By Bethany Remeniuk, PhD, Nicole Couper, and Yi Zheng, PhD
Combination therapies, which target multiple immune checkpoints at once, have the potential to deliver improved outcomes for cancer patients. The development of clinically relevant predictive tools to stratify immune checkpoint inhibition responders from nonresponders will be critical for the advancement of such treatments.
Multiplexed cell and spatial phenotyping of the tumor microenvironment (TME) can provide a deeper understanding of complex interactions between tumors and the immune system, setting the stage for improved patient stratification. Spatial biology provides advantages over other technologies by revealing a clearer and more detailed picture of cellular- and protein-level co-expression, localization, and arrangements within the TME, which in turn can be used to develop prognostic biomarkers called spatial signatures based on the spatial distribution of certain phenotypic features.
PhenoCode™ Signature Panels simplify spatial biomarker assay development and validation when used in combination with Akoya’s PhenoImager® platform. Each of the customizable multiplex panels include key markers for comprehensive mapping of the TME and immune status, providing a rapid, quantitative, end-to-end spatial phenotyping workflow.
Fast and flexible PhenoCode Signature Panels
PhenoCode Signature Panels are designed in a customizable format, allowing easy integration of an additional immune cell or checkpoint marker to a preset five-plex panel (Figure 1). Each panel focuses on distinct areas of tumor biology and potential response to therapy that are of greatest interest to translational and clinical researchers. These multiplex immunofluorescence panels combine Akoya’s patented barcode chemistry with Opal-tyramide signal amplification, providing similar accuracy and sensitivity to the gold-standard chromogenic immunohistochemistry. The time required to develop and validate new spatial signatures using the panels is reduced three-fold compared to conventional assay development.
The panels are used in a seven-step procedure to answer key questions and interrogate the TME. The first three steps follow the traditional formal-fixed paraffin-embedded sample preparation, with steps one through three corresponding to slide preparation (baking and dewaxing), antigen epitope retrieval, and blocking. In the fourth step, the slides are stained with a primary antibody cocktail in which each antibody has been conjugated to a given barcode.
In step five, a single antibody is revealed at a time, beginning with the hybridization of a complementary oligo barcode conjugated to horseradish peroxidase. Tyramide signal amplification is used in step six to amplify immunohistochemistry detection by covalently depositing an Opal fluorophore near the targeted antigen. Once signal amplification is complete, step seven begins. The horseradish peroxidase–conjugated oligo is dehybridized. Steps five, six, and seven are repeated for each antibody, labeling the markers with the different dyes until all have been revealed.
Spatial signatures for NSCLC
In the study outlined below, a PhenoCode Signature Immuno- Contexture Human Protein Panel was used to accelerate identification of spatial signatures in non-small cell lung cancer (NSCLC) that may reliably predict response to immune checkpoint inhibition.
NSCLC patients can have impaired immune responses within the TME, leading to tumor growth progression and poor prognosis. Accurate cell phenotyping combined with spatial phenotyping can provide a better understanding of complex cellular interactions underpinning the tumor-immune response.
PhenoCode Signature Panels and associated artificial intelligence (AI)-powered image analysis methods were used to identify populations of immune cells and their functional status, as well as their interactions within the TME in a set of NSCLC tissue cores from patients treated with first-line standard-of-care and second-line immuno-oncology treatment. Patient groups included responders (R—full responders, partial responders, and stable disease) and nonresponders (NR).
Formalin-fixed paraffin-embedded NSCLC tissue microarrays (TMAs), comprising n = 38 cores containing a range of carcinomas and pathological Tumor Node Metastasis (pTNM) stages, were stained using the PhenoCode Signature Immuno-Contexture Human Protein Panel. This panel includes markers for T cells (CD8 and FoxP3), macrophages (CD68), checkpoint inhibitors (PD-1 and PD-L1), and PanCK as a tumor marker.
Stained TMAs were scanned at 20× magnification on a PhenoImager HT multiplex imaging system. A total of 36 cores passed image QC and progressed to image analysis. Deep learning algorithms were developed to segment each core into tumor and stroma regions of interest (ROIs) and to accurately detect and classify different cell populations. A DeepLabv3+ neural network was used to develop the classifier using DAPI and PanCK. A customized cell analysis algorithm was trained using a U-Net neural network to detect individual cell lineages and subsequent phenotypes of interest.
A hierarchical approach detected CD8, CD68, tumor cells, and then DAPI cells. Staining variance for CD8, CD68, and DAPI was overcome by generating training labels for the three cell types across the TMA cores and using the three markers as input channels for deep learning training. Spatial analysis was performed using an OracleBio proprietary program to calculate readouts for mean nearest neighbor distances between cell populations, as well as neighborhood analysis for selected phenotypes (Figure 2).
Immune cell counts, phenotypes and spatial interactions were identified within the tumor and stroma ROI per core. Data included total and negative cell phenotype counts, cell density in tumor and stroma, as well as average cell distances between specified phenotypes and neighboring spatial interactions in each of the 36 cores in the TMA set.
Immune cell subsets quantified included FoxP3+, CD8+/PD-1+, CD8+/FoxP3+/PD-1+, and CD68+/PD-L1+. Tumor cells of interest included PanCK+ and PanCK+/PD-L1+. Results indicated single FoxP3 per mm2 was significantly lower in the tumor ROI of the R group vs. the NR group (p ≤ .05). A trend was observed in the ratio between CD8 (single and PD-1 dual combined populations) and FoxP3 (single population), where there was a higher proportion of CD8 phenotypes in both tumor ROI and stroma ROI of the R group vs. the NR group. Spatial interactions between phenotypes varied across individual cores, and although trends were observed, no significant differences were found between the R group and the NR group (Figure 2).
The combination of high-quality, spatial phenotyping data provided by the PhenoCode Signature Panel, coupled with deep learning quantitative image analysis techniques, enabled detailed characterization of the complex cellular interactions, at both the functional and spatial levels, within the TME of immuno-oncology-treated NSCLC tissue.
Conclusion
Biomarker discovery based on spatial biology establishes a path toward the use of multiplexed imaging in the clinic, as technologies and workflows become more practical, high throughput, and analytically robust. PhenoCode Signature Panels provide an off-the-shelf, flexible six-plex option that allows more thorough interrogation of the TME with minimal user development requirements.
The ability to deploy signature panels supported by PhenoCode chemistry can accelerate the identification of spatial signatures with the potential to reliably predict response to immune checkpoint inhibition therapy in clinical trials.
Bethany Remeniuk, PhD, is the associate director of laboratory applications at Akoya Biosciences. Nicole Couper is a deputy clinical operations manager at OracleBio. Yi Zheng, PhD, is a director of reagent development at Akoya Biosciences.
Comments are closed.