Why Antigen Expression Alone Cannot Predict Response to B7-H3-Targeted Cancer Therapy

One of the most hotly pursued targets in cancer medicine right now is a molecule called B7-H3, also known as CD276. It sits on the surface of cells as a member of the B7 immune-regulatory family, and it has become attractive for a simple reason: it is found across a strikingly broad range of solid tumors, while it is relatively restricted in many normal tissues. That combination—abundant on cancer, scarce on healthy tissue—is exactly what drug developers look for in a target. It has fueled a wave of monoclonal antibodies, antibody-drug conjugates, chimeric antigen receptor T cells, T-cell engagers, natural killer-cell engagers, and radioimmunotherapy programs, all aimed at the same molecular flag.

But a new review published in the Journal of Translational Medicine by Longhao Zhang, Kai Zhang, Zhihong Chen, Haitao Zhao, and Xin Lu of Peking Union Medical College Hospital argues that the field is making a dangerous assumption. The team, writing from the Department of Liver Surgery, defines what they call the B7-H3 companion diagnostic paradox: the same nominal biomarker—B7-H3 positivity, usually determined by a simple immunohistochemistry stain—is being applied across therapeutic platforms whose mechanisms of action are fundamentally different and whose determinants of response are therefore also fundamentally different. Antigen expression, they conclude, is required for a drug to engage its target, but it is not sufficient to predict whether a patient will actually benefit.

The clinical evidence behind that claim is sobering. In studies of B7-H3-directed antibody-drug conjugates, baseline membrane expression of the antigen on tumor cells has repeatedly failed to correlate with objective response. Patients whose tumors light up brightly on an IHC slide do not necessarily shrink on treatment, and patients with apparently modest staining sometimes respond dramatically. For a class of drugs that is otherwise among the most exciting in oncology, this is a translational problem of the first order: the diagnostic test being used to select patients is measuring the one thing that matters least for predicting the payload’s actual effect.

The reason, the authors argue, lies in what antibody-drug conjugates actually do. An ADC is a two-part machine: an antibody that delivers it to cells bearing the target, and a cytotoxic payload that does the killing. Once the payload arrives, the tumor’s own biology takes over. Patient-derived models indicate that payload-specific vulnerabilities—replication stress, RB1 deficiency, the state of the DNA-damage response machinery, and drug-tolerance programs—can dominate whether the drug works. A tumor with abundant B7-H3 but an intact DNA-damage response may shrug off a payload designed to exploit DNA repair deficits, while a tumor with lower antigen density but a crippled RB1 pathway may be exquisitely sensitive. The biomarker that predicts delivery says nothing about the biology that determines lethality.

Cellular therapies present a parallel but distinct version of the same paradox. For CAR-T cells and related engineered cell products, antigen density on the tumor surface does matter—there must be enough target for the engineered cell to recognize and engage. But the review emphasizes that antigen density is only one entry on a longer list of equally decisive factors: whether the cells actually infiltrate the tumor, whether an effective immune synapse can form between the therapeutic cell and the cancer cell, whether stromal barriers and cancer-associated fibroblasts physically block access, whether the cells persist in the patient or collapse into exhaustion, and even the route by which the cells are administered. A patient can be B7-H3-positive by every staining criterion and still fail a cellular therapy because the tumor microenvironment refuses the cells entry.

What makes the paradox especially consequential is that B7-H3 is being positioned as a pan-cancer target. Its expression on both tumor cells and tumor-associated vasculature, and its accessibility to both antibody-based and cell-based platforms, have encouraged developers to treat a single IHC assay as a universal gatekeeper across indications and across modalities. The authors argue this is a category error. A biomarker that predicts response to a PBD-payload ADC—where DNA cross-linking and replication stress are the operative mechanisms—cannot be the same biomarker that predicts response to a CAR-T construct, where trafficking, persistence, and immune synapse mechanics dominate. The review’s central prescription is therefore that B7-H3 companion diagnostics should be platform-specific rather than universal: each therapeutic modality needs its own validated predictive test, matched to that modality’s actual mechanism of action.

The implications reach beyond B7-H3 itself. Companion diagnostic development in oncology has long favored binary readouts—positive or negative, above or below a staining threshold—because they are simple to standardize, regulate, and deploy. But the biology of modern therapeutics rarely respects binary categories. The review calls for the next phase of B7-H3 translation to move beyond binary positivity toward spatially resolved, functionally validated, and cancer-context-aware response prediction. Spatially resolved means understanding not just whether a tumor expresses the antigen but where—in which compartments, at what distance from vasculature, adjacent to which stromal barriers. Functionally validated means testing whether the predicted vulnerability is actually operative in a given tumor, for example through patient-derived models that interrogate payload sensitivity directly. Cancer-context-aware means recognizing that the same antigen in pancreatic tissue and in lung tissue may sit inside entirely different microenvironments that reshape the therapeutic equation.

For the growing community of researchers working on functional precision oncology, the review lands as both a warning and a roadmap. The warning is that trials of B7-H3-targeted agents risk enrolling patients on a biomarker that cannot discriminate responders from non-responders, generating disappointing results that may be misread as failures of the drugs rather than failures of the diagnostic. The roadmap is to build companion diagnostics that reflect mechanism: for ADCs, assays that capture DNA-damage response state, RB1 status, and replication stress alongside antigen density; for cellular therapies, measures of tumor infiltration, stromal composition, and immune contexture that go beyond surface staining. The tumor microenvironment, in this framing, is not background noise—it is a primary determinant of response that the current diagnostic paradigm ignores.

The authors are careful to note that B7-H3 remains a genuinely attractive target, and nothing in their argument diminishes the clinical promise of the agents now in development. Broad expression across solid tumors, relative restriction in normal tissues, and accessibility to multiple platforms are real advantages that justify the investment flowing into the field. The point is narrower and sharper: target engagement is a necessary condition for response, not a sufficient one, and conflating the two wastes patients, trials, and years. As antibody-drug conjugates and engineered cell therapies move deeper into the clinic, the B7-H3 paradox may prove to be the clearest early lesson in a broader truth about the next generation of cancer therapeutics—the diagnostic test that gets a drug into the right patients is not the same test that measures whether the drug can kill what it reaches.

Subject of Research: B7-H3/CD276-targeted therapy and companion diagnostics

Article Title: The companion diagnostic paradox of B7-H3/CD276-targeted therapy: why antigen expression is insufficient to predict ADC and cell therapy response

Article References: Zhang, L., Zhang, K., Chen, Z., Zhao, H., & Lu, X. (2026). The companion diagnostic paradox of B7-H3/CD276-targeted therapy: why antigen expression is insufficient to predict ADC and cell therapy response. Journal of Translational Medicine. https://doi.org/10.1186/s12967-026-09096-9

Image Credits: AI Generated

DOI: 10.1186/s12967-026-09096-9

Keywords: B7-H3, CD276, antibody-drug conjugate, CAR-T, companion diagnostics, biomarker, companion, diagnostic, paradox, CD276-targeted, therapy, antigen