In this webinar, Aditya Pratapa and Lorcan Sherry present a new workflow for analyzing multiplex immunoflurescence images.
Spatial Signatures are a new class of highly predictive biomarkers that measure the interactions and cellular densities of tumor and immune cells that compose the tumor microenvironment. Based on multiplex immunofluorescence, spatial signatures provide a deeper understanding of complex interactions between tumors and the immune system, enabling improved patient stratification for immunotherapies. A significant hurdle to date has been in developing a data analysis workflow that is straightforward and user-friendly to transform the data rich images into meaningful quantitative spatial signatures.
In this webinar, Aditya and Lorcan review the key features of the new PhenoImager HT 2.0 data analysis workflow. This workflow introduces a simplified framework from scanning to analyzing spectrally unmixed multiplex immunofluorescence images generated on the PhenoImager HT platform. The ready-to-analyze data can be directly imported into image analysis software such as Visiopharm. This presentation covers key aspects of data analysis elements such as image QC, segmentation, phenotyping, and verification – all essential for creating outputs that support the development of a spatial signature.
Key Topics Include:
- Understand Akoya’s new HT 2.0 data analysis workflow
- The challenges in multiplex immunofluorescence analysis and the use of AI and cell
lineage segmentation considerations
- Explore OracleBio’s image analysis workflow incorporating Visiopharm
- Evaluation of analysis data to facilitate spatial profiling and interpretation
7. Quantifying PD1/PDL1 activity in the tumors
• Immunohistochemistry (IHC)
• Tumor Mutation Burden (NGS)
• Gene expression profiling (RNA)
Use any of the above or their combination as a companion
diagnostic, but how good are they in predicting patient response?
7
8. 8
Towards Achieving a Target AUC of 0.81
Source: Lu S, Stein JE, Rimm DL, et al. Comparison of Biomarker Modalities for Predicting Response to PD-1/PD-L1 Checkpoint Blockade:
A Systematic Review and Meta-analysis. JAMA Oncol. 2019;5(8):1195–1204. https://doi.org/10.1001/jamaoncol.2019.1549
The IO biomarker gap
Ideal biomarker
1. Šimundić AM. Measures of Diagnostic Accuracy: Basic Definitions. EJIFCC. 2009;19(4):203-211.
9. Feng, Z. et al. Fox, B., Multiparametric immune profiling in HPV- oral squamous cell cancer. JCI Insight 2, (2017). 9
What about multiplexed imaging?
10. Cells Without Context Provide Limited Information
Feng, Z. et al. Fox, B., Multiparametric immune profiling in HPV- oral squamous cell cancer. JCI Insight 2, (2017). 10
12. Spatial Phenotyping Provides the Highest Predictive Value
12
Towards Achieving a Target AUC of 0.81
Source: Lu S, Stein JE, Rimm DL, et al. Comparison of Biomarker Modalities for Predicting Response to PD-1/PD-L1 Checkpoint Blockade:
A Systematic Review and Meta-analysis. JAMA Oncol. 2019;5(8):1195–1204. https://doi.org/10.1001/jamaoncol.2019.1549
Protein Spatial Phenotyping is
the only biomarker above the
target threshold1 (AUC of 0.8)
…even when other modalities
are used in combination
Spatial phenotyping is poised to address the IO biomarker gap
Ideal biomarker
1. Šimundić AM. Measures of Diagnostic Accuracy: Basic Definitions. EJIFCC. 2009;19(4):203-211.
13. Biomarker discovery
Ultrahigh-plex panels
Biomarker validation
Targeted panels
Translational / Clinical use
Large scale studies
13
PhenoCycler-Fusion PhenoImagerFusion PhenoImagerHT
Clinical
Translational
Discovery
Solutions Spanning the Spatial Biology Continuum
100+ Biomarkers per slide 100+ Slides per week
14. Biomarker discovery
Ultrahigh-plex panels
Biomarker validation
Targeted panels
Translational / Clinical use
Large scale studies
14
PhenoCycler-Fusion PhenoImagerFusion PhenoImagerHT
Clinical
Translational
Discovery
Solutions Spanning the Spatial Biology Continuum
100+ Biomarkers per slide 100+ Slides per week
15. Are the lymphocytes activated?
Is the tumor proliferating?
Are there TAMs?
Are they M1 or M2?
Where are the Tregs?
Are the T cell exhausted?
Presence
Distribution
Subtype
&
Status
Asking the right questions enables systematic analysis
of the tumor immune response
15
Comprehensively phenotype the tumor microenvironment for better stratification
Characterize the TME to develop highly predictive biomarkers
Where are the immune cells located in the TME?
Better Stratification
Personalized treatment
Precision Medicine
Is the tumor “hot” or “cold”?
16. Asking the right questions enables systematic analysis
of the tumor immune response
16
Is the tumor “hot” or “cold”?
Where are the immune cells located in the TME?
Are the lymphocytes activated?
Is the tumor proliferating?
Are there TAMs?
Are they M1 or M2?
Where are the Tregs?
Are the T cell exhausted?
Immuno-
contexture
Panel
CD8, CD68,
PD-L1,
FoxP3,
PanCK
Immune
Profile
CD8,
CD68,
CD3,
CD20,
PanCK
Activated TIL Status
CD8, CD3, Ki67, Grz B,
PanCK
M1/M2 Polarization
CD8, CD68, PD-L1, PD-1,
CD163
Exhaustion
CD8, CD4, CD20, FoxP3,
PD-1
Comprehensively phenotype the tumor microenvironment for better stratification
18. 18
Providing the Flexibility to Ask Your Specific Question
+ 1 Open Position
A La Carte Markers
+
Labeling Kit
CD8 CD3 CD4
CD163 PD-1 PD-L1
FoxP3 Ki67 Granzyme B
CD20 PanCK CD68
CD45RO SMA
5-Plex Panels + 1 Open Position
Immune
Profile
CD8, CD68,
CD3, CD20,
PanCK + 1
Immuno-
contexture
CD8, CD68,
PD-L1, FoxP3,
PanCK + 1
Activated
TIL Status
CD8, CD3,
Ki67, Grz B,
PanCK + 1
M1/M2
Polarisation
CD8, CD68,
PD-L1, PD-1,
CD163 + 1
Exhaustion
CD8, CD4,
CD20, FoxP3,
PD-1 + 1
19. Flexibility to Answer a Myriad of Questions
19
Map additional phenotypes
CD4 Where are the Helper T cells?
PD-1 Are the T cells exhausted?
CD20 Where are the B cells?
GrzB Where are the activated immune cells?
Which cell types are proliferating?
Ki67
? Marker of choice for specific research question
19
CD8, CD68,
PD-L1, FoxP3,
PanCK
IMMUNO-CONTEXTURE PANEL
20. Answer More Questions Quickly
20
Flexibility allows for easy integration of one additional marker
PD-1
CD20
“Hot” Tumor A
No signs of TLS
(low density B cells)
“Hot” Tumor B
Signs of TLS formation
(high density B cells)
Where are the B
cells in the TME?
Are the T cells
exhausted?
CD8, CD68,
PD-L1, FoxP3,
PanCK
I
M
MUNO-CONTEXTURE PAN
E
L
+
+
0
300
600
900
1200
1500
Tumor A Tumor B
Density of B cells
(cell/mm2)
0
100
200
300
400
500
Exhausted (PD1+) Active (PD1-)
CD8+ T cell Subtypes
CD8, CD68,
PD-L1, FoxP3,
PanCK
I
M
MUNO-CONTEXTURE PAN
E
L
21. PhenoCode Signature Panels offer excellent Reproducibility
Panel 1: PD-1 Panel 2: CD20
Integration of different markers does not impact reproducibility
CD20 Panel 2
CD8, CD68,
PD-L1, FoxP3,
PanCK
I
M
MUNO-CONTEXTURE PAN
E
L
+
PD-1 Panel 1
+
CD8, CD68,
PD-L1, FoxP3,
PanCK
I
M
MUNO-CONTEXTURE PAN
E
L
22. PhenoCode Signature Panels provide excellent Specificity
BarcodedAntibodiesofferSpecificitywithFlexibility
24. Speed Up Spatial Signature Development by 3X
24
Overcomes the barrier of expertise needed to develop 6-plex assays
Custom Built
6-plex panel
PhenoCode Signature
6-plex panel
3X Reduction
From Sample to Data at least 3X Faster!
TIME
DEVELOPMENT & OPTIMIZATION
25. Two Solutions for Rapid Multispectral Imaging
25
Medium throughput (100+ slides/week)
Fusion HT
High throughput (300+ slides/week)
• 7-colour whole-slide imaging
• Brightfield whole-slide scanning
• Single-cell resolution
• Autofluorescence removal and spectral unmixing
• Fully enclosed, touchless automation for 80 slides
• Up to 9 colour multispectral imaging capability
• 7-colour whole-slide imaging
• Brightfield whole-slide scanning
• Single-cell resolution
• Autofluorescence removal and spectral unmixing
• 4 slide automation
• PhenoCycler-compatible (100+ biomarkers)
in 18 min
in 9 min
in 12 min
in 6 min
26. PhenoImager HT Whole Slide Workflow
PhenoImager HT Phenochart inForm Analysis Solutions
QuPath
Phenoptr &
phenoptrReports
inForm
Scan Slides View Unmixed
Preview
Select Regions
for Unmixing
Open Selected Regions
and Unmix
Analyse Individual
Regions in inForm
Advanced Analysis
with phenoptrReports
Export Unmixed
Regions to 3rd-party
Software
Stitch Regions
Together to
Generate WSI
Analyse WSI
3rd-party Software
25-75minutes/ slide
12 minutes/ slide
26
27. Introducing The PhenoImager HT 2.0 Whole Slide Workflow
27
PhenoImager HT Phenochart inForm Analysis Solutions
QuPath
Phenoptr &
phenoptrReports
Scan and
Unmix Slides
View WSI
Select Fields for
inForm Analysis
Open Each of the Fields
Analyse Individual
Regions
Advanced Analysis
with phenoptrReports
Analyse WSI
3rd-party Software
inForm
12 minutes scanning
per slide
+ 8 minutes
unmixing per run
28. 28
PhenoImager HT 2.0 File Formats
DAPI
CD68
Ki67
PanCK
CD8
CD20
PD-1
2.52 GB
HT 1.0
Raw 8-bit
4.86 GB
HT 2.0
Raw 16-bit
2.46 GB
HT 2.0
Raw 8-bit
HT 2.0
Unmixed 16-bit
3.45 GB
Folder size:
2.54 GB
Folder size:
2.49 GB
Folder size:
8.35 GB (contains both raw and unmixed images)
29. HT 2.0 offers three data formats to accommodate different
research needs
Format Description Feature Benefit Compatible with
Data Size
Differential
Extended
Range
(new
default)
16-bit file format
Provides 3-fold margin
for samples that are
brighter than
contemplated in the
protocol
Improve first-pass
success for FL
scans
Few or no
saturation rescans.
HT 2.0
Phenochart 2.0
inForm 3.0
2-3x
Unmixed
Data
Unmixed
QPTIFF from 16-
bit format
(Extended
Range)
Greatly simplified
workflow for 4+ plex
scans
HT 2.0 + PSP is the
easiest workflow for
6 plex analysis
HT 2.0
Phenochart 2.0
inForm 3.0
3-5x
Standard
(Legacy)
8-bit file format Compact file size
Compatibility with
existing SOPs
HT 1.0 and 2.0
All versions of
inForm and
Phenochart
1x
NEW
NEW
29
NEW
FEATURES
30. 30
Akoya’s proprietary file
compression algorithm
Simplified On-Instrument Image Processing & File Compression
GBs
Standardized & Compressed Files, Without Data Compromise, Allow for Flexible Data Transfer
s
31. 31
A Comprehensive Framework for Spatial Applications
Comprehensive
SpatialPhenotyping
SpatialSignatures
Signatures that correlate
with clinical outcomes
through highthroughput
studies
5
SpatialFunctionalState
Reveal functional spatial biology with
m
etabolic& proteinexpression mapping
3
SpatialPhenotyping
Identify cells in-situ with
singlecellresolution
1
AI-basedCellDiscovery
Accuratedetectionof morphologically
distinct cell types
2
SpatialNeighborhoods
Discover how spatial neighbors
self-organize to drive tissue biology
4
33. Objectives
• Key considerations in implementing a robust mIF image analysis workflow
• Using image analysis software in the workflow
• Analysis, data management and spatial profiling examples
34. From Images to Information
Your Quantitative Digital Pathology Experts
OracleBio supplies industry leading image analysis services to Pharma and Biotech worldwide.
Data Output
39. Image QC & Annotations
Pathologist annotated tumour
microenvironment
Invasive margin at the tumour /
healthy tissue interface
Integration of anatomical pathologists
working alongside Image analysts to
support IA workflow
40. Considerations for Tissue
Segmentation:
• Representative training images
• # Tissue types
• Pathology heterogeneity
• ROI Mark-up Annotations
• Incorporating context
• Pathologist input
• Using AI
• Optimal Neural Network
• Deep Learning settings
AI powered Tissue & Cell Segmentation
Mark-up training annotations
Annotations Key
Stroma
Glass
Artefact
Tumor
42. ROI Key: Tumor Stroma
A DeepLabv3+ neural network in VIS was used to develop the classifier using DAPI and PanCK
AI powered Tissue & Cell Segmentation
58. Cell counts, phenotypic data and
spatial data is generated directly
within IA software
Export of cell object data files
for post processing outside of
IA software
Use programming scripts (i.e. MATLAB,
Python) to interrogate data files for
deeper, or more complex, spatial profiling
Spatial data can include infiltration
analysis, cell-cell proximity, minimum /
maximum distance relationships
Exported cell object data can
include mean stain intensity, x,y
vector coordinates, phenotype,
cells size, shape etc
Data Management & Spatial Profiling
59. Example spatial analysis was performed on cores from each
cancer type using the cell object data (x, y coordinates) generated
from the Visiopharm software.
Exported data was processed using a proprietary Python script
and demonstrated that across both cancer types, there was closer
proximity of CD8+ cells to macrophages of the M2 subtype
compared to the M1 subtype.
A – Breast Core
B – Lymphoma Core
Data Management & Spatial Profiling
60. IT Infrastructure & Image Management
Requirements for an effective mIF Quantitative Digital Pathology
IT Environment:
• Ability to manage and store increasing image volumes and file sizes
• Capable of handling complex image analysis tasks at speed and scale
• Efficiently integrate new software updates and new capabilities
• Easily accessible with enabled remote working across multiple locations
• Cost effective & address green credentials
61. IT Infrastructure & Image Management
Virtual computers
launch with Image
Analysis software
Batch
Processing
Unlimited
Image
Storage
Scalable access
to CPU / GPU for
parallel image
processing
User connects via
web portal from
home/office
User configures
virtual computer
type/size
Data
management