A macrophage-checkpoint drug program whose organoid readouts measure T cells

What the work claims

This is an observational platform study registered as a drug-discovery program, so the claims are about intent and capability, not results. The stated goal is to create novel small-molecule inhibitors of the CD47-SIRPalpha interaction, designed by structure-based methods, to reverse tumor-associated macrophage, TAM, mediated immune suppression and restore anti-tumor immunity in colorectal cancer. The registry summary explicitly frames the approach as aiming for greater efficacy and lower toxicity than CD47-targeting antibodies, and names patient-derived organoids from colon cancer biopsies as the model system, on the grounds that tumoroids preserve the patient's tumor stroma, including myeloid cells, and therefore model tumor-immune interaction accurately¹.

The registered facts are these. Institut Paoli-Calmettes in Marseille is the sponsor. The study is observational, actually started on 9 January 2023, estimating 115 participants, with estimated completion on 9 January 2027, and it is recruiting. The primary outcome is the establishment of patient-derived organoids, minimum n=30, resembling the primary tissue sample, judged by a histology score comparing sample characteristics, over 18 months. Secondary outcomes are a validated organoid cryobank, minimum n=30, over the same 18 months; a T-cell biobank built by FACS isolation of CD3-positive cells, over 48 months; and an assessment of immunotherapy's effect on the antitumor activity of autologous T cells from peripheral blood, also over 48 months. Eligible participants are adults with metastatic or non-metastatic colon cancer, or recurrence, for whom excision surgery has been proposed¹.

How it works

The biological mechanism is the myeloid checkpoint axis. CD47 is a surface protein, often called the do-not-eat-me signal, that healthy cells display to avoid being phagocytosed by macrophages; many tumors overexpress it. SIRPalpha is its receptor on macrophages. When a tumor cell's CD47 engages macrophage SIRPalpha, phagocytosis is suppressed. Blocking the interaction is meant to release that brake and let macrophages eat tumor cells, which is why the field calls CD47-SIRPalpha an innate immune checkpoint and why the tumor-associated macrophage sits at the center of the rationale¹.

The small-molecule bet has a specific logic. CD47 is also expressed on red blood cells and other normal tissues, and systemic antibody blockade of such a broadly distributed antigen produces on-target toxicity, most visibly anemia, which has plagued the antibody class. A small molecule, in principle, can be dosed to occupy the tumor microenvironment preferentially, or designed against the SIRPalpha side, with the ambition of dissociating efficacy from the anemia sink. That ambition is the registry's stated motivation: greater efficacy and lower toxicity than the antibodies¹. The clinical backdrop is verifiable and sobering. Gilead has discontinued or placed on hold multiple phase III trials of magrolimab, the leading anti-CD47 antibody, since January 2022, citing adverse effects or futility across trials including NCT04313881, NCT04778397, and NCT05079230², and a phase III combination trial in higher-risk myelodysplastic syndromes reported in mid-2026 found no benefit for the magrolimab arm³. Whether those failures reflect the delivery format, the target, or both is exactly the question a microenvironment-restricted small molecule must answer.

The organoid layer is where the program's methodological claim lives. If colon tumoroids genuinely retain stromal and myeloid components from the biopsy, then exposing them to a SIRPalpha-CD47 inhibitor and watching tumor cell death would be a human, patient-specific, mechanism-relevant screen, closer to the tumor microenvironment than any monoculture. The cryobank and T-cell biobank turn the platform into reusable infrastructure for immunotherapy evaluation.

Where a skeptic should push

The most load-bearing assumption is the registry's own: that patient-derived colon organoids preserve the patient's tumor stroma, including myeloid cells, for long enough to serve as the assay system for a macrophage checkpoint drug. The registered fidelity check does not test that assumption. The histology score that anchors the primary outcome compares the organoid to the primary tissue sample on tumor characteristics; it is not stated to score immune or stromal composition. An organoid can pass that fidelity bar while having lost the macrophages the drug is meant to act on, and the well-known tendency of tumor organoids to drift toward epithelial purity with passage makes that loss the default expectation rather than a remote risk. My phrasing of that tendency is field background, not a fact from this record, and the record offers no data against it.

Second, and more sharply: the registered immune endpoints do not measure the drug's mechanism. The two 48-month secondary outcomes are a CD3-based T-cell biobank and an assessment of immunotherapy's effect on autologous T-cell antitumor activity. The target biology is macrophage phagocytosis; the registered readouts are T-cell killing. A program that screens SIRPalpha-CD47 inhibitors on T-cell assays would be screening for T-cell adjuvants while believing it is selecting phagocytosis restorers. If the organoids truly retain myeloid stroma, the obvious decisive readout is a phagocytosis or eat-me restoration assay; its absence from the registered endpoints is the single largest gap in the record¹.

Third, the sampling frame. Eligible patients must have proposed excision surgery, so the platform is built from resectable primary tumors, while the stated disease target includes metastatic colorectal cancer. The immune microenvironment of a resectable primary is not the microenvironment of a metastatic lesion, and CD47-axis biology is known from the broader literature to vary across compartments and disease stages. A screen run on surgical primaries may select compounds that fail where the clinical need is. Finally, the familiar platform-gate pattern: the primary outcome of a drug-discovery program is again organoid establishment, minimum n=30, which is honest but means the registry as written promises infrastructure, not efficacy, through at least its first 18 months.

Why a myeloid target needs a myeloid readout

For organoid-based drug discovery, this entry captures both the opportunity and the characteristic failure mode of the tumor-immune organoid wave. The opportunity is that innate-immune checkpoint drugs, myeloid modulators, phagocytosis enhancers, myeloid-cell engagers, have no good legacy screening system: mouse models misread human myeloid biology, and standard tumor cell lines contain no immune compartment at all. Patient-derived organoids with retained stroma and immune cells are the only scalable human system that could fill that gap, which is why programs like this one keep being registered. A validated myeloid-competent colon tumoroid platform would be genuinely scarce infrastructure with buyers across immuno-oncology.

The threat is readout-mechanism mismatch at the platform level, which generalizes far beyond this one program. As organoid screens become the default layer for tumor-microenvironment drug discovery, a screen whose endpoint cannot see the drug's mechanism does not merely underperform, it systematically selects the wrong compounds, and it does so credibly, because the platform looks right and the number it produces is clean. The fix is unglamorous and cheap relative to what is at stake: anchor these platforms on mechanism-matched assays, phagocytosis for myeloid checkpoints, and publish the immune-composition fidelity data, not just the histology score. If this Marseille group measures whether its organoids still contain macrophages, and scores eat-me restoration directly, its platform claim becomes testable and possibly decisive for a whole drug class that the antibodies burned. If it stays with T-cell endpoints, the program will tell us about T cells, whatever the title says.

The bottom line

NCT05955196 is a well-motivated response to a real clinical failure: the CD47 antibody class stumbled in phase III, and small-molecule, microenvironment-focused approaches are a rational next move. The registered record, though, promises a platform whose decisive assumptions, retained myeloid stroma and a mechanism-matched readout, are exactly the ones the endpoints do not check. No efficacy claim should attach to this program until phagocytosis data appear on a platform with demonstrated immune fidelity. What would confirm the thesis is a published correlation between SIRPalpha-CD47 small-molecule activity in the organoid assay and macrophage-mediated tumor killing in the same samples; what would break it is the now-default outcome, epithelial-drifted organoids passing fidelity scores while the macrophages the drug needs are gone. The antibody class's failures bought this field a question worth answering; the readouts chosen will decide whether it gets answered.