a collection of histopathology specimens into a functional database that allows the pathologist to query on demand and to evaluate metrics not previously contemplated without the effort collecting the material for review. The upfront cost of this on-demand image database includes the 8-week embargo from the date of biopsy until request for material for submission for trial review. The scanning and data management efforts, including quality control require 2 weeks of time, resulting in a minimum of 3 months from date of biopsy until the material is uploaded for the primary evaluation and classification in the online system, after which it is subsequently annotated for glomeruli, before it can be subject to classification. Currently, the average time from biopsy to classification in NEPTUNE is 6 months. Many factors, including work load of the study coordinators, affect this number. More difficult to quantify is the time required to complete annotation of each case because multiple variables are involved including the total number of glomeruli, proper labeling of slides by the histotechnologist after right sequence of levels, and experience. In general, the process requires less time as experience grows, with annotation time fluctuating from 1.5 to 2 hours initially to ≤1 hour. Assignment of cases to annotate or score is performed digitally by the committee chair. Annotator and reader pathologists retrieve and complete assigned tasks and mark them as annotated or scored. As a result, we are only now generating sufficient cases for scoring to evaluate the knowledge we are gaining from this approach.
Concurrently the application of WSI review is aimed to overcome the well acknowledged issues of poor intraobserver and interobserver reproducibility of scoring systems. It has been recently shown that the application of the Banff classification for scoring renal transplant biopsies results into a higher reproducibility if digital pathology is used over conventional LM (21,22). An additional innovative aspect of the NEPTUNE pathology protocol is the introduction of annotation of glomeruli before scoring, with the ultimate goal to further reduce intraobserver and interobserver reproducibility. A similar digital pathology annotation protocol was recently applied to scoring peritubular capillary for GL3 inclusions in a Fabry’s disease clinical trial, which demonstrated reduction in intraobserver and interobserver variability (23).
The WSI-based NPSS includes creating a library of descriptors, specific defined glomerular, and tubulointerstitial and vascular morphologic lesions, assembled to ultimately provide a reproducible and quantifiable method for renal morphologic analysis to the pathology community. With this approach, the NPSS also strives to generate a set of morphologic data (morphologic phenotype) that can be better integrated with molecular phenotypes. It has become apparent that current morphologic classifications of NS are insufficient to capture the complexity of this group of diseases. Monogenic renal disorders can have variable morphology, and conversely a single pattern of glomerular injury may result from different etiologies (24,25). Correlation with molecular profiling data is best accomplished with a granular descriptive approach, rather than summary diagnostic terms, especially in the setting of disease heterogeneity. The ultimate NPSS goal is to provide a morphologic phenotyping system that can be integrated with system biology approaches. As our knowledge of renal disease becomes more sophisticated, it is apparent that on-demand and reproducible morphometric analysis of parenchymal structures will be required to correlate “quantitative” morphology with molecular, clinical, proteomic, and other patient data to further our understanding of the pathogenesis and mechanisms of renal injury. The NPSS provides a platform for performing such morphometric studies on multi-level kidney biopsies, and a model for designing future pathology study protocols. A morphometric analysis of glomerular area has been initiated within the NEPTUNE study. Within the viewer software, an annotation in the form of a grid is placed on each glomerulus. The grid has a uniform structure and fixed spacing that creates a constant reference when counting podocytes within the glomeruli area. Similarly, morphometric analysis of other parameters such as interstitial fibrosis can be implemented.
Another novel feature of the NEPTUNE protocol is the application of WSI to create correlated multi-level representations of renal biopsies, a unique strategy for analysis of glomeruli with segmental and focal lesions (26). Conventional LM analysis typically is two-dimensional, which may result in misinterpretation of widespread lesions as focal, lack of recognition of segmental lesions, and an inaccurate assessment of the frequency of such lesions due to sampling on different levels (Figure 4) (27) In contrast, multidimensional representation using WSI allows multi-level stacking for assessment of different lesions within a single glomerulus, thus reducing the likelihood of sampling error and providing a more accurate representation of the number and nature of glomerular lesions. To date, annotation of cases has revealed that glomeruli often have segments of sclerosis at different levels (Figures 4 and 5). With this technique, an accurate determination of the percentage of glomeruli with segmental lesions can be made, which may augment studies of diseases in which this is relevant such as focal sclerosis, lupus nephritis and immune complex diseases complicated by segmental sclerosis.
Finally, scoring of digitally annotated glomeruli offers full transparency of the raw data for other investigators and regulatory agencies. This ensures data accuracy and the ability to collect additional information or to perform ancillary studies as new research questions arise. As further data are obtained with more extensive NEPTUNE digital slide analysis, there will be further confirmation of the cost, reproducibility, accuracy, and access advantages, as well as means to address technical issues, which will be subsequently reported.
Disclosures
L.B. has served as a speaker for Aperio.
Acknowledgments
This work was supported by NEPTUNE study grants from the National Institutes of Health Office of Rare Diseases Research (U-54DK083912), NephCure, and the Halpin Foundation and by the Intramural Research programs of the National Cancer Institute and the National Institute of Diabetes and Digestive and Kidney Diseases.
Footnotes
Published online ahead of print. Publication date available at www.cjasn.org.
This article contains supplemental material online at http://cjasn.asnjournals.org/lookup/suppl/doi:10.2215/CJN.08370812/-/DCSupplemental.
- Copyright © 2013 by the American Society of Nephrology
References
In this issue
Jump to section
More in this TOC Section
Cited By...
- Ultrastructural Characterization of Proteinuric Patients Predicts Clinical Outcomes
- Deep Learning-Based Histopathologic Assessment of Kidney Tissue
- Inflammatory and JAK-STAT Pathways as Shared Molecular Targets for ANCA-Associated Vasculitis and Nephrotic Syndrome
- Redefining Nephrotic Syndrome in Molecular Terms: Outcome-associated molecular clusters and patient stratification with noninvasive surrogate biomarkers
- Blood Pressure and Visit-to-Visit Blood Pressure Variability Among Individuals With Primary Proteinuric Glomerulopathies
- APOL1-G0 or APOL1-G2 Transgenic Models Develop Preeclampsia but Not Kidney Disease
- Defining Glomerular Disease in Mechanistic Terms: Implementing an Integrative Biology Approach in Nephrology
- Integrative Genomics Identifies Novel Associations with APOL1 Risk Genotypes in Black NEPTUNE Subjects
- Complete Remission in the Nephrotic Syndrome Study Network
- Tissue transcriptome-driven identification of epidermal growth factor as a chronic kidney disease biomarker