Emerging biomarker identification strategies are now designed to overcome limitations of IHC (Table 1). An ideal target should exhibit high tumor-specific expression with limited normal tissue expression, allowing for a homogeneous distribution within the tumor and efficient internalization after binding. In addition, it should reflect relevant tumor biology, be reproducible and easy to measure, capture temporal dynamics of disease evolution and demonstrate a strong predictive value for therapeutic response [15]. Unfortunately, such ideal targets do not exist in clinical practice, although novel quantitative assessments are in development using transcriptomic, liquid biopsy, immunofluorescence-based assay, mass spectrometry, and digital pathology [16]. These assays can also be integrated with machine learning and artificial intelligence (AI) technologies, which have the potential to capture intra-tumor antigen gradients, overcome interobserver variability and reduce time expenditure, having shown a strong concordance with expert pathologists [17]. AI can also infer multiplex phenotypes from routine hematoxylin and eosin (H&E) by learning from multi-omics and imaging data, thus predicting biomarker status [18,19]. Similarly, digital pathology can refine patient stratification by capturing spatial variability of antigen expression and determining whether the expression is membranous or cytoplasmic, as well as the degree of heterogeneity in the sample [20,21]. Indeed, digitalized TROP2 IHC-stained whole-slide images (WSI) of tissue samples from patients with NSCLC have been used to develop the Quantitative Continuous Scoring (QCS) model, a deep learning algorithm trained on pathologists’ annotations to identify tumor areas and cellular compartments (membrane and cytoplasm) of each tumor cell across the WSI [22]. The QCS model measured TROP2 expression in the membrane relative to the cytoplasm, resulting in a normalized membrane ratio (NMR). In this sub-analysis of the TROPION-Lung01, datopotamab-deruxtexan (dato-DXd) demonstrated greater efficacy in patients with TROP2 QCS-NMR positive non-squamous (n-sq) metastatic NSCLC without actionable genomic alterations (AGA), demonstrating potential as a predictive biomarker. Notably, TROP2 QCS-NMR positivity was higher in patients with n-sq histology [22].

An additional strategy could involve molecular functional imaging with radiolabeled antibodies, which can enable the visualization of biomarker expression throughout the body. This strategy can assess TROP2 expression in real-time and non-invasively, offering guidance for targeted therapies. Studies have shown that TROP2-targeted radiotracers exhibited superior imaging performance in TROP2-positive tumors compared to stand-of-care examinations, demonstrating great clinical potential [23].

Another key approach is transcriptomic and proteomic profiling of tumor biology, including drug efflux and DNA-repair pathway activation, that can shape ADC susceptibility but may be invisible to IHC. Integrating these layers into models of disease progression and therapy response may discover biomarkers that go beyond just the presence or absence of an antigen [24]. Transcriptomic profiling can capture resistance pathways better than IHC. For instance, researchers analyzed the transcriptomic tumor profile of six drug efflux pump genes generated from pre-treatment biopsies collected from patients with metastatic breast cancer (MBC) before ADC treatment to elucidate biomarkers of response and resistance to sacituzumab-govitecan (SG) and T-DXd [25].

A further strategy is represented by circulating tumor DNA (ctDNA) and circulating tumor cells (CTCs) assessed through liquid biopsy, which can track dynamic tumor evolution (antigen loss, clonal selection, and resistance mechanisms) across tumor sites at different time points [12,24]. Patients responding to SG have significantly higher TROP2–positive CTCs at baseline than non-responding patients. Also, the number of TROP2–positive CTCs in responder patients significantly decreased during treatment. Separately, ctDNA analysis found a patient with a TOP1 gene mutation acquired during progression on SG, confirming that TOP1 gene mutations can lead to resistance to SG, as demonstrated in MBC [26].

In summary, future development of biomarkers for ADCs will likely require multi-modal approaches integrating IHC, transcriptomics, and liquid biopsy with the help of AI-based technologies for patient selection and monitoring of treatment response and resistance. However, challenges remain, including accessibility to biomarker testing and disparities in diagnostic infrastructure [18]. Therefore, equally important will be the collaboration between oncologists, pathologists, and bioinformaticians to accelerate translation into clinical practice and personalize ADC therapy [18]. A summary of the emerging biomarker workflow is reported in Fig. 1.

An ideal ADC should combine a highly specific antibody for a tumor-specific target with efficient internalization; the linker should be stable in the bloodstream, ensuring controlled intratumoral payload release; eventually, the payload should be sufficiently potent, chemically optimized to avoid aggregation, with high conjugate homogeneity and appropriate DAR to prevent off-target clearance and toxicity [27]. An appropriate DAR is particularly important because ADCs with a low DAR may not achieve the desired clinical efficacy, while ADCs with a high DAR may achieve higher plasma concentrations and exhibit greater off-target toxicity [28]. Consequently, variability in DAR has implications in terms of predictable pharmacokinetic profiles, efficacy and tolerability in real world settings. Beyond the numerical value of DAR, the distribution of DAR within an ADC critically determines its behavior. Early ADCs were made by non-site specific conjugation, which was accomplished by modifying lysine ε-amines or interchain cystine thiols in the conjugation site. This produced mixtures of molecules with variable DAR values ranging from 0 to 8 and a broad polydispersity index (PDI). A higher PDI reflects a wider particle size and molecular distribution, which is associated with reduced stability, increased aggregation and variable clearance [29,30]. This structural variability modifies solubility, charge distribution and hydrophilic–lipophilic balance, thereby affecting ADC performance: high DARs tend to exhibit increased hydrophobicity and aggregation, faster plasma clearance and hepatic uptake, whereas low DARs were less potent. Furthermore, structural variability can promote the formation of ADC immune complexes, thus accelerating clearance and further compromising pharmacokinetic stability [31]. The resulting early ADCs displayed unpredictable pharmacokinetics and narrow therapeutic windows that threatened clinical utility. For example, gemtuzumab-ozogamicin, an early ADC used to treat acute myeloid leukemia, was withdrawn from the market due to safety concerns attributed to its broad DAR distribution, unstable hydrazone linker, and highly potent payload calicheamicin, demonstrating the importance of reducing DAR variability [31].

DAR distribution is deeply connected to the conjugation strategy. Early lysine coupling was associated with greatly variable DARs because each IgG1 antibody typically possesses more than 20 potential attachment sites. In contrast, cysteine coupling was preferred because IgG1 and IgG4 antibodies possess four pairs of interchain disulfide bonds, enabling greater precision in conjugation [15]. However, conjugation remains partially stochastic when native interchain disulfides are reduced, still generating a broad distribution of DAR species. Taking into account considerations such as chemo- and regio-selectivity, additional methods were sought to enable controlled attachment of measured payload to specific positions on antibodies to create homogenous ADCs with consistent DARs, which brought about site-specific conjugation methods [30]. Such methods include engineered cysteine residues that provide selective conjugation sites, unnatural amino acid incorporations for precise coupling, and enzymatic ligation methods to create site specific attachment points. Engineered cysteine approaches introduce defined reactive thiol residues, although introduction of free sulfhydryl groups may predispose to unintended disulfide scrambling, requiring careful structural validation. Unnatural amino acid coupling introduces non-natural amino acids into the antibody sequence, allowing precise drug attachment and homogenous DAR. However, the potential immunogenicity of non-natural residues can carry a greater risk of immune response [31]. Enzymatic ligation strategies employing transglutaminase, formylglycine-generating enzyme or Sortase A enable selective modification of engineered sequences on the antibody structure. These approaches achieve regio-selective conjugation with minimal off-target modification, resulting in homogeneous ADCs with narrow DAR distributions and reduced physicochemical variability [31]. In parallel, click chemistry reactions (e.g., strain-promoted azide–alkyne cycloaddition) allow rapid, high-yield and stereoselective coupling under mild conditions, minimizing structural variation of the antibody and contributing to improved stability [31]. Collectively, these site-specific methods shift ADC engineering from a stochastic conjugation model toward a controlled molecular architecture, resulting in the development of more stable ADCs with improved efficacy [30].

Historically, ADCs were designed to target tumor antigens with high expression in tumor cells (e.g., HER2), but both antigen heterogeneity and target affinity may influence efficacy and safety [32]. High antibody affinity may improve tumor uptake but also increase the likelihood of normal tissue trapping in cases of shared antigen expression, thus increasing toxicity. The field has therefore moved toward prioritizing truly tumor-restricted antigens and refining affinity thresholds to balance penetration and selectivity [32].

In parallel, linker design has also evolved. Indeed, ADC toxicity seemed to be more related to linker stability and payload rather than to the antibody specificity [33]. Fu et al. and Wang et al. further supported that the chemical property of payloads (hydrophobicity, membrane permeability, and bystander effect) can affect efficacy, with hydrophobic payload and linker combinations leading to aggregation, clearance and narrow therapeutic windows, thus emphasizing the importance of payload and linker selection to achieve a better therapeutic index [1,31]. In contrast, hydrophilic linkers can improve stability, pharmacokinetics and reproducibility, thus reducing inter-patient variability and expanding the therapeutic window [31]. Linker advancements have also evolved from cleavage prone linkers that led to early plasma release to a modern linkers with better plasma stability and triggered payload release only in the tumor or after antibody internalization [34]. Bargh et al. classified cleavable linkers based on their cleavage mechanisms: acid cleavable, which exploits endosomal or lysosomal pH-like hydrazones, reduction-sensitive disulfides that respond to intracellular glutathione concentration, enzyme cleavable dipeptides relying on lysosomal proteases to enable selective intracellular release. Cleavable linkers enable payload release and potent bystander effect but may lead to off-target release and systemic toxicity [34]. On the other hand, non-cleavable linkers keep the drug attached until antibody degradation, giving better plasma stability but requiring full internalization and degradation, thus limiting efficacy in antigen-low tissues [34]. These considerations underscore why next-generation ADCs should improve linker selection by integrating tumor biology [33].

The payload choice remains central in ADC performance. Early payloads used tubulin-disrupting agents like maytansinoids (DM1) and monomethylauristatin (MMAE), which had potent cytotoxicity but were limited by systemic toxicity and resistance mechanisms such as upregulation of efflux pumps [35]. The introduction of topoisomerase I (topo I) inhibitor payloads, such as DXd, has dramatically revolutionized ADCs. Metabolites from these payloads are membrane permeable with a more potent bystander effect that can kill neighboring tumor cells with low or absent antigen levels, thus addressing tumor heterogeneity [36]. The spheroid imaging data, a sensitive platform using tumor spheroids and a pharmacodynamic marker to quantitatively map the penetration of cytotoxic bystander payloads, showed that payloads like MMAE and calicheamicin D had a deeper penetration within the spheroid compared to DM1 and calicheamicin G which only affected the targeted cells and those immediately adjacent to it [37]. Khera et al. also described how to quantify the theoretical bystander penetration of payloads by capturing the ratio of cell uptake versus tissue diffusion of payload, to predict optimal payload physicochemical properties [37]. Payload penetration within cells also depends on lipophilicity, charge, hydrogen bonding capacity and molecular weight. Greater permeability can increase uptake by bystander cells, but if too high can fail to reach deeper regions, and if low, the payload would be washed out before achieving intracellular cytotoxic concentrations. Hence, new ADCs should address both the chemical properties of membrane permeability and the optimal radius of payload effect needed for effective bystander killing of tumor cells [37].

However, the future of payloads extends beyond traditional cytotoxic agents and is heading towards immune-modulating and PROteolysis TArgeting Chimera (PROTAC) payloads, transforming ADCs into multimodal agents for direct tumor killing and immune system activation [38,39]. A summary of the modifications across ADC generations is reported in Table 2 and Fig. 2.

Eventually, off-target toxicities remain a major limitation. ADC-related toxicities can arise from target-dependent and target-independent mechanisms, and are strongly influenced by the properties of the payload, linker and antibody. While low level expression of the target in normal tissues can result in receptor-mediated uptake and toxicity, this is often poorly predictive of ADC safety. Instead, target-independent mechanisms, including non-specific endocytosis, Fc receptor–mediated uptake, premature linker cleavage, bystander effect and systemic release of payloads, play a crucial role in many toxicities [40]. For example, neutropenia is a common toxicity of ADCs largely caused by premature payload release due to linker instability and subsequent bystander effect on hematopoietic cells. The cytotoxic effect of MMAE was noted in differentiating neutrophils, likely mediated by serine proteases mediating extracellular cleavage, while, in patients with colorectal cancer, decreased SN-38 glucuronidation was associated with neutropenia. Ocular toxicity is thought to be driven by macropinocytosis in corneal epithelial cells, independent of antigen expression. Similarly, hepatic toxicity and thrombocytopenia have been associated with non-specific clearance mechanisms involving Kupffer cells, sinusoidal endothelial cells, Fc receptor–expressing cells, and possibly mannose receptor–mediated uptake [40]. FcγR-mediated uptake into differentiating megakaryocytes has been implicated in thrombocytopenia with T-DM1, despite a lack of HER2 expression, highlighting the contribution of Fc-dependent internalization to hematologic off-target toxicity. T-DXd, associated with interstitial lung disease (ILD), was primarily localized in alveolar macrophages but not pulmonary epithelial cells, again suggesting target-independent uptake [40].

In order to minimize adverse effects, it is especially important to ensure predictable pharmacokinetics and payload-linker stability. Linker instability, high DARs, and increased hydrophobicity enhance systemic payload exposure and non-specific clearance, reinforcing the direct link between ADC engineering and toxicity patterns. Comparisons of ADCs with cleavable and non-cleavable linkers have shown reduced toxicity with non-cleavable linkers for specific targets [31]. While increased membrane permeability can augment the bystander effect in ADCs and improve tumor killing, the increased cellular permeability required to achieve this effect can worsen off-target toxicity. Many drug-linker combinations are hydrophobic, and in agents with a high DAR, this can contribute to ADC aggregation and non-specific clearance by Kupffer cells [31]. Advancements in ADC engineering aimed at reducing these toxicity mechanisms while preserving antitumor efficacy. As stated previously, the development of more stable linkers to minimize premature payload release, optimization of DAR to reduce hydrophobic clearance, selection of payload with reduced membrane permeability to limit bystander toxicity, as well as site-specific conjugation technologies, can effectively reduce off-target toxicities [15]. In parallel, conditionally active ADCs, particularly probody therapies utilizing masking peptides that are exclusively cleaved by tumor-specific proteases, are able to restrict payload release to the TME [15]. Additional engineering solutions include Fc modifications to reduce hepatic uptake and modulation of surface charge to limit non-specific endocytosis, further highlighting how ADC design is evolving toward complex integrated platforms [40].

Among these, c-MET ADCs have advanced most rapidly. c-MET is dysregulated in NSCLC via overexpression (approximately 25–39%), exon 14 skipping (2–4%) or amplification (5%), the latter linked to resistance in EGFR-mutant disease. Teliso-V consists of an anti-MET antibody linked to MMAE. In the LUMINOSITY trial, a phase II study including patients with EGFR-wildtype (wt) c-MET-overexpressing NSCLC, Teliso-V demonstrated an ORR of 28.6% up to 34.6% in IHC-high tumors, with PFS of 5.7 months. Neuropathy and peripheral edema were the main adverse events [56]. Its recent accelerated approval by the FDA in May 2025 underscores the feasibility of biomarker-driven ADCs in lung cancer; meanwhile, a phase III trial comparing Teliso-V to docetaxel (NCT04928846) is ongoing [72]. In addition, new ADCs targeting MET are currently being tested in phase I clinical studies [73,74,75].

ADCs targeting other emerging antigens have also been investigated, including PTK7, B7-H3, integrin β6, TfR1, AXL, FOLR1, CEACAM5, NaPI2b, TF, LY6E, ROR2, ALCAM, 5T4, and LYPD3 (Table 3) [52,53,59,60,61,62,63,64,65,66,67,68,69,70,71,76,77,78,79,80,81]. Some clinical programs have been discontinued due to limited efficacy or safety concerns, as in the case of enapotamab- or lifastuzumab-vedotin [64,68]. Others, instead, showed significant clinical potential.

PTK7 is a pseudokinase enriched in NSCLC with poor prognosis. Cofetuzumab-pelidotin (Cofe-P) consists of an anti-PTK7 antibody conjugated to a microtubule inhibitor. In a phase Ib trial, Cho et al. included 65 patients with NSCLC expressing PTK7 and reported an ORR of 18% and a median PFS of 5.5 months [59].

B7-H3 is a transmembrane glycoprotein that promotes immune evasion and tumor progression [8]. MCG018, an ADC targeting B7-H3, was tested in advanced solid tumours and demonstrated moderate safety and efficacy, particularly in metastatic castration-resistant prostate cancer [76,82]. Recently, Ifinatamab-deruxtecan (DS-7300) showed interesting clinical activity in advanced solid tumors [77]. A phase I/II study reported an ORR of 4 (31%) of 13 patients with squamous NSCLC [60]. A phase II study in 183 patients with extensive-stage small cell lung cancer (SCLC) reported an impressive response rate of 48.2% and a median PFS of 4.9 months [61].

Another interesting target is represented by integrin β6, a cell surface receptor implicated in tumor invasiveness. The combination of sigvotatug-vedotin (SV) and pembrolizumab showed an ORR of 57% (4 of 7 patients) in a recent phase I study [62]. Based on these preliminary results, a phase III study is currently comparing this agent to pembrolizumab in patients with PD-L1 high NSCLC (NCT06758401) [62].

AXL is a cell-surface receptor tyrosine kinase that is expressed in several solid tumor types. Increased AXL expression is associated with tumor resistance to chemotherapy, PD-1 inhibitors, molecular targeted therapy, and radiation therapy. Mecbotamab-vedotin (BA3011) is an anti-AXL ADC. BA3011 alone or in combination with nivolumab demonstrated an ORR of 28% [65]. In the case of enapotamab vedotin, which initially showed meaningful activity in 5 of 8 AXL-selected sarcoma PDX models, the effect did not correlate with AXL expression and subsequent findings suggested that the response was likely influenced by additional downstream determinants such as tumor sensitivity to MMAE payloads. These findings point to biomarker and ADC design limitations creating an engineering mismatch between target selection and payload effectiveness [83]. In a phase I/II study, clinical activity likewise did not clearly correlate with AXL expression, further supporting efficacy limiting factors like internalization and/or intracellular release of the cytotoxic payload, as well as a payload resistance mechanism with suggestions that increasing the payload to antibody ratio could potentially mitigate these mechanisms and increase efficacy, particularly among patients with NSCLC, as free MMAE can generate bystander toxicity [84]. The key engineering problem was therefore not merely patient selection, but also the payload design, which is susceptible to heterogeneous delivery and resistance.

Tusamitamab-ravtansine (TUSA) is an ADC that targets cells expressing CEACAM5. In a phase Ib study including 64 patients with high and 28 patients with moderate CEACAM5 expression and n-sq NSCLC, the confirmed partial responses (PR) were 20% and 7% in high and moderate expressors, respectively. Interestingly, the most common grade 3 treatment-related adverse event was keratopathy [81]. However, Besse et al. found no difference in median OS with tusamitamab-ravtansine vs. docetaxel (12.8 vs. 11.5 months) [53]. An exploratory analysis of patients with long-term treatment exposure has revealed that responses to TUSA could not be related to CEACAM5 expression by IHC [85].

Eventually, TF, also known as thromboplastin or factor III, is a transmembrane glycoprotein that is overexpressed in many types of cancer. Tisotumab-vedotin is an ADC that targets TF and was approved by the FDA for the treatment of recurrent or metastatic cervical cancer based on the innovaTV 301 trial [86]. The phase I/II trial in patients with advanced or metastatic solid tumors reported an ORR of 13% in 2 of 15 patients with NSCLC [70].