Immuno-cytotoxic convergence: integrating antibody‒drug conjugates and immunofusion proteins to overcome resistance in gastrointestinal cancers

Abstract

Gastrointestinal (GI) cancers account for nearly one-third of cancer-related deaths worldwide and often remain difficult to treat because of pronounced molecular heterogeneity and strongly immunosuppressive tumor microenvironments (TMEs). Antibody–drug conjugates (ADCs) deliver highly potent cytotoxic payloads to antigen-positive cells and may partially address intratumoral heterogeneity through bystander killing. In contrast, immunofusion proteins (IFPs)—including cytokine–antibody fusions and T-cell–redirecting modalities such as bispecific T-cell engagers—are designed to localize immune activation and/or retarget immune effectors within immunologically “cold”, stroma-rich tumors. In this review, we integrate recent clinical and translational advances in ADCs and emerging IFP platforms across gastric, colorectal, pancreatobiliary and hepatocellular cancers, with particular attention to organ-dependent efficacy–toxicity trade-offs (e.g., interstitial lung disease (ILD) associated with DXd-based ADCs; cytokine release syndrome with T-cell engagers) and convergent resistance mechanisms, including antigen loss, impaired payload processing, immune exhaustion, and stromal exclusion. We further propose “immunocytotoxic convergence” as a hypothesis-generating and testable working model: under specific conditions, ADC-driven cytoreduction, immunogenic stress signatures consistent with immunogenic cell death (ICD), and/or stromal remodeling may transiently improve immune accessibility and thereby create a window for subsequent immune amplification by IFPs. Direct clinical evidence for an explicit ADC→IFP Prime–Amplify sequence in GI cancers remains limited. We therefore summarize the current evidence base, define key failure modes and safety constraints, and outline biomarker-enabled experimental and early-phase trial approaches needed to validate—or falsify—this sequencing concept.

Background

Gastrointestinal (GI) cancers are among the deadliest cancers. Globally, they cause approximately one in three of all cancer deaths [1]. Colorectal cancer (CRC) remains a leading cause of global oncological morbidity, whereas pancreatic ductal adenocarcinoma (PDAC) is one of the most lethal malignancies, with a five-year survival rate that is persistently less than 10% [2]. The therapeutic recalcitrance of these advanced malignancies stems largely from their biological complexity. Notably, their molecular heterogeneity, such as RAS/RAF mutations, causes CRCs to be resistant to anti–epidermal growth factor receptor (EGFR) therapy [3]. Furthermore, the heavily immunosuppressive tumor microenvironment (TME) in PDAC [4], characterized by a dense desmoplastic stroma, poses a formidable physical barrier to drug delivery. This underscores the critical need for dual-modality strategies capable of simultaneously eliminating malignant cells and remodeling their protective stromal niches.

Recent advances in targeted and immune therapies have begun to address these challenges. On the cytotoxic front, antibody-drug conjugates (ADCs) such as trastuzumab deruxtecan (T-DXd) have demonstrated breakthrough efficacy in molecularly defined subsets, achieving an objective response rate (ORR) of 51% in chemotherapy-refractory, HER2-positive gastric cancer, thereby validating the delivery of potent payloads to overcome antigen heterogeneity [5, 6]. Concurrently, immunofusion proteins (IFPs), including bispecific T-cell engagers (BiTEs) and cytokine-antibody fusions, have emerged as potent tools to redirect immune effectors towards immunologically “cold” tumors, offering a mechanism to bypass classical immune evasion pathways such as MHC downregulation [7].

Despite these individual successes, significant gaps remain. ADC monotherapy often fails to sustainably remodel the immunosuppressive stroma, while IFPs struggle to penetrate and activate immune responses in densely fibrotic, antigen-heterogeneous environments. Furthermore, shared resistance mechanisms, such as antigen loss for ADCs and T-cell exclusion for IFPs, limit the durability of response. Therefore, a critical, unresolved issue is whether and how these two potent modalities can be rationally integrated to achieve synergistic and durable efficacy.

In this review, we synthesize clinical and translational data on ADCs and IFPs across major GI cancers and critically compare their resistance landscapes, pharmacologic constraints, and toxicity liabilities. Based on these observations, we use the term “immunocytotoxic convergence” to describe a context-dependent framework in which cytotoxic precision and immune engagement may be mechanistically coupled in GI malignancies. We further outline a testable “Prime–Amplify” sequencing hypothesis. In this model, selected ADCs may create a transient window of immune accessibility, reflected by measurable pharmacodynamic (PD) signals consistent with ICD and/or stromal remodeling. This putative “primed” state could increase the likelihood of benefit from subsequent IFP-mediated immune activation, but requires sequence-explicit validation. Throughout, we clearly distinguish established evidence from extrapolations and hypotheses, while also proposing the necessary biomarker readouts and study designs to evaluate this sequencing concept in GI oncology.

Search strategy

Relevant publications were identified through structured searches of PubMed, Web of Science and Google Scholar for the period from 1 January 2019 to 25 December 2025 via combinations of the following terms: “antibody–drug conjugate”, “ADC”, “immunocytokine”, “fusion protein”, “bispecific”, “gastrointestinal cancer”, “gastric”, “colorectal”, “pancreatic”, “biliary”, “hepatocellular”, “clinical trial”, “immunotherapy”, and “checkpoint inhibitor”. Additional data on ongoing trials were retrieved from ClinicalTrials.gov, the Chinese Clinical Trial Registry (ChiCTR), and abstracts from major oncology conferences (ASCO, ESMO). Only original research articles and trial reports with clearly reported clinical or preclinical efficacy and/or mechanistic outcomes were included. Review articles and editorials were excluded. Conference abstracts were included only when they reported clearly interpretable efficacy and/or safety data and no full publication was available. Given the rapidly evolving nature of the field, key preclinical studies illustrating novel ADC and IFP platforms or ADCs and immunotherapy combinations in GI-relevant models have also provided unique mechanistic insights. To distinguish established findings from inference, we applied an evidence-grading framework (Sect.  3). Key claims, supporting data types, and limitations are integrated into an evidence map presented later in the manuscript.

Evidence framework and terminology

To reduce conceptual overreach and to clearly separate established observations from inference, we adopt an evidence-based framework to discuss “immunocytotoxic convergence”. In this review, immunocytotoxic convergence is defined as a mechanistically plausible coupling between cytotoxic precision (e.g., ADC-mediated tumor and/or stromal debulking) and immune engagement (e.g., IFP-mediated immune activation or redirection). In selected contexts, such coupling may improve antitumor activity in GI malignancies. We distinguish convergence as an observation (two modalities showing complementary mechanisms and partly non-overlapping resistance constraints) from convergence as a sequencing strategy (a deliberate ADC→IFP Prime–Amplify regimen). The latter remains hypothesis-generating in GI oncology and requires sequence-explicit validation.

Evidence grading (GI-context anchored)

Given the heterogeneity of data across GI tumor types and biologic platforms, we categorize supporting evidence into three pragmatic grades anchored to GI relevance and causal strength: Grade A, direct GI clinical evidence and/or GI-relevant in vivo evidence with therapeutic effect plus mechanistic PD readouts supporting the claim; Grade B, GI-relevant but indirect evidence consistent with the claim yet lacking sequence-explicit validation, reverse-order controls, or relying on related modalities (e.g., ADC→ICI rather than ADC→IFP); and Grade C, extrapolative or hypothesis-level evidence derived mainly from non-GI contexts, in vitro systems, or conceptual reasoning without GI-specific validation. This grading system does not imply that claims designated as Grade B/C are incorrect; rather, it clarifies where evidence is supportive versus inferential and highlights which components require prospective testing.

Evidence map for the Prime–Amplify model

The Prime–Amplify hypothesis involves multiple linked steps: ADC-driven cytoreduction and immunogenic stress, increased immune accessibility and antigen presentation, followed by IFP-driven effector amplification. Evidence supporting one step should not be interpreted as proof of the entire sequence. Consequently, an evidence map is provided that systematically organizes each claim (A–N) by specifying its scope of support, GI-relevant context, evidence type and source, representative examples, limitations, and an evidence grade (A for strong, B for supportive but indirect, C for extrapolative/hypothetical). Throughout the manuscript, this framework is adopted to distinguish ADC–ICI synergy from evidence supporting ADC→IFP sequencing, to clarify the dependence of conclusions on indirect or extrapolated data, and to define the minimum experimental and clinical datasets required for validating or refuting the directionality, timing, and safety parameters of sequential regimens in GI malignancies.

Mechanisms of action and design principles

Antibody‒drug conjugates (ADCs)

ADCs represent a major advance in GI cancer therapy by combining the targeting specificity of monoclonal antibodies with the potency of cytotoxic payloads. Three essential components include a tumor-targeting antibody, an engineered linker, and a potent payload (Fig. 1). The mechanism proceeds via a sequence of coordinated events, which include specific binding to tumor-associated antigens, efficient receptor-mediated endocytosis following antigen binding, lysosomal trafficking with linker cleavage by lysosomal enzymes such as cathepsin β, and final release of the cytotoxic warhead [8].

Fig. 1.

Mechanism of Action and Bystander Effect of ADCs. Schematic illustrating the four-step process of ADC-mediated cytotoxicity: (1) Specific binding to tumor-associated antigens (TAAs) such as HER2; (2) Receptor-mediated endocytosis and intracellular trafficking; (3) Lysosomal cleavage of the cleavable linker (e.g., GGFG in T-DXd) and release of the membrane-permeable payload (DXd); (4) Induction of DNA damage and cell death, which in some contexts may be accompanied by immunogenic stress signals consistent with immunogenic cell death (ICD). Key components including antibody, cleavable linker, potent payload, and drug-antibody ratio (DAR) are labeled

Released payloads such as DXd and SN-38 induce cell death by damaging DNA [9, 10]. Moreover, MMAE and DM1 hinder the assembly of microtubules [11, 12]. A critical feature is the bystander effect mediated by membrane-permeable payloads such as DXd [13]. Nevertheless, the same response leads to on-target, off-tumor toxicity for the target antigen, which shows baseline expression in healthy tissue (e.g., CLDN18.2 shows baseline expression in the gastric mucosa) [14]. Clinical data from DESTINY-CRC02 revealed that DXd-based ADCs such as T-DXd were associated with adjudicated drug-related interstitial lung disease (ILD) in 8% of patients treated at 5.4 mg/kg and 13% at 6.4 mg/kg, including one fatal event at the higher dose, highlighting a clear dose–toxicity gradient [15]. In contrast, nonpermeable payloads such as DM1, as used in T-DM1, reduce systemic off-tumor toxicity but yielded only a 16% ORR in the GATSBY trial [16], underscoring that drug-to-antibody ratio (DAR) optimization alone is insufficient and must be complemented by TME-tailored activation strategies in heterogeneous GI tumors.

Continuous component optimization has driven ADC evolution. Humanized IgG1 frameworks reduce immunogenicity, and advances in conjugation techniques help to improve DAR homogeneity, maintaining a balance between systemic stability and efficient release of the active agent inside cells. Linkers can be designed in either cleavable formats (e.g., the GGFG peptide in T-DXd) or noncleavable formats (e.g., the thioether linker in T-DM1) [17, 18]. Payload selection is increasingly shifting toward ultrapotent warheads (picomolar to nanomolar IC50) to maximize the cytotoxic impact per internalization event. Optimizing the DAR is essential, as T-DXd showed better cytotoxicity with a DAR of ~ 8, SHR-A1811 maintained efficacy with improved safety with a DAR of ~ 6, and T-DM1 (DAR of 3.5) is underperforming in gastric cancer [19, 20]. Additionally, bispecific ADCs (BsADCs) targeting multiple antigens (e.g., CDH17×GUCY2C) appear to improve tumor specificity and internalization, as observed in CRC [21]. Even with these engineering advances, organ-specific toxicity continues to hinder the development of ADCs for GI tumors [22]. More fundamentally, although modern ADCs are highly effective at direct cytotoxicity, they often struggle to reverse the profound immunosuppression of the GI stroma independently. This intrinsic limitation means that a complementary immunomodulatory partner is needed to sustain the antitumor response.

Immunofusion proteins (IFPs)

Whereas ADCs deliver cytotoxic payloads, IFPs aim to reshape immunosuppressive TMEs by localizing immune stimulation and/or redirecting effector cells. This makes them conceptually attractive in immune-cold, stroma-rich GI tumors [23]. However, a significant challenge persists: the activity of immune engagers is frequently impeded by physical and cellular barriers in GI malignancies, including dense fibrosis, abnormal vasculature, and myeloid-dominant suppressive niches that restrict T-cell access and function. Consequently, the clinical potential of IFPs may depend on upstream interventions that improve tissue accessibility, such as tumor debulking and/or stromal remodeling, rather than on immune activation alone. In this review, we use “IFPs” as an umbrella term for antibody-fusion formats that localize immune modulation (e.g., cytokine–antibody fusions and multifunctional checkpoint fusions) as well as T-cell–redirecting formats (e.g., BiTE-like engagers), while acknowledging their distinct pharmacokinetics and safety profiles.

Conceptually, IFPs can be subdivided into three major classes (Fig. 2): (i) cytokine–antibody fusions, in which potent cytokines (e.g., IL-2, IL-12, and IL-15) are fused to tumor- or stroma-targeting antibodies to concentrate immune stimulation within the TME; (ii) bispecific or multispecific T-cell engagers, which bridge tumor antigens (e.g., CEA and CLDN18.2) with CD3ε on T cells to form MHC-independent immunological synapses; and (iii) multifunctional checkpoint fusions, exemplified by bintrafusp alfa and ABL503, which combine checkpoint blockade with additional immunoregulatory activities (e.g., TGF-β trapping or 4-1BB agonism).

Fig. 2.

Three Major Classes and Mechanistic Action of Immunofusion Proteins (IFPs). (A) Cytokine–antibody fusions: Localize cytokine signaling (e.g., IL-2v, IL-15) to the tumor microenvironment (TME) via tumor/stromal targeting (e.g., FAP) to selectively activate CD8⁺ T and NK cells without systemic toxicity. (B) Bispecific T-cell engagers (BiTEs): Bridge tumor antigens (e.g., CEA, CLDN18.2) to CD3ε on T cells to form MHC-independent immunological synapses and trigger T-cell cytotoxicity. (C) Multifunctional checkpoint fusions: Combine checkpoint blockade (e.g., PD-L1) with additional immunomodulatory activities (e.g., TGF-β trap, 4-1BB agonism) to remodel the suppressive TME

Cytokine‒antibody fusions

Cytokine‒antibody fusion is an innovative immunotherapy that covalently links highly potent cytokines to monoclonal antibodies capable of targeting tumors, resulting in localized immune activation within the TME and reduced systemic exposure [23]. For example, FAP-IL2v targets fibroblast activation proteins on cancer-associated fibroblasts (CAFs) in pancreatic cancer and CRC [24, 25]. The reengineered IL-2 variant (R38E/F42A) has lower CD25 affinity but binds to IL-2Rβγ, which allows for CD8 + T-cell and NK cell expansion with no increase in Tregs [24, 26].

A novel agent called anti-PD-L1-IL-15 combines blockade of checkpoints with the support of IL-15 for memory T and NK cells [27]. CD8 + T-cell interference and antitumor efficacy were greater with anti-PD-L1-IL-15 than with PD-L1 blockade alone in a model of hepatocellular carcinoma (HCC) [27]. These fusion proteins are optimized pharmacokinetically through half-life enhancement strategies such as albumin fusion and PEGylation to prolong circulation in the body while ensuring retention in the tumor and reducing off-target toxicity [28].

Bispecific T-cell engagers (BiTEs)

BiTEs (including other T-cell redirecting formats) are an innovative class of immunotherapies that induce T-cell cytotoxicity by binding simultaneously to a tumor antigen and CD3ε on T cells to form an MHC-independent immunological synapse and overcome a major immune evasion mechanism in GI cancers. Early clinical experience with the CEA×CD3 bispecific antibody cibisatamab (CEA-TCB) in mCRC demonstrated modest efficacy (ORR ≈ 14%) and a high incidence of cytokine release syndrome (CRS ~ 60%) with features of T-cell exhaustion, indicating that current T-cell engagers require further engineering to mitigate toxicity while enhancing activity in immunologically “cold” TMEs [29].

Through carefully controlled optimization, AZD5863 was shown to bind with high affinity to CLDN18.2 while simultaneously engaging less with CD3 to reduce cytokine-related risks [30]. The construct demonstrates significant antitumoral activity in gastric and pancreatic models, thus causing minimal induction of cytokines, and is currently being evaluated clinically (NCT06005493). Tetrameric designs and knob-into-hole technologies enable enhanced binding avidity and T-cell activation under stroma-rich conditions [31].

Next-generation designs therefore prioritize conditional activation (pH-/protease-sensitive masking, low-affinity CD3 arms) to expand the therapeutic window in solid tumors [30, 32, 33]. In contrast to cytokine fusions such as FAP-IL2v, the brief half-life of BiTEs aids in tissue penetration but raises concerns about increased dosing frequency and potential toxicity [34, 35]. Therefore, incorporating adjustable designs, such as AZD5863’s CD3 binding with low affinity, is advised to manage these aspects, although additional randomized controlled trials are necessary to validate their prolonged advantages in treating GI malignancies.

Immune checkpoint fusions

Immune checkpoint fusions represent a new class of IFPs that combine checkpoint blockade and tumor specificity to localize immunomodulatory effects and limit systemic toxicity. A representative example is bintrafusp alfa (M7824), which blocks PD-L1 while also neutralizing TGF signaling [36, 37]. Both pathways are involved in biliary tract and pancreatic cancers. Even though the results of the early phases are promising [37], phase II/III trial failure highlights the complexity of dual pathway inhibition and necessitates the selection of patients on the basis of tumor microenvironmental characteristics [38].

Next-generation structures such as ABL503 use conditional activation [39]. This means that they only engage 4-1BB upon binding PD-L1 [39]. The goal is to increase antitumor activity in HCC while minimizing systemic toxicity. The clinical setbacks of bintrafusp alfa highlight that the mechanistic rationale alone is insufficient. Success likely requires precise spatiotemporal control of dual-pathway modulation, supported by biomarker-based patient selection and safety-oriented engineering. As a result, recently developed designs contain more protease-cleavable linkers and pH-sensitive domains to enhance stromal penetration while preventing actions toward other GI tumors [40]. The unique immunomodulatory properties of IFPs complement the cytotoxic characteristics of ADCs, laying the foundation for their combined application.

Comparative ADC/IFP pharmacology

ADCs and immune-engaging biologics, including IFP formats and immune checkpoint inhibitors (ICIs), offer complementary paradigms for GI cancers: ADCs deliver targeted cytotoxic pressure, whereas immune-directed biologics aim to localize immune activation or redirect effectors within suppressive TMEs. Their integration is mechanistically appealing, particularly in immune-cold settings, such as PDAC and MSS CRC [41, 42]. However, the extent to which ADC-induced immunogenic stress, including ICD-associated features, translates into clinically meaningful immune priming in GI tumors remains context-dependent and incompletely defined. This uncertainty underscores the need to move from static antigen detection toward multidimensional assessment of TME state, drug distribution, and on-treatment PD biomarkers when evaluating combinations and sequencing.

For example, the prolonged half-life of ADCs (7–21 days) ensures continuous cytotoxic effects but renders them vulnerable to antigen loss (seen in 10–30% of instances) [43], whereas the immune stimulation of IFPs targets “cold” tumors but poses a CRS hazard (> 50% in BiTEs) [29]. These differences could contribute to cross-resistance in certain settings, for example when immune engagement fails to capitalize on ADC-associated immunogenic stress (Table 1). This constraint may reduce the efficacy of PDAC therapy. ADCs and conditional IFPs have different mechanisms of action. ADCs provide sustained cytotoxic pressure, and conditional IFPs enable adjustable immune redirection. This difference creates a unique opportunity for sequencing in oncology.

Table 1.

Comparative pharmacological properties of ADCs and IFPs

Parameter	ADCs	IFPs
Core Mechanism	Precision cytotoxic delivery via tumor-specific targeting; bystander effect	TME reproGradeamming; immune effector activation/redirection
Primary Target	Tumor-associated antigens (HER2, CLDN18.2, TROP2)	Tumor antigens + immune checkpoints/cytokine receptors/stromal markers
Therapeutic Window Driver	Linker stability; payload potency; antigen specificity	Immune activation localization; conditional targeting
Key Toxicity Profile	On-target/off-tumor effects; myelosuppression; ILD (DXd-based)	CRS; irAEs; neurological toxicity (rare)
Resistance Drivers	Antigen loss; impaired internalization; payload efflux	Immune exhaustion; checkpoint adaptation; stromal exclusion
Optimal Tumor Type	Antigen-positive; heterogeneous; moderate TME immunosuppression	Immunologically “cold”; stroma-rich; checkpoint-positive
Pharmacokinetic Trait	Long half-life (7–21 days); Fc-mediated circulation	Variable (BiTEs: short; Fc-fusions: long); tissue penetration-dependent
Representative clinically advanced agents	T-DXd, Sacituzumab govitecan, Disitamab vedotin (China), CMG901 (phase III ongoing)	Tebentafusp (only approved TCE, uveal melanoma), Cadonilimab (PD-1/CTLA-4 BsAb, GC/GEJ cancer China approved)

ADC, antibody‒drug conjugate; IFP, immunofusion protein; TME, tumor microenvironment; ILD, interstitial lung disease; CRS, cytokine release syndrome; irAEs, immune-related adverse events; BiTEs, bispecific T-cell engagers

Clinical breakthroughs and setbacks

The implementation of ADC and IFP mechanisms in clinical practice is currently transforming the treatment landscape for GI cancers. This journey has been marked not only by success but also by failure, highlighting the importance of target selection, effective molecular design and patient selection.

Gastric and Gastroesophageal Junction (GC/GEJ) Cancer

The treatment protocol for GC/GEJ cancer has undergone a major shift owing to the targeting of HER2 and CLDN18.2, thus moving from a one-size-fits-all approach to a molecularly subdivided approach.

HER2-targeted therapies

The initial success of trastuzumab in HER2-targeted therapy for gastric cancer paved the way for advanced therapies [44]. While trastuzumab has established HER2 as a valid target, there are some limitations associated with the development of ADCs. The failure of the first-generation ADC T-DM1 in the GATSBY trial was attributed to suboptimal molecular design, characterized by a low DAR (3.5), a noncleavable linker, and a payload lacking membrane permeability [16]. This informed the optimized design of T-DXd, which has a high DAR (8), cleavable linker and membrane-permeable DXd payload with potent bystander activity [45]. This design achieved landmark success in DESTINY-Gastric01, with a 51% ORR and 12.5-month OS in pretreated patients [46]. Later trials confirmed efficacy across populations and showed activity in HER2-low tumors, expanding treatable populations beyond the classic HER2-positive definitions [6]. The simultaneous development of bispecific antibodies, such as zanidatamab and zenocutuzumab, offers alternative strategies in this field, but their clinical progression in GC lags behind that of ADCs [47, 48].

Taken together, the transition from trastuzumab to T-DM1 and subsequently to T-DXd in GC/GEJ underscores that the ADC design must be tailored to the antigen heterogeneity and TME of each disease rather than extrapolated from breast cancer experience. High DARs, cleavable linkers and membrane-permeable payloads can overcome intratumoral HER2 variability in GC but introduce new, context-dependent toxicities. Because HER2-directed ADCs are already being evaluated in combination with immune checkpoint blockade in the first-line setting, GC/GEJ provides a practical clinical context in which immune–cytotoxic interactions can be interrogated with serial biomarkers. However, whether these combinations reflect true “priming” as opposed to additive activity remains to be established, and mechanistic readouts (e.g., ICD-associated signatures, antigen presentation, immune infiltration, stromal remodeling) are critical to interpret clinical outcomes beyond response rates.

CLDN18.2-targeted therapies

The expression of the tight junction protein CLDN18.2 is found in 30–50% of GC/GEJ cancers and represents a parallel axis for precision therapy [49]. Zolbetuximab, a chimeric monoclonal antibody (mAb) that exerts effects mainly via antibody-dependent cellular cytotoxicity, was associated with a significant survival advantage when combined with chemotherapy in the phase III SPOTLIGHT and GLOW trials, leading to regulatory approval [50, 51].

The therapeutic landscape of CLDN18.2 is rapidly evolving beyond that of traditional monoclonal antibodies. In early phase trials, ADCs, including CMG901 (AZD0901) and tecotabart vedotin (LM-302), both conjugated to MMAE, had ORRs of 29% and 32.7%, respectively [52, 53]. At the IFP level, CLDN18.2×CD3 T-cell engagers are being engineered to reduce the risk of CRS [30].

The CLDN18.2 story in GC/GEJ has established a non-HER2 molecular subtype in which antibody-based therapy can reshape first-line standards and ADCs can further increase activity in pretreated disease. Moreover, target expression in normal gastric mucosa and early signs of GI toxicity associated with CLDN18.2-ADCs highlight the importance of carefully balancing efficacy with on-target/off-tumor effects. The parallel development of CLDN18.2-directed BiTEs provides an opportunity to test stromal-focused and antigen-focused ADC–TCE or ADC–IFP combinations, linking this axis directly to the “Prime–Amplify” strategies outlined in Sect.  7.

Colorectal cancer (CRC)

CRC ADC development involves a two-pronged strategy involving ultraspecific targeting of molecular subsets plus broader targeting of commonly expressed antigens.

HER2-targeted therapies

HER2 amplification is observed in approximately 2–5% of patients with metastatic CRC [54]. This finding suggests that HER2 amplification defines a distinct subset of mCRC that is intrinsically resistant to anti-EGFR therapy [55]. The DESTINY-CRC01 trial evaluated T-DXd in a cohort with a median of three prior lines of therapy; the agent achieved an ORR of 45.3% and a median OS of 15.5 months [56]. An important lesson learned through this experience is the increased value of biomarker selection. The ORR was far superior in those with HER2 amplification (57.5% ORR) versus those with HER2 overexpression only (20.0% ORR), strongly underlining the necessity of in situ hybridization in this disease [56]. The DESTINY-CRC02 trial validated these findings, with an ORR of 37.1% throughout the study population, but 41% (34/83) of patients in the 5·4 mg/kg group and 49% (19/39) in the 6·4 mg/kg group experienced grade 3 or worse drug-related treatment-emergent adverse events [57].

In contrast to those in GC, HER2-targeted ADCs in mCRC benefit only a small, biomarker-enriched subset and are constrained by a higher incidence of drug-related treatment adverse events. The requirement for ISH-confirmed HER2 amplification and the narrower therapeutic window of T-DXd in CRC emphasize that organ-specific biology strongly shapes ADC risk–benefit profiles. These differences argue for CRC-specific optimization of the dose, schedule and combination partners and motivate cross-cancer comparisons and toxicity considerations that follow.

TROP2-targeted therapies

Unlike HER2, which defines only a small molecular subset, TROP2 is overexpressed in 80–90% of CRC cases and thus represents a rational target for ADCs in the broader mCRC population [58]. In the phase II basket trial of TROPiCS-03 (NCT03964727), which evaluated sacituzumab govitecan (SG), an ADC delivering the active metabolite of irinotecan SN-38, SG showed modest single-agent activity in refractory patients with unselected mCRC, underscoring the need for better biomarkers beyond TROP2 IHC [59]. Although CRC-specific outcomes from TROPiCS-03 remain limited, these data, together with meta-analytic estimates of SG across solid tumors, support the rationale for further evaluation of TROP2-ADC in mCRC, ideally in randomized trials with chemotherapy comparators [60].

TROP2-directed ADCs exemplify a broad-spectrum precision approach in mCRC, which targets a nearly ubiquitous antigen with potent cytotoxic payloads. However, current evidence is derived mainly from single-arm basket cohorts, and the reliance on SN-38- or DXd-based warheads raises concerns about cross-resistance with prior irinotecan and other topoisomerase I-directed agents. These issues, revisited in the resistance section, highlight the need for randomized CRC-specific trials and for payload-aware sequencing when integrating TROP2-ADCs into treatment algorithms or combinations.

GUCY2C-targeted therapies

GUCY2C is emerging as a CRC-preferring antigen with the potential to improve the therapeutic index of ADCs by reducing off-tumor exposure [21, 61]. Early-phase data with GUCY2C-targeted antibodies (PF-07062199) have revealed that safety and efficacy still evolve, but this target provides an attractive scaffold on which to test more sophisticated payloads and architectures, such as ferroptosis-inducing or PROTAC-ADCs [62]. As such, it serves as a bridge between current clinical programmes and next-generation platforms.

Pancreatobiliary cancers

Pancreatic and biliary malignancies are among the most difficult cancers to treat in GI oncology. The development of the IFP and ADC strategies is being scaled up against dense stroma and extreme immune suppression.

Pancreatic Ductal Adenocarcinoma (PDAC)

Therapeutic delivery in PDAC is severely compromised by the dense desmoplastic stroma. ADC techniques are now cleverly trying to make this barrier the target. For example, an MUC1-C ADC reversed resistance to KRAS inhibitors in preclinical models and reduced tumor growth by 75–85% [63]. In the same manner, a CEACAM6-targeted ADC attached to the novel BET degrader 84-EBET produced strong bystander killing of both cancer cells and CAFs in patient-derived organoid models [64, 65].

Furthermore, an emerging biomarker target in PDAC is claudin-6 (CLDN6). A claudin-6-targeted ADC with a DXd payload has shown promising preclinical activity, leading to tumor regression in 50% of preclinical PDAC models, potentially paving the way for future PDAC therapy [66].

Across PDAC models, ADCs against MUC1-C, CEACAM6 and CLDN6 consistently demonstrated that effective delivery of potent payloads to both tumor cells and CAFs can partially overcome the dense, fibrotic stroma and reverse resistance to targeted agents. Although clinical data are still lacking, these preclinical results suggest that certain ADC designs targeting tumor and/or stromal compartments may partially overcome delivery barriers in PDAC and may be compatible with downstream immunotherapy. Importantly, whether such cytotoxic or stromal effects translate into a reproducible immune-accessible window suitable for subsequent IFP-mediated amplification remains to be demonstrated in immune-competent models and biomarker-embedded early-phase clinical studies. PDAC therefore represents a prototypical setting in which the Prime–Amplify hypothesis could be tested prospectively.

Biliary tract cancers (BTCs)

BTCs exhibit significant genomic heterogeneity and have a modest response to ADCs. In the DESTINY-PanTumor02 trial, T-DXd showed activity in HER2-positive subsets, with an ORR of 22%, which increased to 56.3% for those with the strongest (IHC 3+) expression [67].

Another target being investigated is B7-H3, a biomarker in BTCs. The ADC DS-7300, which targets B7-H3, has completed phase I clinical trials in BTC patients. The failure of bintrafusp alfa in biliary tract cancer highlights the difficulties associated with dual-pathway inhibition [38]. The findings highlight that multifunctional agents must have more clarity on the crosstalk of pathways and precision stratification of patients [38]. Together, these experiences underscore both the opportunity and the risk of multifunctional biologics in a genomically heterogeneous disease. Without robust biomarkers of pathway activation and the TME context, rationally designed agents may deliver limited benefit and excess toxicity.

Hepatocellular Carcinoma (HCC)

Unlike pancreatobiliary cancers, which are characterized by dense stroma, HCC presents unique challenges imposed by underlying liver cirrhosis and impaired hepatic metabolism, which significantly narrows the therapeutic window for ADCs. These features pose specific challenges to the development of ADCs and IFPs (Table 2). Therefore, the creation of ADCs and IFPs for HCC must be organ-specific to maximize the therapeutic window for this vulnerable patient population.

Table 2.

Key advances and setbacks of ADCs and IFPs in GI cancers

Cancer Type	Target	Agent (Class)	Key Trial/Phase	Efficacy Highlights	Key Context/Limitation
GC/GEJ cancer	HER2	T-DM1 (ADC)	GATSBY [16]/Phase III	OS: 7.9 vs. 8.6 m (taxane); ORR: 16%	Failed due to suboptimal design (low DAR, nonpermeable payload), underscoring ADC engineering importance.
	HER2	T-DXd (ADC)	DESTINY-Gastric01 [46]/Phase II	ORR: 51%; OS: 12.5 vs. 8.4 m (chemo)	Practice-changing; high DAR & bystander effect overcome heterogeneity. Subsequent studies suggest activity beyond classical HER2-positive definitions.
	CLDN18.2	Zolbetuximab (mAb)	SPOTLIGHT [50]/Phase III	OS: 18.23 vs. 15.54 m (chemo); PFS: 10.61 vs. 8.67 mo	New 1 L standard for CLDN18.2 + disease; establishes non-HER2 subtype-driven paradigm.
	CLDN18.2	CMG901 (ADC)	KYM901 [52]/Phase I (Phase III ongoing as AZD0901)	ORR: 29% (incl. CRs)	Promising activity in pretreated pts; potential option postzolbetuximab.
CRC	HER2	T-DXd (ADC)	DESTINY-CRC01 [56]/Phase II	ORR: 45.3%; OS: 15.5 m (HER2-amplified)	Highly active in molecularly defined subset; HER2 amplification (not IHC) is key predictive biomarker.
	TROP2	Sacituzumab Govitecan (ADC)	TROPiCS-03 [60]/Phase II	OS: 10–12 m	Broadly expressed target in mCRC, but current evidence is derived mainly from basket cohorts and biomarker selection remains suboptimal.
	CEACAM5	Tusamitamab Ravtansine (ADC)	NCT02187848 [68]/Phase I	Limited efficacy (ORR < 15%)	Despite high target prevalence, underscoring the importance of payload and platform optimization
BTC	HER2	T-DXd (ADC)	DESTINY-PanTumor02 [67]/Phase II	ORR: 22% (56.3% in IHC 3+)	Viable for HER2 + subset; highlights need for biomarker identification in heterogeneous disease.
	PD-L1 /TGF-β	Bintrafusp alfa (IFP)	INTR@PID BTC 055 [38]/Phase II	No OS benefit vs. chemotherapy	Pivotal failure; underscores complexity of dual immuno-pathway inhibition and need for precise biomarkers.
PDAC	Various (MUC1-C, CEACAM6)	CEACAM6-ADC/84-EBET (ADC)	Preclinical [64, 65]	Significant tumor Gradeowth inhibition; reversal of KRAS resistance	Focus on stromal/stemness targets & novel payloads (e.g., BET deGradeaders) to overcome therapeutic barriers.
HCC	PD-1/CTLA-4	Cadonilimab (BsAb)	COMPASSION-08 [69]/Phase Ib/II	ORR: ~35.7%; OS: 27.1 m (with Lenvatinib)	Competitive with SOC; demonstrates efficacy of dual-checkpoint blockade via a single agent in 1 L.
	GPC3	GPC3xCD3 BiTEs (IFP)	Preclinical [70]	Potent cytotoxicity in vitro/in vivo	Lead modality for premier HCC-specific target; ADC development lags, favoring immune-redirection strategies.
	PD-L1/4-1BB	ABL503 (BsAb)	Preclinical [40].	Reactivation of exhausted CD8 + TILs; synergy with anti-PD-1	“Conditional activation” design mitigates historical liver toxicity of 4-1BB agonists, a key safety advance.

GC/GEJ, gastric/gastroesophageal junction; CRC, colorectal cancer; BTC, biliary tract cancer; mAb, monoclonal antibody; ADC, antibody‒drug conjugate; IFP, immunofusion protein; ORR, objective response rate; OS, overall survival; DAR, drug‒antibody ratio; 1 L, first-line

HCC studies have moved beyond GPC3 targeting to tissue-restricted antigens such as CD44 variant 5 and LGR5, which have shown promising preclinical ADC activity without safety concerns in impaired livers [71, 72]. Both bispecific immunotherapies and ADCs have achieved notable clinical progress in treating HCC [40]. In first-line treatment, cadonilimab targeting PD-1 plus CTLA-4 with lenvatinib achieved an ORR of 35.5–35.7% and a median OS of 27.1 months in the phase Ib/II COMPASSION-08 trial, positioning dual PD-1/CTLA-4 blockade plus antiangiogenic therapy as a competitive first-line option in HCC [69]. Novel designs that could be conditionally activated, such as ABL503 and simultaneous target agents, represent sophisticated ways to overcome the immunosuppressive barrier found in HCC [40].

HCC illustrates how underlying organ dysfunction can dominate the therapeutic landscape: despite promising preclinical data, ADC development remains cautious, whereas bispecific immunotherapies and IO–TKI combinations have already produced competitive first-line outcomes. The coexistence of cirrhosis, altered drug metabolism and immunogenic tumor biology suggests that immunocytotoxic convergence in HCC patients must prioritize liver-sparing strategies and carefully tailored dosing. This organ-specific context connects naturally to the toxicity management and AI-based dose-optimization frameworks outlined.

Comparative efficacy in common GI targets

The organ-specific differences in T-DXd pharmacology between GC/GEJ and CRC (Table 3) argue against pan-tumor ADC deployment and mandate disease-specific optimization.

Table 3.

Organ-specific pharmacology of trastuzumab deruxtecan in gastric versus colorectal cancer reveals divergent biomarker thresholds and toxicity profiles

Parameter	Gastric/Gastroesophageal Junction Cancer	Colorectal Cancer
Target Prevalence	15–22% (HER2-positive); HER2-low populations may further expand the potentially treatable Gradeoup [6, 46]	2–4% (HER2-amplified, ISH+); HER2-low subsets remain under investigation [56]
Key Predictive Biomarker	HER2 protein expression (IHC); HER2-low (IHC 1 + or 2+/ISH−) is clinically actionable	HER2 gene amplification (ISH+) is critical; protein expression alone is insufficient
Efficacy (Later-line)	ORR: 51% (DESTINY-Gastric01, HER2-positive); 38% in HER2-low (DESTINY-Gastric02) [46, 73]	ORR: 45.3% (DESTINY-CRC01, HER2-amplified); median PFS: 6.9 months [56]
Primary Safety Concern	GI toxicity (e.g., nausea, vomiting, diarrhea: 30–40% Gradeade ≥ 3), myelosuppression [6, 46]	ILD: 10.5% incidence (3.2% Gradeade ≥ 3); fatal in 1.6% [56]
Clinical Implication	Expands treatable population via HER2-low classification; potential for first-line combination strategies	Mandates rigorous HER2 ISH testing; proactive ILD monitoring (e.g., baseline/serial chest imaging)

ORR, objective response rate; OS, overall survival; PFS, proGradeession-free survival; IHC, immunohistochemistry; ISH, in situ hybridization; ILD, interstitial lung disease

Cross-trial comparisons of T-DXd across GC/GEJ and CRC further emphasize the organ-specific nature of ADC pharmacology. While HER2-low expression has become a clinically actionable category in GC, enabling expansion and potential first-line combination strategies, HER2-low CRC remains investigational, and HER2 amplification by in situ hybridization (ISH) is still the dominant predictor of benefit. In addition, the incidence and severity of ILD appear to be greater in mCRC than in GC/GEJ cancers, suggesting differences in drug metabolism, lung susceptibility or stromal biology between these diseases. These observations strongly argue against a “one-size-fits-all” approach to ADC deployment in GI oncology and support disease-specific optimization of dosing, partner agents and biomarker cut-offs, particularly when ADCs are combined with ICIs or IFPs.

Major challenges

The advancement of ADCs and IFPs in clinical applications for GI cancers is hindered by three interrelated challenges, including complex resistance, class-specific toxicity and a lack of predictive biomarkers. Maneuvering through the complex landscape is only possible with a detailed cancer type-specific approach and the management of GI cancer.

Drug resistance

Resistance to ADCs and IFPs is a dynamic process, driven by the complex interplay of cell-intrinsic adaptations and microenvironmental stresses. Close examination of these pathways reveals obstacles and therapeutic vulnerabilities that could be exploited therapeutically. The complex network is shown in Fig. 3. This figure illustrates how the multistep ADC life cycle can be disrupted at multiple levels and how immune escape adaptations promote resistance. It also highlights potential strategies that can be employed to counter these resistance mechanisms.

Fig. 3.

Interconnected Networks of Resistance Mechanisms to ADCs and IFPs in GI Cancers. Left: ADC resistance pathways include antigen loss/downregulation, impaired internalization/lysosomal processing, payload efflux/mutation, and stromal exclusion. Right: IFP resistance pathways include MHC-I downregulation, T-cell exhaustion, checkpoint adaptation, and stromal-mediated immune exclusion. Central: Shared TME-driven resistance (desmoplasia, TGF-β signaling, CAF accumulation) drives cross-resistance and limits efficacy of monotherapies

Mechanisms of ADC resistance

ADC resistance typically emerges from flaws in the multistep cytotoxic process. A primary mechanism is antigen-dependent escape, wherein tumors evade targeting through heterogeneous antigen loss or profound downregulation. The loss of HER2 occurs in 10 to 30% of T-DXd-refractory GC/GEJ cancers [74], whereas the downregulation of CLDN18.2 can be observed in 15 to 20% of patients after receiving CLDN18.2-targeted therapy [75]. This loss of antigen expression often leads to a 25–40% decrease in the ORR, especially in the HER2-low subset [76], highlighting the importance of continuous surveillance in GI malignancies for enhancing targeted treatment approaches and representing a limitation of single-target therapy.

Another major pathway involves defects in intracellular processing, such as impaired clathrin-mediated endocytosis or reduced lysosomal enzyme (e.g., cathepsin B) activity, which hinder the release of the payload [77, 78]. Tumor cells develop resistance to payloads through efflux pumps such as MDR1 for MMAE or through mutations in molecular targets such as topoisomerase I for DXd-based ADCs [79, 80]. The resistance pathways illustrated in Fig. 3 are rarely isolated events; rather, they function as interconnected networks that reinforce therapeutic evasion. In GI cancers with high TROP2 expression, ADCs exploit the bystander effect to address tumor diversity. However, they encounter hurdles in combating resistance mediated by MDR1, leading to a notable increase in the IC50 [80]. This resistance complements IFP resistance, such as T-cell exhaustion, and may exacerbate fibrosis in PDAC by reshaping the TME [81]. These intricate interactions highlight the constraints of monotherapy and underscore the need for BsADCs or AI models to predict and address resistance evolution effectively. In addition to exerting a cell-autonomous mechanism, the TME exerts physical and biochemical constraints. The dense stroma of GI cancer limits access to antibodies, which are further inhibited by TGF-β from CAFs [82]. Overall, counteracting these resistance mechanisms requires a disease-specific approach; for instance, optimizing bystander-capable payloads may be prioritized for heterogeneous CRCs, whereas stroma-penetrating or stroma-modulating designs warrant evaluation in PDAC.

Mechanisms of IFP resistance

Resistance to immune checkpoint therapies stems from different immune evasion mechanisms, especially in immunologically “cold” tumors. Key resistance mechanisms include disruption of antigen presentation due to downregulation of MHC class I or mutations in antigen-processing components (e.g., β2-microglobulin), which shield tumors from T-cell recognition [83]. Tumors also demonstrate checkpoint adaptation. For example, blocking PD-L1 leads to the upregulation of other pathways, such as TIM-3 or LAG-3. This is observed with bintrafusp alfa in biliary tract cancer patients [84]. This includes T-cell exhaustion because of chronic exposure to antigens in MSS CRC by cibisatamab [29] and the physical exclusion of T cells by the dense stromal architecture of tumors such as PDAC [85]. Recent studies have shown that T-cell exhaustion occurs in 40–60% of MSS CRC patients, leading to a 3–6 month reduction in PFS [86]. These resistance features motivate the rational testing of combinations and sequencing in which cytotoxic debulking and/or stromal remodeling might improve immune accessibility, while immune engagers may help control residual antigen-low clones. Nonetheless, the clinical benefit is not guaranteed and may be constrained by toxicity, timing, and the possibility of compensatory immunosuppressive rebound. Therefore, sequence-explicit studies with on-treatment PD biomarkers are essential.

Resistance evolution modeling

The resistance mechanisms outlined above seldom occur in isolation; rather, they form interconnected networks that evolve under therapeutic pressure. The remodeling of the TME, which is characterized by advanced fibrosis in the stroma, irregular formation of blood vessels, and the accumulation of immune-suppressing cytokines, plays a key role in this system, both hindering the infiltration of IFPs and preventing the entry of active T cells [87]. In PDAC, for example, CAF-driven deposition of collagen and hyaluronan not only reduces antibody access to tumor cells but also reinforces myeloid-dominant, T-cell-poor niches that blunt the impact of IFPs and ICIs [88]. Similarly, although less extreme, patterns are observed in subsets of MSS CRC and BTC [89, 90]. These stromal and cytokine “nodes” thus represent shared vulnerabilities amenable to intervention. Strategies combining stromal-targeted ADCs, IFPs armed with hyaluronidase, or antiangiogenic agents could potentially delay or mitigate the emergence of cross-resistance to ADCs and IFPs. It will be crucial to include longitudinal biopsies, ctDNA, and imaging biomarkers in clinical trial protocols to monitor these evolving adaptations effectively and adjust treatment strategies preemptively to prevent the emergence of irreversible cross-resistance. Conceptually, partially non-overlapping resistance drivers (e.g., antigen loss versus stromal/immune exclusion) raise the possibility that appropriately designed combinations or sequences could constrain tumor escape through dual pressures. However, the viability of such a “pincer” effect without unacceptable toxicity, and the comparative efficacy of ADC→IFP sequencing versus concurrent or reverse dosing, remain open, testable questions in the field of GI oncology.

Toxicity management

The side effects that occur due to ADCs and IFPs are directly related to their modes of action, which require specific management.

ADC-associated toxicity

ADCs toxicity profiles are consistent with their three-component architecture. On-target/off-tumor effects occur when target antigens have baseline expression on healthy tissues, leading to GI adverse events with CLDN18.2-directed ADCs and cardiotoxicity with HER2-targeted agents [52, 91]. Owing to linker instability causing untimely release of the payload, off-target toxicities were identified in 50–70% of patients receiving MMAE-conjugated ADCs with neutropenia and others presenting with a range of GI side effects [73]. Another concern is idiosyncratic reactions, particularly ILD, which are observed in 10–15% of patients on DXd-containing ADCs [73]. The higher rates of CRC than of GC suggest that either the drugs have cancer-specific differences in metabolism or that there are differences in stromal biology. However, further validation of the underlying mechanisms is still needed [5, 67]. This research calls for constant refinement of linker stability and payload to improve the therapeutic window.

IFP-associated toxicity

The toxicities caused by IFPs are mostly from overactivation of the immune system, which gives rise to CRS along with immune-related adverse events (irAEs) resembling those associated with ICIs [92]. Moreover, neurologic events also occur less frequently [93]. To reduce risks, molecular designs are evolving toward conditionally activated constructs. One example is the pH-sensitive BiTEs, which limit the engagement of CD3 in the acidic TME, thus localizing immune activation [94]. However, the long-term safety profiles of GI cancer treatment platforms are still an ongoing question.

Biomarker development

Biomarker development in this field remains challenging because it still relies heavily on fixed, single-parameter markers. Next-generation biomarkers should capture not only target expression but also the dynamic relationships among tumor cells, the surrounding TME and drug distribution.

Biomarkers for ADCs

Moving beyond antigen expression is critical. Methods to assess ADC suitability are moving beyond static antigen expression and increasingly incorporate quantitative measures of antigen density, internalization efficiency, and payload sensitivity. Emerging powerful tools include liquid biopsies; circulating tumor DNA (ctDNA) can detect emergent resistance (e.g., HER2 loss), and exosomal HER2 mRNA levels correlate with the T-DXd response [95]. Novel biomarkers consist of spatial transcriptomic signatures of antigen distribution and ctDNA methylation patterns that are predictive of payload sensitivity [96, 97]. Spatial transcriptomics profiling of resistant GC revealed distinct resistance programs across HER2-targeted modalities: approximately one-third of trastuzumab-resistant tumors showed activation of the endoplasmic reticulum–associated degradation (ERAD) pathway, whereas T-DXd-resistant tumors were characterized by human leukocyte antigen (HLA) loss and upregulation of oxidative phosphorylation, underscoring the value of longitudinal molecular monitoring to detect emergent, therapy-specific resistance mechanisms [74]. However, multicenter validation is necessary to reduce the false-positive rate.

Biomarkers for IFPs

Biomarker strategies for IFPs must target a dynamic immune state. The density and spatial distribution of CD8 + T cells, coexpression patterns of immune checkpoints and imaging of stromal contents provide more information than does static PD-L1 IHC [64]. Real-time assessment of the response can be performed via liquid biopsies that track immune cell subsets or cytokine levels [65]. Spatial transcriptomics is a breakthrough method that allows visualization of T-cell infiltration hotspots and checkpoint coexpression in the TME [66].

Biomarker decision tree for ADC/IFP patient stratification

Figure 4 illustrates a systematic method for choosing patients for ADC and IFP therapies. The target antigens HER2, CLDN18.2, and TROP2, as well as PD-L1 status, are assessed by IHC and in situ hybridization. Spatial transcriptomics is used to map the immune landscape and radiomics for the stromal assay in the subsequent TME characterization. The framework employs longitudinal monitoring via liquid biopsies to track antigen loss via ctDNA and immune dynamics through cytokine and T-cell profiling.

Fig. 4.

Prevalence and Distribution of Key Tumor-Associated Antigens as Therapeutic Targets Across GI Cancer Types. Anatomical summary and expression frequency of clinically actionable targets (HER2, CLDN18.2, TROP2, GPC3) in gastric/gastroesophageal junction, colorectal, pancreatic, biliary tract, and hepatocellular carcinomas. Target prevalence is indicated as percentage ranges, guiding biomarker-driven patient selection for ADC and IFP therapies

Lessons from Failed Trials

The bintrafusp alfa programme in BTCs provides a cautionary example of the complexity of dual-pathway inhibition. In the phase 2 INTR@PID BTC 047 study, bintrafusp alfa monotherapy produced durable responses in a small subset of pretreated BTC patients but did not meet its prespecified primary endpoint [98]. In the subsequent randomized phase 2/3 INTR@PID BTC 055 trial (NCT04066491), bintrafusp alfa plus gemcitabine–cisplatin failed to improve overall survival over gemcitabine–cisplatin alone (median OS 11.5 vs. 11.5 months; HR 1.23, 95% CI 0.66–2.28) and was associated with higher rates of bleeding events and early deaths [38]. These results suggest that, without biomarkers identifying TGF-β–addicted TMEs, the broad application of PD-L1/TGF-β dual blockade may reduce benefits while increasing toxicity.

TME Heterogeneity and pathway context dependency

The bintrafusp alfa experience highlights a crucial distinction that was overlooked in earlier designs. The presence of a target does not necessarily indicate pathway addiction. Simply inhibiting an overexpressed protein is futile if the signaling network is not the main driver of tumor survival in that particular microenvironment [38]. Consequently, the field must pivot from binary biomarkers to functional pathway profiling. Future trials should utilize multiomics to identify specific TME archetypes, such as TGF-β-driven immune exclusion, ensuring that multifunctional agents are deployed only in populations dependent on the relevant pathway rather than being diluted across broad, unselected cohorts.

Target plasticity and adaptive resistance

Secondary resistance often develops in patients receiving ADC therapy, revealing a significant temporal shortcoming in current treatment strategies [99]. Traditional protocols adhere to a single targeted agent until clear radiological progression is observed. However, this approach can unintentionally create an opportunity for antigen-low subclones to proliferate unrestrainedly when exposed to specific payload pressures [100]. Future clinical strategies must move beyond the reactive switching of therapies and adopt a proactive approach to antigen cycling. This involves moving beyond basic monitoring to implement adaptive trial designs. Early molecular signs of target plasticity, identified through liquid biopsy, can prompt an immediate transition to a non-cross-resistant ADC or IFP. This approach can effectively counter clonal evolution before clinical failure occurs.

Immune editing and checkpoint redundancy

The low response rate of MSS CRC to cibisatamab is likely due to the tumor’s capacity for immune editing, which involves downregulating targeted antigens or upregulating other checkpoint pathways (TIM-3, LAG-3) to escape T-cell attack [101]. Logical combinations of IFPs with agents targeting complementary immune checkpoints or stromal barriers are needed. Quantitative analysis revealed that bintrafusp alfa failure did not increase OS (HR 1.1), as more than 70% of resistance rates were linked to TGF-β pathway diversity, emphasizing the need for initial biomarker screening to increase treatment efficacy [38].

Prioritizing translation in clinical practice involves profiling the TME and detecting targets for identifying drug-sensitive contexts. Therapeutics can be enhanced with pH-sensitive conditional activation, and adaptive trial designs should be adjusted to evolve biomarkers. Patient-derived organoids provide crucial preclinical evidence for combination strategies. Together, these strategies improve precision targeting in GI cancer.

Next-generation platforms and rational sequencing strategies

The evolution of ADCs and IFPs has gone beyond mere iteration, embracing a synthesis of precision engineering, rational combination, therapy sequencing, and AI-integrated multiomics.

Next-generation adc design and engineering

ADCs are accelerating their development beyond the standard formats to very customized architectures. These are meant to overcome the particular biological barriers of GI cancers. The aim of these platforms is to overcome limitations in stability, penetration, and heterogeneous target expression, which are best encapsulated in Fig. 5.

Fig. 5.

Next-Generation ADC Platforms Overcoming Limitations of Conventional ADCs. Left: Limitations of conventional ADCs including heterogeneous conjugation, unstable linkers, low drug-antibody ratio (DAR), poor stromal penetration, and off-target toxicity. Right: Innovative solutions including bispecific ADCs (dual-antigen targeting), antibody–DNA nanoconjugates, bottlebrush prodrug conjugates (ultra-high DAR), and stroma-penetrating designs to enhance efficacy, safety, and access to stroma-rich GI tumors

Precision engineering of components

The optimization of ADCs uses precise engineering techniques to address the unique pathogenic features of GI cancer subtypes. The linker and payload selection should be tailored to the TME properties. For example, enzyme-cleavable linkers that are cleaved by stromal proteases can be used. Furthermore, membrane-permeable payloads can be selected to maximize bystander effects in heterogeneous tumors [102]. The development of payloads is increasingly directed against vulnerabilities in the tumors themselves. BET degraders target the resistant stroma of PDAC [64]. Ferroptosis inducers exploit metabolic dependencies in models of CRC [21].

Novel engineering technologies are expanding targeted precision cancer therapy. PROTAC-ADCs enable snowballing efficacy by targeting relevant antigens and catalytically degrading superfluous proteins [103]. Bottlebrush prodrug conjugates of antibodies achieve ultrahigh drug-to-antibody ratios to penetrate barriers in stroma-rich tumors such as pancreatic cancer [104]. BsADCs counteract antigen heterogeneity by engaging with two antigens, while more advanced designs also incorporate stroma-binding moieties to facilitate drug delivery [21]. Antibody‒DNA nanoconjugates have modular frameworks through which agents can be loaded to induce apoptosis and necroptosis, which is useful against antigen-heterogeneous malignancies such as CRC [105]. These platforms address ongoing challenges in treating GI cancers.

Radiolabeled ADCs

Pretargeted radioimmunotherapy, which targets glycoprotein A33, accurately localizes to CRC liver metastases [106]. Moreover, the [¹⁷⁷Lu]-labeled anti-CLDN18.2 antibody is a highly effective cytotoxic payload-independent chemical for advanced GC [107]. These radioimmunotherapies act as a bridge or salvage agent for ADC resistance, although their combination with other modalities needs further research. Early clinical data from [¹⁷⁷Lu]-CLDN18.2 radioimmunotherapy in GC patients revealed a 60% ORR with manageable myelosuppression [108].

Bispecific Immunomodulatory-ADCs

Immuno-cytotoxic convergence is now being engineered into single molecular entities. One example is the bispecific ADC JSKN027, which targets PD-L1 and VEGFR2 while delivering a cytotoxic payload. Its investigational new drug (IND) application has been accepted by China’s National Medical Products Administration (NMPA) [109], representing a critical step toward clinical testing. This integrated design aims to enhance tumor selectivity and reduce off-target toxicity by targeting commonly co-expressed antigens. Although these constructs test convergence directly at the molecular level, they complement, rather than replace, the need for optimized combinatorial regimens that use distinct ADC and IFP agents. The following sections therefore focus on rational combination strategies that operationalize the Prime–Amplify paradigm through the deliberate sequencing of separate therapeutics.

Advanced IFP platforms and delivery systems

The subsequent generation of immunotherapy focused on ensuring spatial specificity to maximize antitumor immunity while restricting systemic toxicity in IFPs. This draws directly from previous clinical failures.

TME-responsive activation

IFPs of the next generation are capable of precision targeting owing to TME-responsive activation. Examples of new constructs, including the combination of MMP-cleavable linkers with pH-sensitive domains, can trigger selective IL-12 in PDAC [110]. This dual mechanism has resulted in significant tumor shrinkage in preclinical models and minimal systemic toxicity, making it a promising strategy for improving therapeutic efficacy and safety in GI cancers.

IFP-CAR combination therapies

The strategic combination of IFPs with chimeric antigen receptor (CAR) therapies could help overcome the dual challenges of stromal exclusion and immune exhaustion in immunologically cold tumors. Recent studies have shown that FAP-targeted IL-2v IFPs can significantly increase the ability of CAR-T cells to infiltrate and persist in PDAC mice [111]. This increases tumor regression rates from 20% to 65% through the smart remodeling of the stroma and the supply of local cytokines [111]. Additionally, αPD-L1-IL-12-engineered CAR-T cells enhance T-cell migration and infiltration into tumors, directing IFN-γ release, modifying the TME, and promoting antitumor reactions while minimizing systemic inflammation-related adverse effects [112].

TME-modulating nanoparticles

Utilizing complete nanoparticle systems rather than binding antigens directly can alter the immunosuppressive TME. Targeting valosin-containing protein (VCP) or disrupting neutrophil extracellular traps (NETs) with nanoparticle platforms efficiently transforms immunologically inactive CRC into active states, facilitating the effectiveness of immune checkpoint blockade (ICB) or ICD [113].

Rational combination therapies

The distinct yet potentially complementary modes of action of ADCs and IFPs provide a rationale for combination development in GI cancers (Fig. 6). ADCs debulk antigen-expressing tumor cells and, depending on payload class, target compartment, and host context, ADC treatment may be accompanied by immunogenic stress signals consistent with ICD and altered antigen availability. Meanwhile, IFP formats (e.g., cytokine–antibody fusions or T-cell engagers) aim to amplify or redirect antitumor immunity but are often constrained by immune exclusion, myeloid suppression, and toxicity. In stroma-rich, immune-cold settings such as PDAC and MSS CRC, where drug penetration and T-cell access are major barriers, immunocytotoxic convergence is best viewed as a context-dependent, biomarker-testable framework.

Fig. 6.

Prime–Amplify as a testable two-outcome working model for immunocytotoxic convergence between ADCs and IFPs in GI cancers. This schematic summarizes a hypothesis-generating, falsifiable sequencing model rather than an established clinical strategy. Prime phase (ADC): after antigen binding and internalization, ADC payload release can induce tumor cell death and, in selected contexts, trigger immunogenic stress signals consistent with ICD (e.g., calreticulin exposure; ATP/HMGB1 release) and/or stromal remodeling, thereby increasing antigen availability and immune accessibility. These processes are proposed to facilitate dendritic cell (DC) uptake, maturation, and antigen presentation, leading to expansion of tumor-reactive CD8⁺ T cells in lymphoid tissues. Amplify phase (IFP): subsequent administration of immune-engaging IFP formats (e.g., cytokine–antibody fusions or T-cell engagers) is hypothesized to amplify intratumoral effector activity by increasing T/NK infiltration and function to eliminate residual antigen-low or heterogeneous tumor clones. Alternative (failure) path: priming may fail if ICD-associated pharmacodynamic (PD) signatures are absent, if suppressive stromal/myeloid programs persist or rebound (e.g., MDSC/TAM enrichment, CAF/TGF-β dominance), or if treatment induces lymphodepletion/exhaustion, in which case subsequent IFP treatment may yield limited benefit and/or increased toxicity (e.g., CRS or overlapping inflammatory adverse events). The model therefore generates testable predictions and motivates sequence-explicit studies incorporating serial PD biomarkers (ICD/DAMPs, antigen presentation, immune infiltration/exhaustion, stromal remodeling) to validate—or refute—the Prime–Amplify concept in GI malignancies

On this basis, several classes of combinations are being explored in GI oncology, including ADCs with ICIs, ADCs with antiangiogenic agents, IFPs with stromal-modulating drugs, and sequential ADC–IFP “Prime–Amplify” strategies, which are detailed in the following sections.

ADCs with ICIs

Early-phase clinical studies now provide preliminary evidence that combining HER2-targeted ADCs with ICIs can achieve greater antitumor activity than either modality alone in patients with HER2-expressing GI cancers. In a multicenter, open-label phase 1 trial (NCT04280341), disitamab vedotin (RC48) plus the PD-1 inhibitor toripalimab achieved a confirmed ORR of 50% (11/22; 95% CI 28–72) in patients with HER2-expressing GC/GEJ cancer treated at the recommended phase 2 dose, with a manageable safety profile and no dose-limiting toxicities [114]. Notably, clinical benefit was observed in both the HER2-positive subgroup and the HER2-low subgroup, with ORRs of 56% and 46%, respectively, suggesting that ADC-mediated bystander killing and PD-1-mediated immune reactivation can benefit patients in combination beyond classical HER2-overexpressing disease [114]. A recent multicenter retrospective study compared RC48 plus PD-1 inhibitors with RC48 monotherapy in pretreated patients with HER2-overexpressing (IHC 2+/3+) advanced GC/GEJ cancer [115]. After propensity score matching, the combination group achieved a significantly greater ORR and longer PFS (5.3 vs. 3.8 months; hazard ratio (HR) = 0.51, 95% CI 0.31–0.85; p = 0.010) and OS (10.0 vs. 6.8 months; HR = 0.45, 95% CI 0.27–0.77; p = 0.003) than did the monotherapy group, without a disproportionate increase in grade 3–4 treatment-related adverse events [115]. These preliminary findings appear promising but are constrained by small cohorts, nonrandomized study designs, and enrollment of previously treated patients. Further validation in larger and more definitive first-line studies remains essential (Table 4).

Table 4.

ADC–immunotherapy combination trials with reported clinical data in GI cancers

Trial ID/Phase	ADC (Target; Payload)	Combination Partner(s)	Tumor Type / Line	N (Evaluable)	Key Efficacy Data	Key Safety Data	Ref
NCT05980481, phase II/III (phase II data reported)	Disitamab vedotin (RC48) (HER2; MMAE)	Toripalimab + CAPOX with or without trastuzumab	HER2‑expressing (pos/low) la/m GC/GEJ; 1 L	HER2‑pos: 51; HER2‑low: 93	HER2‑pos: ORR 82.4% (EG2) vs. 68.8% (CG1); HER2‑low (Stage 1): ORR 70.8% (EG1) vs. 47.8% (CG1), PFS HR 0.67; HER2‑low (Stage 2): ORR 76.9% (EG2)	Any‑Gradeade / Grade ≥ 3 TRAEs: 100% / 73.3–100%; dose‑reduced CAPOX improved tolerability	ASCO 2025 LBA4012 [116]
NCT04379596 (DESTINY‑Gastric03), phase Ib/II (multi‑cohort)	T‑DXd (HER2; DXd; 5.4 or 6.4 mg/kg)	Pembrolizumab + FP	HER2‑positive unresectable/metastatic GC/GEJ/esophageal adenocarcinoma; 1 L	Arm D: 43; Arm F: 32	Time‑matched: Arm F (5.4 mg/kg): ORR 59.4%, mPFS 5.8 mo; Arm D (6.4 mg/kg): ORR 41.9%, mPFS 6.4 mo	Higher toxicity at 6.4 mg/kg; dose de‑escalation performed; mOS not reached	ASCO GI 2025 [117]
NCT05586061 (RCTS), phase II (single‑arm, multicenter)	Disitamab vedotin (RC48) (HER2; MMAE)	Tislelizumab + S‑1	HER2‑overexpressing la/m GC/GEJ; 1 L	57 (ITT); 55 (PP)	ORR 89.4% (ITT); mPFS 12.7 mo; 18‑mo OS 72.7%; HER2‑pos: ORR 92.1%; CPS ≥ 1: ORR 92.3%	Grade 3–4 TRAEs: 63.2%; neutropenia, fatigue, leukopenia; no treatment‑related deaths	ASCO 2025 [118]
NCT05502393, phase I (dose‑escalation/expansion)	JS107 (CLDN18.2; MMAE)	Toripalimab + XELOX	CLDN18.2‑positive la/m GC/GEJA; 1 L	45 (CLDN18.2+); 30 (high)	CLDN18.2‑high: ORR 86.7%, DCR 100%, mPFS 11.14 mo; median follow‑up 7.1 mo	≥ Grade 3 TRAEs: 55.6–58.3%; neutropenia, thrombocytopenia; 1 DLT at 3 mg/kg	ESMO Asia 2025 LBA5 [119]

For NCT05980481: EG1 = DV + toripalimab + CAPOX; EG2 (HER2‑pos) = DV + toripalimab + trastuzumab; EG2 (HER2‑low Stage 2) = DV + toripalimab + reduced‑dose CAPOX; EG3 = reduced‑dose DV + toripalimab + reduced‑dose CAPOX; CG1 (HER2‑pos) = toripalimab + trastuzumab + CAPOX; CG1 (HER2‑low) = toripalimab + CAPOX; CG2 = toripalimab + CAPOX. Data cut‑off: February 7, 2025

For DESTINY‑Gastric03: Time‑matched analysis between Arm D (DCO October 27, 2022) and Arm F (DCO May 6, 2024); median follow‑up 4.1 and 4.6 months, respectively; results remain immature

For RCTS (NCT05586061): 71.9% HER2 IHC 3+; median follow‑up 11.8 months

For JS107 (NCT05502393): CLDN18.2‑high = ≥ 40% tumor cells at ≥ 2 + staining; dose expansion used JS107 2 mg/kg with 75% or 100% XELOX

ADC, antibody–drug conjugate; IO, immunotherapy; GC/GEJ, gastric/gastroesophageal junction; GEJA, gastroesophageal junction adenocarcinoma; la/m, locally advanced or metastatic; HER2‑pos, HER2-positive; 1 L, first‑line; DV, disitamab vedotin; Tor, toripalimab; Tra, trastuzumab; CAPOX/XELOX, capecitabine + oxaliplatin; FP, fluoropyrimidine + platinum; ORR, objective response rate; DCR, disease control rate; mPFS, median proGradeession‑free survival; mOS, median overall survival; HR, hazard ratio; CI, confidence interval; DLT, dose‑limiting toxicity; TRAE, treatment‑related adverse event; ITT, intent‑to‑treat; PP, per‑protocol; CPS, combined positive score; EG, experimental Gradeoup; CG, control Gradeoup; Grade, Gradeade; DCO, data cut‑off

To contextualize the currently available clinical evidence, Table 4 summarizes ADC–immunotherapy combination studies in GI cancers with reportable efficacy and/or safety data, which currently center on HER2- and CLDN18.2-directed strategies and include both MMAE- and DXd-based ADC platforms. These include two RC48-based regimens (NCT05980481, first-line randomized; and the RCTS trial, NCT05586061 [118]), the DESTINY-Gastric03 T-DXd triplet (NCT04379596), and early conference-reported data for a CLDN18.2-targeted ADC–IO combination (JS107 plus toripalimab plus XELOX [119]). Across these studies, three recurring themes emerge. Response rates across these investigations have been promising, dose optimization plays a critical role in achieving a favorable balance between efficacy and safety, and biomarker-stratified subgroups reliably identify patients most likely to derive clinical benefit.

Notably, all ADC–ICI combinations with reported clinical data remain confined to upper GI cancers (Table 4). In contrast, ADC–immunotherapy combinations in PDAC, MSS CRC, BTC and HCC have not yet reported clinical efficacy data, although early-phase programs such as EBC-129 (NCT05701527, CEACAM5/6-ADC plus pembrolizumab) and PYX-201 (NCT06795412, EDB-fibronectin-ADC plus pembrolizumab) are underway. This organ-specific imbalance reflects the earlier developmental stage of ADCs in stroma-rich, immune-excluded GI malignancies.

From a translational perspective, GC/GEJ now provides a proof-of-concept template suggesting that ADC–IO combinations spanning HER2 and CLDN18.2 targets can achieve ORRs ranging from approximately 60% to 89% in selected dosing regimens within biomarker-selected populations. Extending this paradigm to PDAC, MSS CRC, BTC and HCC, where no ADC–IO combination has yet reported efficacy data, represents a key priority and motivates the biomarker-driven sequencing strategies discussed below.

ADCs with Anti-Angiogenic Agents

Direct clinical data on the combination of ADCs with antiangiogenic agents in GI cancers are currently lacking, but several preclinical studies have illustrated how antiangiogenic and cytotoxic mechanisms can be mechanistically integrated within an ADC framework relevant to HCC and other solid tumors. Li and colleagues designed a bevacizumab-based ADC by conjugating the anti-VEGF mAb to the microtubule inhibitor MMAE via a cathepsin-cleavable Val-Cit-PABC linker [120]. In vitro, bevacizumab vedotin showed potent antiproliferative and proapoptotic activity against multiple tumor cell lines, including HepG2 HCC cells, and retained the canonical antiangiogenic effects of bevacizumab by blocking VEGF/VEGFR signaling and inhibiting endothelial tube formation [120]. These data support a “built-in” dual-function concept in which an anti-VEGF antibody serves both as the targeting moiety and as an antiangiogenic effector, in which the conjugated payload provides high-potency cytotoxicity [120].

In parallel, stromal and vascular targeting can also be achieved by directing ADCs against extracellular matrix (ECM) components associated with angiogenesis. An ADC specific for the extra domain-B splice variant of fibronectin (EDB + FN), a stromal ECM protein linked to tumor growth and neovascularization, demonstrated robust tumor growth inhibition and complete regression across multiple xenograft and syngeneic models, including pancreatic cancer PDX models [121]. Although not combined with a separate antiangiogenic drug, EDB-ADC localized to the tumor stroma and indirectly remodeled the vasculature and immune infiltration, providing a functional analog of antiangiogenic ADCs. Taken together, these studies suggest that in GI malignancies with dense stroma and abnormal vasculature (e.g., HCC and PDAC), ADCs incorporating anti-VEGF or ECM-targeting backbones could, in principle, recapitulate the vascular normalization paradigm while simultaneously delivering potent cytotoxic agents. Prospective GI-specific trials combining such constructs with established antiangiogenic agents have not yet been reported, and an important translation gap remains.

IFPs with Stroma-Modulating Agents

Experimental work over the past five years has begun to validate the strategy of combining IFPs with agents that actively degrade or remodel the fibrotic stroma in GI cancers [122]. A key example is the development of antibody–enzyme (AbEn) fusion molecules that couple stromal targeting with hyaluronidase-mediated ECM degradation. Zhou et al. engineered a trispecific AbEn, TAVO423, that simultaneously recognizes FAP and leucine-rich repeat-containing 15 (LRRC15) on CAFs and delivers a recombinant human hyaluronidase (HYAL) payload [123]. In HA-rich CRC models, TAVO423 induced deeper intratumoral hyaluronan depletion and more pronounced tumor growth inhibition than untargeted hyaluronidase did, confirming that stromal-anchored enzymatic degradation can overcome the diffusional barrier imposed by a dense ECM.

Crucially, TAVO423 was evaluated in combination with multiple immune and targeted modalities that are directly relevant to GI oncology. In an HA-rich colorectal model, adding TAVO423 to 5-fluorouracil, an anti-PD-L1 mAb, a PD-L1×CD3 T-cell engager, or a CD318-targeting ADC increased tumor growth inhibition (TGI) to 49–67%, compared with 1–28% for each monotherapy [123]. In a pancreatic cancer model expressing high HA, TAVO423 combined with a 5T4×CD3 bispecific T-cell engager improved the TGI from 73% to 92% and was associated with marked increases (6–9-fold) in intratumoral CD8⁺ T-cell density [123]. These data provide direct preclinical evidence that hyaluronidase-armed IFPs can dismantle the fibrotic stromal barrier in PDAC and CRC, thereby increasing T-cell infiltration and increasing the efficacy of both checkpoint-targeted immunotherapies and T-cell-engaging immunotherapies. Although no GI-specific clinical trials of TAVO423-like IFPs have been registered, these findings strongly support further translational development of FAP/LRRC15-targeted immunofusion constructs as partners for existing GI immunotherapies.

ADC‒IFP sequencing as a testable Prime–Amplify model (current evidence and constraints)

The “Prime–Amplify” concept hypothesizes a two-step therapeutic sequence. First, initial ADC exposure (“Prime”) debulks the tumor and, in specific contexts, may increase immune accessibility through immunogenic stress signatures consistent with ICD and/or stromal remodeling. Second, these changes may create a temporal window for subsequent IFP administration (“Amplify”) to effectively redirect and potentiate antitumor immune responses. Direct clinical evidence explicitly evaluating ADC→IFP sequencing in GI malignancies remains limited. Current rationale primarily derives from two converging lines of indirect preclinical and translational research: stromal-/ECM-targeted ADCs that sensitize tumors to downstream immunomodulation (typically tested with ICIs), and stroma-modifying immunofusion-like platforms that improve intratumoral distribution and efficacy of cytotoxic or immune-engaging agents.

Preclinical studies—particularly those in GI models—substantiate stromal-/ECM-targeted ADCs as potential priming agents. In TMEs with dense stroma and ECM, ADCs targeting these components can disrupt suppressive niches and reprogram immune cell composition. In PDAC models, a FAP-targeted ADC conjugated to exatecan (FAP-ADC) achieved stromal debulking. When combined with anti-PD-L1, it yielded superior tumor control compared with monotherapy [124]. Mechanistic profiling confirmed increased M1-polarized macrophages, reduced myeloid-derived suppressor cells (MDSCs) and Tregs, and enhanced CD8⁺ T-cell infiltration [124]. These phenotypic changes indicate improved immune accessibility. Similarly, an EDB-fibronectin-targeted ADC (EDB-ADC) augmented CD3⁺ T-cell infiltration and PD-L1 expression. Its combination with anti-PD-L1 enhanced treatment durability in syngeneic models [121]. Although these studies used ICIs instead of classical IFPs, they support a key premise of the model: stromal-targeted cytotoxic interventions can induce quantifiable immune remodeling. This remodeling could plausibly enhance the efficacy of subsequent immune-engaging biologics.

Complementary evidence for the “amplifier/enabler” component comes from stroma-targeted antibody–enzyme fusion strategies. Zhou et al. engineered a trispecific AbEn construct (TAVO423). It co-targets FAP and LRRC15 on CAFs while delivering HYAL. This depletes hyaluronan and mitigates ECM-mediated diffusion barriers [123]. In hyaluronan-rich CRC models, TAVO423 improved tumor growth inhibition when combined with 5-fluorouracil (5-FU), anti-PD-L1, a PD-L1×CD3 T-cell engager, or a CD318-targeting ADC [123]. In a hyaluronan-high pancreatic cancer model, combining TAVO423 with a 5T4×CD3 engager enhanced tumor growth inhibition. It also increased intratumoral CD8⁺ T-cell density by 6–9-fold [123]. These experiments mechanistically demonstrate that stromal barrier disruption can augment immune infiltration. It can also amplify the activity of both immune-engaging biologics and cytotoxic platforms in clinically relevant GI tumor models.

Sequence directionality, timing, and safety remain critical unresolved variables. The Prime–Amplify model does not claim all ADCs will prime immunity. Nor does it assert all IFPs will inevitably amplify responses after ADC exposure. Priming success or failure depends on the ADC’s target compartment (tumor cell versus stroma), payload class, baseline immune contexture, and administration timing. ADC exposure may also trigger compensatory myeloid cell recruitment, lymphodepletion, or inflammatory toxicity. These effects could compromise IFP tolerability or efficacy.

Thus, sequence directionality and timing should be treated as experimentally testable variables, not fixed assumptions. Definitive validation requires GI-relevant preclinical models and early-phase clinical trials. These should incorporate reverse-order controls (IFP→ADC), serial PD biomarkers of immune and stromal remodeling, and vigilant monitoring for overlapping toxicities. Such toxicities include myelosuppression or ILD from ADCs, and CRS or irAEs from immune engagers.

Mature GI-specific clinical trials directly testing ADC→IFP sequencing remain lacking. However, converging preclinical data indicate that stromal-/ECM-targeted cytotoxic interventions can induce immune remodeling consistent with a priming effect [121, 124]. Additionally, stroma-modifying immunofusion-like approaches can facilitate immune infiltration and amplify the activity of immune-engaging modalities [123]. These findings justify the Prime–Amplify concept as a testable working model. They also provide a compelling rationale for implementing biomarker-embedded, sequence-explicit early-phase studies in GI malignancies. The integrated evidence map for immunocytotoxic convergence and the Prime–Amplify hypothesis is shown in Table 5.

Table 5.

Evidence map for the immunocytotoxic convergence framework and Prime–Amplify hypothesis in GI cancers

Claim / model step	What is supported (scope)	Key GI context	Evidence type (data source)	Representative examples cited in this review	Major limitations / What is NOT proven	Evidence Gradeade#
A. ADC cytoreduction can be clinically meaningful in selected GI subsets	High ORR in biomarker-defined GI populations	GC/GEJ; CRC; BTC	GI clinical trials	T-DXd in HER2 + GC/GEJ (DESTINY-Gastric01, ORR 51%) [46]; T-DXd in HER2-amplified mCRC (DESTINY-CRC01, ORR 45.3%) [56]; T-DXd in HER2 + BTC subset (DESTINY-PanTumor02) [67]	Cytoreduction ≠ immune priming; durability limited by antigen loss and TME suppression	A
B. ADC pharmacology and toxicity are organ-specific in GI cancers	Same ADC shows different benefit–risk profiles across GI organs	GC/GEJ vs. CRC	GI clinical cross-trial observation	Different biomarker thresholds and ILD/toxicity patterns for T-DXd in GC vs. CRC (Table 3; DESTINY-Gastric/CRC studies) [46, 56, 73]	Mechanisms underlying organ-specific ILD/GI toxicity not fully established; cross-trial confounding	A/B
C. ADC design features (DAR, linker, payload permeability) affect efficacy and safety	Engineering parameters influence bystander killing and toxicity	GC/GEJ	GI clinical + mechanistic rationale	T-DM1 (low DAR, noncleavable linker, nonpermeable payload) underperforms in GC (GATSBY) [16] vs. T-DXd (high DAR, cleavable linker, permeable DXd payload) success [46]	These features do not directly demonstrate ICD or immune priming; may increase off-tumor toxicity	A/B
D. ADC resistance in GI cancers frequently involves antigen loss/downregulation	Antigen loss contributes to post-ADC resistance and reduced ORR	GC/GEJ; CLDN18.2 + disease	GI clinical/translational	HER2 loss after T-DXd in GC/GEJ (10–30%) [74]; CLDN18.2 downregulation after CLDN18.2-directed therapy (15–20%) [75]	Antigen loss is not the only driver; needs longitudinal sampling; does not imply immune accessibility	A
E. ADCs can be accompanied by immunogenic stress signatures (ICD-like features) in some contexts	ICD-associated markers may occur depending on payload/context	GI cancers (context-dependent)	Indirect GI translational; general mechanistic inference	ICD and DAMPs discussed as mechanistic possibility (CRT exposure, ATP/HMGB1 release); clinical ADC + IO synergies as indirect support	Not uniformly shown across ADCs in GI; ICD markers rarely measured serially in GI trials; causality not established	B/C
F. Stromal-/ECM-targeted cytotoxic interventions can remodel immune context in stroma-rich GI models (primer component)	Immune remodeling toward increased accessibility	PDAC/CRC models	GI-relevant preclinical	FAP-ADC in PDAC models [124]; EDB-ADC [121] in PDAC/CRC models	ICI (not IFP) partner; no sequence-explicit control; model-dependent	B
G. ADC + ICI combinations show early clinical activity in GI cancers (indirect “priming” support)	Combination responses exceed historical monotherapy in some cohorts	GC/GEJ	GI early clinical (nonrandomized)	Disitamab vedotin (RC48) + toripalimab ORR 50% (NCT04280341) [114]; retrospective PSM: improved PFS/OS vs. RC48 alone [115]	Not IFP; nonrandomized/small; cannot prove priming or optimal sequencing; biomarker PD readouts often absent	B
H. IFP/T-cell engager activity in GI cancers is constrained by immune exclusion and toxicity (CRS), limiting amplification	Demonstrated but modest efficacy and frequent CRS in cold GI tumors	mCRC (MSS)	GI early clinical	CEA×CD3 cibisatamab in mCRC: ORR ~ 14%, CRS ~ 60%, exhaustion features [29]	Shows amplification is not guaranteed; CRS constrains dose/intensity; may require conditioning/priming	A
I. Conditional activation / affinity tuning may improve TCE/IFP therapeutic window	Next-gen designs aim to reduce CRS while retaining activity	GC/PDAC (preclinical/early clinical)	Translational + early clinical proGradeam	AZD5863 CLDN18.2×CD3 tuned CD3 binding to reduce cytokines; clinical eval ongoing (NCT06005493) [30]	Clinical efficacy/safety in GI not yet established; sequencing with ADC not tested	B
J. Stromal barrier removal can amplify efficacy of immune-engaging biologics and cytotoxic platforms (amplifier/enabler component)	ECM deGradeadation increases penetration and CD8 infiltration in GI models	HA-rich CRC; PDAC	GI-relevant preclinical	TAVO423 (FAP/LRRC15 + HYAL) improves TGI with PD-L1×CD3 engager or CD318-ADC; ↑intratumoral CD8 density (6–9×) [123]	Not classical “cytokine–Ab fusion”; not direct ADC→IFP clinical sequencing; timing/sequence variables incompletely explored	B
K. Full ADC→IFP Prime–Amplify sequencing superiority is not yet demonstrated in GI cancers	Proposed as testable hypothesis; requires validation	GI cancers	Evidence gap / hypothesis	Evidence is indirect (ADC→ICI; stromal IFP-like + other agents) supporting components	No mature GI clinical trials explicitly testing ADC→IFP sequence; directionality & interval unknown; toxicity overlap possible	C (as proven strategy); B (as hypothesis supported by converging components)
L. Overlapping toxicity is a practical constraint for convergence	ADC ILD/myelosuppression + IFP CRS/irAEs may stack	GI cancers	Clinical pharmacology reasoning + trial observation	ILD rates reported for DXd-based ADCs [15, 56, 73]; CRS in TCEs [29]	Combination safety and schedule mitigation require prospective study; not predictable from monotherapy alone	B
M. Dynamic biomarkers (ctDNA/spatial/radiomics) can help stratify “primer-ready” and “amplifier-ready” contexts	Composite biomarkers outperform single IHC; can monitor resistance	GC; CRC	GI translational/clinical	ctDNA and tissue/plasma HER2 metrics in DESTINY-CRC01 biomarker analysis [125]; radiomics LRS/MRS in GC for immune context [126]; ctDNA mutation signatures in GC ICI cohorts [95]	Mostly not validated for ADC→IFP sequencing; standardization and prospective validation needed	B
N. AI/multiomics may accelerate design and stratification, but GI ADC–IFP sequencing is not yet an AI-validated use case	Promising direction rather than ready tool	Cross-tumor	Extrapolative methods	ADCNet and structure prediction tools [127, 128]	Training data often non-GI; not specific to ADC–IFP sequencing; risk of over-interpretation	C (implementation readiness); B (directional potential)

# Evidence Gradeade definition

A: direct GI clinical and/or GI-relevant in vivo evidence with therapeutic effect plus mechanistic PD readouts supporting the claim

B: GI-relevant but indirect evidence consistent with the claim yet lacking sequence-explicit validation/reverse-order controls or relying on related modalities

C: extrapolative or hypothesis-level evidence mainly from non-GI contexts, in vitro systems, or conceptual reasoning without GI-specific validation

ADC, antibody–drug conjugate; IFP, immunofusion protein; GI, gastrointestinal; TME, tumor microenvironment; ICD, immunogenic cell death; DAMPs, damage-associated molecular patterns; CRT, calreticulin; DC, dendritic cell; MHC-I, major histocompatibility complex class I; TCR-seq, T-cell receptor sequencing; CRS, cytokine release syndrome; irAEs, immune-related adverse events; ILD, interstitial lung disease; TCE, T-cell engager; BiTE, bispecific T-cell engager; ICI, immune checkpoint inhibitor; CAF, cancer-associated fibroblast; ECM, extracellular matrix; MDSC, myeloid-derived suppressor cell; TAM, tumor-associated macrophage; PDAC, pancreatic ductal adenocarcinoma; MSS, microsatellite-stable; CRC, colorectal cancer; GC/GEJ, gastric/gastroesophageal junction; BTC, biliary tract cancer; HCC, hepatocellular carcinoma; ORR, objective response rate; PFS, proGradeession-free survival; OS, overall survival; IHC, immunohistochemistry; ISH, in situ hybridization; PD, pharmacodynamic(s); IFN, interferon; STING, stimulator of interferon genes; HYAL, hyaluronidase; HA, hyaluronan

Falsifiable predictions and minimal PD criteria. The Prime–Amplify hypothesis yields testable predictions that can be evaluated in GI-relevant models and biomarker-embedded early-phase trials: (i) if ADC treatment does not increase ICD-associated PD signals and antigen-presentation markers (e.g., DAMPs/IFN signatures, DC activation, MHC-I upregulation), subsequent IFP administration is unlikely to improve efficacy and may primarily add toxicity; (ii) if ADC exposure induces lymphodepletion or accelerates T-cell exhaustion, immune amplification by T-cell–redirecting IFPs is expected to be attenuated; and (iii) if stromal or myeloid exclusion programs persist (e.g., CAF/TGF-β dominance, MDSC/TAM enrichment), increased immune infiltration and durable control are unlikely without additional stroma-/myeloid-targeted interventions.

Integration of multiomics and artificial intelligence (AI)

Accurate patient stratification and drug design are crucial for optimizing ADC–IFP combination strategies and sequential regimens, and multiomics profiling and artificial intelligence (AI) technologies can significantly enhance these processes. The rapid expansion of multiomics profiling and AI is reshaping the development and deployment of ADCs, IFPs and ICIs in GI cancers. Contemporary studies integrate genomic, transcriptomic, radiomic and digital pathology features into multivariate models that predict response and resistance with increasing accuracy rather than relying solely on single-parameter biomarkers such as HER2 or PD-L1 immunohistochemistry [80, 129]. Moreover, AI-driven platforms for protein structure prediction and sequence design are accelerating the engineering of ADCs and IFPs [130], whereas multimodal machine learning models support the stratification of patients and the individualization of doses for complex regimens [131]. Although these studies were conducted in the context of ICI monotherapy, the same ctDNA- and radiomics-based frameworks can be readily adapted to guide patient selection and on-treatment monitoring in HER2-positive/low or MSI-H GI cohorts receiving ADC–ICI combinations.

Multiomic Biomarker Discovery

Recent work in advanced GC illustrates how ctDNA-based and imaging-based multiomics can guide ICI use and, by extension, inform patient selection for ADC–ICI strategies [95]. In a group of 47 patients with advanced GC lacking HER2 expression treated with ICI therapy, next-generation sequencing was conducted on ctDNA samples collected at baseline and during treatment [95]. This analysis revealed a total of 658 somatic mutations spanning 203 genes. Certain genetic mutations present at baseline, such as those in MEN1, MLH1, CEBPA, ATR, GNAQ, and FOXL2, are more common in individuals who exhibit a positive response to treatment [95]. Furthermore, specific cooccurring mutations, such as IRS2/CEBPA, IRS2/POLD1, TP53/PIK3CA, and POLD1/CEBPA, were linked to prolonged PFS, whereas mutations in CDKN2A and the CDKN2A/MSH6 comutation were associated with prolonged OS. Changes in ctDNA levels during treatment, particularly elevated levels or frequencies of variant alleles, were correlated with inferior treatment outcomes. Additionally, newly acquired mutations in POLE, FGFR2, and MDC1 potentially indicate the development of resistance to ICIs [95]. These findings demonstrate that liquid biopsy mutation signatures and ctDNA dynamics can function as multigene predictors of ICI benefit in HER2-negative GC patients and could, in principle, also be used to stratify patients for ADC–ICI trials targeting the same cohort.

Complementary multimodal imaging work has come from large radiogenomic efforts. In a cohort of 2,600 patients with GC, the integration of radiomics from computed tomography (CT), immune context from immunohistochemistry, and transcriptomics resulted in the identification of two imaging biomarkers: the lymphoid radiomic score (LRS) and the myeloid radiomic score (MRS). These biomarkers provide noninvasive estimations of the densities of lymphoid and myeloid cells infiltrating the tumor [126]. In 261 patients treated with anti-PD-1 therapy, high LRS and low MRS were independent predictors of better disease-free survival, prolonged OS, and increased ORR, which varied from 10.2% to 53.3% among the four immune subtypes classified by the imaging biomarkers [126]. This work highlights how radiomics-derived immune signatures can serve as surrogate markers of the tumor immune microenvironment and ICI sensitivity and provides a template for incorporating similar imaging-immune scores into ADC–ICI response models in GC.

Multiomics approaches have also been applied intrinsically to the ADC response. An exploratory biomarker analysis was conducted on Cohort A (53 patients) from the DESTINY-CRC01 study in metastatic CRC with HER2 expression [125]. This analysis examined correlations between tissue and plasma HER2 metrics and treatment outcomes in patients receiving T-DXd. Higher baseline values for HER2 immunohistochemistry/in situ hybridization scores, HER2/CEP17 ratios, HER2 H-scores, plasma ERBB2 amplification statuses, and adjusted plasma copy numbers were associated with superior ORRs (45.3%, 95% CI 31.6–59.6), prolonged PFS, and improved OS. Baseline ctDNA profiling also provided evidence of T-DXd activity, even among patients with activating mutations in RAS, PIK3CA, or HER2 [125]. This study demonstrated that multilayered HER2 read-outs across tissue and blood more accurately capture “functional HER2 addiction” to T-DXd than does IHC alone and offers a blueprint for building composite ADC-specific biomarker panels in GI cancers.

Finally, several radiomics-driven signatures have been developed to noninvasively infer MSI status and ICI benefit in patients with GC [132]. CT radiomics, in conjunction with mRNA sequencing, facilitated the creation of a nine-feature signature for the prediction of high microsatellite instability (MSI-H). This signature had AUC values of 0.851 in the training cohort and 0.816 in the validation cohort. Immunotherapy outcomes were successfully stratified into two distinct anti-PD-1 treatment cohorts, which included 132 and 43 patients. Compared with the other groups, the “low-radscore” group presented significantly prolonged PFS, improved OS, and higher complete response (CR) and partial response (PR) rates [132]. Similarly, smaller studies have shown that CT-based radiomic nomograms can predict PD-1 benefit in advanced GC better than can clinical factors alone [133]. Although these models were built around ICIs rather than ADCs, the same multiomic pipelines (ctDNA + radiomics + transcriptomics) can be readily extended to ADC–ICI combinations in HER2-positive, HER2-low or MSI-H GI populations.

AI-Driven drug design

While most multiomic biomarker studies in GI oncology focus on patient selection, parallel advances in AI are reshaping how ADCs and IFPs themselves are designed and optimized [134, 135]. Deep learning frameworks such as ADCNet integrate protein language models for antibody and antigen sequences (e.g., ESM-2) with small-molecule language models for linkers and payloads (e.g., FG-BERT), together with scalar features such as DAR, to predict ADC activity on curated datasets of thousands of conjugates [127]. ADCNet achieves an area under the ROC curve (AUC) of approximately 0.93 for ADC activity classification on curated datasets and is available as a public web server [127]. This enables in silico screening of new antigen–antibody–linker–payload combinations before committing to synthesis. Although current training sets are dominated by breast, lung and hematologic indications, the underlying architecture is agnostic to the tumor site, and HER2- and CLDN18.2-directed ADCs for GC or CRC can, in principle, be optimized via the same platform.

On the structural side, next-generation protein-structure predictors such as AlphaFold-Multimer and AlphaFold3 have markedly improved the ability to model antibody‒antigen complexes, Fc glycosylation and even linker‒payload conjugated domains at near-atomic resolution [128]. These tools are increasingly combined with antibody-specific deep-learning models (e.g., DeepAb, IgFold) to identify aggregation-prone motifs, surface-exposed residues suitable for site-specific conjugation, and paratope regions that must be preserved to maintain HER2 or CLDN18.2 recognition [128]. In practice, this means that AI-assisted in silico pipelines can now propose conjugation sites and linker geometries that minimize steric clashes with the antigen-binding site while preserving Fc effector function and manufacturability, substantially shortening the design–test cycle for ADCs that will eventually be combined with ICIs in GI oncology.

Although there are not yet GI-specific examples of AI-designed ADCs entering clinical trials, the same AI-driven principles are starting to influence IFP engineering. For example, structural models of FAP-targeted immunocytokines or CLDN18.2×CD3 bispecific engagers can be used to simulate how domain orientation and epitope spacing affect synapse formation, cytokine release and bystander activation, guiding the design of next-generation IFPs that cooperate more effectively with priming ADCs via “Prime–Amplify” strategies [35, 136].

AI-powered patient stratification and dose prediction

In addition to drug design, AI is increasingly being deployed to stratify GI cancer patients for complex biologic regimens, including potential ADC–ICI combinations, and to inform dosing strategies.

In GC, multiple groups have independently shown that CT-based radiomic models coupled with machine learning classifiers can noninvasively predict PD-1 benefits and the underlying immune context. One study developed a radiomic nomogram on the basis of data from 87 patients with advanced gastric cancer who were receiving PD-1 inhibitor treatment [137]. The model achieved an area under the curve (AUC) of 0.865 in the training cohort and 0.778 in the validation cohort and was also able to divide patients into low-risk and high-risk groups with notably different PFS outcomes: 6.5 months versus 3.2 months, respectively [137]. Another study built on this line of research by combining CT radiomics with MSI testing and messenger RNA sequencing [132]. The results indicated that a nine-feature signature could predict high MSI status. The signature could also distinguish between complete or partial response and stable or progressive disease following PD-1 inhibitor treatment and identify immune-inflamed and immune-suppressed microenvironments [132]. A multicenter study on a larger scale developed lymphoid and myeloid radiomic scores from 2,600 GC patients [126]. These findings confirmed that these imaging biomarkers, together with the four-class radiomic immune subtype, could independently predict patient survival and response to PD-1 inhibitor treatment. Patients with high lymphoid radiomic scores and low myeloid radiomic scores achieved an ORR of approximately 53.3% [126]. In contrast, patients with a low lymphoid radiomic score and a high myeloid radiomic score had an ORR of only approximately 10.2% [126]. These outcomes emphasize the value of AI-derived imaging signatures in stratifying large populations who could benefit from immunotherapy. While these studies did not focus specifically on ADCs, they established a research framework for multimodal AI models. This framework integrates CT, radiomics, histology, transcriptomics, and ctDNA features to define ADC-suitable and ADC-resistant populations.

For ADCs themselves, integrative analyses, such as the DESTINY-CRC01 biomarker study, demonstrate that combining tissue IHC/ISH, plasma HER2 amplification and ctDNA mutational profiles can refine predictions of T-DXd benefit beyond single markers [125]. These multiomic datasets are natural inputs for AI models, including graph-based and transformer architectures, that jointly encode antigen density, copy number status, codriver mutations and microenvironmental features. While most such ADC-focused AI pipelines have thus far been built and validated in non-GI tumors, there is already active work in developing multimodal, often federated, AI models that integrate histopathology images, genomics and clinical data to predict ADC response and toxicity across institutions without sharing raw patient data [129]. Extending these approaches to GI cohorts receiving T-DXd or disitamab vedotin, alone or in combination with ICIs, would enable data-efficient, privacy-preserving model training and help overcome the fragmentation of ADC datasets across centers.

Notably, true AI-based platforms for predicting the doses of ADCs and IFPs in GI oncology are still in the conceptual phase. Several of the above studies provide insights into relevant input factors. These include hepatic and renal function, radiomic-defined tumor burden, immune context, ctDNA-based clearance kinetics, and germline variants in drug-metabolizing enzymes. While still in the conceptual phase, integrating these factors into Bayesian or machine-learning frameworks warrants future investigation to guide the adaptive dosing of ADCs and immune biologics. This would be particularly relevant for frail HCC or PDAC patients, for whom therapeutic windows are narrow. This area continues to be an important focus for future methodological and clinical research.

Conclusion and future perspective

The treatment landscape of GI cancers is being reshaped by ADCs and a growing set of immune-engaging biologics, including IFP formats. The clinical impact of T-DXd underscores how careful ADC engineering can translate into meaningful benefit in selected molecular subsets. At the same time, mixed outcomes with multifunctional immune biologics remind us that mechanistic elegance does not automatically translate into clinical value without context-aware biomarkers and disciplined toxicity control.

This review frames immunocytotoxic convergence as a context-dependent, hypothesis-generating concept built on two practical observations. First, many GI tumors, particularly those that are stroma-rich and immune-excluded, often require deeper cytoreduction than immunotherapy alone can deliver. Second, durable control is unlikely without addressing stromal and myeloid programs that restrict immune access and promote dysfunction. Within this framework, Prime–Amplify is presented as a testable sequencing hypothesis, not a validated treatment algorithm. We hypothesize that, in selected contexts, appropriately chosen ADCs may induce localized immunogenic stress and/or stromal remodeling, potentially creating a transient window that could be leveraged by subsequent IFP-driven immune redirection. Our intent is not to recommend a sequencing regimen, but to define evidence thresholds, biomarkers, and failure modes that can determine whether sequencing is warranted in specific GI contexts.

Turning this concept into patient benefit will require prospective, biomarker-embedded studies. Priorities include: (1) defining minimal PD criteria for “priming” (e.g., ICD-associated signals, antigen-presentation markers, and spatially resolved changes in immune infiltration); (2) identifying failure modes such as myeloid rebound, lymphodepletion, or persistent stromal exclusion that predict low benefit or excess toxicity; and (3) running sequence-explicit early-phase trials that incorporate serial biopsies and, where feasible, reverse-order comparators. Recently reported ADC–IO combination trials in GC/GEJ across HER2- and CLDN18.2-directed platforms provide practical settings in which to embed these mechanistic readouts and explore whether timing, payload class, target selection, and baseline TME state determine the probability of meaningful immune amplification. If the hypothesis is confirmed, immunocytotoxic convergence may guide rational combination and sequencing strategies beyond GI oncology; if not, the same datasets will still clarify when—and when not—cytotoxic precision and immune engagement can be productively coupled.

Author contributions

SHW, RC and LY conceived the study and designed the review framework. SHW, KNZ and ZY performed the systematic literature search and data extraction. SHW, ZZY and WWC conducted the data analysis and created the figures/tables. The initial manuscript was drafted by SHW and RC. KNZ, ZY, ZZY and WWC participated in critical revision of the manuscript for important intellectual content. LY supervised the entire project. All authors read and approved the final manuscript.

Funding

This research was funded by the Macao Science and Technology Development Fund, grant number: FDCT0148/2022/A3 and 0019/2024/RIA1.

Data availability

All summary data generated or analysed during this study (including the curated tables and figures) are included in this published article. The original data underlying the cited studies are available from the corresponding publications referenced.

Declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.