From Nature Med, Nat Rev Clin Oncol, and NEJM.
##
These papers support a substantial opportunity for digital pathology manufacturers—but the opportunity is to predict benefit from a particular ADC in a particular clinical setting, not simply to measure its target more precisely. TROP2 computational pathology is an especially important example because subcellular localization may contain information that conventional expression scoring misses. However, the papers also explain why many successful ADCs will not require a target-expression companion diagnostic.
For a manufacturer anticipating FDA CDx approvals, the strongest strategy is to establish a reliable platform for drug-specific diagnostic claims, while developing the ability to incorporate spatial biology, payload sensitivity, and treatment history. The likely future is a portfolio of validated applications sharing infrastructure, rather than one universal “ADC response score.”
The three articles are reviews or perspectives, so the appropriate synthesis is a critical narrative review rather than a pooled statistical meta-analysis. The assessment below separates established clinical uses, investigational biomarkers, and commercial forecasts.
The three papers address complementary questions.
Hurvitz explains the clinical demand: an expanding treatment menu is making selection and sequencing harder.
“Advances in Systemic Treatment of Breast Cancer,” NEJM, October 8, 2026, describes how targeted agents, ADCs, and immunotherapy are moving across breast-cancer subtypes and into earlier treatment settings. The article itself is dated 2026, despite the uploaded filename.
Its importance to a diagnostics manufacturer is the growing number of consequential decisions. Clinicians increasingly must choose among therapies with overlapping indications, different toxicities, and uncertain optimal sequencing. HER2 assessment now informs treatment across a wider expression range, while TROP2-directed ADCs can be used in settings without requiring TROP2-expression selection.
Hurvitz also makes clear that earlier treatment is not automatically better for every patient. Moving a potent ADC forward can increase exposure to continuous cytotoxic therapy, and progression-free survival gains do not settle every question about overall survival, quality of life, or optimal sequencing.
Manufacturer implication: demand will extend beyond identifying eligibility. There is an emerging need to determine who warrants early intensification, who can defer it, and which therapy remains useful after prior ADC exposure. Those are attractive future diagnostic claims, but they require evidence beyond an association between staining and response.
Mosele and colleagues explain the therapeutic pipeline: new constructs will change what needs to be measured.
“The next generation of antibody–drug conjugates,” Nature Medicine, August 3, 2026, argues that innovation in ADC chemistry and combinations is outpacing the capacity to evaluate them clinically.
The authors envision improvements in antibody architecture, linker stability and cleavage, payloads, drug-to-antibody ratios, and combinations. Their longer-term proposal includes libraries of ADC constructs selected according to an individual tumor’s biology.
Particularly relevant is their proposed composite predictor: an AI-assisted IHC assay measuring target expression and localization, combined with an RNA- or protein-based measure of payload sensitivity. They also consider liquid biopsies, molecular imaging, and functional testing in patient-derived models.
This is a research agenda, not evidence that a comprehensive predictor is ready for clinical deployment. The authors emphasize how much remains unknown about drug delivery, intracellular processing, and the quantitative contribution of bystander killing in patients.
Manufacturer implication: the diagnostic platform must accommodate changing drug biology. A score optimized for one antibody–linker–payload construct cannot be presumed to work for another ADC merely because both bind the same antigen.
Liu and colleagues provide the most directly actionable diagnostic-development framework.
“A modular framework for antibody–drug conjugate biomarkers,” Nature Reviews Clinical Oncology, October 2026, organizes biomarkers around target biology, linker processing, and payload susceptibility, modified by the tumor microenvironment and host factors.
This is the most relevant paper for planning a digital pathology portfolio. It explicitly discusses quantitative continuous scoring, membrane versus cytoplasmic localization, spatial heterogeneity, multiplex assays, cross-resistance, specimen timing, regulatory thresholds, and implementation.
Its most commercially useful recommendation is also restrained: use broad profiling selectively during discovery, then reduce the result to the smallest assay package that can inform a defined clinical or development decision.
The authors do not claim that a comprehensive modular predictor has been validated. Target expression is established for selected ADCs; most other modules remain exploratory, and linker-processing biomarkers are particularly immature.
Manufacturer implication: build a sophisticated discovery capability, but aim to commercialize focused, reproducible assays with clearly defined uses.
The LinkedIn commentary identifies the right question: when does selection improve the therapeutic decision, and when does it exclude patients unnecessarily?
Several different situations can produce clinical benefit without a mandatory target-expression test:
- The target may be sufficiently prevalent that a cutoff adds little discrimination.
- Conventional IHC may measure the wrong feature—abundance when localization or internalization matters more.
- Payload sensitivity or resistance may dominate the response.
- Bystander effects or extracellular payload release may weaken the relationship between staining and benefit.
- A trial may establish favorable benefit–risk in a clinically defined population without demonstrating the need for further selection.
These explanations have different implications. Failure of an H-score to predict benefit does not establish that the target is biologically irrelevant. Conversely, an improved association using a more elaborate assay does not establish that withholding treatment from assay-negative patients improves care.
A CDx opportunity arises when the test becomes important to the safe and effective use of the drug—not merely when it produces an interesting biological measurement. FDA’s companion-diagnostic framework emphasizes this relationship between the diagnostic and the therapeutic product. FDA
TROP2 QCS is particularly significant because it changes the biological question.
A conventional TROP2 assessment asks how much protein is present. The investigational normalized membrane ratio, or NMR, assesses its subcellular distribution.
In the reported exploratory TROPION-Lung01 analysis, positivity meant that at least 75% of tumor cells had an NMR at or below 0.56. This should not be paraphrased as “TROP2-high.” The measure reflects relatively greater cytoplasmic localization.
The sponsor reported median progression-free survival of 6.9 versus 4.1 months for datopotamab deruxtecan versus docetaxel in the biomarker-positive population, with an HR of 0.57. However, the cutoff was optimized using trial outcome data in a selected subgroup. That makes independent, prospective evaluation especially important. astrazeneca-us.com
Liu’s discussion of ICARUS-LUNG01 adds biological support: greater internalization capacity and predominantly cytoplasmic TROP2 distribution were associated with datopotamab activity. Nevertheless, a static FFPE localization score is a potential surrogate for relevant trafficking biology; it is not a direct measurement of live-cell internalization.
For a digital pathology manufacturer, this is a more differentiated proposition than automating an existing manual score. It uses computational analysis to extract a potentially treatment-relevant feature that routine scoring does not adequately capture.
The clinical claim must remain specific, however. Evidence for datopotamab in a defined NSCLC population cannot automatically establish the same cutoff for breast cancer, another TROP2 ADC, or a different treatment line.
The public development program provides concrete milestones, but does not support treating approval as certain or assigning an assured launch year.
The prospective TROPION-Lung17 trial compares datopotamab with docetaxel in previously treated, TROP2-NMR-positive nonsquamous NSCLC without actionable genomic alterations. The sponsor describes central testing with the investigational VENTANA TROP2 RxDx device and dual primary endpoints of progression-free and overall survival. daiichisankyo.us
TROPION-Lung10 is also important: its published design places progression-free and overall survival in the TROP2-NMR-positive population among the primary endpoints for the first-line combination comparison. This extends the biomarker question into a different treatment setting. PubMed
For timing, UCLA’s public Lung17 listing gives an estimated completion in January 2029. That is not a forecast of CDx approval, and other datasets could inform regulatory development earlier. It does mean that Lung17 should not be represented as completed prospective confirmation. ucla.clinicaltrials.researcherprofiles.org
The current DATROWAY prescribing information reviewed here includes breast-cancer indications and an EGFR-mutated NSCLC indication without a TROP2-QCS testing requirement. The lung indication selects on EGFR mutation status. An approved TROP2-directed drug therefore does not, by itself, create a mandatory TROP2 testing market. accessdata.fda.gov
Roche’s public description identifies the TROP2 device as an integrated combination of staining assay, algorithm, image-management system, and specified scanners, with pathologist quality assessment. Its Breakthrough Device Designation is a development milestone, not marketing approval. diagnostics.roche.com
The broader marker portfolio should be ranked by clinical maturity and by how much digital analysis adds.
| Marker or application | What is established or emerging | Implication for a digital pathology manufacturer |
|---|---|---|
| HER2, including low/ultralow expression | Established treatment-selection roles; expanding clinical settings | Large existing clinical need. Demonstrate reliable categorization and useful improvements at difficult boundaries; do not assume finer quantification automatically improves treatment selection. |
| TROP2 localization/QCS | Exploratory clinical evidence, mechanistic support, and prospective phase III evaluation | A leading opportunity for an algorithm-dependent CDx. Keep claims specific to drug, tumor, treatment setting, and validated workflow. |
| FRα/FOLR1 | Established CDx for mirvetuximab | A mature target for reproducibility and workflow improvements. Existing manual assay approval does not automatically authorize a digital implementation. |
| MET protein expression | Established CDx for telisotuzumab vedotin | A relevant adjacent assay opportunity; distinguish protein overexpression from MET genomic alterations. |
| CLDN18.2 | Existing non-ADC selection and an advancing ADC pipeline | Particularly important for drug-specific thresholds and assay bridging; the same antigen can support different eligibility definitions. |
| HER3, B7-H3, nectin-4 and other targets | Active development with variable evidence for expression-based selection | Valuable discovery partnerships, but testing demand must be established program by program. |
| Payload and spatial-response signatures | Strong rationale; mostly exploratory clinical evidence | Develop alongside drug sponsors. These may become future products, but should not be treated as mature CDx revenue today. |
For context, FDA’s FOLR1 device approval uses a cutoff of at least 75% viable tumor cells with moderate or strong membrane staining. Its MET CDx listing specifies at least 50% of tumor cells with strong membrane and/or cytoplasmic staining. Those are clinically and analytically distinct measurement tasks. accessdata.fda.gov
CLDN18.2 supplies a particularly instructive current example of what may be coming next.
In July 2026, AstraZeneca announced phase III CLARITY-Gastric01 results for sonesitatug vedotin in gastric/GEJ and related cancers. Enrollment used CLDN18.2 expression in at least 25% of tumor cells at any staining intensity. The announcement reported an overall-survival benefit, while the co-primary progression-free-survival endpoint did not reach statistical significance. The study also evaluates the clinical performance of the VENTANA SP455 assay.
This is sponsor-reported trial evidence, not an FDA approval. Nevertheless, it is a concrete development opportunity rather than a purely conceptual marker. astrazeneca.com
The existing CLDN18 CDx for zolbetuximab—a non-ADC therapy—uses a different selection framework, including a 75% tumor-cell threshold. The commercial unit is the validated assay–drug–indication relationship, not “CLDN18.2 testing” in the abstract. FDA
The next major expansion is likely to involve response and resistance features beyond antigen abundance.
The papers point toward several developments, with different levels of readiness.
Subcellular and spatial measurements are the most natural extension of digital pathology. Membrane localization, heterogeneous expression, target-positive cell neighborhoods, stromal barriers, vascular proximity, and immune architecture can all affect delivery and response. Liu describes exploratory HER2 and HER3 studies in which architecture contributed information beyond expression alone.
However, a feature associated with response in a treated cohort may also reflect tumor type or prognosis. It must be tested for the intended treatment decision.
Payload susceptibility is likely to become increasingly important as patients receive multiple ADCs. Changing from a HER2-directed ADC to a TROP2-directed ADC changes the delivery target, but may retain a related topoisomerase-I-inhibitor payload. That creates a plausible route to cross-resistance.
Candidate markers include SLFN11, RAD51 and other DNA-repair features, TOP1 alterations, and drug-efflux mechanisms. These are development candidates, not established tests that can reliably assign the next ADC in routine practice.
This has a direct design implication: the next generation of diagnostic models should preserve treatment history as a relevant input. A pretreatment image may not contain enough information to identify resistance acquired during prior therapy.
Linker-processing biomarkers remain less mature. Cathepsin L and other tumor-associated proteases are mechanistically interesting, including for extracellular release. But protein abundance, enzyme activity, and actual drug cleavage are different measurements. Liu explicitly notes the lack of prospective validation for linker-processing biomarkers used in treatment selection.
A manufacturer should maintain discovery capability here without assuming that adding another IHC stain will yield a clinically useful product.
Multiplex selection will become more attractive as treatment choices proliferate. Both reviews contemplate combining multiple target measurements with payload-related features. Measuring several proteins on one section could conserve tissue and avoid sequential testing.
But technical multiplexing is easier to establish than comparative treatment selection. A panel showing HER2, TROP2, and HER3 does not necessarily identify which corresponding drug will benefit the patient most. That stronger claim requires appropriately comparative evidence.
Earlier-stage treatment will increase the stakes of diagnostic errors. FDA has already approved T-DXd uses in HER2-positive early breast cancer, confirming that this movement is underway. FDA
As potentially curable patients receive ADCs, avoiding unnecessary exposure becomes more important. This could support future predictive or de-escalation tests. It does not mean every earlier-stage drug indication generates a new assay: some will use existing HER2 testing.
The most important scientific limitation is that these papers share evidence; they do not independently validate every proposed biomarker.
Mosele and Liu draw on overlapping translational programs. Their convergence strengthens the intellectual case for multidimensional assessment, but should not be counted as independent clinical replication.
There are also useful differences in emphasis. Hurvitz documents major changes in treatment practice. Mosele argues that, outside particularly successful settings, ADC benefit still falls short of its potential. Liu provides a framework for improving selection while repeatedly emphasizing incomplete validation.
Read together, they support a credible research and product-development direction. They do not support assuming that a complex multimodal assay will necessarily outperform a simpler test in clinical utility, access, or cost.
For a manufacturer, the development plan should distinguish three separate claims.
| Claim | Evidence needed |
|---|---|
| The assay measures its intended feature reproducibly | Analytical performance across relevant specimens, staining variation, scanners, sites, operators, and software versions |
| The feature identifies differential treatment benefit | Appropriate clinical comparisons, independent validation, and analysis of predictive versus prognostic effects |
| Using the assay improves treatment decisions | Evidence supporting the intended selection strategy, including consequences of false-positive, false-negative, failed, and unavailable results |
A positive-only randomized trial can establish benefit in an enriched population. By itself, it cannot establish that biomarker-negative patients derive no benefit or quantify the overall advantage of test-guided treatment over treating everyone. That distinction should shape both the evidence plan and the eventual claims.
For QCS, practical validation should pay particular attention to tumor segmentation and the membrane–cytoplasm boundary. Fixation, staining intensity, section quality, optical characteristics, and segmentation errors can affect the measured ratio. Near a clinical cutoff, apparently small shifts may change eligibility.
This makes the complete measurement system commercially important. A broadly cleared scanner or image-management platform does not automatically validate every CDx algorithm or combination of components.
The manufacturer’s strongest commercial position will come from joining drug development early.
The most valuable partnerships begin while the sponsor can still collect suitable specimens, preserve outcome-linked images, investigate plausible biomarkers, and prospectively validate the selected approach.
Waiting until phase III results are available can leave the diagnostic developer with inadequate tissue, a retrospectively optimized cutoff, and limited opportunity to distinguish biological signal from statistical selection.
The practical priorities would be:
Secure the lead TROP2 claim. Define the intended population, deployment configuration, specimen requirements, algorithm, threshold, and clinical validation plan with the therapeutic sponsor.
Develop adjacent assays with established treatment-selection demand. HER2, FRα, MET, and selected CLDN18.2 programs provide different balances of existing market, technical difficulty, and incremental digital value.
Build a shared discovery platform for spatial and payload-related features. Preserve continuous measurements and cell-level information so future research is not limited to today’s positive/negative result.
Partner for measurements that images cannot reliably supply. Genomics, transcriptomics, pharmacogenetics, liquid biopsy, and functional assays may be complementary. The platform need not own every measurement technology.
Validate a deployable service. Turnaround time, tissue adequacy, assay failures, pathologist review, site transfer, and technical support can determine whether an approved CDx is actually used.
The papers’ recommendations for OASIS, ADC-MATCH, and other translational initiatives also point toward useful partnership environments. These are opportunities to develop and test hypotheses, rather than proof that broad ADC-selection panels are ready for clinical adoption.
The commercial forecast should be built from testing episodes and labeled uses, not from aggregate ADC sales. Drug revenue can grow because patients remain on therapy longer without generating additional diagnostic tests. Conversely, more indications and more candidate patients can expand testing even when the biomarker-positive proportion is relatively small.
For each anticipated product, the manufacturer should model the population requiring testing, tissue availability, likely test completion, central versus decentralized deployment, and whether the result changes treatment. FDA approval, reimbursement, and economical laboratory implementation are separate requirements.
My assessment is therefore favorable toward drug-specific computational CDx, particularly TROP2 localization and difficult quantitative IHC applications; more cautious toward near-term universal ADC-selection panels. The near-term advantage belongs to manufacturers that can deliver a reproducible assay tied to persuasive clinical evidence. The longer-term opportunity is to help select among ADC constructs and treatment sequences—but that will require integrating information about the tumor, the payload, and prior therapy, not simply adding more antigen scores.
