We've been hearing for a couple years that important new antibody-payload drugs will need IHC diagnostics, rather than the genomic ones we're used to (KRAS, etc).
Now we have a pivotal paper about the diagnostic TROP2 for datroway / datopotamab deruxtecan. See Garassino et al, Nature Medicine, November 11, 2027.
Key CDx point: They tried regular scoring 1+ 2+ 3+, and it didn't work. They used elaborate software applications...and it did.
Here's a Chat GPT 6 summary.
![]() |
| Computer can read this symbol, Pathologist can't... |
# # #
TROP2’s New Companion Diagnostic Measures
Where the Protein Is
November 18, 2026
TROP2 has been a frustrating biomarker for TROP2-directed drugs. About 90% of non-small-cell lung cancers express the protein, yet conventional immunohistochemistry has done little to identify which patients benefit from datopotamab deruxtecan (Dato-DXd, marketed as Datroway).
- A new Nature Medicine paper reports a more useful question for the pathologist’s slide: How is TROP2 distributed between the tumor cell membrane and its cytoplasm?
- Garassino et al., 2026
That question requires sophisticated image analysis. AstraZeneca developed a proprietary computational pathology platform called quantitative continuous scoring (QCS). Roche is developing the accompanying VENTANA TROP2 companion diagnostic, which combines an immunohistochemistry assay, slide scanning, digital pathology software, and an algorithm incorporating QCS. The resulting measure is the normalized membrane ratio, or NMR.
Roche received FDA Breakthrough Device Designation for the system in 2025; the designation is not FDA approval. diagnostics.roche.com
Why the familiar H-score failed (1+ 2+ 3+)
An H-score turns the pathologist’s familiar 0, 1+, 2+, and 3+ staining assessments into one weighted number, usually by multiplying the percentage of tumor cells at each intensity by its intensity grade. It is a common term of art for a structured version of visual IHC scoring. It describes how much staining a tumor shows.
In the paper’s initial Dato-DXd study, that was the wrong measurement for predicting benefit. Response rates were approximately 18% in the H-score-high group and 17% in the H-score-low group. Progression-free survival was likewise similar. The investigators also found no useful association between manual TROP2 scoring and response in their subsequent phase III trial analysis. A tumor could stain strongly for TROP2 and still fail to respond.
QCS examines digitized IHC slides at a different level. It identifies tumor regions, delineates individual tumor cells, and measures TROP2 staining separately at each cell’s membrane and in its cytoplasm. NMR expresses membrane staining relative to membrane plus cytoplasmic staining. In this analysis, a lower ratio—more cytoplasmic staining relative to membrane staining—was the favorable pattern. The selected positive classification required at least 75% of tumor cells to have an NMR of 0.56 or less. This is a precise, cell-by-cell threshold; it is not something a pathologist can reliably reproduce by assigning a whole slide a 1+, 2+, or 3+.
There is a plausible biological explanation. An antibody–drug conjugate must bind its target and deliver a toxic payload. The authors’ cell-line experiments linked the lower NMR pattern with greater Dato-DXd internalization. But an IHC image is a snapshot: cytoplasmic staining does not, by itself, prove that a particular patient’s tumor is actively internalizing the drug. The clinical outcome data are the stronger evidence for the biomarker.
The result that makes this a predictive biomarker story
The investigators developed the NMR concept using samples from an early Dato-DXd study, then evaluated it retrospectively in the randomized TROPION-Lung01 trial of Dato-DXd versus docetaxel. In the paper’s principal group—221 biomarker-evaluable patients with nonsquamous lung cancer and no actionable genomic alteration—median progression-free survival among Dato-DXd recipients was 7.2 months for NMR-positive patients versus 4.0 months for NMR-negative patients. Median overall survival was 17.0 versus 10.8 months.
The docetaxel arm supplies the crucial comparison. Its median progression-free survival was 4.1 months in NMR-positive patients and 4.4 months in NMR-negative patients. The treatment-by-biomarker interaction for progression-free survival was P = 0.0098. In other words, NMR did not merely pick out patients who tended to do better regardless of treatment; it distinguished outcomes specifically associated with Dato-DXd in this analysis.
The comparison also keeps us from overstating the result. Some NMR-negative patients responded to Dato-DXd: the reported response rates in this lung cancer group were 36.8% for NMR-positive and 22.5% for NMR-negative patients. NMR is a tool for selecting a group with greater expected benefit, not a claim that every negative patient is incapable of benefiting.
An approved drug, but a companion diagnostic still in development
Datroway already has three U.S. indications: previously treated EGFR-mutated advanced lung cancer; unresectable or metastatic triple-negative breast cancer in patients who are not candidates for PD-1/PD-L1 inhibitor therapy; and previously treated HR-positive, HER2-negative unresectable or metastatic breast cancer. None of those indications requires a TROP2 NMR result. The current lung indication selects by EGFR mutation, not by this TROP2 assay. accessdata.fda.gov
The proposed companion diagnostic addresses a different opportunity: identifying patients with previously treated nonsquamous lung cancer without an actionable genomic alteration who are most likely to benefit from Dato-DXd. Roche describes that as the device’s intended use, while stating that the product remains in development and is not yet commercialized in the United States. An FDA decision in 2027 is a plausible prospect, but its timing and outcome have not been established. diagnostics.roche.com
Nor does this lung cancer publication validate NMR as a selection assay for breast cancer. Future biomarker-defined breast indications are conceivable, but they would need their own clinical evidence and regulatory review. AstraZeneca’s current pipeline does include a phase III program in NMR-positive nonsquamous lung cancer, illustrating how the assay could move from retrospective analysis into prospective patient selection. AstraZeneca
This is an important publication, with a real limitation: the clinical biomarker analysis was retrospective, and the final threshold was selected and tested using data from TROPION-Lung01, with resampling and cross-validation to assess robustness. Prospective results will matter. Still, the paper shows what computational pathology can contribute when the measurement is tied to the drug’s mechanism.
- Ordinary staining intensity missed the signal; the location of TROP2 within individual tumor cells revealed it.
#
Sidebar | A LinkedIn horizon scan for TROP2
Measure what the drug needs to do. A commentary on the Garassino paper by Luis Canu argues that the failure of H-score was more than a problem of pathologists disagreeing about 1+, 2+, and 3+ staining. Even a perfectly reproducible H-score would still measure total staining, while NMR asks where TROP2 sits within each tumor cell. The post suggests that the cytoplasmic pattern may reflect internalization. That is a plausible interpretation, not something the slide alone can prove.
The target is only one part of an ADC. In a separate discussion of metastatic breast cancer, Ermelinda Damko emphasizes payload biology and early changes in circulating tumor cells over baseline TROP2 or HER2 abundance. Her argument concerns a different study and should not be treated as validation—or refutation—of Roche’s NMR assay in lung cancer. It does reinforce the question raised by Garassino and colleagues: which aspect of target biology actually predicts that this drug will work? linkedin.com
More TROP2-directed designs are coming. Yasin Oduk describes an early, experimental breast cancer formulation that would use TROP2-binding peptides to help deliver an encapsulated metabolic inhibitor. It is far from an approved therapy, but it illustrates why “TROP2-positive” may never be a universal biomarker: different drugs can bind the same target yet depend on different payloads and delivery mechanisms.
The common thread: these posts are useful prompts and contexts for drug and assay developers, rather than additional clinical evidence for NMR.



