A failed immune checkpoint gets a second run in a dish

What the work claims

The program's stated goal is to create selective small-molecule inhibitors of the SIRPalpha-CD47 interaction, designed from the proteins' structures, that reverse macrophage-mediated immune suppression in colorectal cancer and restore anti-tumor immunity. The record claims these candidates can hit the tumor microenvironment with greater efficacy and lower toxicity than CD47-targeting antibodies¹. The testing substrate is the claim that matters to this site: organoids derived from colon-cancer biopsies, which the record says preserve the patient's tumor stroma, including myeloid cells, and thereby provide an accurate in vitro model of tumor-immune interaction for evaluating immunotherapies¹.

Read the endpoints and the real deliverable comes into focus. The primary outcome is not a drug result at all: it is the establishment of patient-derived organoids, a minimum of 30, resembling the primary tissue sample, scored by comparing the histology of primary tumor and organoid¹. Secondary deliverables are a validated cryobank of those organoids, tested for growth before and after cryopreservation and through four passages, plus a biobank of autologous T-cells sorted by CD3, with which the effect of immunotherapy on anti-tumor activity can be assayed from peripheral blood¹. The study samples 24 ml of blood and matched tumor with adjacent normal tissue from surgical specimens, plans an estimated 115 participants, and is observational and case-only¹. This is, up front, an infrastructure build: a stroma-preserving colon-tumoroid bank with matched immune-cell reagents, intended to serve a drug program that is still preclinical.

How it works

CD47 on a tumor cell engages SIRPalpha on a macrophage and delivers an inhibitory signal that suppresses phagocytosis; blocking the interaction lets macrophages eat the tumor. Tumor-associated macrophages are abundant in colorectal cancer, and their suppressive activity is one reason colorectal tumors exclude effective immunity, so the target is rational. The record's premise is that antibodies failed, at least partly, for pharmacologic reasons: a systemically administered antibody against a ubiquitous surface antigen produces an antigen sink and hematologic liabilities, and the Marseille team bets that a small molecule engaging the SIRPalpha-CD47 interface can be dosed more selectively¹.

The magrolimab precedent is the sobering backdrop, and it is verifiable. On 2024-02-07, Gilead announced discontinuation of the phase 3 ENHANCE-3 study of magrolimab in acute myeloid leukemia after an interim analysis showed futility and an increased risk of death, primarily from infections and respiratory failure; the FDA placed all magrolimab studies in myelodysplastic syndromes and acute myeloid leukemia on full clinical hold; and Gilead stated that the primary analyses of the ENHANCE trial in higher-risk MDS and ENHANCE-2 in AML with TP53 mutations had also demonstrated futility with increased risk of death in the magrolimab arms². Gilead said it would not pursue further development of magrolimab in hematologic cancers².

The model innovation is the other half of the mechanism claim. Conventional tumor-organoid culture is epithelial-selective: growth conditions favor tumor epithelium and shed immune and stromal components within the first passages. A colon organoid that keeps its myeloid stroma would let the team watch, in a patient's own tissue, whether SIRPalpha-CD47 blockade re-arms macrophages and whether that restores T-cell killing in the matched autologous co-culture¹. That is precisely the readout antibody programs never had before entering patients.

Where a skeptic should push

The single most load-bearing assumption is the sentence on which the whole model rests: that biopsy-derived colon organoids preserve the patient's tumor stroma, including myeloid cells¹. Stroma preservation is the field's most overclaimed property. It is achievable, but only under non-standard culture regimes that are fussy about medium, matrix, oxygen, and passage discipline, and even then myeloid composition drifts. The record provides no method: no medium, no matrix, no passage schedule, and, crucially, no immune-composition readout. The primary endpoint judges resemblance by histology score, which measures architecture, not immune content. A tumoroid can look like the primary tumor on histology and have lost most of its macrophages. The cryobank validation amplifies organoids through four passages with growth as the score¹, and passage number is exactly where stromal and immune content die. The program's central claim is thus asserted, not demonstrated, and the instrument chosen to demonstrate it cannot see the variable that matters.

Second, the chemistry bet is stated without data. The SIRPalpha-CD47 interface is a large, flat protein-protein contact, a class of target where potent, selective small-molecule blockade is notoriously difficult; the record asserts greater efficacy and lower toxicity than antibodies as the design goal, not as a result¹. Goals are cheap; the antibody program also had a mechanistically tidy rationale, and it still produced futility and excess deaths in three phase 3 trials². Part of that failure was on-target biology rather than format, including immune dysregulation; a cleaner pharmacologic window does not by itself resolve whether CD47 blockade benefits patients at all.

Third, the arithmetic is soft. An estimated 115 participants are sampled to yield a minimum of 30 established organoids¹, implying the team expects, or at least plans around, a substantial establishment failure rate, and the record does not say what happens to the drug-testing question if the bank lands closer to 30 than to 115. The study is observational and case-only, with no comparator of any kind, and 48-month timelines on the immune-assay secondary endpoints¹. Nothing here is wrong; it is simply slower and more preliminary than the framing of an inhibitor program suggests.

Colon organoids that keep their immune cells

For organoid models of human organs and the drug-discovery work built on them, the opportunity is infrastructural and large. An immune-competent, stroma-preserving, biobanked colon tumoroid with matched autologous T-cells is close to the most-wanted model in immuno-oncology: a screenable, patient-specific replica of the tumor-immune interface. If the Marseille team publishes the immune-composition data to back the histology claim, including macrophage lineage, abundance, and function across those four cryobank passages, the platform becomes general-purpose infrastructure that any myeloid-targeting, macrophage-engager, or combination-immunology program could run on. Drug discovery has lacked a cheap ex vivo gate for immunotherapies; this design, validated properly, is a candidate for one.

The threat is that the claim outruns the assay. If stroma preservation turns out to be morphology language for drifted co-cultures, then screens run on this bank will systematically mis-rank macrophage-directed compounds, and the field will have rebuilt the same generalization failure it knows from epithelial-only models, one level deeper: a dish that claims to speak for the immune microenvironment while quietly speaking for a reconstruction of it. The magrolimab history sharpens the point. A model that had genuinely predicted the antibody program's failure would have been worth more than any target; a model that cannot see immune toxicity will keep clearing candidates that die the same way. The corrective is unglamorous and cheap relative to the stakes: report myeloid composition and function, not just histology, at establishment, at passage four, and after thaw. If those numbers hold, this is a genuinely important platform. If they are never reported, the word tumoroid is doing work the data cannot.

The bottom line

Established from the registry record: a recruiting, observational, case-only study at Institut Paoli-Calmettes estimating 115 participants, with the primary deliverable of at least 30 colon-cancer organoids resembling primary tissue by histology score, a validated four-passage cryobank, an autologous CD3-sorted T-cell biobank, and a preclinical program of structure-designed small-molecule SIRPalpha-CD47 inhibitors to be tested in these stroma-preserving tumoroids¹. Established from the primary corporate record: the antibody flagship of the same pathway, magrolimab, failed three phase 3 trials and was withdrawn from hematologic-cancer development in 2024 after futility and excess deaths, with the FDA placing its remaining MDS and AML studies on full clinical hold². Not established: any inhibitor potency or selectivity data, any demonstrated immune-composition preservation in the organoids, or any drug result of any kind. What would confirm the approach: published myeloid and T-cell composition with functional phagocytosis and killing assays that survive passage four and a thaw cycle. What would break it: histology-concordant organoids with depleted macrophages, which would reduce the platform to a conventional epithelial model wearing an immune-competent label.