Research advances on CD47 in digestive system cancers: from molecular mechanisms to clinical translation

Abstract

CD47 is a major innate immune checkpoint molecule that interacts with signal regulatory protein α (SIRPα) to deliver a “don’t eat me” signal and restrain macrophage-mediated phagocytosis. Accumulating evidence indicates that dysregulated CD47 expression is associated with immune escape, treatment resistance, and adverse clinicopathological features in several digestive system cancers, although the strength of these associations varies across tumor types. This narrative review provides an integrative overview of CD47 structure, ligand interactions, signaling and immune-regulatory functions, and summarizes reported expression patterns and available evidence regarding regulatory mechanisms and relevant mechanistic contexts in esophageal, gastric, colorectal, hepatocellular, pancreatic, and biliary tract cancers. We further review therapeutic strategies targeting the CD47–SIRPα axis, including anti-CD47 antibodies, SIRPα fusion proteins, bispecific antibodies, and emerging cell-based and nanotechnology-based platforms, with particular attention to recent clinical developments. Finally, we critically discuss translational challenges, including hematologic toxicity and antigen sink, assay standardization and biomarker selection, resistance mechanisms, optimization of combination regimens, and the need for prospective validation of hypothesis-generating biomarker signals. By integrating recent 2025–2026 clinical developments with mechanistic and engineering advances, including lessons from unsuccessful or discontinued clinical programs, this review provides a balanced perspective on the opportunities and remaining barriers to the clinical translation of CD47-directed therapy in digestive system cancers.

Introduction

Digestive system cancers account for a substantial proportion of global cancer incidence and mortality. According to GLOBOCAN 2022 data, approximately 4.9 million new cases and 3.32 million deaths from digestive system cancers were recorded, accounting for nearly one-quarter of all new cancer cases worldwide (Danpanichkul et al. 2026). Esophageal, gastric, colorectal, liver, and pancreatic cancers are among the major contributors to cancer-related mortality (Danpanichkul et al. 2026). Pancreatic cancer, in particular, continues to have a poor prognosis despite advances in diagnosis and treatment (Rahib et al. 2025). In recent years, PD-1/PD-L1 blockade has improved outcomes in several digestive system cancers, although the magnitude of benefit varies across tumor types and molecular subgroups (Peshin et al. 2025). In colorectal cancer, clinical benefit from PD-1 blockade is particularly evident in microsatellite instability-high/mismatch repair-deficient (MSI-H/dMMR) tumors, whereas microsatellite-stable/mismatch repair-proficient (MSS/pMMR) tumors are generally much less responsive to immune checkpoint blockade (Peshin et al. 2025). These limitations have increased interest in innate immune checkpoints, particularly the CD47–SIRPα axis, which plays an important role in regulating macrophage-mediated phagocytosis (Montero and Isenberg 2023).

CD47 is a transmembrane glycoprotein of the immunoglobulin superfamily that is widely expressed on many cell types (Matozaki et al. 2009). By engaging SIRPα on myeloid cells, CD47 provides a physiological “self” signal that limits inappropriate phagocytic clearance (Oldenborg et al. 2000). In tumors, increased CD47 expression can reduce macrophage-mediated phagocytosis and contribute to immune evasion (Willingham et al. 2012). Available studies report increased CD47 expression in several digestive system cancers and associations with adverse clinicopathological features, chemotherapy resistance, or metastatic potential, although the strength and nature of these associations vary by tumor type and assay (Chen et al. 2024; Liu et al. 2023; Hu et al. 2020). Preclinical studies have further shown that disruption of the CD47–SIRPα axis can enhance tumor-cell phagocytosis and may also influence downstream adaptive immune responses (Liu et al. 2023; Fan et al. 2023). This biological rationale has supported the clinical development of therapeutic strategies targeting CD47 or SIRPα (Chen et al. 2024).

This article is a narrative review. We use the term “digestive system cancers” to include malignancies of the gastrointestinal tract as well as hepatobiliary and pancreatic cancers. The review covers esophageal cancer, gastric cancer, colorectal cancer, hepatocellular carcinoma, pancreatic cancer, and biliary tract cancers, for which disease-specific evidence on CD47 expression, biological function, or therapeutic development is available. Studies from other tumor types are discussed only when they are relevant to broader aspects of CD47 biology and are identified accordingly. A recent review summarized the biological rationale and early translation of CD47-targeted therapy in digestive system cancers (Chen et al. 2024). The present review complements this earlier work by focusing on recent 2025–2026 clinical developments, exploratory biomarker analyses relevant to patient selection, approaches designed to reduce treatment-related toxicity, mechanisms underlying limited clinical activity and resistance, and lessons from unsuccessful or discontinued clinical programs. A schematic overview of upstream CD47 regulation and the classical CD47–SIRPα signaling axis is presented in Fig. 1.

Fig. 1.

Upstream regulation of CD47 and the CD47–SIRPα signaling axis in digestive system cancers. A Examples of CD47 regulation in digestive system cancers include Galectin-3 in gastric cancer, the LINC00460/miR-186-3p/MYC network in colorectal cancer, and miR-133a in esophageal cancer. B CD47 binding to SIRPα induces ITIM phosphorylation and recruits SHP-1/2, thereby inhibiting macrophage phagocytosis. Reduced recruitment and activation of non-muscle myosin IIA is one downstream effect. C CD47 blockade can increase macrophage-mediated tumor-cell phagocytosis. Tumor-antigen uptake and MHC class I cross-presentation by dendritic cells may also promote CD8 + T-cell priming

CD47 molecular structure and primary interacting ligands

Molecular structural characteristics of CD47

CD47, also known as integrin-associated protein (IAP), is a transmembrane glycoprotein of approximately 50 kDa encoded by the CD47 gene (Brown and Frazier 2001). Its structure comprises an N-terminal extracellular immunoglobulin V-type (IgV) domain, five membrane-spanning segments, and a short C-terminal cytoplasmic tail (Fenalti et al. 2021). The extracellular IgV domain mediates interaction with SIRPα, whereas the unusual five-pass transmembrane region contributes to receptor organization and CD47-associated signaling (Fenalti et al. 2021; Liu et al. 2007; Brown 2001).

Alternative splicing generates four cytoplasmic-tail isoforms of CD47 with tissue-specific expression (Fenalti et al. 2021). The full-length human CD47 structure reported by Fenalti et al. further defined how the extracellular IgV domain is connected to the transmembrane bundle and provided structural information relevant to receptor organization and therapeutic targeting (Fenalti et al. 2021).

Primary ligands interacting with CD47

CD47 interacts with several ligands and is involved in cell adhesion, migration, survival, and immune regulation (Fenalti et al. 2021; Wu et al. 2024). SIRPα is the best-characterized inhibitory receptor for CD47 and is expressed predominantly on myeloid cells, including macrophages, granulocytes, and subsets of dendritic cells (Qu et al. 2022). The cytoplasmic tail of SIRPα contains two immunoreceptor tyrosine-based inhibitory motifs (ITIMs) (Qu et al. 2022; Liu et al. 2023).

After CD47 binding, phosphorylation of these ITIMs recruits SHP-1 and SHP-2, which inhibit signaling events required for efficient phagocytosis (Qu et al. 2022; Liu et al. 2023).

CD47–SIRPα signaling can also reduce the recruitment and activation of non-muscle myosin IIA at the phagocytic synapse (Qu et al. 2022; Liu et al. 2023; Tsai and Discher 2008).

Thrombospondin-1 (TSP-1) is another important ligand of CD47. VVM-containing sequences within the C-terminal region of TSP-1 have been implicated in CD47 interaction (McDonald et al. 2003; Rath et al. 2006). The TSP-1-CD47 interaction regulates angiogenesis, apoptosis, and inflammatory responses (Zhao et al. 2023; Weng et al. 2025; Lawler and Lawler 2012). By inhibiting the nitric oxide/cGMP signaling pathway and VEGFR-2 phosphorylation, it affects the function of endothelial cells and smooth muscle cells (Isenberg et al. 2009; Kaur et al. 2010). The interaction between integrins and CD47 was first observed in the αvβ3 integrin complex, which is why CD47 was originally referred to as an “integrin-associated protein” (Brown and Frazier 2001; Brown et al. 1990; Isenberg et al. 2008). CD47 can form cis-complexes with various integrins (such as αIIbβ3, α2β1, and α4β1), promoting integrin activation, cell adhesion, and migration (Chung et al. 1999; Wang and Frazier 1998; Brittain et al. 2004; Podolnikova et al. 2023). In some cases, the CD47-integrin complex can also form a trimeric signaling unit with TSP-1 or inhibitory G proteins (Gi), thereby regulating the cAMP signaling pathway (Brown and Frazier 2001; Isenberg et al. 2008; Wang et al. 1999).

Immune regulatory mechanisms of CD47 in the tumor microenvironment

"Don't eat me" signals in macrophage phagocytosis

The CD47-SIRPα axis is a key signaling pathway that regulates phagocytic function in macrophages. Under normal physiological conditions, CD47 is highly expressed on the surface of blood cells such as red blood cells and platelets, preventing their premature clearance by macrophages in the spleen and liver (Oldenborg et al. 2000; Olsson et al. 2005). Cancer cells exploit this physiological mechanism to evade macrophage-mediated innate immune surveillance by upregulating CD47 expression (Jaiswal et al. 2009; Chuang et al. 2024). The phagocytic function of macrophages against tumor cells depends on a dynamic balance between “eat me” signals and “Don’t Eat Me” signals. “Eat me” signals include calreticulin and phosphatidylserine, which can be recognized by corresponding receptors on the surface of macrophages, thereby promoting phagocytosis (Xiao et al. 2024). In contrast, the “don’t eat me” signal represented by CD47–SIRPα inhibits phagocytosis (Polara et al. 2024). Tumor cells often disrupt this balance by upregulating CD47, leading to immune evasion (Polara et al. 2024).

Regulation of dendritic cell function

In addition to macrophages, dendritic cells also express SIRPα; CD47-SIRPα signaling can inhibit the maturation and function of dendritic cells (Braun et al. 2006). Research has shown that blocking CD47 enhances the ability of dendritic cells (DCs) to phagocytose apoptotic tumor cells, promotes cross-presentation of antigens by MHC class I molecules, and thereby activates tumor-specific CD8 + T cells (Liu et al. 2015).

Indirect regulation of T-cell responses

SIRPα is expressed predominantly on myeloid cells, whereas its expression and direct signaling in T cells appear to be limited (Sato-Hashimoto et al. 2011). The effects of the CD47–SIRPα axis on adaptive immunity therefore seem to depend mainly on antigen-presenting cells (Liu et al. 2015; Hendriks et al. 2020). CD47 blockade can enhance tumor-antigen uptake and cross-presentation by dendritic cells and promote CD8 + T-cell priming (Liu et al. 2015; Hendriks et al. 2020). CD47 can also influence T-cell function through other ligand interactions. For example, recent evidence suggests that TSP1–CD47 signaling contributes to T-cell exhaustion in tumors (Weng et al. 2025). These findings support combining CD47-directed strategies with treatments that enhance adaptive antitumor immunity.

Regulation of angiogenesis and fibrosis

TSP1–CD47 signaling can inhibit endothelial-cell proliferation and migration and suppress nitric oxide and VEGFR2 signaling, thereby limiting angiogenesis (Zhao et al. 2023; Isenberg et al. 2009; Kaur et al. 2010). CD47 also contributes to vascular responses after tissue injury. In experimental liver ischemia–reperfusion injury, loss or blockade of CD47 improved tissue perfusion (Isenberg et al. 2008). In chronic liver injury, CD47–SIRPα signaling has been implicated in fibrotic remodeling. In experimental NASH, increased CD47 on necroptotic hepatocytes and SIRPα on liver macrophages impaired macrophage-mediated clearance of damaged hepatocytes, whereas blockade of CD47 or SIRPα enhanced hepatocyte clearance, reduced hepatic stellate-cell activation, and attenuated liver fibrosis (Shi et al. 2022). Chronic inflammation and progressive fibrosis or cirrhosis are well-established pathological backgrounds for hepatocellular carcinoma. Thus, dysregulated CD47–SIRPα signaling during chronic liver injury may contribute to an immune and fibrotic microenvironment relevant to HCC development. However, these findings do not establish CD47 as a direct or independent driver of hepatocarcinogenesis.

Metabolic–immune crosstalk relevant to CD47-directed therapy

Beyond its canonical role in suppressing phagocytosis, CD47 may also exert tumor-cell-intrinsic metabolic effects. In colorectal cancer, CD47 was reported to interact with and stabilize enolase 1 (ENO1) by limiting its ubiquitin-mediated degradation, thereby enhancing aerobic glycolysis and ERK signaling and promoting tumor-cell growth and metastasis (Hu et al. 2020). This finding provides a disease-specific link between CD47 and tumor metabolism, although the reported mechanism was largely immune-independent. In addition to such tumor-intrinsic metabolic regulation, metabolic features of the tumor microenvironment may influence immune responses relevant to CD47-directed therapy. In the OSCC model cited here, lactate accumulation and extracellular acidification were associated with immune evasion. Astragaloside IV targeted MCT1, reduced MCT1 mRNA stability, lactate secretion, and extracellular acidification, and enhanced CD8 + T-cell antitumor activity (Zhao et al. 2026). Although this study did not directly investigate CD47 blockade and was conducted outside the digestive system cancer setting, it suggests that lactate metabolism may influence the immune conditions required for effective macrophage- and T-cell-mediated tumor control.

Lipid signaling may also be relevant in this context. Diacylglycerol kinases (DGKs) regulate the conversion of diacylglycerol to phosphatidic acid and have important roles in T-cell receptor signaling and immune-cell activation (Liu et al. 2025). However, current evidence does not support a direct mechanistic link between DGKs and CD47 signaling. Rather, DGK-dependent signaling may influence the functional state of immune cells and thereby indirectly affect responses to therapies targeting the CD47–SIRPα axis. Further studies are needed to determine whether modulation of lactate or lipid metabolism can improve the efficacy of CD47-directed therapy in digestive system cancers.

Expression and clinical significance of CD47 in digestive system cancers

CD47 expression and its reported regulatory mechanisms vary among digestive system cancers. Table 1 summarizes representative studies and lists the assay and scoring methods when these were reported in the original studies. Because the methods and patient cohorts differed, the reported expression rates are not directly comparable.

Table 1.

Reported CD47 expression patterns, assay/scoring approaches, regulatory mechanisms or mechanistic context, and clinical associations across digestive system cancers

Cancer type	Reported CD47 expression (cohort)	Assay/scoring approach	Reported regulatory mechanisms or mechanistic context	Clinical associations and references
Esophageal squamous cell carcinoma	n = 100; CD47-high in 47% by tumor proportion score (TPS) and 50% by combined positive score (CPS)	IHC; CD47 expression was evaluated using tumor proportion score (TPS) and combined positive score (CPS), with a cutoff of 10 used to define high expression (TPS ≥ 10%; CPS ≥ 10)	miR-133a directly suppresses CD47 expression in ESCC; loss of miR-133a can therefore increase CD47 (Suzuki et al. 2012). IFN-γ can induce CD47 through JAK–STAT1–IRF1 signaling across several cancer cell types, but an ESCC-specific role has not been established (Ye et al. 2021).	High CD47 TPS/CPS was associated with shorter overall and relapse-free survival in the n = 100 cohort (Li et al. 2026). High CD47 expression was also associated with lymph-node metastasis and adverse prognosis in an independent ESCC study (Suzuki et al. 2012).
Gastric cancer	n = 453 across discovery and validation cohorts; tumors were classified as CD47-high or CD47-low using the study-defined H-score cutoff	IHC on tissue microarrays; semiquantitative H-score. The median H-score (70) was used as the high/low cutoff in both cohorts	Galectin-3 functionally cooperates with CD47 to suppress phagocytosis in gastric adenocarcinoma with peritoneal metastasis (Fan et al. 2023). In the clinical cohort, higher CD47 mRNA was enriched in MSI tumors and associated with ARID1A mutation and FGFR2-pathway activation (Shi et al. 2021). These findings represent biological associations rather than proof of a single upstream regulator	High CD47 expression was associated with poorer overall survival and reduced apparent benefit from fluorouracil-based adjuvant chemotherapy in retrospective subgroup analyses; CD47 expression was also associated with macrophage-related features (Shi et al. 2021)
Colorectal cancer	56.1% positive expression (55/98)	IHC on tissue microarrays; membranous staining scored semiquantitatively by staining intensity and area, with the study median IHC score used to define negative versus positive expression	The LINC00460/miR-186-3p/MYC feedback network increases CD47 and PD-L1 expression and promotes CRC immune escape (Luo et al. 2024).	CD47 positivity was associated with shorter overall survival and lymph-node involvement; CD47 remained an independent adverse prognostic factor in multivariable analysis (Aktepe et al. 2025).
Hepatocellular carcinoma	21.7% positive expression (36/166)	IHC on tissue microarrays; membranous staining intensity and the percentage of positive tumor cells were combined into an H-score. A ROC-derived H-score ≥ 10 was used to define positive expression (Kim et al. 2021).	A tumor-cell-intrinsic upstream regulator was not established in the cited HCC expression cohort (Kim et al. 2021). In NASH models, CD47–SIRPα signaling influences macrophage-mediated hepatocyte clearance and liver fibrosis, providing relevant liver-disease context rather than a direct HCC expression mechanism (Shi et al. 2022).	CD47 positivity was associated with large-vessel invasion, more advanced AJCC stage, and higher Ki-67 proliferation index (Kim et al. 2021).
Pancreatic ductal adenocarcinoma	54.1% positive expression (53/98)	IHC on tissue microarrays; tumor-cell membrane staining evaluated by H-score (0–300). H-score ≥ 10 defined positive expression	Radixin has been shown to regulate the cell-surface localization of CD47 in human PDAC cells; radixin silencing reduced surface CD47 expression without substantially altering CD47 mRNA levels (Kobori et al. 2023). Preclinical studies further indicate that CD47 blockade can remodel tumor-infiltrating immune-cell compartments and enhance antitumor immune responses (Pan et al. 2019; Chen et al. 2025; Song et al. 2020).	Positive CD47 expression was associated with advanced disease features and poorer overall and disease-free survival; it remained an independent adverse factor for overall survival in multivariable analysis (Amare et al. 2025).
Gallbladder cancer	17.4% positive expression in gallbladder cancer (16/92)	IHC was reported in the available conference abstract; detailed antibody/scoring/cutoff information was not provided in the accessible report	Specific upstream regulatory mechanisms of CD47 remain insufficiently characterized in gallbladder cancer	CD47 expression was associated with vascular and perineural invasion in gallbladder cancer (Jung and Kim 2022).
Cholangiocarcinoma	CD47 was detectable in 50/54 CCA tissues and was higher than in the comparator HCC cohort	IHC; CD47 expression was evaluated by H-score. Surface CD47 expression was also assessed by flow cytometry in CCA cell lines	Blockade of the CD47–SIRPα interaction with anti-CD47 B6H12.2 or anti-SIRPα enhanced macrophage-mediated phagocytosis of CCA cells; anti-CD47 treatment also reduced CCA colonization in the spleen and liver in an experimental metastasis model (Vaeteewoottacharn et al. 2019).	Clinical outcome associations were not evaluated in this study; disease-specific clinical evidence remains limited (Vaeteewoottacharn et al. 2019).

Reported expression frequencies were derived from independent studies that used different antibodies, tissue platforms, scoring systems, cutoff definitions, and patient populations. They should therefore not be interpreted as directly comparable prevalence estimates across cancer types. Where a tumor-specific upstream regulator has not been demonstrated, the table explicitly distinguishes mechanistic context from direct regulation of CD47 expression

Esophageal cancer

In esophageal squamous cell carcinoma (ESCC), high CD47 expression is associated with poor prognosis. In a cohort of 100 patients, 47% of tumors were classified as CD47-high by tumor proportion score (TPS) (Li et al. 2026). CD47 staining was detected in both tumor cells and immune cells, and these compartments should be considered separately in IHC assessment (Li et al. 2026). miR-133a is downregulated in ESCC and directly suppresses CD47 expression. Reduced miR-133a expression may contribute to CD47 upregulation and tumor progression (Suzuki et al. 2012). IFN-γ can induce CD47 expression in several tumor models, but this has not been specifically demonstrated in ESCC (Ye et al. 2021).

Gastric cancer

In gastric cancer, a 453-patient IHC study used a median H-score of 70 to classify tumors as CD47-high or CD47-low. High CD47 expression was associated with poorer survival (Shi et al. 2021). CD47 expression was also associated with M1-polarized macrophage infiltration and reduced benefit from fluorouracil-based adjuvant chemotherapy in retrospective analyses (Shi et al. 2021). In gastric cancer models, Galectin-3 has been shown to regulate CD47 expression and cooperate with CD47 in suppressing antitumor immunity (Fan et al. 2023). Higher CD47 mRNA levels have also been associated with MSI, ARID1A mutation, and FGFR2 pathway activation (Shi et al. 2021). In an exploratory analysis of ASPEN-06, higher membranous CD47 expression appeared to be associated with greater benefit from evorpacept-based therapy in HER2-positive gastric/GEJ cancer, but the cutoff has not been prospectively validated (Wainberg et al. 2025).

Colorectal cancer

In a cohort of 98 patients with primary colorectal cancer (CRC), 55 (56.1%) were classified as CD47-positive using a median IHC score of 4 as the cutoff. CD47 positivity was associated with lymph-node involvement and shorter overall survival and remained independently associated with poor prognosis in multivariable analysis (Aktepe et al. 2025). CD47 and CD44 expression levels were correlated, and both were associated with distant metastasis (Fujiwara-Tani et al. 2019). Other studies have reported differences in CD47 expression across tumor compartments (El Dein Mohamed et al. 2024). Single-cell analysis showed higher SIRPα expression in tumor-associated macrophages and granulocytic myeloid-derived suppressor cells in CRC (Huang et al. 2024). Data on CD47 regulation in CRC remain limited. The LINC00460/miR-186-3p/MYC feedback loop was shown to increase CD47 and PD-L1 expression and promote immune escape (Luo et al. 2024).

Hepatocellular carcinoma

In a cohort of 166 patients with hepatocellular carcinoma (HCC), 21.7% of tumors were CD47-positive, which was associated with major-vessel invasion and a higher Ki-67 proliferation index (Kim et al. 2021). In experimental NASH, increased CD47 on damaged hepatocytes and SIRPα on liver macrophages impaired macrophage-mediated hepatocyte clearance, whereas blockade of CD47 or SIRPα reduced hepatic stellate-cell activation and liver fibrosis (Shi et al. 2022). Because HCC often develops in fibrotic or cirrhotic liver, these findings may be relevant to the setting in which HCC develops, but they do not show that CD47 directly drives hepatocarcinogenesis (Shi et al. 2022; Otuagomah et al. 2025).

Pancreatic cancer

In a tissue microarray study of 98 patients with pancreatic ductal adenocarcinoma (PDAC), 53 tumors (54.1%) were CD47-positive using an H-score cutoff of 10. CD47 positivity was associated with advanced disease and shorter overall and disease-free survival and remained associated with overall survival after multivariable adjustment (Amare et al. 2025). In PDAC cells, radixin regulates the cell-surface localization of CD47. Radixin silencing reduced surface CD47 expression without a marked change in CD47 mRNA levels (Kobori et al. 2023). In PDAC models, CD47 blockade altered tumor-infiltrating immune-cell populations and enhanced antitumor immune responses (Pan et al. 2019; Song et al. 2020).

Biliary tract cancers

Evidence for CD47 in biliary tract cancers remains relatively limited compared with that in gastric, colorectal, hepatocellular, and pancreatic cancers. In a conference abstract reporting a 92-patient gallbladder cancer cohort, CD47 expression was detected in 17.4% of cases (16/92) and was associated with vascular and perineural invasion (Jung and Kim 2022). Because detailed information on the antibody, scoring method, and cutoff was not available in the accessible report, these findings should be interpreted cautiously.

Additional evidence supports a role for CD47 in cholangiocarcinoma (CCA).

In a study of 54 CCA and 22 hepatocellular carcinoma tissues, CD47 expression assessed by H-score was higher in CCA, with detectable expression in 50 of 54 CCA samples. Surface CD47 expression was also demonstrated in several CCA cell lines. Functional experiments showed that blockade of the CD47–SIRPα interaction with anti-CD47 B6H12.2 or anti-SIRPα enhanced macrophage-mediated phagocytosis of CCA cells. Anti-CD47 treatment also reduced CCA colonization in the spleen and liver in an experimental metastasis model (Vaeteewoottacharn et al. 2019). These findings provide preclinical support for targeting the CD47–SIRPα axis in CCA, although disease-specific clinical evidence in biliary tract cancers remains limited.

Therapeutic strategies targeting CD47 and advances in clinical research

Given the central role of the CD47-SIRPα axis in tumor immune evasion, various targeted therapeutic strategies have been developed and have entered the clinical trial phase. These strategies primarily include anti-CD47 monoclonal antibodies, SIRPα fusion proteins, bispecific antibodies, and anti-SIRPα antibodies (Li et al. 2026).

Primary treatment strategies and mechanisms of action

Anti-CD47 monoclonal antibodies were among the first CD47-directed agents tested clinically

They block CD47–SIRPα signaling and promote macrophage-mediated phagocytosis of tumor cells. However, binding to CD47 on normal blood cells and Fc-mediated effector activity can also cause hematologic toxicity (Sikic et al. 2019; Behrens et al. 2022).

Magrolimab (Hu5F9-G4) has been evaluated in both hematologic malignancies and solid tumors. In the phase III ENHANCE-3 trial in untreated AML, adding magrolimab to venetoclax and azacitidine did not improve overall survival or complete remission, and the study was stopped at a prespecified interim analysis for futility. Fatal treatment-emergent adverse events were more frequent in the magrolimab arm, mainly because of grade 5 infections (Daver et al. 2025). In CRC, magrolimab plus cetuximab produced an ORR of 6.3% in patients with KRAS-wild-type tumors and no objective responses in those with KRAS-mutant tumors in the phase 2 cohorts (Eng et al. 2025). The randomized phase 2 ELEVATE CRC study subsequently evaluated magrolimab plus FOLFIRI/bevacizumab and enrolled 77 patients before being terminated by the sponsor (ClinicalTrials.gov. n.d.). These findings also highlight the need to reduce CD47 binding to normal blood cells or restrict CD47 blockade to tumors. Early strategies included priming-dose schedules and antibodies with reduced red-blood-cell binding (Sikic et al. 2019; Sallman et al. 2023; Berlin et al. 2020).

Fc-inactivated SIRPα fusion proteins and tumor-directed bispecific antibodies are other approaches used to reduce off-tumor effects (Lakhani et al. 2021; Zhang et al. 2024).

Fc-inactivated SIRPα fusion proteins are designed to block CD47 while limiting Fc-mediated toxicity

Evorpacept (ALX148), a high-affinity CD47-blocking fusion protein with an inactive Fc region, has been evaluated alone and in combination with other anticancer therapies (Lakhani et al. 2021). Another engineered SIRPα–Fc fusion protein, HCB101, was designed to reduce red-blood-cell binding while maintaining CD47 blockade (Wang et al. 2025). Early results from an ongoing phase Ib/IIa study showed dose-dependent antitumor activity when HCB101 was combined with ramucirumab and paclitaxel in second-line gastric cancer (Yu et al. 2026). These findings remain preliminary and require confirmation in larger cohorts.

Bispecific antibodies are designed to improve tumor selectivity by simultaneously targeting CD47 and tumor-associated targets, such as HER2, PD-L1, Claudin 18.2, DLL3, and mesothelin

Representative agents include IBI322 (CD47/PD-L1) (Wang et al. 2021), IMM2902 (CD47/HER2) (Zhang et al. 2024), AK132 (CD47/Claudin 18.2) (Pang et al. 2024), PT217 (CD47/DLL3) (Sands et al. 2025), and NI-1801 (CD47/mesothelin) (EU Clinical Trials Register n.d.). PT886 (spevatamig), a bispecific antibody targeting CD47 and Claudin 18.2, has shown preliminary activity in first-line metastatic pancreatic ductal adenocarcinoma (mPDAC). In the phase 1/2 TWINPEAK trial (NCT05482893), the 2 mg/kg cohort reported an ORR of 48%, a DCR of 90%, a median PFS of 7.3 months, and a median OS of more than 13 months, with a median follow-up of 8.9 months (Saeed et al. 2026). Cytopenia rates were not higher than those expected with gemcitabine plus nab-paclitaxel, and no grade ≥ 3 nausea or vomiting was reported (Saeed et al. 2026). Responses were observed across different RAS mutation status and Claudin 18.2 expression levels (Saeed et al. 2026). These results are preliminary and require confirmation in larger cohorts.

Anti-SIRPα monoclonal antibodies

Anti-SIRPα antibodies, such as OSE-172 (BI 765063), target SIRPα on myeloid cells rather than CD47 on erythrocytes, thereby avoiding direct binding to red blood cells. Phase I data have shown tolerability and dose-dependent pharmacodynamic activity in patients with advanced solid tumors (Champiat et al. 2021).

Advances in clinical trials for digestive system cancers

Table 2 summarizes representative CD47-directed clinical studies in digestive system cancers. Planned/registered enrollment is listed separately from the population included in reported efficacy analyses.

Table 2.

Representative clinical trials targeting the CD47–SIRPα axis in digestive system cancers

Trial	Status	Cancer	Agent/regimen	Phase	Planned/registered enrollment	Efficacy-evaluable population	Key reported results	Refs
NCT05002127 (ASPEN-06)	Active, not recruiting; phase III portion not pursued by sponsor	HER2-positive gastric/GEJ cancer	Evorpacept + trastuzumab + ramucirumab + paclitaxel	Randomized phase II component of phase II/III program	Original planned enrollment N = 450; actual enrollment N = 127	Randomized phase II population N = 127; CD47 data available N = 119; HER2-confirmed N = 90; CD47-high/HER2-confirmed subgroup n = 43	Overall ORR 40.3% vs 26.6%; exploratory CD47-high subgroup ORR 65% (13/20) vs 26% (6/23), PFS HR 0.37	(Wainberg et al. 2025; ClinicalTrials.gov. n.d.; Shitara et al. 2025)
NCT05276310	Recruiting	Advanced solid tumors, including HCC and biliary tract cancer expansion cohorts	IMC-002 monotherapy (dose escalation); IMC-002 + lenvatinib in HCC/BTC expansion cohorts	Phase I	N = 62 (estimated)	Published dose-escalation cohort N = 12	DCR 50.0%; CBR 33%; no treatment-related thrombocytopenia reported	(Ahn et al., n.d.)
NCT05960955	Recruiting (last verified November 2023)	Resectable gastric/GEJ adenocarcinoma	AK117 + cadonilimab + chemotherapy vs cadonilimab + chemotherapy	Phase II	N = 90 (estimated)	Not reported	Efficacy results not reported	(ClinicalTrials.gov 2023)
NCT05167409	Terminated (accrual terminated due to safety concerns)	Refractory MSS metastatic CRC	Evorpacept + cetuximab + pembrolizumab	Phase II	Registry enrollment target N = 48	19 patients enrolled; 16 patients treated and included in the clinical efficacy analysis	ORR 6.3%; DCR 12.5%; mPFS 2.3 months; mOS 10.9 months; enrollment stopped after treatment-related grade 5 HLH and CRS (one each)	(ClinicalTrials.gov 2021; Lentz et al. 2025)
NCT05482893 (TWINPEAK)	Recruiting	Advanced gastric/GEJ adenocarcinoma, biliary tract carcinoma, and pancreatic ductal adenocarcinoma; reported efficacy from the first-line mPDAC cohort	Spevatamig (PT886) + gemcitabine/nab-paclitaxel in the reported mPDAC cohort	Phase I/II	N = 258 (estimated)	Reported 2 mg/kg QW spevatamig + GnP first-line mPDAC cohort N = 22	ORR 48%; DCR 90%; mPFS 7.3 months; mOS > 13 months; median follow-up 8.9 months	(Saeed et al. 2026; ClinicalTrials.gov 2022)
NCT05805956	Unknown status (last verified March 2023)	HER2-expressing advanced solid tumors	IMM2902	Phase I/II	N = 105 (estimated)	Not reported	Results not reported	(ClinicalTrials.gov 2023)

Planned/registered enrollment refers to the enrollment target or total enrollment reported in the trial registry and may differ from the population included in published or conference efficacy analyses. Trial status was checked on ClinicalTrials.gov on August 13, 2026. Where shown, “last verified” refers to the Last Verified date reported in the registry, rather than the date of our status check. In the exploratory ASPEN-06 biomarker analysis, CD47-high was defined as ≥ 10% of tumor cells with high-intensity membranous CD47 staining

Theoretical foundations and practical applications of combined treatment strategies

Single-agent CD47-directed therapy has generally shown limited activity in solid tumors. Combination strategies are therefore being explored to strengthen pro-phagocytic signaling, support adaptive immune responses, or alter the tumor microenvironment.

Combination with targeted antibodies

Tumor-targeting antibodies can opsonize cancer cells and engage FcγRs on macrophages, thereby promoting antibody-dependent cellular phagocytosis (ADCP). Blocking CD47–SIRPα can further enhance this FcγR-dependent phagocytic response. This mechanism has been demonstrated with trastuzumab- and cetuximab-based approaches in preclinical models (Upton et al. 2021; Hao and He 2024).

Combination with chemotherapy

Chemotherapy can induce immunogenic cell death and increase pro-phagocytic signals such as surface calreticulin, while also releasing damage-associated molecular patterns and tumor antigens. By blocking the CD47–SIRPα inhibitory signal, CD47-directed therapy may increase macrophage phagocytosis of chemotherapy-treated tumor cells (Chen et al. 2024; Ye et al. 2023).

Combination with PD-1/PD-L1 inhibitors

CD47 blockade can increase tumor-antigen uptake and cross-presentation by antigen-presenting cells, which may support CD8 + T-cell priming (Liu et al. 2015; Hendriks et al. 2020). Combining CD47-directed therapy with PD-1/PD-L1 blockade may therefore enhance both myeloid and T-cell responses. However, clinical benefit from this approach has not been established across digestive system cancers, and greater immune activation may also increase toxicity.

Combination with radiotherapy

Radiotherapy can induce immunogenic cell death and may also increase CD47 expression in tumor cells. In colorectal cancer models, irradiation increased CD47 through an ATR-dependent mechanism, which limited antitumor immune responses (Hsieh et al. 2022). Combining radiotherapy with CD47 blockade may therefore help counter this treatment-induced inhibitory signal, although clinical evidence in digestive system cancers remains limited (Hsieh et al. 2022; Rauf et al. 2025).

Across these approaches, CD47 blockade removes an inhibitory phagocytic signal, while the partner treatment may provide tumor opsonization, induce pro-phagocytic changes, or enhance adaptive immune responses. The main mechanisms of these combination strategies are summarized in Fig. 2.

Fig. 2.

Combination strategies with CD47 blockade and a conceptual sequential scheduling approach. A Chemotherapy and radiotherapy can induce immunogenic cell death, with increased calreticulin exposure and release of DAMPs such as HMGB1. In colorectal cancer models, radiotherapy can also increase CD47 expression through an ATR-dependent mechanism. B Tumor-targeting antibodies such as cetuximab and trastuzumab can promote FcγR-dependent antibody-dependent cellular phagocytosis (ADCP). C PD-1/PD-L1 blockade may help sustain antitumor T-cell responses. D The figure also shows a conceptual sequential strategy in which CD47 blockade plus cetuximab is followed by PD-1 blockade. This strategy has not been clinically validated, and whether sequential administration can reduce HLH, CRS, or other immune-related toxicities remains unknown

Outlook

Addressing toxicity issues

CD47 is widely expressed on normal hematopoietic cells, including red blood cells and platelets, which contributes to on-target hematologic toxicity (Pei et al. 2025). Anemia and thrombocytopenia have been reported in early clinical studies. To address this issue, current solutions include: (1) Priming-dose regimens to reduce the risk of anemia (Guo et al. 2025); (2) Developing selectively binding antibodies, such as lemzoparlimab (TJC4) (Berlin et al. 2020); (3) Using Fc-inactivated SIRPα fusion proteins, such as evorpacept (Lakhani et al. 2021); (4) Bispecific antibodies designed to improve tumor selectivity (Zhang et al. 2024; Nan et al. 2025); (5) Anti-SIRPα antibodies avoid direct binding to red blood cells (Voets et al. 2019; Kuo et al. 2020). These strategies have demonstrated the potential to improve safety in preclinical or early clinical trials.

Recent studies have shown that erythrocyte toxicity of anti-CD47 antibodies depends not only on CD47 expression on red blood cells but also on antibody valency and binding properties. Bivalent anti-CD47 antibodies can bind CD47 molecules simultaneously on the erythrocyte surface, increasing avidity and promoting red-cell crosslinking and agglutination. Monovalent binding reduces this crosslinking effect and is associated with less erythrocyte damage. In one study, bivalent Hu5F9 caused marked anemia even in the absence of intact CD47–SIRPα signaling, whereas a single-arm Hu5F9 construct caused only mild RBC destruction, indicating that antibody bivalency itself can contribute to hematologic toxicity (Zhang et al. 2026). Reducing RBC binding or crosslinking may therefore improve hematologic safety. The HuNb1-IgG4 nanobody-based construct, for example, showed lower binding to human RBCs and did not induce detectable RBC agglutination over the tested concentration range (Ma et al. 2020). Monovalent formats, constructs with reduced RBC binding, and Fc-inactivated SIRPα-based agents may therefore reduce on-target hematologic toxicity (Lakhani et al. 2021; Zhang et al. 2026; Ma et al. 2020).

Biomarkers for treatment response and patient selection

In ASPEN-06, an exploratory analysis suggested that higher tumor-cell membranous CD47 expression may be associated with greater benefit from evorpacept-based therapy in HER2-positive gastric/GEJ cancer. CD47 expression data were available for 119 of 127 randomized patients, and 90 had confirmed HER2 positivity based on fresh biopsy or circulating tumor DNA assessment (Wainberg et al. 2025). Among patients with confirmed HER2 positivity and ≥ 10% of tumor cells showing high-intensity membranous CD47 staining, the ORR was 65% (13/20) with evorpacept-containing therapy versus 26% (6/23) with control, and the PFS HR was 0.37 (95% CI 0.2–0.8) (Wainberg et al. 2025). Similar trends were observed across several CD47 cutoffs (Wainberg et al. 2025). The ≥ 10% threshold has not been prospectively validated for patient selection.

Several issues remain before CD47 expression can be used for patient selection. As summarized in Table 1, published studies have used different IHC methods, scoring systems, and cutoffs, limiting comparisons across cohorts. CD47 membrane localization may also be important, although its clinical value remains unclear. In HCC, AFP/HuR signaling promotes CD47 translocation to the cell membrane, whereas in PDAC cells, radixin silencing reduces surface CD47 expression with little effect on CD47 mRNA levels (Kobori et al. 2023; Pan et al. 2025). Preclinical studies also indicate that tumor-intrinsic type I interferon signaling can influence the response to CD47–SIRPα blockade (Zhou et al. 2024). Prospective studies are needed to determine whether combining CD47 expression with other tumor or immune features can improve patient selection.

Combination strategies and safety considerations

CD47-directed therapy is often combined with treatments that provide a pro-phagocytic signal or stimulate adaptive immunity (Ye et al. 2023). Such combinations include tumor-targeting antibodies, chemotherapy, radiotherapy, and checkpoint inhibitors. However, greater immune activation may also increase toxicity. In the phase II study of evorpacept, cetuximab, and pembrolizumab in refractory MSS CRC, 16 patients were treated and the ORR was 6.3%. Accrual was stopped after a second treatment-related grade 5 event; the two fatal events were hemophagocytic lymphohistiocytosis (HLH) and cytokine release syndrome (CRS) (Lentz et al. 2025). The investigators considered the HLH event probably related and the CRS event possibly related to the three-drug regimen, although the contribution of each individual agent could not be determined. Progressive disease with a high tumor burden may also have contributed to the HLH event (Lentz et al. 2025). Cetuximab can provide an FcγR-dependent opsonizing signal, evorpacept relieves CD47–SIRPα-mediated inhibition of phagocytosis, and pembrolizumab relieves PD-1-mediated inhibition of T-cell responses. Concurrent activation of myeloid and T-cell responses may have contributed to the inflammatory toxicity, although this remains uncertain (Lentz et al. 2025).

The MSS CRC study illustrates that combining agents that act on different immune pathways may increase toxicity without necessarily improving efficacy. Future studies should carefully evaluate dose, schedule, and patient selection. Sequential administration could also be explored, but there is currently no clinical evidence that it reduces HLH, CRS, or other immune-related toxicities.

Mechanisms underlying limited clinical activity and resistance

The limited and variable activity of CD47-directed therapy in solid tumors may reflect several biological barriers. Tumor-intrinsic factors may include heterogeneous CD47 expression and insufficient pro-phagocytic signals (Liu et al. 2023). CD24–Siglec-10 and MHC-I–LILRB1 are additional anti-phagocytic pathways that suppress macrophage-mediated clearance and may therefore limit responses to CD47 blockade (Liu et al. 2023; Barkal et al. 2019; Barkal et al. 2018).

Macrophage phenotype and FcγR signaling may also affect response. Blocking CD47–SIRPα removes an inhibitory signal, but phagocytosis still requires activating signals. In antibody combinations, tumor-bound IgG can engage FcγRs and promote ADCP, which is particularly relevant for CD47-directed agents with attenuated Fc activity (Lakhani et al. 2021; Tsao et al. 2019). Macrophage phenotype may add further variability; in a pancreatic cancer model, anti-CD47 treatment increased pro-inflammatory macrophage populations and reduced anti-inflammatory macrophage populations (Pan et al. 2019). CD47 on erythrocytes and other normal cells also creates an antigen sink and contributes to hematologic toxicity (Sikic et al. 2019; Pei et al. 2025; Guo et al. 2025). These effects can narrow the therapeutic window.

Physical barriers may also reduce the activity of CD47-directed therapy in digestive system cancers. In PDAC, dense stroma can limit drug delivery and immune-cell access (Alausa et al. 2022). In chronic liver disease, CD47–SIRPα signaling has been associated with impaired macrophage clearance of damaged hepatocytes and fibrosis, but this has not been shown to cause resistance to CD47-directed therapy in HCC (Shi et al. 2022; Otuagomah et al. 2025). Durable tumor control may also depend on antigen cross-presentation and T-cell priming after phagocytosis (Liu et al. 2015; Hendriks et al. 2020). If these responses are weak, increased phagocytosis may not translate into sustained antitumor activity.

Exploration of novel therapeutic modalities

Beyond antibodies and fusion proteins, cellular engineering and tumor-localized delivery are being investigated as alternative ways to limit systemic CD47 blockade. In a murine colon cancer model, anti-Trop2 CAR-T cells engineered to secrete SIRPα-Fc showed greater antitumor activity than conventional CAR-T cells (Chen et al. 2022). CAR macrophages have also been studied in gastric cancer. The pArg1-CD47 CAR-Mφ platform uses an Arg1-responsive promoter to restrict CAR expression to the tumor microenvironment. In preclinical breast and gastric cancer models, these macrophages showed antitumor activity and reduced erythrocyte cytotoxicity compared with constitutive CD47-targeting CAR expression (Du et al. 2025). These cellular approaches remain preclinical.

Nanotechnology has also been used to localize CD47-directed treatment. In hepatocellular carcinoma, a dual-targeted nanosystem directed against CD47 and CDC7 suppressed tumor growth in mouse models (Gong et al. 2024). Another approach targets the cancer-associated integrin αvβ3–CD47 complex rather than blocking CD47 systemically and has shown preclinical antitumor activity with less anemia in model systems (Yu et al. 2026). Tumor-microenvironment-responsive nanocarriers are being developed for a similar purpose by limiting systemic exposure and increasing drug delivery to tumors (Liao and Niu 2022).

CD47 overexpression has also been used in hypoimmunogenic cell-engineering strategies. In regenerative medicine, pluripotent stem cells and pancreatic islet platforms have been engineered using HLA class I/II depletion together with increased expression of immunoregulatory molecules, including CD47, to reduce immune recognition and graft rejection (Shalaby and Abdelalim 2025). These approaches were developed for transplantation rather than cancer therapy, but they provide additional evidence that increased CD47 expression can protect cells from innate immune clearance. Selected engineering strategies to reduce CD47-related toxicity and improve tumor selectivity are shown in Fig. 3.

Fig. 3.

Engineering strategies to reduce CD47-related toxicity and improve tumor selectivity. A Bivalent anti-CD47 antibodies can promote RBC crosslinking and agglutination, whereas monovalent or reduced-RBC-binding formats may reduce RBC binding and erythrocyte toxicity. B An Arg1-responsive promoter is being explored to drive TME-responsive CAR expression in macrophages. C Interfering with integrin αvβ3–CD47 co-stabilization can increase macrophage phagocytosis in preclinical models. D TME-responsive nanocarriers may localize the release of CD47-directed agents and reduce systemic exposure. The approaches shown in panels B–D remain preclinical

Conclusion

CD47 regulates macrophage-mediated phagocytosis and can contribute to tumor immune evasion. Its expression and regulation vary among digestive system cancers, and reported expression rates are not directly comparable because studies have used different IHC methods, scoring systems, cutoffs, and patient cohorts. CD47–SIRPα is the best-characterized inhibitory pathway, but TSP1 signaling, metabolic regulation, and other anti-phagocytic checkpoints may also influence tumor immunity.

These biological features have shaped the development of CD47-directed therapies. CD47-directed therapy has expanded beyond anti-CD47 monoclonal antibodies to include Fc-inactive SIRPα fusion proteins, tumor-targeted bispecific molecules, and engineered cellular or nanotechnology approaches. The limited activity of magrolimab in CRC, together with mixed results and the discontinuation of several magrolimab programs, highlights the limitations of broad systemic CD47 blockade and the need for a wider therapeutic window. Evorpacept and newer tumor-directed approaches are designed to reduce hematologic toxicity or improve tumor selectivity, often in combination with antibodies that provide an opsonizing signal.

Among the available clinical data, ASPEN-06 provides one of the clearest examples in digestive system cancers. In HER2-positive gastric/GEJ cancer, the randomized phase 2 ASPEN-06 study showed clinical activity with evorpacept-based therapy and also included an exploratory analysis of tumor CD47 expression. Higher membranous CD47 expression appeared to be associated with greater benefit in this exploratory analysis, but the ≥ 10% high-intensity staining threshold has not been prospectively validated (Wainberg et al. 2025; Shitara et al. 2025).

In refractory MSS CRC, evorpacept combined with cetuximab and pembrolizumab showed limited efficacy, and accrual was stopped after fatal treatment-related HLH and CRS events (Lentz et al. 2025). These findings support more careful patient selection and closer attention to the safety of combination regimens.

These clinical findings also highlight several barriers to broader efficacy. CD47 expression on normal cells contributes to an antigen sink and hematologic toxicity, while tumor heterogeneity and alternative anti-phagocytic pathways may reduce treatment response. Macrophage phenotype and FcγR signaling can influence ADCP, and dense stroma may limit drug delivery and immune-cell access. Sustained tumor control may also depend on effective adaptive immune responses. More tumor-selective therapies and better biomarkers for patient selection are still needed.

Future studies should clarify how CD47 is regulated in different digestive system cancers and whether membrane localization affects treatment response. Biomarker assays also require prospective validation and better standardization. Reducing hematologic toxicity remains important, and combination strategies should be guided by tumor biology, including alternative anti-phagocytic pathways, macrophage phenotype, stromal features, and adaptive immune activity. CD47 is a well-established phagocytosis checkpoint, but the clinical benefit of CD47-directed therapy remains uncertain in digestive system cancers. Recent trials have shown both positive and negative results, and benefit has not been consistent across tumor types. CD47-directed therapy may therefore have a role in selected patients and selected combination regimens, but it cannot yet be considered a broadly effective strategy across digestive system cancers.

Authors’ contributions

Haolin Sun contributed to conception and design. Chenyu Wei, Jianglan Long and Xiaojia Liu contributed to literature search. Haolin Sun contributed to manuscript writing. Bangwei Cao and Quanfu Li approved the final version of the manuscript, and agree to be accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved.

Funding

Supported by the Capital Health Research and Development of Special (grant no. 2026–2-1031).

Data availability

No datasets were generated or analysed during the current study.

Declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.