Targeting CD44 Variant 5 with an Antibody–Drug Conjugate Is an Effective Therapeutic Strategy for Intrahepatic Cholangiocarcinoma Open Access

Intrahepatic cholangiocarcinoma (ICC) is a group of malignancies of the biliary epithelium (cholangiocytes) that arise in the intrahepatic biliary tree. Because of the steadily increasing prevalence, ICC is the second most frequent type of primary liver cancer, accounting for 15% of cases (1). Most patients with ICC often present at late stages with either locally advanced and unresectable or metastatic disease at diagnosis. Even patients who undergo curative resection have poor outcomes due to the high rate of tumor recurrence (2). The 5-year survival rate for these patients is less than 30% (3). Standard therapy, such as gemcitabine and a platin-based drug, only confers minimal survival benefit (4). The major reason why the majority of patients with ICC fail to respond to standard therapy is the heterogeneity of this cancer type (5). To date, several comprehensive genomic analyses have been performed and revealed a diverse mutational landscape in ICC (6). Alterations in some genes, such as FGFR2 fusions and mutations in IDH1 and IDH2, occur in patients with ICC and contribute to oncogenesis and malignant progression (5, 7, 8). On the basis of these findings, some novel targeted therapies have been developed and have shown promising clinical efficacy against ICC. The selective pan-FGFR kinase inhibitor showed impressive antitumoral activity in a phase II study of patients with ICC with FGFR2 fusion (9). Moreover, the IDH1 inhibitor ivosidenib has been proved by FDA for patients with ICC with IDH1 mutation (10). However, patients with these aforementioned mutations account for a small percentage of patients with ICC [IDH1 is 16% (2), and FGFR2 fusion is 15% (11)], which limits the clinical application of these promising targeted therapies. Therefore, novel targeted therapeutic strategies to combat ICC are urgently sought.

Antibody–drug conjugates (ADC) are a therapeutic modality that combines the targeting properties of mAbs with an anticancer agent (also known as the payload or warhead) via a chemical linker (12). The ADC binds a tumor-associated antigen with high selectivity, internalizes, and subsequently delivers the cytotoxic drug preferentially into tumor cells, thereby improving the efficacy and reducing systemic exposure and toxicity. Currently, over 10 ADCs have been approved by the FDA for clinical application for various types of cancer treatments, and more than 80 ADC candidates are under clinical development worldwide (13). Therefore, an ADC-based therapeutic strategy could potentially be applied for ICC treatment.

For the success of an ADC, selecting an appropriate target antigen is the first critical step. CD44 is a single transmembrane glycoprotein receptor and is the most common surface marker of both normal stem and cancer stem cells (CSC) in several types of cancer (14–16). The full-length CD44 gene consists of 20 exons in mice but 19 exons in humans (exon 6 is missing in humans; ref. 17). Exons 6–14 of the CD44 gene in humans (known as variable exons, CD44v2-v10) undergo extensive alternative splicing (AS) via excision or inclusion in various combinations in the membrane-proximal stem region to generate splicing variants (CD44v isoforms). Importantly, recent studies have demonstrated that the standard isoform (CD44s) is expressed on most vertebrate cells with stem-like properties, whereas alterative splicing producing CD44v occurs only under specific conditions; this isoform is predominantly expressed in epithelial-type carcinomas, particularly those in advanced stages (18). In ICC, CD44v was found to be newly expressed during the carcinogenesis of intrahepatic biliary cells (19). High expression of CD44v9 was demonstrated to correlate with human liver fluke Opisthorchis viverrini-related ICC. Knockdown of CD44v8–10 suppresses cell growth and triggers autophagic cell death in cholangiocarcinoma (15). Bivatuzumab mertansine (BIWI 1) consists of a humanized IgG1 kappa mAb conjugated to mertansine, which is the first ADC to enter into clinical trials for treatment of CD44v6-positive tumors (20, 21). Unfortunately, the clinical trials were abruptly ended because of the on-target off-tumor skin-related toxicities, although some clinical benefits were observed (21). CD44 variant isoforms remain potential therapeutic targets for ICC.

In this study, we verified that CD44v5 (a variant of CD44 containing exon 9 in humans) was specifically upregulated in over 30% of ICC tumors versus normal tissues according to analysis based on both The Cancer Genome Atlas (TCGA) database and clinical specimens. Interestingly, unlike CD44v6, CD44v5 exhibited little expression in normal skin tissues, leading to the hypothesis that it could make a tractable therapeutic target for an ADC in ICC treatment. Then, we developed a humanized antibody suitable for an ADC approach with consideration of target specificity, affinity, and internalization into target-expressing cells. Next, we present preclinical data on the activity of a CD44v5-directed ADC (H1D8–DC, H1D8–drug conjugate) composed of a humanized anti-CD44v5 mAb, a cleavable valine–citrulline-based linker, and MMAE. H1D8–DC specifically eliminated CD44v5-positive ICC cells both in vitro and in vivo. Furthermore, potent antitumoral activity was observed in CD44v5-positive ICC patient-derived xenografts (PDX), with favorable safety profiles. The results presented here support further clinical investigation of this CD44v5-targeted ADC with potential utility in CD44v5-expressing ICC.

RBE, HCCC-9810, and HuCCT1 cells (originally purchased from the typical cell culture collection of the Committee of the Chinese Academy of Sciences Library) were cultured in RPMI-1640 medium (Invitrogen) supplemented with 10% FBS (Bioind) and 1% penicillin‒streptomycin at 37°C with 5% CO₂ according to the supplier's instructions. CD44s-OE and CD44v-OE cells were generated in our laboratory, representing the transfected RBE cell lines stably expressing human CD44s and CD44v, respectively. All cells tested negative for Mycoplasma and were authenticated by short tandem repeat DNA fingerprinting at the Key Laboratory of Pharmaceutical Biotechnology and The Comprehensive Cancer Center of Nanjing University (Nanjing, China). No cell lines were passaged for more than 6 months after resuscitation in our study.

Mouse polyclonal antibodies against human CD44v5 were generated as described in our previous work (22). Briefly, mice were immunized with a purified variant region of CD44, and hybridomas were generated by using standard protocols (23). Then, clones were screened for binding to CD44v5 protein and CD44v5-expressing cells by comprehensive application of ELISA, WB, FCM, and HIC (hydrophobic interaction chromatography).

The 1D8 mAb was humanized (defined as H1D8) using a dual complementary determining region (CDR) grafting method (24). Briefly, the VH and VL sequences were searched against the human germline sequence databases with IgBLAST (https://www.ncbi.nlm.nih.gov/projects/igblast/) and IMGT/V-QUEST (https://www.imgt.org/IMGT_vquest/input), and the most similar human germline Fv sequence and J region were identified. The residues within the CDR were grafted onto the framework regions of templates. Biacore analysis of antigen binding affinity was accomplished using standard techniques on a Biacore T200 (GE Healthcare).

The humanized VH and VL sequences of H1D8 (Table. S3) were fused by using the T2A peptide, which was subsequently cloned into the pcDNA3.4 backbone vector to generate the expression plasmid. The antibody was produced in the human HEK293F cell line by transient transfection with polyethylenimine (PEI 25 K_d linear, Yeasen Biotech). The secreted humanized IgG in the culture supernatant was purified by affinity chromatography with a protein A column (GE). The purified H1D8 was tested by SDS-Page.

Total RNA of indicated cells was extracted by using TRIzol reagent (Invitrogen). 2-μg RNA was reverse transcribed into cDNA via the 5 × All-In-One RT Master Mix kit (Abm Cat#G486 Code Q111–02). cDNA (20–50 ng) was subjected to PCR analysis with primers targeted to exons 5 and 16 of human CD44 (forward: 5′-ATCCCAGACGAAGACAGTCCC-3; reverse: 5′-TGTTTGCTCCACCTTCTTGAC-3′). The positions of PCR products were shown as indicated (CD44s: 160 bp; CD44v: 289–3200 bp). GAPDH expression was used as a loading control.

The CD44v-targeted ADC is composed of the H1D8-humanized IgG1 mAb conjugated to the cytotoxic agent MMAE via a valine–citrulline (mc–vc—PAB, maleimidocaproyl–valine–citrulline–para-aminobenzyloxycarbonyl) linker (25). Bioconversion was accomplished by derivatizations of the side chains of cysteine residues. Partial reduction of these disulfide linkages provides a distribution of free thiols that can be functionalized with the maleimide handle on the linker. Specifically, the H1D8 mAb was partially reduced by the addition of 3.2 eq. Tris-(2-carboxyethyl)-phosphine hydrochloride (TCEP-HCl) and stirred mildly for 2 hours at 25°C. Then, the resulting reaction mixture, dimethylacetamide (6% v/v) and 7.0 eq. of mc–vc–PAB–MMAE were sequentially added and stirred mildly for 2 hours at 25°C, followed by incubation with a 3-fold excess of N-ethylmaleimide to cap the unreacted thiols and addition of 25 eq. N-acetyl-L-cysteine to quench any unreacted linker-payload. The final mixture was purified using Zeba spin desalting columns (Thermo Fisher Scientific). The ADC was further characterized by size exclusion chromatography, HIC, and reversed-phase (RP)-high-performance liquid chromatography (HPLC). The protein concentration was determined by ultraviolet spectrophotometry.

The raw sequencing data of project PRJNA422089 were downloaded from European Nucleotide Archive (ENA). Quality control and trimming were performed on the RNA-seq data using FastQC and Trim galore. The data were aligned to the GRCh38 human reference genome by HISAT2 (26). After alignment and quantification, FeatureCounts was used to generate raw counts for AS analysis. The count files were analyzed using DEXSeq R package (27), with GRCh38 genome gff file plus the experimental design using default parameters. The differentially expressed exons between control and ICC samples have been selected with an adjust P value cutoff of 0.01. The AS of CD44 was visualized using plotDEXSeq function.

The CD44 exons expression and transcriptome data in ICC were downloaded from TCGA Splicing variants Database (TSVdb) and cBioPortal. The relationship between CD44 exons and expression patterns were clarified by using Pearson correlations analysis. The coexpression genes were selected with an r² value cutoff of 0.5. The co-expression networks of each CD44 exon were further analyzed by using the Venn diagram.

For the binding assay, single-cell suspensions of RBE, HCCC-9810, and HuCCT1 cells were obtained with EDTA, blocked for 30 minutes on ice in PBS (with 0.2% IgG-free BSA and 2 mmol/L EDTA), and subsequently incubated with increasing concentrations of H1D8 for 30 minutes on ice. The cells were washed and incubated with a goat anti-human secondary antibody labeled with PE (eBioscience) for 30 minutes on ice. The cells were washed and detected by FCM. Background-subtracted mean fluorescence intensities were normalized and plotted as a function of antibody concentration. Data were fit to a single binding site equation using GraphPad Prism, from which apparent K_d values were derived.

For internalization analysis, single-cell suspensions of RBE-CD44s and RBE-CD44v cells were incubated with h1D8 (10 μg/mL) at 37°C for 30 minutes. Then, the cells were washed with PBS to remove free antibody. The cells were cultured at 37°C with 5% CO₂ for the indicated times. Then, the cells were analyzed using BD FACSCalibur 4 CLR.

The internalization and trafficking of the ADC in cholangiocarcinoma were investigated by confocal microscopy. For internalization analysis, HuCCT1 cells were incubated with h1D8 for the indicated times. Then, the cells were fixed in 4% paraformaldehyde for 30 minutes at room temperature. The fixed samples were permeabilized, and the ADC was visualized using a FITC-labeled secondary antibody (green). Nuclei and lysosomes were stained with anti-Lamp1 (red) and Hoechst 33342 (blue), respectively. For trafficking analysis, HuCCT1 cells were incubated with Cy5-labeled h1D8 and maintained under 5% CO₂ at 37°C. Nuclei were stained with Hoechst 33342. The trafficking of the ADC was analyzed by a live cell imaging system (PerkinElmer Operetta CLS).

A total of 155 primary human ICC tumor tissues and 5 corresponding nontumor adjacent tissues and human normal tissues (including liver, n = 2; lung, n = 5; heart, n = 3; spleen, n = 2; skeletal muscle, n = 7; skin, n = 3; thyroid, n = 4; tongue, n = 3; esophagus, n = 4; gastric mucosa, n = 6; rectum, n = 6; ileum, n = 3; vermiform appendix, n = 5; colon, n = 2; pancreas, n = 2; trachea, n = 3; prostate, n = 2; testis, n = 7; bladder, n = 4; telencephalon, n = 3; epencephalon, n = 2; brainstem, n = 1) collected from the National Human Genetic Resources Sharing Service Platform (No. 2005DKA21300) were used as tissue array samples. This tissue microarray (TMA), including samples of 87 cases in early stages (stage I/II) and 68 cases in late stages (III/IV) with corresponding overall survival (OS) time data, was constructed as previously described using a tissue arrayer (Beecher Instruments). The TMA was further analyzed by IHC staining using H1D8 mAb. The immunostained slides were quantified by Histoscore and TissueGnostics software. Written informed consent for the use of the collected samples was obtained from all participants. This study was approved by the Institutional Review Board and Ethics Committee of the Nanjing Drum Tower Hospital (IRB Review Approval Documents: 2020–072–01).

Cells were seeded into 96-well plates at a density of a total of 5×10⁴ cells/well (100 μL) and incubated for 24 hours. Then, the cells were treated with different concentrations of H1D8–DC and Isotype-DC. The effective concentration of ADC that killed 50% of cells relative to untreated cells (EC₅₀) was calculated by logistic nonlinear regression on GraphPad Prism software. Three or more biological replicates were performed for cytotoxicity experiments.

Cells were seeded into 96-well plates (JETBIOFIL, cat no. TCP011096) at a concentration of 2,500 cells per well. After overnight culture, ADC was added to the wells at different concentrations (0.1–1000 ng/mL). Cell proliferation was measured using a Cell Counting Kit 8 (CCK-8; Beyotime, cat. no. C0037) according to the manufacturer's instructions. The cytotoxicity is presented as the IC₅₀ value, which is the ADC concentration that reduced cell proliferation by 50% compared with cells that were treated with isotype antibody only.

The induction of apoptosis or necrosis was investigated in RBE, HuCCT1, RBE-CD44s, and RBE-CD44v cells using a Calcein-AM and propidium iodide double staining kit (Beyotime, cat no. C2013M). Briefly, a total of 5×10³ cells were seeded per well in a 96-well confocal cassette. Twenty-four hours after seeding, the cells were treated with ADC at the indicated concentration. Isotype antibody was used as the control treatment. After 3–5 days of incubation, the drug-containing medium was removed, and the cells were washed and stained with Calcein-AM/propidium iodide (PI) according to the manufacturer's instructions. The samples were analyzed using a confocal microscope (Leica TCS SP8).

The objective of in vivo efficacy studies was to evaluate the activity of H1D8–DC in ICC models, including RBE cell line xenografts and PDX models. The sample size (n = 4 to 6 mice per group) was determined on the basis of consistency and homogeneity of tumor growth in the various models and was sufficient to determine statistically significant differences in tumor response between the various treatment groups. Animals were randomized on the basis of tumor size and were treated with various treatments until the average tumor volumes reached 150 mm³. All animals were purchased from the Model Animal Research Center of Nanjing University (Nanjing, China). The animal studies were approved by the Laboratory Animal Welfare and Ethics Committee of Nanjing University (IACUC-2109010) and carried out at Nanjing University. All animal experiments conformed to the guidelines of the Animal Care and Use Committee of Nanjing University. All efforts to minimize suffering were made.

For subcutaneous tumor models, a total of 3×10⁶ ICC cells were resuspended in 200 mL PBS with 30% Matrigel (cat no. 356234, BD Biosciences) and injected subcutaneously into the flanks of 6-week-old male BALB/c nude mice. ICC PDX models were initiated as previously described (28). Briefly, tumor fragments from patients with ICC were implanted subcutaneously into 5- to 7-week-old male NCG mice. The animals were treated intravenously with isotype (5 mg/kg) or H1D8–DC (5 mg/kg), respectively. Tumor volumes were measured using an electronic caliper and calculated using the following formula: Volume (mm³) = L×W²×π/6, where L and W represent the largest and smallest diameters, respectively. The animals were sacrificed in a CO₂ chamber in accordance with Institutional Animal Care and Use Committee guidelines. Tumors were harvested and used for histological analysis or rapidly frozen.

IHC was performed on deparaffinized formalin-fixed paraffin-embedded sections as previously reported (23). Briefly, 5-mm slides were deparaffinized and rehydrated and then subjected to heat-induced epitope retrieval with citrate butter at 100°C for 20 minutes. The following antibodies and conditions were used: Anti-H1D8 (prepared in this study, 1:400 dilution), anti-γH2AX (#80312, CST; 1:400 dilution), anti-activated caspase-3 (ab32042, Abcam; 1:400 dilution), anti-Ki-67 (#9449, CST; 1:400 dilution), and anti-CD133 (66666–1-lg, Proteintech; 1:1,000 dilution). Hematoxylin solution was used for counterstaining. The expression of CD44v was evaluated using the Histoscore (H-score; ref. 29).

H1D8 and H1D8–DC were intravenously injected into mice (n = 3) at two dosages (1 or 5 mg/kg). Serum was collected at 1, 24, 48, 72, 96, 120, 144, and 168 hours. ADC and mAb concentrations were determined by an enzyme-linked immunosorbent assay-based approach. Pharmacokinetic parameters, including C_max, exposure (AUC_(0-∞)), and half-life (T_1/2), were calculated by WinNonlin 8.0 (Certara).

All statistical analyses were performed by GraphPad Prism 5 (GraphPad). TCGA RNAseq_V2 gene expression data were downloaded from the cBioPortal website. Differences between two groups were calculated by the Student t test. Multiple comparisons between two populations were conducted by multiple t tests with type 1 error correction. Differences among multiple groups were calculated by one- or two-way ANOVA. Differences in survival were calculated by the log-rank Mantel‒Cox test. Differences between tumor-growth curves were determined by repeated measures two-way ANOVA. Significance was set at a P value of less than or equal to 0.05. For all figures, *, P < 0.05; **, P < 0.01; ***, P < 0.001. Unless noted in the figure legend, all data are shown as the mean ± SEM. Generally, all experiments were carried out with n ≥ 3 biological replicates.

The CD44 exons expression and transcriptome data in ICC were downloaded from TCGA Splicing Variants Database (TSVdb) and cBioPortal (CHOLANGIOCARCINOMA, TCGA, PanCancer Atlas, 36 samples), respectively. The raw sequencing data of project PRJNA422089 were downloaded from ENA. All data associated with this study are present in the article or the Supplementary Materials. Reagents developed in this study will be made available to the scientific community through a material transfer agreement by contacting P. Shen.

AS is an important regulator of most hallmarks of cancer (30–32) and has recently been investigated as a novel source of neoantigens (33). To achieve this goal, we performed pancancer analysis by using TSVdb based on TCGA database as described before (34), focusing on AS events. Our preliminary study revealed that by querying 30 Pan-Cancer datasets, variable exons of CD44 were specifically elevated in primary tumor tissues compared with adjacent normal tissues (Supplementary Fig. S1A), especially in cholangiocarcinoma (Fig. 1A and B; Supplementary Fig. S1B and S1C), as indicated by significantly elevated mean mRNA expression. In addition, splicing isoforms, such as isoform_2, _3, and _4 containing variable exons (Fig. 1A), were preferentially expressed in primary tumor tissues rather than in corresponding normal tissues (Supplementary Fig. S1D). Thus, we speculated that CD44v has great potential as a tumor-specific antigen in ICC. For further confirmation, we conducted de novo assembly and AS analysis of next-generation RNA-seq data (from GSE107943, which include information on 30 primary tumor tissues and 26 matched control tissues from the surrounding normal liver; ref. 35). According to the Sashimi plot shown in Fig. 1C, we observed that the splicing pattern in normal tissues greatly differed from that in primary tumors. As expected, the variable exons (6 to 14) were almost undetectable in adjacent normal tissue, whereas the number of variable splicing events was significantly increased in ICC tumor tissues (Fig. 1D and E). Because of the lack of essential mAbs that specifically recognize each variable exon of CD44, we next prepared clinical specimens and conducted an RT‒PCR assay for CD44v identification (36). As shown in Fig. 1F, the results showed that CD44v was expressed in many ICC specimens but was seldom observed in adjacent normal liver tissues. Interestingly, CD44v was also widely expressed in multiple non-ICC cancer cell lines, including breast cancer cell lines (4/7), lung cancer cell lines (4/8), colon cancer cell lines (6/8), gastric cancer cell lines (4/6), liver cancer cell lines (2/8), and pancreatic cancer cell lines (5/7; Supplementary Fig. S2A). These data strengthen the hypothesis that CD44v isoforms could be potential tumor-specific targets.

In line with previous findings (17, 37), we noticed a high heterogeneity of CD44 variants in different cancers (Supplementary Fig. S1A and S1B; Supplementary Fig. S2). The expression patterns of each CD44v isoform seem to vary by cancer type (37). To determine the exact expression pattern of each CD44 variant in ICC, we comprehensively conducted transcriptomic analysis in parallel with each exon expression (Supplementary Table S1) and subsequently determined the correlation gene-expression signature of each exon of CD44. The Venn diagram showed that CD44v5 exhibited a distinct expression pattern comparing with other splicing variants (Supplementary Fig. S2B), indicating a special role of CD44v5 in ICC.

To further assess protein expression and determine whether CD44v5 is specifically on the surface of ICC cells, anti-CD44v5 antibodies were generated and characterized (38). Briefly, recombinant variable exon 5 of the CD44-encoded peptide, synthesized by GenScript, was used as an immunogen to produce mouse Abs. Then, we examined the specificity and binding activity of hybridoma-derived murine mAbs by comprehensively using ELISA, western blotting, and flow cytometry as indicated in Supplementary Fig. S3A. The results showed that subclone 1D8 exhibited potent binding to CD44v5-expressing cells, cross-species reactivity (human and rhesus monkeys) and good specificity (Supplementary Fig. S3B–S3D), rendering it an appropriate candidate for further application. For clinical development, the selected mouse mAb, 1D8, was humanized via CDR grafting of the murine variable regions onto the human immunoglobulin G1 (IgG1)/k constant regions (Supplementary Fig. S4A; and Supplementary Table S2; ref. 24). Humanized H1D8 was expressed via a HEK 293F-based eukaryotic expression system and purified by protein A (Supplementary Fig. S4B and S4C). Humanized H1D8 showed a similar affinity to the parental mouse mAb 1D8 against the CD44v5 antigen, as defined by surface plasmon resonance (Supplementary Fig. S3D).

To examine the binding activity of humanized H1D8, we performed flow cytometry analysis using ICC cell lines, including RBE, HCCC-9810, and HuCCT1. CD44v5 expression in these cell lines was confirmed by using RT‒PCR as described above. Consistent with RT‒PCR analysis, H1D8 bound to CD44v-positive HCCC-9810 and HuCCT1 cells but not to RBE cells (Fig. 2A). Furthermore, H1D8 exhibited binding activity to the membrane of HCCC-9810 and HuCCT1 cells starting at 20 ng/mL until 1,000–10,000 ng/mL, whereas no significant binding was observed on RBE cells at the highest concentrations (Fig. 2B). For further determination of the binding selectivity of H1D8, the pair of prepared isogenic RBE cells transgenically expressing either the CD44v (CD44v-OE) or CD44s isoform (CD44s-OE) were used (Fig. 2C). Similar to previous findings, H1D8 specifically bound to CD44v-OE RBE cells (Fig. 2C). The probable H1D8-binding epitope of CD44v was confirmed again by using the CD44v truncation test (Supplementary Fig. S5A). These data indicated that humanized H1D8 could bind to variant 5 on the membrane of ICC cells with high specificity and affinity.

With the available anti-CD44v5 mAb, we retrospectively investigated CD44v5 expression and cellular localization in clinical ICC tumors. To this end, we prepared a TMA from a set of 5 nontumor adjacent tissues and 155 human ICC tumor tissues, including tissues from 87 cases in early stages (stage I/II) and 68 cases in late stages (III/IV), and determined the expression of CD44v5 via IHC. For further quantification, surface CD44v5 expression was subsequently quantified by converting the staining intensity (range, 0 to 3) and the percentage of cells with expression to a histoscore (range, 0 to 300) as described previously (29). Interestingly, we noticed that CD44v5 was predominantly expressed on the surface of ICC tumor cells versus nontumor adjacent cells (Fig. 2D and E). In addition, ICC tumors in the late stages showed higher expression of CD44v5 than those in the early stages (Fig. 2F). Generally, 66% (103 of 155) of ICC specimens had membrane staining, and over 16% of them had high expression of CD44v5 (Fig. 2G). More importantly, no normal tissues showed significant positive H1D8 staining (histoscores were lower than 10; Fig. 2H and I; Supplementary Fig. S6A). Particularly in the normal skin tissues, CD44v5 is less detectable than CD44v6 (Fig. 2I; Supplementary Fig. S6B and S6C), where the main on-target/off-tumor toxicity occurred for the commercial anti-CD44v6 ADC (bivatuzumab mertansine; ref. 21). No significant differences were observed between IDH1-mutated ICC and IDH1 wild-type ICC (Fig. 2I; Supplementary Fig. S6B). Among patients with ICC, those with high expression of CD44v5 showed a shorter OS time (Fig. 2J). These clinical data indicated the prognostic role of CD44v5 in tumor development and progression, consistent with previous findings (31, 39, 40). Collectively, these data show that H1D8 can target ICC tumor cells by binding CD44v5 with high specificity, indicating the potential clinical application of H1D8 as a candidate mAb for ADC establishment.

For a successful ADC, internalization is also a very important capacity in addition to the high specificity and affinity of the selected mAb. To evaluate whether CD44v5 was internalized after H1D8 engagement, we prepared Cy5-labeled H1D8 (H1D8–Cy5 conjugate) through the reaction of Cy5-derived N-hydroxysuccinimide esters with primary amines on the mAb protein N-terminus and on lysine residues. The time-lapse confocal data demonstrated that the H1D8–Cy5 conjugate could initially bind to the membrane of CD44v-OE cells after incubation (Fig. 3A). H1D8–Cy5 was observed to rapidly move from the cell membrane into the cells in a time-dependent manner (Fig. 3A). Unlike immunotoxin, which requires trafficking to the endoplasmic reticulum to release Pseudomonas exotoxin, ADC should travel to lysosomes to release the payload. To verify this phenomenon, we costained lysosomes and the mAb H1D8 and examined the colocalization by using immunofluorescence. After incubation with mAb H1D8 for the indicated time, the confocal imaging data showed that mAb H1D8 was internalized by CD44v-OE cells and specifically localized to the late endosomal/lysosomal LAMP-1–positive compartments, as indicated by the overlap of green and red color (manifested as yellow/orange; Fig. 3B). These data suggested that H1D8 was efficient in inducing internalization after conjugation with CD44v5 and was a suitable candidate for further ADC construction.

Next, we generated an ADC based on the prepared mAb H1D8 after partial reduction of native disulfide (Supplementary Fig. S7A), as described previously (41). The structure of the H1D8–DC is shown in Fig. 3C. The linker mc–vc–PAB was used to attach the hydrophobic cytotoxic drug monomethyl auristatin E (MMAE) to the H1D8 mAb, with a mean drug-to-antibody ratio of 3.7 (Supplementary Fig. S7B). The dipeptide-based linker valine–citrulline was selected because of its reasonable stability under physiological conditions; however, it underwent rapid hydrolysis in the presence of cathepsin B (CTSB) and liberated the toxic payload in lysosomes (42). Drug conjugation had little effect on the binding affinity and specificity of H1D8 (Fig. 3D). To further confirm whether drug conjugation had an influence on internalization, the dynamics of H1D8–DC internalization were quantitatively measured by using the FACS method. The results showed that approximately 50% of H1D8–DC was internalized within 3 hours (Fig. 3E), which was similar to the naked mAb H1D8 properties in previous findings (Fig. 3A). These data illustrated that the prepared H1D8–DC could bind to CD44v-positive cells and was rapidly internalized and trafficked to lysosomes.

Finally, we set out to explore the in vitro cytotoxicity of the established H1D8–DC. To this end, ICC cells with different CD44v5 expression levels were seeded and incubated with increasing concentrations of the H1D8–DC. Cell viability was monitored by using Calcein-AM/PI double staining and quantitatively measured 3 days later via CCK-8 assay. Exposure to H1D8 mediated the killing of HuCCT1 and CD44v-OE RBE cells but not RBE or CD44s-OE RBE cells (Fig. 3F and G). Moreover, the H1D8–DC exhibited durable significant cytotoxicity against CD44v-OE cells in a concentration-dependent manner but had no effects on CD44s-OE cells (Fig. 3H and I). To further confirm the induction of apoptosis following H1D8–DC treatment, we assessed PARP and caspase-3 cleavage as apoptosis markers. The levels of cleaved PARP and cleavage of caspase-3 were elevated in CD44v-OE ICC cells in the presence of H1D8–DC, indicating the proapoptotic role of H1D8–DC (Fig. 3J).

In addition to target expression and target internalization, lysosomal CTSB promotes toxin release via cleavage of the valine–citrulline peptide linker of ADCs (43). Indeed, elevated CTSB expression partially facilitated H1D8–DC efficacy against CD44v5-positive tumor cells (Supplementary Fig. S8A–S8J). Interestingly, we observed that bivatuzumab mertansine, rather than H1D8–DC, exhibited higher cytotoxicity in human keratinocyte cell line (HaCaT; Supplementary Fig. S8K), indicating the potential advantage of H1D8–DC for reduction of the risk of on-target off-tumor side effects. In line with previous findings (44), we also observed that CTSB was specifically overexpressed in multiple types of tumors, rather than in corresponding normal tissues (Supplementary Fig. S9A–S9D), indicating the potential to minimize risk in further clinical investigation. Furthermore, with pretreatment with CTSB, H1D8–DC induced a marked increase in toxicity against CD44v5-negative ICC cells (Fig. 3I and J), indicating the potential bystander effect triggered by the secretion of CTSB from cancer cells in the tumor microenvironment (45). These results not only indicated the potent cytotoxicity but also demonstrated the specificity of H1D8 against CD44v5-positive tumor cells.

H1D8–DC showed good in vitro stability (Supplementary Fig. S10A). Before moving to the in vivo antitumoral assay, we tested H1D8–DC for serum pharmacokinetics and in vivo stability, as described previously (46). Briefly, for serum pharmacokinetics, H1D8–DC and its parental mAb, H1D8, were intravenously injected into nude mice. Serum was collected at the indicated times, and the antibody concentrations were evaluated by ELISA. Pharmacokinetic studies demonstrated that drug conjugation had little effect on H1D8 exposure time, and the half-life (T_1/2) of H1D8–DC was nearly 4 days (Supplementary Fig. S10B). Furthermore, sera collected at different times were further used to treat CD44s/CD44v-OE RBE cells at the same final concentration of H1D8–DC (5 μg/mL). The results showed that their tumor-killing activities were generally comparable, indicating good stability in vivo (Supplementary Fig. S10C).

To further assess the tumor selectivity of H1D8 in vivo, we simultaneously implanted CD44s-OE cells and CD44v-OE cells in the opposite sites of each nude mouse to perform xenograft experiments as shown in Fig. 4A; then, we intravenously injected the mice with a prepared Cy5-labeled H1D8 (H1D8–Cy5 conjugate, 1 mg/kg) or isotype IgG (Isotype–Cy5 conjugate, 1 mg/kg), respectively, to monitor tissue localization. In line with the in vitro analysis, the H1D8 mAb selectively accumulated in the grafted tumors formed by CD44v-OE cells within 6 hours (Fig. 4B). We also carried out detailed tissue distribution analysis of H1D8 using tumor-bearing mouse necropsy samples. A significantly higher H1D8 signal was evident in CD44v-OE cell-derived tumors versus control tumors (Fig. 4C); signals were also observed in the lungs and livers (Fig. 4D and E), similar with previous findings (47, 48). Interestingly, the signals in the lungs and livers rapidly decreased within 24 hours after H1D8–Cy5 conjugate injection, whereas the signal in CD44v-positive tumor tissues remained high and was still detectable after 48 hours (Fig. 4F). These data implied that H1D8 accumulates preferentially in CD44v5-positive tumor tissues in vivo, and the nonspecific accumulation in the liver and lung was rapidly attenuated. In support of these findings, we prepared formaldehyde-fixed paraffin sections and performed IHC assays by using horseradish peroxidase–conjugated goat anti-hIgG (human IgG) antibody to investigate the tissue distribution of H1D8. As predicted, H1D8 was specifically enriched in CD44v-positive tumors and distributed from the boundary region to the core region (Fig. 4G). In the magnified view of tumor tissue, H1D8 was mostly localized in the cell membrane, whereas it could also be detected in the cytoplasm of some cells, which indicated the internalization of H1D8 in vivo. However, no visible staining was observed in other normal tissues (Fig. 4H). These findings in animal models further strengthen the findings regarding the selective anchor receptor CD44v5-mediated retention of H1D8 antibody in grafted tumors.

Next, we evaluated the in vivo efficacy of H1D8–DC. The prepared xenograft mouse models were established as described previously (23). The procedure to assess the in vivo antitumoral efficacy of H1D8–DC is shown in Fig. 5A. Briefly, tumor formation was observed on day 14 post tumor implantation, and tumors were allowed to grow until the tumor volume reached 150 mm³ before treatments. Then, mice bearing tumors were randomized to groups for treatment with intravenous injection of H1D8–DC (5 mg/kg) or isotype ADCs (5 mg/kg) on days 16, 20, and 24. Tumor volume was monitored after treatment. Compared with isotype ADC (Isotype-DC) treatment, H1D8–DC exhibited a significant inhibitory effect against tumors formed by CD44v-OE cells (Fig. 5B–D). Furthermore, although the potent MMAE warhead was cytotoxic to both CD44s-OE and CD44v-OE cells in vitro (Fig. 3I) and presumably toxic in vivo, H1D8–DC preferentially inhibited the growth of tumors formed by CD44v-OE cells versus CD44s-OE cells (Fig. 5B–D). Then, the tumor tissues were isolated, and γH2AX antigen staining, a commonly used genotoxicity assessment method, together with hematoxylin and eosin (H&E) staining, was performed to further investigate the tumor suppression efficiency of H1D8–DC. As shown in Fig. 5E, even though the tumor size in H1D8–DC-treated CD44v-OE tumor was much smaller than those in other groups, a notable increase in tumor necrosis was observed (asterisk). For γH2AX staining, we also found numerous instances of γH2AX-positive cells in CD44v-OE tumors from H1D8–DC-treated xenografted mice (Fig. 5F). Interestingly, the expression of the CD44v5 antigen was still high after H1D8–DC treatment, which might benefit H1D8–DC-mediated durable tumor killing. More importantly, there was no obvious weight loss in the H1D8–DC-treated group throughout the experiment, indicating low potential for systematic toxicity (Fig. 5G). Finally, the toxicities related to major organs were evaluated by histological analysis with H&E staining and serum biochemical assays. As expected, there was no observed serious damage in H&E-stained sections of the lung, liver, spleen, kidney, heart, and brain or in serum biochemical assays (Fig. 5H and I; Supplementary Table S3).

PDX models accurately reflect the genetic and biological diversity required to decipher tumor biology and therapeutic vulnerabilities. To further evaluate the antitumoral efficacy of H1D8–DC in ICC treatment, we generated PDX models by directly engrafting fresh tumor tissue acquired during ICC patient surgery into immunodeficient mice. Elevated expression of CD44v5 was confirmed in two of these primary ICC tumors (Donor 1, medium expression, H-score = 62; Donor 2, high expression, H-score = 127) and the corresponding low-passage PDX tumors examined by IHC (Fig. 6A and B), so these groups were applied for further investigation. Then, the mice were randomized into groups of five mice once the tumor volumes reached approximately 100 to 200 mm³. Each group was intravenously injected with H1D8–DC or isotype as indicated (Fig. 6A). As shown in Fig. 6C, CD44v5-targeted H1D8–DC induced sustained tumor regression in all PDX models, and in one PDX mouse, xenograft was undetectable after H1D8–DC treatment. However, no significant systemic toxicities were observed (Fig. 6D). In line with previous findings, H1D8–DC administration had no impacts on the expression of CD44v5 antigen (Fig. 6E) and induced significant apoptosis of CD44v5-positive tumor cells (Fig. 6F). Moreover, the expression of ICC CSC marker CD133 was decreased after H1D8–DC treatment (Fig. 6F). Taken together, these preclinical models showed that H1D8–DC demonstrated robust antitumoral activity in CD44v5-expressing ICC xenograft models, with no observed toxicity.

AS is an important regulator of most hallmarks of cancer (30–32) and has recently been investigated as a novel source of neoantigens (33). In this study, according to immunohistochemical analysis of an ICC TMA, we have identified that AS of CD44 dominantly occurs in ICC tumor tissues rather than in normal human tissues, indicating the potential role of CD44v as the therapeutic target of ICC. To this end, we generated and tested a panel of CD44v antibodies and subsequently developed a CD44v5-targeted ADC that was proven to induce potent cytotoxicity in vitro against CD44v5-expressing cells and elicit robust antitumor activity in ICC PDX tumors. These data provide a rationale for performing CD44v5-targeted testing and support the clinical efficacy of H1D8–DC for patients with CD44v5-positive ICC.

Targeted therapy aims to deliver drugs to cancer cells based on targeting particular genes or proteins that are specific to cancer cells or the tumor microenvironment while sparing normal cells, and this approach has shown great advantages in clinical cancer intervention. However, the selection of the targeting antigen remains the most important determining factor for targeted therapy. In ICC, recent progress in the development of targeted therapies has been made possible based on comprehensive genetic analysis. Multiple clinical trials that target tumor-specific molecular and genetic aberrations, such as IDH mutations (NCT02073994 and NCT02273739) and FGFR2 fusions (NCT01752920), have been performed. However, their low incidence remains a limitation in the clinical application of these targeted therapies. Recently, Kahles and colleagues (49) provided a comprehensive picture of splicing events and revealed approximately 251,000 exon‒exon junctions that predominantly occurred in tumors rather than in normal tissues. Claudin-18 splice variant 2 (Claudin 18.2), a highly selective gastric lineage marker, has been identified as a pancancer target in primary and metastatic lesions (50), and the first-in-class Claudin 18.2-targeted mAb zolbetuximab obtained significantly superior efficacy and safety comparing with controlled in clinical treatment (51). These data indicate that tumor-specific AS is a good source of neoantigens. We and others have demonstrated that the levels of CD44 splice variants are elevated in cancer cells (23, 31, 39, 52). Similar results were proven in ICC in this study, and the incidence of CD44v5-positive tumors was unexpectedly over 50%, much more than that of either IDH mutation (∼15%) or FGFR2 fusion (∼10%). Furthermore, no significant CD44v5 expression was detectable in normal human tissues, suggesting that CD44v5 splice isoforms have great potential as ideal targets for ICC clinical treatment.

CSCs are a pivotal group of tumor cells with the capacity to self-renew and are responsible for tumorigenesis and malignant transformation. An increase in the abundance of CSCs serves as an early event in the carcinogenesis of cholangiocarcinoma. In most solid cancers, CSCs constitute less than 3% of the total tumor mass, but CSCs constitute over 30% of the tumor mass in ICC (53), which indicates that CSCs might be an emerging target for ICC treatments. To date, many surface markers of CSCs in ICC, such as CD133, CD24, EpCAM, CD117, and CD44, have been revealed. Encouragingly, mAb-based therapeutic agents targeting CSC-associated surface biomarkers have shown great clinical potential.

CD44, as a common marker of CSCs, is expressed in several cell types physiologically. Targeting and inhibiting CD44 through neutralizing antibodies (54), peptide mimetics (55), aptamers (56), and pharmacological inhibitors (57) have shown potent antitumoral activity in many preclinical studies, and some of these strategies have undergone clinical trials (NCT01358903). Increasing studies validate that CD44 variants, rather than the CD44 standard isoform, are more specifically expressed in a variety of cancers (37), particularly those in advanced stages (39, 40, 41), and are of great potential as clinical prognostic indicators (58–60). Tumor cells with CD44v isoform expression but not CD44s isoform expression have enhanced tumor-initiation capacity (61) and resistance to ROS stress (52), thus promoting tumor growth (52) and metastasis (39). Mechanistically, the extra domain formed by variant exons after distinct AS and assembly is able to interact with and sequester different growth factors, as well as cytokines, thus endowing CD44 with additional functions (17). Targeting tumor-specific CD44v via an antibody-based approach (62), a specific peptide (63), and shRNA (64) exhibited promising antitumoral efficacy. In addition, we observed that suppressing AS of CD44 from CD44s to CD44v enhanced immunotherapeutic efficacy in triple-negative breast cancer (23). These studies demonstrate that CD44v might be a promising therapeutic target and shed light on ICC-targeted therapy options.

Unfortunately, owing to the high heterogeneity of CD44 splicing variants and technical difficulties in CD44v detection, analysis and manipulation, little progress has been made in CD44v-targeting therapy. Theoretically, over 800 membrane-bound CD44 isoforms can be generated because of the differential utilization of the variant exons. However, thus far, only a few of them have been identified. Interestingly, the expression patterns of each CD44v isoform seem to vary by cancer type (37). Todaro and colleagues (31) have revealed that CD44v6 predicts a poor prognosis in colon cancer; CD44v8–10 is a specific marker for gastric CSCs (65), whereas CD44v3 is specifically expressed in head and neck cancer CSCs (66, 67). The explanation for this phenomenon is still unclear. The lack of commercially available well-characterized CD44s and various CD44v identification mAbs with standardized protocols contributes. In this work, we interrogated CD44 splicing across normal and ICC tissues based on the TCGA database and revealed, for the first time, that CD44v5 shares a distinct coexpression signature with other variants, as well as CD44s, in ICC (Supplementary Fig. S1E). Our further characterization of CD44v5 expression in ICC clinical specimens with an established mAb (H1D8) with high specificity and affinity not only supported the pivotal role of CD44v5 as a promising tumor-selective target in ICC but also provided a suitable mAb candidate for CD44v5 detection and CD44v5-targeted ADC establishment.

To date, although over 10 ADCs have been approved by the FDA for the clinical treatment of cancer, many ADC candidates have failed to provide favorable benefit-risk profiles, mainly due to toxicities or limited efficacy at tolerable doses (48). Bivatuzumab mertansine, the first and only CD44v6-targeted commercial ADC, is composed of an IgG-type humanized mAb and tubulin inhibitor mertansine (DM1), conjugated by a disulfide linker; it has been discontinued because of severe on-target/off-tumor skin toxicity (21). Unlike bivatuzumab, the humanized anti-CD44v5 mAb H1D8 established in this work showed no significant staining in skin tissue, as well as other normal tissues. In addition, our H1D8–DC with dipeptide linker was remarkably stable under physiological conditions, even after repeated freeze‒thaw cycles (Supplementary Fig. S10A) and was precisely hydrolyzed in the presence of lysosomal CTSB. Most importantly, CTSB was observed to be overexpressed in ICC (Supplementary Fig. S9), as well as various other malignancies, including breast, lung, prostate, and colorectal cancers (44). The low expression of CTSB limited the cytotoxicity of H1D8–DC in the human keratinocyte cell line HaCaT (Supplementary Fig. S8). Consistently, no significant systemic toxicity was observed in xenograft mouse models (Figs. 5G–I and 6D, and Supplementary Table S3). More importantly, the expression of CD44v5 antigen was not decreased after H1D8–DC treatment in ICC mouse models formed by cancer cell lines or PDX, which resulted in a durable antitumoral efficacy of H1D8–DC. However, only one animal showed full regression, whereas most of the others showed stable disease, indicating that H1D8–DC needs further improvement. The major reason might be that the target cells of H1D8- were CD44v5-positive CSCs. Many strategies have been established for targeting and eliminating CSCs, including CSCs signaling pathway inhibition (68), the epigenetic regulator (69), and ADCs (70, 71). Nevertheless, this remains a major challenge. Intrinsic tolerance to cytotoxicity agents (72) and elevated proliferative capacity (Fig. 6B) might contribute to the limitation of H1D8–DC efficacy. Thus, it is urgent and necessary to screen for novel payloads that are capable of inducing potent cytotoxicity to CSCs. In addition to intrinsic resistance, heterogeneous expressions of CSCs markers also make contributions to the limitation of CSCs-targeting ADC efficacy (72). Similarly, CD44v5 was noticed to be unexpressed in some tumor cells (Fig. 6B and E), potentially promoting tumor cells escaping from H1D8–DC-induced cytotoxicity. More importantly, because of the deficiency of the immune system, ADC-induced indirect antitumoral effects via antibody-dependent cellular cytotoxicity, antibody-dependent cellular phagocytosis, and induced immunogenic cell death cannot be well illustrated in PDX models. Furthermore, the pivotal role of ADC-induced CSCs elimination in remodeling immunosuppressive tumor microenvironment (73) is also ignored. Thus, the bona fide efficacy of H1D8–DC needs to be systemically investigated via further clinical trials. Overall, in this study we provide insights regarding CD44v5 as a potential therapeutic target of ICC and a rationale for further clinical investigation of CD44v5-targeted H1D8–DC-based approaches for ICC treatment.

No disclosures were reported.

Y. Bei: Conceptualization, formal analysis, funding acquisition, validation, investigation, visualization, methodology, writing–original draft, project administration. J. He: Resources, data curation, software, formal analysis, writing–review and editing. X. Dong: Formal analysis, validation, investigation. Y. Wang: Validation, investigation, methodology. S. Wang: Data curation, formal analysis. W. Guo: Validation, methodology. C. Cai: Validation, investigation, visualization. Z. Xu: Visualization, methodology. J. Wei: Resources, supervision, methodology. B. Liu: Supervision, project administration. N. Zhang: Resources, supervision, project administration. P. Shen: Conceptualization, supervision, project administration, writing–review and editing.

The authors thank S.Y. Guo (GemPharmatech) for PDX model construction and L. Wu (Nanjing Drum Tower Hospital) for cell lines. The authors also thank Y. Lu (Nanjing University), Y.H. Huang (Nanjing University), and W. Zheng (Nanjing University) for technical assistance. They also thank NPG Language Editing for editorial assistance. This work was supported in part by the grant from Natural Science Foundation of Jiangsu Province (BK20222009), the Key Research and the Nature Science Foundation of Jiangsu Province (BK20210027), the Jiangsu Innovative and Enterpreneurial Talent Program (JSSCBS20211488), and Guangdong Basic and Applied Basic Research Foundation (2021B1515120016).

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.