AstraZeneca has used artificial intelligence to devise a unique biomarker for its Daiichi Sankyo-partnered datopotamab deruxtecan (Dato-DXd), hoping to differentiate the antibody-drug conjugate (ADC) from competitors while offering an explanation why the TROP2-directed therapy appears to work only in some non-small cell lung cancer (NSCLC) patients.
The new biomarker is called normalized membrane ratio of TROP2 by quantitative continuous scoring (NMR-QCS). Just as its name is complicated, so too are the criteria to determine a patient’s biomarker status.
Unlike traditional biomarkers such as HER2, where biomarker status is determined by cell surface expression of the cancer-related protein, NMR-QCS measures the ratio of TROP2 expression in the membrane relative to the cytoplasm of tumor cells. To be considered NMR-QCS-positive under the company’s current model, a patient’s sample must have at least 75% of tumor cells with a TROP2 normalized membrane ratio of no more than 0.5585.
In a retrospective analysis of the phase 3 TROPION-Lung01 trial in previously treated NSCLC, Dato-DXd showed much better efficacy in those with QCS-NMR-positive tumors than those with negative disease. Compared with the chemotherapy docetaxel, Dato-DXd lowered the risk of tumor progression or death by 43% in QCS-NMR+ patients, whereas the ADC performed worse than docetaxel in QCS-NMR- group, which experienced a 16% higher risk, according to data presented at the 2024 World Conference on Lung Cancer (WCLC).
QCS-NMR+ patients who took Dato-DXd went a median 6.9 months without tumor progression, versus 4.1 months for docetaxel takers. In the QCS-NMR- group, the median numbers were 2.9 months for Dato-DXd and 4 months for control.
AstraZeneca is now partnering with Roche to co-develop and commercialize the TROP2-QCS biomarker companion diagnostic and digital pathology algorithm, which will combine AZ’s proprietary computational pathology platform with Roche’s navify digital pathology image management system.
AI-assisted algorithm
To develop the biomarker, AZ, with the help of AI, looked through thousands of different factors and permutations of those factors before landing on a combination that’s most predictive of Dato-DXd’s progression-free survival (PFS) benefit in second-line nonsquamous NSCLC without actionable genomic alternations, David Fredrickson, executive vice president of AZ’s oncology business unit, explained in an interview with Fierce Biotech.
Initially, the company performed a hypothesis-free exploration of potential features linked to PFS improvements among NSCLC patients in the phase 1 TROPION-PanTumor01 study. After some work, AZ identified QCS-NMR as the most promising feature based on its correlation with PFS. The exact cut points to determine the patient’s biomarker status were optimized with data from cases of nonsquamous NSCLC without actionable mutations in the phase 3 TROPION-Lung01 trial.
AZ’s search for the biomarker comes amid a setback with TROPION-Lung01. The trial previously met one of its dual primary endpoints of PFS. But Dato-DXd’s disease progression benefit was observed only in patients with nonsquamous histology—not squamous. Therefore, Daiichi and AZ filed an application with the FDA for previously treated nonsquamous NSCLC, with a target decision date set for Dec. 20.
But the contrasting efficacy between the two histologies lacked a scientific explanation, and AZ’s examination by traditional cell surface expression of TROP2 didn’t provide the answer. So AZ suspected that TROP2 expression beyond the cell membrane in the cytoplasm matters, too. Because Dato-DXd has a very stable linker that almost always releases the toxic payload after entering the tumor cells, that internalization process becomes critical for the drug’s cancer-killing effect.
Among the biomarker-evaluable patients in TROPION-Lung01, QCS-NMR+ tumors were more prevalent (66%) in nonsquamous disease than in squamous histology (44%), providing an alternative—and potentially better—explanation for the efficacy dichotomy than histology alone.
By magnitude of PFS improvement, Dato-DXd’s 43% advantage over chemo in QCS-NMR+ tumors was better than the 37% recorded in nonsquamous disease.
AZ and TROPION-Lung01 investigators now argue that QCS-NMR has potential to be a predictive biomarker for response to Dato-DXd. But, for now, that conclusion appears to fall in a circular reasoning fallacy; namely, because the biomarker was also trained based on PFS results from TROPION-Lung01, the same data set was used to both modify and confirm the model.
Fredrickson said independent validation was performed on other studies before applying the model to TROPION-Lung01. But he acknowledged that more work is needed to incorporate the biomarker prospectively into studies.
“We are still in the early days of really developing this platform and specifically applying it to data,” Fredrickson said. “But I think what’s really important about what we’ve accomplished with this data set is that we’ve been able to apply this to TL01 and be able to see that it’s predictive.”
AZ is now utilizing the biomarker in the phase 3 AVANZAR trial, which is evaluating Dato-DXd in combination with Imfinzi and chemo as first-line treatment of advanced NSCLC without actionable genomic alterations, Fredrickson said. The trial’s primary endpoints measure PFS and overall survival in “TROP2 biomarker positive population,” according to ClinicalTrials.gov. The study could read out next year.
Regulatory uncertainty
Fredrickson stressed that QCS-NMR “may not be the only path forward,” for Dato-DXd in NSCLC, but that it’s just “one of the shots on goal.”
AZ has put Dato-DXd’s peak sales potential at more than $5 billion, with the “overwhelming majority” tied to the drug’s potential use in first-line NSCLC, according to Fredrickson. The underlying assumption is that “the biomarker can allow us to go above and beyond what we have within our plan,” he said. AVANZAR is currently the only phase 3 trial to which AZ and Daiichi have prospectively adopted the biomarker.
AZ has used the biomarker retrospectively on other large phase 3 studies, including the POSEIDON and NEPTUNE trials for Imfinzi and Imjudo, and found similar prevalence of histology in which the biomarker is enriched in nonsquamous population. But the biomarker is creating a complication for Dato-DXd’s open FDA application in second-line nonsquamous NSCLC.
Previously, Leerink Partners analysts have questioned the approvability of Dato-DXd because the PFS benefit, even in the nonsquamous population, was modest, and because the nonsquamous-versus-squamous situation lacked a biological rationale. The failure of the TROPION-Lung01 trial on its other dual primary endpoint, overall survival, threw more doubt into the equation.
Now, the biomarker might undermine AZ’s nonsquamous bid, even though the histology analysis was prespecified.
According to data presented at the WCLC conference, in a subgroup of nonsquamous patients without actionable genomic mutations who tested negative for QCS-NMR, Dato-DXd was linked to a 22% increased risk of tumor progression or death compared with docetaxel.
On the flip side, the biomarker strategy could help Dato-DXd potentially reach some squamous patients. In TROPION-Lung01, among squamous patients with biomarker-evaluable tumors, 44% were QCS-NMR+. However, investigators didn’t provide a separate efficacy analysis of squamous patients for the biomarker. The squamous data are encouraging, but the dataset is too small to be reliable and needs further validation, an AstraZeneca spokesperson told Fierce Biotech.
These contrasting results based on different cuts of patient populations raise the possibility that the FDA might reject Dato-DXd and ask for a different phase 3 trial with more data.
Fredrickson didn’t offer any update on AZ’s discussions with the FDA. But he argued that a reasonably high level of prevalence of biomarker-positive patients in the nonsquamous histology could support histology as predictive of response.
Real-world application
AZ is rolling out the biomarker even as its TROP2 ADC competitors Gilead Sciences and Merck & Co. didn’t seem to have the same nonsquamous-versus-squamous problem.
When broad patient coverage seems out of reach, strong efficacy in TROP2 biomarker-positive patients could compensate for a narrower indication and increase Dato-DXd’s competitiveness, Leerink Partners analysts have argued. The emphasis is on “strong” efficacy. During a recent interview with Fierce Pharma, Leerink’s Daina Graybosch, Ph.D., predicted that Dato-DXd could risk becoming irrelevant in the TROP2 ADC race if its biomarker-only data weren’t notably better than the histology-agnostic data from Gilead’s Trodelvy or Merck’s Kelun Biotech-partnered sacituzumab tirumotecan (sac-TMT).
The latest biomarker data bode well for Dato-DXd, though. The 43% PFS improvement in QCS-NMR+ patients was markedly better than the 16% figure Gilead’s Trodelvy pulled off against docetaxel in the failed EVOKE-01 study. Leerink has suspected that a relatively low representation of patients with actionable genomic alterations might have cost Gilead a positive trial. TROP2 ADCs have been found to work much better in NSCLC patients with actionable genomic mutations than in those without the abnormalities.
Even in nonsquamous patients without mutations, Dato-DXd’s benefit remained strong at 48%, according to data shared at WCLC.
The question then moves to whether the biomarker test would be too cumbersome for doctors, especially considering that the biomarker may be different for Dato-DXd in other tumor types given that it was trained with NSCLC data.
“We know that whenever we’ve been able to really demonstrate the value of a biomarker, that the willingness of providers […] goes up considerably,” Fredrickson said.
The QCS-NMR test still utilizes a common IHC assay to assess TROP2 expression. The diagnostic process is automated: IHC-stained whole-slide images are analyzed under computational pathology to precisely quantify TROP2 NMR for every tumor cell.
Still, the digitalization of slides requires additional infrastructure, Fredrickson acknowledged. AZ has already started prepping for that and is talking with labs that are interested in digital pathology not only for Dato-DXd.
“I […] believe that if the data show that the QCS biomarker is making a difference, and in particular, if we’re able to bring it across multiple ADCs, that I think that the willingness and the desire to replace classic chemotherapy with something better is high enough, that we will see that the supply side gets adequately built out,” Fredrickson said.