AACR 2025 interview series: Arutha Kulasinghe
At AACR 2025, we were fortunate enough to sit down with Arutha Kulasinghe, Associate Professor and Principal Research Fellow, Clinical-oMx Lab, University of Queensland. He shared his insights into the computational challenges of spatial biology, the power of AI-driven insights, and the exciting potential to democratize spatial tools.
Please introduce yourself
My name is Arutha Kulasinghe, I'm an Associate Professor at the Frazer Institute, University of Queensland. I also lead the Queensland Spatial Biology Centre at the Wesley Research Institute, which is a purpose-built spatial biology unit on a hospital campus – here, we’re using spatial biology to understanding patients, tumors, biopsies, and surgical resections across multiple tumor types. Specifically head and neck cancer, lung, breast, ovarian and skin cancers – we're trying to use cutting-edge spatial biology approaches to understanding every patient's individual tumor.
To do this, we map every cell in a tumor; every cell has an X and Y coordinate, and then we try to digitize those patients tumors. Once we've digitized them, that's then a computational challenge, right? You've got 3-5 million cells in space, allowing us to contrast drug responders with non-responders.
"We can explore the patterns in patients' tumors that are linked to response or resistance to therapy and, ultimately, determine how to build translational, predictive signatures that can be applied in the clinic."
What brings you to AACR?
AACR is my favorite meeting: 30,000 people from around the world, scientists, oncologists, patients, advocates, biotech, pharma, industry – it's just a melting point of ideas. We get to see the new cutting edge tools that are coming onto the market and think about how we can apply them in the context of our disease area. Can we take these new technologies – particularly spatial technologies – and apply them to translational cancer projects to better understand the biology behind resistance to therapies, as well as sensitivity to immunotherapies?
"For example, these treatments can cost anywhere from $300,000 to $500,000 per patient per year, yet they only work in 15 to 20% of cases."
And just like chemotherapy, immunotherapy doesn’t work for most patients. So how do we pinpoint that 15 to 20% across multiple tumor types? And for patients who don’t respond to immunotherapy, are there ways to support or enhance the efficacy of these drugs? Ultimately, it’s about finding the right drug for the right patient at the right time.
What initially drew you to your field?
I did my PhD in the liquid biopsy space and quickly realized that detecting tumor-like material in blood is incredibly challenging due to dilution. Everything I worked on in blood kept pointing me back to the tumor itself. So, in my post doc, despite having no prior experience in the tumor microenvironment, I said "this is what I'm going to build my career on". I started the lab with a small seed grant of $10,000. I could run three slides and 20-plex protein.
"This was back in 2018 so when we got the data back, we were blown away – seeing 20 proteins mapped spatially, revealing distinct expression profiles across different areas of the tissue... It felt like the future."
That realization snowballed, and we built the lab from there – it's been an exciting journey. Just five years ago, the field was working with only 10 or 20 proteins – now, we're talking about thousands. The pace of technological advancement is exponential, which is a huge advantage for us; every few months, we get access to newer tools to test on clinical samples, leveraging hospitals and biobanks to source material. We've got projects across multiple solid cancers, but also infectious disease.
We’re currently doing pandemic studies going back to 1918 – it's incredible! One of our students, Lauren Steele, spent a year in Germany gathering samples, searching museums to collect lungs from patients and soldiers who died in 1918. The goal is to compare adult versus children's flu and contrast it across five pandemics, including COVID.
"Now, we have these lungs in our studies, generating beautiful protein data from 105-year-old samples, truly unlocking the archives of history. Though my background is in cancer biology, the infectious disease space is super exciting too."
How are you using spatial biology in your translational research?
One of the biggest challenges in research right now is determining which drugs to give individual patients. Traditionally, this involves analyzing expression profiles, such as mutational load or PDL1 expression, which serves as a single biomarker. It's probably the worst biomarker we have today. But to make progress, we really need to understand the biology of the disease. The key is to comprehensively characterize the microenvironment, and spatial analysis enables us to do just that – examining it cell by cell. We now have high-plex, single-cell resolution, and the beauty of this approach is that we can computationally analyze tissue atlases in incredible detail. It’s an exceptionally powerful tool, revealing tumor heterogeneity in ways we've never seen before.
Tumor heterogeneity is a concept that we throw around when we can't explain why the patient didn't respond to a drug – when there’s inherent resistance. The beauty of spatial biology today is that we pinpoint where your resistance is likely to be down to single cells. The goal is to mechanistically understand what drives this resistance – and if we can, then we ask: can we target it with a drug? So these are the insights that we're getting.
What excites you about the field and how do you think it has the potential to impact patient health?
We've never had tools capable of examining this level of granularity before. But now, not only can we capture data from individual cells within a patient’s tumor, we can also begin to untangle how these tumors respond to therapy and develop resistance. There's a whole AI play in this, along with the development of foundation models, and we're excited about the future and where it's headed. It’s computationally extremely challenging data to look at; my lab is now almost exclusively computer vision students, because tackling these datasets requires expertise in coding and data interpretation.
When we analyze a dataset, we gather around the table with oncologists, pathologists, immunologists, biologists, computational biologists, biostatisticians, and sometimes patient advocates. Each expert approaches the tissue from a different angle – the immunologist might focus on macrophages, the computational scientist searches for patterns in the data, the oncologist wants to locate the tumor, and the pathologist is interpreting the tissue itself. It’s almost like we’re speaking different languages, yet collectively, we’re working to understand every cell within a patient’s tumor. We’ve never had the tools to do this until now. The future of the field is incredibly exciting, as we’re uncovering entirely new biology. The challenge ahead will be scaling these efforts, shifting from analyzing just a few samples to studying hundreds of patients.
What opportunities or challenges do you see in using spatial biology?
The complexity lies in the data – each slide can range from 10 to several hundred gigabytes, making storage, computation, and analysis highly demanding. Having the right compute infrastructure is really important, and we constantly push university supercomputers to their limits, often facing resistance due to resource constraints. The computational side is still evolving, and we need to explore ways to improve efficiency. Some of the transcriptomic datasets we analyze are around 10 terabytes per slide, meaning they’re impossible to load on a standard laptop. Sometimes, when working with one of our data scientists, we’ll try to open an image, only for it to stall on auto-load, leaving us wondering what’s going on.
"The data challenge is massive – one of the biggest bottlenecks in the field."
Computational analysis is a key hurdle, but we also need to move beyond small-scale studies. Spatial research today involves hundreds of antibodies, making assays extremely expensive. Running a well-powered study with 1,000 samples would cost millions, making scalability a major concern. The goal is to achieve statistically robust datasets while ensuring we can computationally analyze them effectively. Emerging spatial agents and foundation models offer promising solutions, and we’ll continue learning from them as they evolve.
"The future is bright – it’s an exciting time, but these remain some of the key challenges we’re tackling."
Where do you see the field progressing in the next five to 10 years?
That’s the dream, right? I believe we’ll get there as training datasets continue to improve. This is something I’m deeply passionate about, having grown up in South Africa and lived across Africa, where these tools simply aren’t available.