From organoids to the clinic?

One of the joys (and surprises) of being a computational postdoc is the sheer variety of projects that come your way. Back in 2018, while working at the CRUK Cambridge Institute, I found myself with a unique opportunity. Maria Vias, a staff scientist in James Brenton’s group, reached out for help processing DNA sequencing data from ovarian cancer organoids. As the resident DNA copy number expert, I was happy to jump in—especially since I’d already analyzed drug response data for these same organoids for Maria’s upcoming paper.

What made this project particularly exciting was the timing. I had just developed a new method to identify patterns of chromosomal instability (CIN) in DNA—essentially, signatures that could act as biomarkers for the underlying causes of CIN in cancer. We suspected these signatures might also predict how tumors respond to certain therapies. So, I applied my method to the organoid data and started hunting for correlations between CIN signatures and drug response.

The results were striking. Tumors lacking a particular CIN signature related to replication stress were much more sensitive to doxorubicin, a common chemotherapy drug. Even better, we could validate this finding using similar data from short-term cultured patient-derived spheroids. Sensing the clinical potential, James Brenton and I pitched the idea to CRUK’s commercial team, and a patent was soon filed.

From patent to startup: The bumpy road of translation

With a shiny new publication and patent in hand, I naively assumed our discovery would quickly find its way to patients. Reality, of course, had other plans. Translating a biomarker from the lab to the clinic is less about science and more about business—and we needed a serious commercial push.

Enter the Judge Business School at Cambridge, which was hosting a venture creation weekend. My colleague Ania, a genomics postdoc, and I delivered a rapid-fire pitch: could we turn chemotherapy into precision medicine? Our idea was selected, and over an intense weekend, we formed a team (including Jason Yip, now our CEO), built a business plan, and won a spot on the business school’s accelerator program. This early success gave us the confidence to spin out Tailor Bio, where I now serve as part-time CSO alongside founders Jason, Ania, and my mentors James Brenton and Florian Markowetz. We licensed the patent, won a CRUK Accelerator Award, and joined the Illumina Accelerator to develop a diagnostic test.

Scaling up and hitting roadblocks

Around this time, I started my own lab at the Spanish National Cancer Research Centre (CNIO). Despite the chaos of the COVID pandemic, our team—especially PhD student Ruben Drews, who I co-superivised with Florian, and postdoc Barbara Hernando—expanded the CIN signature approach to a pan-cancer scale. This work landed on the cover of Nature and earned us a second patent over the methodology (also licensed to Tailor Bio).

But commercializing a test for older chemotherapies proved tough. Investors were wary, since the main benefit would be cost savings for healthcare systems, not blockbuster profits. So, we pivoted to government funding, securing support from Innovate UK to keep developing our chemotherapy response prediction test.

Lessons from rejection and a change in focus

Our expanded CIN signature study hinted that these biomarkers could also predict responses to platinum-based chemotherapy. Working with Ania in James Brenton’s lab, we retrospectively tested our method on 50 ovarian cancer patients and saw promising results for both doxorubicin and platinum. Confident, we submitted our findings to Nature Genetics.

The reviews were constructive but highlighted a critical flaw: our patient cohort lacked the clinical RECIST response data needed for robust validation of sensitivity prediction. Since this was a retrospective study, there was no way to fill this gap. Around this time, Joe Thompson, who had been working on the technology at Tailor Bio, joined my lab as a PhD student. Together with Barbara, we realized that predicting resistance, rather than sensitivity, might offer greater clinical value and allow us to overcome this roadblock. This strategy was appealing as we didn’t want to mess with the patients currently receiving the chemotherapy that are sensitive, but rather, help those resistant to avoid the toxic side-effects.  

Rethinking clinical trials with real-world data

Meanwhile, Tailor Bio secured another Innovate UK grant and began designing a prospective clinical trial for our doxorubicin response test. The idea was to help decide whether platinum-resistant ovarian patients should receive doxorubicin or paclitaxel first as a second-line therapy. However, when we presented the trial protocol to gynecological cancer experts, the response was lukewarm. There were ethical concerns about treating patients predicted to be resistant, and little enthusiasm for trials involving older drugs.

Just as we were running out of options, we stumbled upon a study that had annotated clinical data for ovarian cancer samples in TCGA, including treatment response endpoints like time to treatment failure. With the rise of real-world data in drug approvals, we saw an opportunity: could we retrospectively validate our resistance predictors using these datasets? Energized, we annotated clinical data for other tumor types in TCGA and gained access to the Hartwig Medical Foundation cohort. In parallel, Barbara also showed that CIN signatures could predict taxane resistance in ovarian cancer, extending our approach to a third chemotherapy.

To our surprise, we now had data on hundreds of patients treated with our drugs of interest—and other standards of care—across multiple cancers. This opened up a new possibility: could we use real-world data to emulate randomised control biomarker trials? Some quick power calculations suggested it was feasible for ovarian, breast, and prostate cancers, and for a non-randomized study in sarcoma.

Success, publication, and the road ahead

This real-world data strategy paid off. Extending the study to include taxane resistance prediction and this more robust performance assessment, we appealed our initial rejection, and after another round of review, our study was accepted for publication.

Looking forward, our goal is to bring this technology into the clinic. We’ve received support from NextGenerationEU funds via the Spanish Ministry of Digital Transformation and Public Service, and are now conducting analytical validation of our test. While we know real-world data can’t fully replace prospective clinical trials, we hope it will enable faster, more practical prospective assessments of clinical utility. Our next step? A prospective trial of our platinum resistance test in lung cancer, planned for 2026.

Translating a biomarker from discovery to patient care is rarely straightforward. But with persistence, collaboration, and a willingness to rethink the rules, it’s possible to turn a chance postdoc project into a technology with real clinical promise.