Pathology Imaging and Informatics as the Cornerstone of Precision Medicine: Two years of Experience with the NIDDK
Kidney Precision Medicine Project (KPMP)
Jeff Hodgin, MD PhDAssistant Professor of Pathology
University of MichiganMay 2019
Disclosures
Receive funding from:AstraZeneca, MedImmune, Moderna, Gilead
Kidney Disease 2019
Acute Kidney Injury Chronic Kidney Disease
• Lack of abundant human kidney biopsy tissue
Understand and treat human kidney diseases• Ethically and safely obtain kidney biopsies from participants with AKIs or CKDs• Create a kidney tissue atlas• Identify critical cells, pathways and targets for novel therapies• Find disease subgroups to stratify patients• Clinical assays to help devise individualized treatments• Improve scientific knowledge base
Goals/Deliverables of KPMP
Change patient management; improve health
Unique and challenging aspects of KPMP• Participant Collaborators (patients) intrinsic to research effort
• Maintain ethical, safety, and rigorous quality control standards• Research biopsies in people with common kidney diseases (hypertension, diabetes, AKI)
• Longitudinal 5‐10+ year biopsy cohorts for AKIs and CKDs• Need large cohorts with deep phenotyping to create personalized medicine metrics
• Finding out what we do not know (dark matter)• Create routine clinical assays for use in clinical practice
• Initially, better use of routine histology; then 20‐40 markers; then ??• Develop new patient categorizations (endophenotypes), find novel disease pathways and therapeutic targets, better predict outcome.
Our journey begins with Version 0.5
Phase 1• Establish clinical protocols, MOPs• Informed consent documents• Optimize/validate tissue processing• Establish Kidney Atlas• Start Biopsy cohorts
Phase 2• Proof of concept studies • Enrich Kidney Atlas• Next generation tissue interrogation assays
• Expand longitudinal cohort studies
Phase 3• Expand to larger cohort studies
2027202220192017
10+ Year KPMP Timeline
Ambitious long term project
Redefining disease categories and classifications
IntegratedPathology
Pathologists create disease categories, based on few pre-selected features,
and then test against outcome
Unsupervised learning and multiaxial datasets defining categories of patients
with similar biological trajectory
Identify Targets for Therapeutic
Intervention
Predict Clinical Outcome
CurrentPathology
“one size fits all” approach
Takes in account individual variability
KPMP: Three distinct but highly interactive activities
*
Goal of 3 kidney cores:• 1 diagnostic processing• 2 research
- Transcriptomics- Proteomics- Metabolomics- Spatial molecular profiling
Research Biopsy Work Flow ‐ Recruitment Sites
Recruitment Site Training
Tissue Interrogation Sites
UTHSA/PNNL/EMBL
IU/OSUUCSF/Stanford
UCSD/WashUUCSF/Stanford
UMich/Broad/Princeton
UCSD/WashUUCSF/Stanford
IU/OSU
IU/OSU
IU/OSUUCSF/StanfordUSCD/WashU
UMich/Broad/Princeton
OCT frozen
Cryopreserved
FFPE
CODEXUCSF
CytometryIU/OSU
3D
2D
Targeted profiling
High throughput unnbiased profiling
123 7
5
4
Assessment of gene expression correlation (by RNA Seq) at single nucleus vs single cell
Comparison of transcriptomes measured by single cell RNA Seq (individual vs pooled)
Correlation analysis of gene expression at single cell level vs region specific (by LCM)
Study of cell populations extracted by bulk RNA Seq vs single cell/nucleus
Generation of proteome map at near single cells and regions of kidney
Meta‐analysis of metabolome data and proteomics data at near single cell level
3D tissue/ whole slide spatial visualization and mapping
Integrative “Omic” analysis of transcriptome, proteome, and metabolome data
6
Integrated single cell functional Kidney Map
8
DART‐FISH UCSD/Wash U MIFISH
UCSF Broad/U of M
3D Imaging
8
7
/EMBL
2D/
PREMIERE TIS
30 cell clusters are identified in kidney tissue by single cell RNA‐seq
Samples: 27 total (16 nephrectomies and 11 biopsies; 3 samples from table 1 not used)
Major cell types: Residential kidney cells of the nephron, interstitium, arterioles, and inflammatory cells
20,986 cells endothelial
Staining for top scRNA-seq markers validates endothelial cell subsets
PREMIERE TIS
In situ hybridization for mRNA
PREMIERE TIS
30 cell clusters are identified in kidney tissue by single cell RNA‐seq
Samples: 27 total (16 nephrectomies and 11 biopsies; 3 samples from table 1 not used)
Major cell types: Residential kidney cells of the nephron, interstitium, arterioles, and inflammatory cells
20,986 cells
Central Hub System Architecture
Data Generators
‐omics
Imaging & Histology
Longitudinal Clinical Data (REDCap)
…
Data Lake
Knowledge Environment
Identity and Access Management: Shibboleth/InCommon Authentication
Derived Data Analysis Pipelines
Data Viewing Tools
Kidney Tissue Atlas
Patient‐centric Tool
...
Raw Data Viewer
Data Integration and Metadata: Ontologies
Data Flow ‐ DVC
Specimen Tracking
KPMP Knowledge Environment
Framework for multiscalar data integration to mechanistically define AKI and CKD‐DKD kidney tissue disease manifestation for targeted therapeutic intervention
The Depth Defining data and metadata
Establishing ontology structures
Integrating molecular data types
Mapping molecular data to spatial data
Software development process
Requirements Gathering and Planning Development Sprint(s) Deployment
Personas & Problem
Statements
Design User Stories
Release Planning
Sprint Kickoff Estimation
Show‐and‐Tell
Sprint Planning
Daily:• Tracking
• Stand‐up Meeting• JIRA updates
•Development• Pairing• CI
• Testing System‐Wide Testing
Application Release
User Feedback
Becky Steck – Michigan Nephrology
Ul Balis – Michigan Path Inf
Personas and Use CasesPersonas are representations of types of real users, which help us: - Understand our audience and answer the question, “Who are we
building for?” - Visualize/Humanize our user research- Prioritize features and align on strategy - Use the same language when we’re discussing our users- Create a shared understanding of how the product would be used by
customers- Provide a reference point throughout the design and development
process
Whole Slide Image Viewer – Patient Access
Whole Slide Image Viewer – Patient Access
Transcriptomics User Research Interviewees
Screenshots from the Mid‐Fidelity Mock‐ups
Gene Search Demo
Standing up an Atlas for theKidney Precision Medicine Project
Atlas of Histology,V. Eroschenko,Wolters Kluwer
Becomes…Matched Clinical Content
DemographicsComorbidities
OutcomeseGFR
ExposuresMedications
Matched Morphological Content
HistologyMolecular MarkersStructural Elements
Structural Ensembles
Matched Molecular Content
Expression PatternsLinkages to PhenotypesMolec. Def. of Disease
Novel Targets
Through the use of relatively recent image‐query and image analysis tools, histology data may now be promoted to serving as one of the primary starting points for interrogation, and not simply reside in the repository as passive information.
Top Level Objectives of the Atlas (from an imaging perspective)
• Images will provide the central anatomic frame of reference, allowing for anchoring of all data types to each other, spatially
• Two key thrusts:• Visualization: for every data type across dimensionality (Resolution and Depth‐agnostic)
• Analytics: Tools will be tailored to personas• Will address the need to interrogate various length scales and biological units:
• Organ structures• Cells• Sub‐cellular spaces
• Image‐based analytics will be integral to the atlas
Prioritization of Atlas Features from a User‐Centric Perspective
• Most critically‐needed capabilities in the initial ensemble of functions:
• Allow for visualization and rendering of all expected TIS image‐based data types
• Provide for an integrated ecosystem of tools, representing the most requested types of quantitative analyses, with candidate examples being:
• VTEA: for tissue cytometry/ image analytics• VIPER Studio: for segmentation• TIS‐Provided Solutions
• Allow for data extraction and compilation of results into standardized formats
• Exchangeability across TISs • Reproducibility (with analysis of metadata)• Dissemination
• Provide access to all raw data and its associated annotations
Descriptor LibraryConventional Diagnosis Library
Vector Library
Diabetic Nephropathy
Mapped –omics Library
Omics mapping on structural normal and abnormal histologic primitives
Pathology
Text Search
Image Search
140 fields of view computed in less than 5 seconds, yielding an AUC of 0.96
Automated Interstitial Region Fractionation – VIPER Studio
University of Michigan, Pathology Informatics:• Ulysses Balis• Jerome Cheng
Proof‐of‐concept re: connecting molecular and spatial data
Identification of spatial gene expressionGene Histology
Identification of spatial gene expressionHistology Genes
Popup Query From 3D image volume to omics – Indiana Univ. TIS
DE AnalysisGeneXGeneY…MetabolomicsMetaboliteXMetaboliteY…LMD transcriptomicsGeneXGeneY
DE AnalysisGeneXGeneY…MetabolomicsMetaboliteXMetaboliteY…LMD transcriptomicsGeneXGeneY
Data Visualization CoreSystems‐Biology
Ontologies
Digital Pathology
Team Science
Project Mgt.
Ravi IyengarMSSM
Olga TroyanskayaPrinceton
Matthias KretzlerMichigan
Laura MarianiMichiganCijiang He
MSSM
Rachel SealfonPrinceton
Evren AzelogluMSSM
Zidong ZhangPrinceton
Jian ZhouPrinceton
Barbara MirelMichigan
Heather AscaniMichigan
Sean MooneyU. Washington
Oliver HeMichigan
Becky SteckMichigan
Ul BalisMichigan
Jeff HodginMichigan
Laura BarisoniDuke
KPMP
Ruth DannenfelserPrinceton
Data Visualization CoreSystems‐Biology
Ontologies
Digital Pathology
Team Science
Project Mgt.
Ravi IyengarMSSM
Olga TroyanskayaPrinceton
Matthias KretzlerMichigan
Laura MarianiMichiganCijiang He
MSSM
Rachel SealfonPrinceton
Evren AzelogluMSSM
Zidong ZhangPrinceton
Jian ZhouPrinceton
Barbara MirelMichigan
Heather AscaniMichigan
Sean MooneyU. Washington
Oliver HeMichigan
Becky SteckMichigan
Ul BalisMichigan
Jeff HodginMichigan
Laura BarisoniU. Miami
KPMP
Ruth DannenfelserPrinceton
Pathology Informatics:• Ulysses Balis• Jerome Cheng• Ross Smith
Nephrology:• Becky Steck• Michael Rose• Rachel Dull• Becky Reamy• Zach Wright
Acknowledgements & Collaborating Groups• Indiana University
• Tarek Ashkar• Seth Winfree• Ken Dunn• Micheal Ferkowicz• Angela Sabo• Pierre Dagher• Michael Eadon
• University of Michigan• Rajasree Menon• Edgar Otto • Celine Berthier• Matthias Kretzler• Keluo Yao
• Duke University• Laura Barisoni , MD
• University of Oklahoma• Christopher Williams, MD
• Funding Sources:1. 1U2CDK114886‐01 NIDDK
Central Hub for Kidney Precision MedicineProject (KPMP)
2. UG3 DK114923‐01 NIDDK TIS3. P30 DK079312 3D Quantitative Core – Indiana O’Brien Center
Kidney PrecisionMedicine Project
Broad Institute• Paul J. Hoover• Tony D. Jones• Thomas Eisenhaure• Shuqiang Li• David Lieb• Nir Hacohen
Princeton• Rachel Sealfon• Olga Troyanskaya
KPMP.org
end
Patients want to know•What do I have?•What will happen to me?•What can I do about it?•What does it mean for my family?
Prefer precision medicine approach•Right intervention for right patient at right time
Transcriptomics User Research Interviewees
Digital Pathology Repository
Overview on pathology work flow
RS
KPMP pathology
team
CRF:PathologyReport
CRF:Dx coreHigh levelassessment
CRF:TIS coreHigh levelassessment
Data Lake
Disease caterogyassignment
KIDNEY ATLAS
TIS
WSI & IF & EM Images
DescriptorMorphologic profile
REDCap
Links all in one place
• Participant web‐based slide viewer: https://mydata.kpmp.org/• Demo landing page: https://demo.kpmp.org/• Pathology web‐based slide viewer: https://demo.kpmp.org/dpr/• Cell type search prototype: https://demo.kpmp.org/atlas/• Gene search prototype: https://demo.kpmp.org/gene‐search/
• Associated mid‐fidelity mock‐ups