Researchers at Johns Hopkins University (JHU) recently published findings from AstroPath, the result of a unique collaboration between the astrophysics and pathology departments at JHU. The platform enables deep whole-slide imaging and spatial profiling of microscopic tumor sections by combining the Phenoptics™ platform for multiplex immunofluorescence with sky-mapping algorithms derived from the Sloan Digital Sky Survey.
Using AstroPath, JHU investigators were able to discover and validate a novel biomarker signature to predict immunotherapy response in advanced melanoma, and their results were published in Science.
Sneha Berry, PhD
Lead Scientist, Akoya Biosciences
Cliff Hoyt
VP of Translational & Scientific Affairs, Akoya Biosciences
We spoke to first author, Sneha Berry, who has since joined Akoya as a Lead Scientist, and Cliff Hoyt, VP of Translational & Scientific Affairs, to learn more about AstroPath and the findings from this groundbreaking study.
What is AstroPath, and how does it facilitate immuno-oncology biomarker discovery?
Sneha: AstroPath is an end-to-end platform that facilitates the capture, curation, and querying of multiplex immunofluorescence imaging. It was created to ensure the generation of high fidelity, large datasets that simultaneously map multiple cell types and proteins. High-plexing, spatial context and improved accuracy, together, extend our ability to study complex tumor microenvironments beyond standard histopathology biomarker discovery tools.
For both astronomy and pathology, spectral information is gathered, objects are defined, be it cells or stars, and these objects are then mapped in relation to each other.
AstroPath was born out of a unique collaboration between astronomers and pathologists at Johns Hopkins. Why did you turn to astronomy to solve your data analysis challenges and how did it lead to AstroPath?
Sneha: We realized that astronomy had many parallels with what we hoped to achieve with multiplex immunofluorescence pathology. For both astronomy and pathology, spectral information is gathered, objects are defined, be it cells or stars, and these objects are then mapped in relation to each other. Additionally, all these aspects are already performed at high scale in astronomy, making the tools and strategies used for mapping stars in the sky perfect for translation to mapping cells in tissues.
How was AstroPath used to discover and validate a spatial biomarker predictive of immunotherapy response in melanoma?
Sneha: We wanted to identify a cell signature that would be able to predict patient outcomes to anti-PD-1/PD-L1 therapies. To achieve this, we designed a panel that targeted cells and proteins of interest, involved in the PD-1/PD-L1 axis. After designing the panel, we integrated it into the AstroPath workflow. Patient samples were stained with a carefully optimized assay and data was collected across the entire tumor specimen, by scanning the whole slide. AstroPath enabled us to handle and manage such massive amounts of data.
Additionally, large datasets are more prone to compounding errors. Therefore, the AstroPath platform was designed to mitigate potential sources of error at all stages, from staining to imaging to analysis. To identify a highly predictive signature, we used a user-independent approach to optimize slide sampling and combined the most informative cell phenotypes. As a result of our focus on generating high quality data, the identified signature proved to be robust enough to validate in an independent cohort.
Melanoma sample stained with a six-plex assay containing PD-1, PD-L1, CD8, FoxP3, CD163, and tumor marker.
How does the Phenoptics workflow contribute to the AstroPath platform?
Cliff: The AstroPath platform is built on top of the Phenoptics workflow. The Phenoptics workflow provides accurate cell-level data across whole slides, and it’s the ability to look at up to six proteins or more on a per cell basis that enables interrogations into the spatial biology. The analytical robustness of the workflow ensures that the data produced by AstroPath is reliable and durable, which is borne out in this study.
The Phenoptics workflow.
Going forward, you’ll find that the leading public journals will require more levels of validation for biomarker-oriented studies.
This study placed a strong emphasis on validation. What can cancer researchers learn from this biomarker development framework as they seek to validate novel spatial biomarkers?
Cliff: It’s now becoming more commonly understood that when you’re dealing with high dimensional data, it’s easy to overfit your data and think you’ve got a reliable assay or signature based on a training set of tens to hundreds of samples. But to really validate a signature – because there’s so many variables that go into these things – you need to try that signature on a fresh new set of cases, preferably from a different institution altogether, to really understand how robust a signature is.
Up until recently, a lot of biomarker-oriented publications that tout powerful new signatures don’t have validation. Then, when they get tried on new sets of samples, the performance falls off. Going forward, you’ll find that the leading public journals will require more levels of validation for biomarker-oriented studies.
Sneha: We were very excited to see the biomarker validate as well, particularly because the second cohort was collected at a different site. There are many pre-analytical variables, such as fixing procedures, that can impact results. We were very excited to see that our assay did hold up against these.
How do you reduce the nearly infinite number of things you could measure in a whole section down to the optimum actionable score? It’s a daunting task.
What are the future directions for AstroPath, and how might this platform impact the future of precision medicine?
Sneha: AstroPath can be used as a biomarker discovery tool for any indication or therapy. In the context of immunotherapy, additional panels that target different markers could provide us with even more information to better understand tumor immune microenvironments and further refine patient classification. As the most predictive sets of markers for various drug regimens are identified, these panels can be transferred to the clinic and used to inform patient care.
Cliff: The future directions of AstroPath, I think, are going to focus mainly on continuing to explore the immune landscape of cancer. They’re just getting started, and it’s a largely undiscovered space. How do you reduce the nearly infinite number of things you could measure in a whole section down to the optimum actionable score? It’s a daunting task.
I expect that AstroPath will, for the foreseeable future, be looking at different indications, different types of therapy, and looking for the optimum predictive signatures. It really is a very powerful and unique discovery platform. I think it’s one of a kind in its ability to explore highly dimensional datasets very quickly. And the idea is that the signatures discovered this way, which is a significant task, can be reduced to a simple algorithm that’s run in a high throughput, standardized clinical workflow.
You don’t need all that computational power once the discovery is done. It will become a canned algorithm, and that’s where the impact on precision medicine will occur. Once the Phenoptics platform is fully clinical ready, the signatures that are discovered can be then deployed in a variety of clinical settings. That’s the vision – that these signatures will become part of the standard of care.
