Therapeutic effects of TM4SF5-targeting chimeric and humanized monoclonal antibodies in hepatocellular and colon cancer models

The transmembrane 4 L six family member 5 (TM4SF5) is aberrantly expressed in hepatocellular and colorectal cancers, and has been implicated in tumor progression, suggesting that it could serve as a novel therapeutic target. Previously, we screened a murine antibody phage-display library to generate a novel monoclonal antibody, Ab27, that is specific to the extracellular loop 2 of TM4SF5. In this study, we evaluated the effects of chimeric Ab27 using cancer cells expressing endogenous TM4SF5 or stably overexpressing TM4SF5 in vivo and in vitro. Monotherapy with Ab27 significantly decreased tumor growth in liver and colon cancer xenograft models, including a sorafenib-resistant model, and decreased the phosphorylation of focal adhesion kinase (FAK), p27Kip1, and signal transducer and activator of transcription 3 (STAT3). No general Ab27 toxicity was observed in vivo. Combination treatment with Ab27 and sorafenib or doxorubicin exerted higher antitumor activity than monotherapy. In addition, we humanized the Ab27 sequence by the complementarity-determining region (CDR) grafting method. The humanized antibody Ab27-hz9 had reduced immunogenicity but exhibited target recognition and antitumor activity comparable with those of Ab27. Both Ab27 and Ab27-hz9 efficiently targeted tumor cells expressing TM4SF5 in vivo. These observations strongly support the further development of Ab27-hz9 as a novel therapeutic agent against liver and colorectal cancers.

The transmembrane 4 L six family member 5 (TM4SF5) is aberrantly expressed in hepatocellular and colorectal cancers, and has been implicated in tumor progression, suggesting that it could serve as a novel therapeutic target. Previously, we screened a murine antibody phage-display library to generate a novel monoclonal antibody, Ab27, that is specific to the extracellular loop 2 of TM4SF5. In this study, we evaluated the effects of chimeric Ab27 using cancer cells expressing endogenous TM4SF5 or stably overexpressing TM4SF5 in vivo and in vitro. Monotherapy with Ab27 significantly decreased tumor growth in liver and colon cancer xenograft models, including a sorafenib-resistant model, and decreased the phosphorylation of focal adhesion kinase (FAK), p27 Kip1 , and signal transducer and activator of transcription 3 (STAT3). No general Ab27 toxicity was observed in vivo. Combination treatment with Ab27 and sorafenib or doxorubicin exerted higher antitumor activity than monotherapy. In addition, we humanized the Ab27 sequence by the complementarity-determining region (CDR) grafting method. The humanized antibody Ab27-hz9 had reduced immunogenicity but exhibited target recognition and antitumor activity comparable with those of Ab27. Both Ab27 and Ab27-hz9 efficiently targeted tumor cells expressing TM4SF5 in vivo. These observations strongly support the further development of Ab27-hz9 as a novel therapeutic agent against liver and colorectal cancers.

INTRODUCTION
Hepatocellular carcinoma (HCC), the most common primary liver cancer, is the second leading cause of cancer-related death. 1,2 The prognosis of HCC is poor, and systemic treatment options are limited. Sorafenib, a multi-kinase inhibitor, is the main systemic therapeutic option in patients with advanced HCC, but the high resistance rate and significant side effects of the drug have significantly limited the benefit of sorafenib therapy. 3 To overcome sorafenib resistance, it will be necessary to identify new therapeutic targets and develop spe-cific therapeutics for advanced HCC. Colorectal cancer (CRC) is among the most lethal and prevalent malignant tumors around the world. 1,4 As with HCC, the prognosis of CRC is not satisfying, especially for patients with metastasis, although there have been advances in treatment options, including the anti-epidermal growth factor receptor (EGFR) antibody cetuximab and the anti-angiogenesis antibody bevacizumab. 4 Effective therapies for both advanced liver cancer and CRC patients are urgently required.
The transmembrane 4 L six family member 5 (TM4SF5) is a member of the transmembrane 4 L six family, similar to the tetraspanin family. 5 TM4SF5 is a cell surface protein with 4 transmembrane domains and 2 extracellular loops. TM4SF5 is highly expressed in diverse cancers, including liver, colon, pancreatic, and esophageal cancers. [6][7][8] Previously, we reported that TM4SF5 plays a critical role in HCC development and metastasis; specifically, it mediates the epithelial-mesenchymal transition (EMT) and proliferation by phosphorylating FAK/c-Src and p27 Kip1 and upregulating the expression of p27 Kip1 , and angiogenesis by upregulating the expression of vascular endothelial growth factor (VEGF). 7,9 It also promotes self-renewal of circulating tumor cells through its association with CD44 and subsequent activation of the c-Src/STAT3/TWIST1/ BMI1 pathway. 10 TM4SF5 crosstalks with integrins a2, a5, and b1 and EGFR to mediate cell migration, tumorigenesis, and drug resistance. 9,11 The interactions of TM4SF5 with integrin a2b1, 12 EGFR, 11 and CD44 10 are probably mediated by its extracellular loop 2 (EC2), which may play a critical role in the promotion of tumors, as suggested by studies of the effects of point mutations of N-glycosylated residues within the EC2 and blocking of the EC2 domain with an anti-TM4SF5 compound, the synthetic chalcone derivative 4 0 -(p-toluenesulfonylamido)-4-hydroxychalcone (TSAHC). 13 These observations suggest that TM4SF5 could be targeted for anticancer therapy using, for example, a TM4SF5targeting monoclonal antibody.
Previously, we identified novel monoclonal antibodies specific to the EC2 domain of human TM4SF5 by screening a murine antibody (single-chain variable fragment; scFv) phage-display library. 14 One of the clones, Ab27, a chimeric antibody of scFv fused to the Fc domain of human immunoglobulin G1 (IgG1) (the scFv-Fc format 15 ), reacted with recombinant TM4SF5 EC2 protein and naive TM4SF5 on the cell surface and was effective in inhibiting cancer cell invasion and proliferation. Ab27 mediated antibody-dependent cell-mediated cytotoxicity. Intratumoral injection of Ab27 efficiently decreased ectopic TM4SF5-expressing tumor growth in vivo. 14 In this study, we evaluated the therapeutic potential of the Ab27 chimeric monoclonal antibody against liver and colon cancer cells expressing endogenous levels of TM4SF5 or stably overexpressing the protein in vitro and in vivo. Furthermore, we generated a humanized monoclonal antibody, Ab27-hz9, and evaluated its reactivity in vitro and in vivo. Ab27 suppressed the tumor growth in liver and colon cancer xenograft models, including sorafenib-resistant HCC, and decreased the phosphorylation of FAK, p27 Kip1 , and STAT3. Combination treatment with Ab27 and sorafenib or doxorubicin resulted in higher antitumor activity than monotherapy. Ab27-hz9 exhibited target recognition and antitumor activity comparable with those of Ab27, indicating that Ab27-hz9 could serve as a starting platform for the further development of clinical treatments for liver cancer and CRC.

RESULTS
Ab27 suppresses TM4SF5-mediated STAT3 activation to inhibit cancer cell growth To determine whether Ab27 specifically recognizes endogenous TM4SF5 on the cancer cell surface, we performed flow cytometry of several types of cancer cells transiently transfected with TM4SF5-specific or negative control small interfering RNA (siRNA). As shown in Figure 1A, Ab27 bound to HCT-116, HT-29 (colon), SNU-398 (liver), SNU-638 (stomach), and C8161 (melanoma) cancer cells expressing endogenous TM4SF5, and suppression of TM4SF5 expression by siRNA decreased Ab27 binding to these cells, confirming the specificity of Ab27. Similarly, immunofluorescence staining revealed that Ab27 substantially stained the cell membrane and cytoplasm of HepG2 ( Figure 1B) and HT-29 ( Figure S1A) cells, but not those of TM4SF5-suppressed cells. In addition, when HCT-116 cells were incubated with Ab27 to allow internalization, the residual level of cell surface-bound Ab27 was substantially reduced ( Figure 1C), indicating that Ab27 was internalized after binding to endogenous TM4SF5 on the cell surface, thereby decreasing the cell-surface level of TM4SF5. This is consistent with our previous observations of ectopically TM4SF5-overexpressing SNU-449Tp cells. 14 In addition, treatment of SNU-449Tp cells with DyLight 488-conjugated Ab27 for 3 h followed by staining with LysoTracker to label lysosomes revealed that Ab27 was localized to lysosomes ( Figure 1D), indicating rapid internalization and lysosomal targeting of Ab27.
Previously, we observed that TM4SF5 activates STAT3 in HCC cells. 10 STAT3 facilitates cancer cell proliferation and survival, tumor spheroid formation, and metastasis. 16 Knockdown of STAT3 in SNU-638, SNU-398, and HT-29 cells decreased cell proliferation (Figure S1B), confirming the role of STAT3 in cell proliferation. In SNU-398, HT-29, HepG2, and C8161 cells, the phosphorylation of STAT3 and subsequent expression of BMI1 were reduced following the suppression of TM4SF5 expression ( Figure 1E), consistent with the results of our previous studies. 10 In addition, analysis of The Cancer Genome Atlas (TCGA)-generated liver hepatocellular carcinoma data (TCGA, PanCancer Atlas) revealed that TM4SF5 mRNA expression was positively correlated with the phosphorylation of STAT3 at Tyr705 ( Figure S1C). Ab27 decreased the phosphorylation of STAT3 and expression of BMI1 in SNU-398 and HT-29 cells ( Figure 1F), and it significantly decreased the anchorage-independent growth of SNU-398 and HT-29 cells in a dose-dependent manner ( Figure 1G). These results indicate that Ab27 suppresses TM4SF5-mediated STAT3 activation, contributing to the reduction of cancer cell growth. Consistent with our previous results, 7,14 the phosphorylation of FAK and p27 Kip1 and the expression of p27 Kip1 were also reduced by TM4SF5 knockdown ( Figure 1E) and Ab27 ( Figure 1F). In addition, Ab27 decreased the expression of vimentin in SNU-398 and HT-29 cells ( Figure 1F), indicating that Ab27 suppresses TM4SF5-induced EMT events.

Ab27 inhibits HCC growth in xenograft mouse models
In a previous study, we observed that the intratumoral injection of Ab27 decreased tumor growth in nude mice bearing TM4SF5-overexpressing SNU-449T 7 (liver cancer) subcutaneous xenografts. 14  (i.p.) injection 3 times per week, for a total of 8 times. In vivo imaging analysis revealed that Ab27 inhibited tumor growth by 64% without affecting body weight (Figure 2A). Similarly, the administration of Ab27 (100 mg/mouse/dose) suppressed tumor growth by 66% when the same tumor cells were subcutaneously injected into the flanks of nude mice ( Figure S2).
We also evaluated the antitumor activity of Ab27 in a sorafenib-resistant SNU-449T 7 cell xenograft model. TM4SF5-overexpressing SNU-449T 7 cells were treated with sorafenib (1-10 mM; 3 culture medium changes per week) for 3 weeks, and the surviving cells (sorafenibresistant SNU-449T 7 ) were injected subcutaneously into the flanks of nude mice. As shown in Figure 2B, the administration of Ab27 (250 mg/mouse/dose) by i.p. injection (3 times per week for a total of 8 times) inhibited tumor growth by 50%, based on mean tumor volume, whereas sorafenib (400 mg/mouse/dose) did not decrease tumor growth, as expected. Immunoblot analysis of xenograft tumor tissues showed that Ab27 decreased the phosphorylation of FAK (moderately), c-Src (moderately), p27 Kip1 , and STAT3, and the expression of BMI1, whereas sorafenib moderately decreased the phosphorylation of FAK, c-Src, p27 Kip1 , and STAT3 and the expression of BMI1 ( Figure 2C). Of note, consistent with our previous observation, 14 the expression of p27 Kip1 was not substantially changed by Ab27 in xenograft models established using SNU-449T 7 cells. This may be due to the phosphorylation of p27 Kip1 at sites other than Ser10, affecting protein stabilization.
We also investigated the antitumor activity of Ab27 in nude mice bearing liver cancer xenografts of SNU-398 cells expressing endogenous TM4SF5. Ab27 (300 mg/mouse/dose), cetuximab (300 mg/mouse/dose), or sorafenib (600 mg/mouse/dose) was i.p. injected at 2-or 3-day intervals a total of 6 times. As shown in Figure 2D, Ab27 inhibited tumor growth by 46% without affecting body weight. The antitumor activity of Ab27 was comparable with that of cetuximab and sorafenib, which inhibited tumor growth by 50% and 51%, respectively. Immunostaining of tumor sections showed that injection of Ab27, cetuximab, or sorafenib significantly decreased the Ki67 proliferation index ( Figure 2E).
Antitumor activity of Ab27 against colon cancer growth in a xenograft model in mice Previously, we observed that high levels of TM4SF5 correlated with worse overall survival of CRC patients, 14 suggesting a potential role for TM4SF5 in colon cancer progression. Therefore, we investigated the antitumor efficacy of Ab27 in nude mice bearing colon cancer xenografts of HT-29 cells expressing endogenous TM4SF5. Ab27 (142 mg/mouse/dose) was i.p. injected twice per week a total of 6 times into xenografted mice. Ab27 administration inhibited tumor growth by 38% without affecting body weight ( Figure S3A). Immunostaining of tumor sections showed that the number of proliferative Ki67 + cells (outside the region of cell death) was significantly lower in tumors from mice injected with Ab27 than in tumors from control mice (Figure S3B). In addition, Ab27-treated tumors exhibited a larger area of cell death within the tumor core than control tumors ( Figure S3C), indicating that Ab27 can accelerate necrosis or apoptosis induced by hypoxia. 17 Consistent with this, Ab27 promoted apoptosis in HT-29 cells grown under suspension conditions in vitro ( Figure S3D).
These results indicate that Ab27 slows tumor growth in vivo by inhibiting tumor cell proliferation and promoting tumor cell death. Immunoblot analysis of tumor tissues revealed that Ab27 decreased the phosphorylation of p27 Kip1 and STAT3 and the expression of p27 Kip1 and BMI1, but had a less substantial effect on the phosphorylation of extracellular signal-regulated kinase 1/2 (ERK1/2) (Figure S3E). These results indicate that Ab27 inhibits tumor growth in nude mice bearing TM4SF5 + liver and colon cancer xenografts and that this occurs concomitantly with a reduction in the phosphorylation levels of STAT3 and p27 Kip1 .

Antitumor efficacy of combined treatment with Ab27 and sorafenib or doxorubicin
Sorafenib is a standard regimen for advanced HCC, and doxorubicin is an anticancer agent used for transcatheter arterial chemoembolization (TACE) of HCC. For combined treatment with Ab27 and sorafenib or doxorubicin, we performed preliminary experiments to evaluate the antitumor efficacies of sorafenib and doxorubicin in the SNU-449T 7 xenograft mouse model. In the next experiment, submaximal schedules or doses of Ab27, sorafenib, and doxorubicin were injected into the SNU-449T 7 xenograft model to examine whether combined treatment with Ab27 and the drug of interest would inhibit tumor growth to a greater extent than Ab27 or drug alone. Ab27 (270 mg/mouse/ dose), sorafenib (1200 mg/mouse/dose), doxorubicin (20 mg/mouse/ dose), a combination of Ab27 (270 mg/mouse/dose) and sorafenib (1200 mg/mouse/dose) or doxorubicin (20 mg/mouse/dose), or saline (control) was i.p. injected twice per week a total of 6 times into nude mice (n = 6 per group) bearing SNU-449T 7 xenografts. As shown in Figure 3A, Ab27, sorafenib, and doxorubicin single treatment inhibited tumor growth relative to the control by 25%, 42%, and 26%, respectively, whereas combined treatment with Ab27 and sorafenib inhibited tumor growth by 54%, and combined treatment with Ab27 and doxorubicin inhibited tumor growth by 52%. Combined treatments inhibited tumor growth to a significantly greater extent than single treatments. Neither combined nor single-agent treatment affected body weight ( Figure 3B). Immunoblot analysis of tumor lysates revealed that, relative to single treatment, combined treatment with Ab27 and sorafenib increased the inhibition of phosphorylation of FAK and p27 Kip1 and that combined treatment with Ab27 and doxorubicin increased the inhibition of the phosphorylation of FAK, p27 Kip1 , and STAT3 ( Figure 3C). Notably in this regard, sorafenib abolished phosphorylation of STAT3, so it was not feasible to determine the effect of combined treatment. Reduction of BMI1 by Ab27 was confirmed, although additive reduction was not observed following combined treatment. These results indicate that combined treatment with Ab27 and sorafenib or doxorubicin inhibits tumor growth more strongly than treatment with antibody or drug alone.

In vivo toxicity of Ab27
We next examined the cross-reactivity of Ab27 with mouse TM4SF5. ELISA showed that Ab27 did not bind to recombinant mouse www.moleculartherapy.org   TM4SF5 EC2 protein as efficiently as it did to recombinant human TM4SF5 EC2 protein ( Figure S4). By contrast, flow cytometry and immunofluorescence analyses showed that Ab27 recognized TM4SF5 on CT26 mouse colon cancer cells ( Figures 4A and 4B). In addition, we transiently transfected PC3 human prostate cancer cells, which normally exhibit low endogenous TM4SF5 expression, with mouse TM4SF5 expression vector. Flow cytometry revealed that Ab27 bound to PC3 cells transfected with mouse TM4SF5 expression vector more efficiently than to PC3 cells transfected with empty vector ( Figure 4C), confirming reactivity with mouse TM4SF5. Therefore, the mouse is an appropriate species for studying safety in vivo.
To evaluate the in vivo toxicity of anti-TM4SF5 antibody, we administered a single dose of Ab27 (48 mg/kg) to ICR mice (female, n = 5) via intravenous (i.v.) injection and analyzed blood samples 28 days after antibody injection. TM4SF5 mRNA is expressed in the liver (https://genecards.org), but we did not detect TM4SF5 protein in this tissue (S.K., unpublished data). Liver function was determined by measuring serum concentrations of general parameters, including alanine aminotransferase (ALT), aspartate aminotransferase (AST), alkaline phosphatase (ALP), g-glutamyl transpeptidase (GGT), bilirubin, and albumin. We observed no general toxicities, including significant changes in liver function or body weight ( Figure 4D), indicating that Ab27 is not significantly toxic in vivo.
Generation and characterization of humanized monoclonal antibody Ab27-hz9 For clinical applications of monoclonal antibodies, murine sequences must be humanized to decrease immunogenicity in humans. 18 Therefore, we constructed a humanized Ab27 antibody by grafting murine complementarity-determining regions (CDRs) into similar human germline sequences. Germline genes have fewer intraclonal somatic hypermutations, which can be recognized as immunogenic. 18 The human germline V genes IGHV1-2 and IGKV3-20 were selected as human acceptor frameworks for the grafting of murine CDRs. However, such CDR grafting often results in partial or complete loss of affinity of the humanized antibody. Therefore, some residues from the murine framework sequences need to be retained by replacing human residues at the corresponding positions to restore some of the lost affinity. 19 Therefore, we also engrafted murine framework sequences H71 and H73 into the human framework. The final humanized antibody was named Ab27-hz9. In silico immunogenicity analysis predicted that the immunogenicity of Ab27-hz9 VH would be lower than that of adalimumab and omalizumab, which are currently used in the clinic, while the immunogenicity of the Ab27-hz9 lightchain variable domain (VL) was predicted to be comparable to those of adalimumab and omalizumab ( Figure S5).
We produced humanized antibody Ab27-hz9 in Expi293 cells (Figure S6) and evaluated its reactivity. The binding affinities (K D ) of Ab27 and Ab27-hz9 for human TM4SF5 (recombinant human TM4SF5 EC2-GST fusion protein) were 2.4 and 4.8 nM ( Figure 5A), respectively, as measured by competitive ELISA. Next, we assessed the antigen-binding kinetics of Ab27 and Ab27-hz9 for human TM4SF5 by surface plasmon resonance (SPR) analysis. The binding affinities of Ab27 and Ab27-hz9 were 5.92 and 8.73 nM ( Figure 5B), respectively, indicating that the affinity of Ab27-hz9 was 1.47-fold lower than that of Ab27. Flow cytometry of SNU-449Cp and SNU-449Tp cells revealed that Ab27-hz9 and Ab27 bound more efficiently to SNU-449Tp cells than to SNU-449Cp cells ( Figure 5C). Immunofluorescence staining also showed that Ab27 and, particularly, Ab27-hz9 stained the membrane edges of SNU-449Tp cells, but not those of SNU-449Cp cells ( Figure 5D). These results indicate that Ab27-hz9 recognizes TM4SF5 on the cell surface at least as strongly as Ab27.
Internalization analysis by flow cytometry revealed that the residual levels of cell surface-bound Ab27-hz9 and Ab27 were substantially reduced after binding to TM4SF5 ( Figure 5E), suggesting that Ab27-hz9 decreased the level of cell-surface TM4SF5 as effectively as Ab27. These findings indicate that the humanized antibody Ab27-hz9 is fully reactive to TM4SF5 protein and recognizes its target on the cell surface more efficiently than the original antibody Ab27.
In vivo antitumor activity of Ab27-hz9 in a liver cancer xenograft model Next, we evaluated the antitumor efficacy of Ab27-hz9 in the SNU-449T 7 xenograft model. Ab27-hz9, Ab27, cetuximab, or normal human IgG (300 mg/mouse/dose) was i.p. injected 3 or 4 times per week (total 12 times) into nude mice bearing the SNU-449T 7 xenograft. Ab27-hz9 and Ab27 inhibited tumor growth by 66% and 62%, respectively, without affecting body weight ( Figure 5F), which is comparable to the 59% inhibition induced by cetuximab. Immunostaining of tumor sections revealed that the level of proliferative Ki67 + cells was significantly lower in tumors from mice injected with Ab27-hz9 or Ab27 than in tumors from control mice ( Figure 5G), indicating that Ab27-hz9 decreased tumor growth by suppressing tumor cell proliferation as efficiently as Ab27.
In vivo tumor targeting of Ab27 and Ab27-hz9 in xenograft models To validate the tumor-targeting ability of anti-TM4SF5 antibodies, we evaluated the distributions of Ab27 and Ab27-hz9 after injection into the SNU-449Tp xenograft model. Ab27, Ab27-hz9, and control human IgG were conjugated with a fluorescent dye (DyLight 755) and i.v. injected into nude mice bearing SNU-449Tp cell-derived tumors.
After 96 h, the distribution of the dye-labeled antibody was quantified by measuring the total photon flux of the fluorescence. Ab27 and Ab27-hz9 were predominantly localized in the tumor, whereas control IgG was mostly in the liver ( Figure 6). In addition, we examined in vivo tumor targeting of Ab27 and Ab27-hz9 using the SNU-398 xenograft model. As shown in Figure 6, both antibodies were predominantly localized to the endogenous TM4SF5-expressing tumor, whereas control IgG was mainly detected in the liver ( Figure S7). These results suggest that Ab27 and Ab27-hz9 can target tumor cells expressing TM4SF5 in vivo.

DISCUSSION
In our previous study, 14 we evaluated the antitumor effect of intratumorally injected Ab27 in a xenograft model established using the  Figure 1C. Of note, Ab27 (0.3 mg/mL) and Ab27-hz9 (0.2 mg/mL) were used to maintain a similar extent (legend continued on next page) Molecular Therapy: Oncolytics stably TM4SF5-expressing liver cancer cell line SNU-449T 7 . We observed that Ab27 significantly inhibited tumor growth, demonstrating the antitumor efficacy of Ab27. However, this study was limited by the administration method and the use of a xenograft model established using the TM4SF5-overexpressing stable cancer cell line. To improve the clinical potential of this antibody, in this study, we evaluated the therapeutic potential of Ab27 using endogenous or stably TM4SF5-expressing liver and colon cancer cells in vivo and in vitro. Our results demonstrate that systemic administration of Ab27 significantly suppressed tumor growth in both endogenous and stably TM4SF5-expressing HCC and colon cancer xenograft models.
In particular, Ab27 efficiently reduced tumor growth in a sorafenibresistant model. Combination treatment with Ab27 and sorafenib or doxorubicin increased antitumor activity relative to monotherapy. Ab27 did not exert general toxicity or interfere with normal liver function in vivo. Consistent with this, TM4SF5-KO (Tm4sf5 À/À ) mice generated in our previous study 20 did not exhibit any abnormal phenotypes at ages up to 24 months (J.W.L., unpublished data). These results suggest that TM4SF5 has potential as a therapeutic target for the treatment of HCC and CRC, and that it is likely that therapeutic blocking of TM4SF5 is safe.
In addition, the humanization process via CDR grafting and backmutation of Ab27 successfully generated Ab27-hz9, which exhibited immunogenicity lower than or comparable to those of antibodies used in the clinic while retaining reactivity to TM4SF5 protein. Therefore, Ab27-hz9 could serve as a platform for the further development of clinical treatments for liver cancer and CRC. We are generating improved antibodies by performing affinity maturation of Ab27-hz9 through mutation of the CDR sequences after conversion of the antibody to a whole-IgG format. The final optimized antibodies will be reported upon completion.
Sorafenib, a multi-kinase inhibitor that inhibits Raf-1, B-Raf, VEGF receptors, and platelet growth factor receptor, is the first-line systemic therapy that has demonstrated a significant increase in mean overall survival in patients with advanced HCC. However, its clinical benefit is modest due to the high resistance rate. 2 Potential mechanisms responsible for the acquisition of sorafenib resistance include EMT, the presence of cancer stem cells, EGFR activation, and autophagy. 3,21 In our previous studies, we showed that TM4SF5 mediates EMT, cell migration, tumorigenesis, and drug resistance by associating with integrins a2, a5, and b1, as well as EGFR. 9,11 TM4SF5-mediated EMT may have an important function in the resistance of lung cancer cells to EGFR kinase inhibitors (e.g., gefitinib). 11 The anti-TM4SF5 compound TSAHC also inhibits TM4SF5-mediated tumor growth in a manner that is distinct from the antitumor effect of sorafenib. 22 Therefore, blocking TM4SF5 may have the potential to overcome sorafenib resistance. In this study, we observed that Ab27 exhibited anti-tumor activity comparable with that of sorafenib and decreased tumor growth in a sorafenib-resistant HCC xenograft model. In addition, combination treatment with Ab27 and sorafenib or doxorubicin exerted higher antitumor activity than either drug alone, suggesting the potential benefit of combination treatments in systemic injection or TACE treatment. Sorafenib inhibits the phosphorylation of STAT3 at Tyr705, thereby suppressing growth and inducing apoptosis in HCC, 23 indicating that STAT3 contributes to the antitumor effects of sorafenib on HCC. In addition, sorafenib-mediated upregulation of STAT3 in HCC cells contributes to sorafenib resistance, 24 and combination treatment with a STAT3 inhibitor and sorafenib synergistically increases the antitumor effect of sorafenib on HCC cells. 25 Thus, STAT3 is not only an indirect target of sorafenib but it also mediates sorafenib resistance. In a previous study, we showed that TM4SF5 activates STAT3 signaling to promote the spheroid formation and self-renewal of HCC. 10 Here, we showed that an anti-TM4SF5 antibody suppressed STAT3 phosphorylation, contributing to antitumor activity in a sorafenib-resistant model and increasing antitumor activity in combination with sorafenib. Therefore, blocking TM4SF5 is a candidate strategy against HCC cells resistant to sorafenib.
EGFR is a receptor tyrosine kinase involved in multiple cancers. EGFR-inhibiting agents such as EGFR-blocking antibodies or EGFR tyrosine kinase inhibitors have been approved to treat various cancers, including colorectal and non-small cell lung cancers. 26 In addition, blocking EGFR could be used to overcome the sorafenib resistance of HCC. 3 TM4SF5 is aberrantly expressed in liver, colon, lung, pancreatic, and esophageal cancers, and its high expression is associated with a poor prognosis. 8,27 We observed that Ab27 and Ab27-hz9 suppressed in vivo tumor growth to the same extent as cetuximab. The anti-TM4SF5 strategy may have therapeutic advantages for the treatment of metastatic CRC and HCC because it targets different pathways from the current therapeutics, and Ab27-hz9 could be exploited as a starting platform to establish anti-TM4SF5 strategy. Further evaluation of Ab27-hz9 in CRC, lung, and pancreatic cancer models, alone or in combination with current therapeutics, would be worthwhile.
Antibody-drug conjugates (ADCs) are monoclonal antibodies conjugated to highly cytotoxic small molecules through chemical linkers and have emerged as a potential strategy for the treatment of several cancers. ADCs enable the targeting of cancer cells and selective delivery of potent cytotoxic drugs. 28 To develop an ADC, the antibody must target a cell-surface receptor that is highly expressed on the targeted cancer cells; in addition, it must be internalized to deliver the drug inside the cells. 28 We observed that Ab27 and Ab27-hz9 bound to cell-surface TM4SF5 and were internalized in tumor cells. In addition, these antibodies efficiently targeted TM4SF5 + tumors in vivo. Based on the internalization and lysosomal targeting of Ab27, we generated an ADC using Ab27 and a DNA crosslinking agent. The Ab27-based ADC exhibited dramatic antitumor activity in vitro and in a colon cancer xenograft model (S.K., unpublished data), suggesting that Ab27-based ADCs could serve as potent therapeutics for the treatment of TM4SF5 + cancers. In addition, immune cell therapy with adoptively transferred chimeric antigen receptor (CAR)-T cells is being actively investigated for the treatment of solid tumors after the reported success of CD19-targeting CAR-T cell therapy against B cell malignancies. 29,30 Thus, it would be interesting to generate and evaluate the effects of anti-TM4SF5 CAR-T cells based on the Ab27-hz9 scFv sequence in HCC and CRC models.
In summary, we evaluated the therapeutic potential of the anti-TM4SF5 chimeric antibody Ab27 using liver and colon cancer cells expressing TM4SF5 in vitro and in vivo. Ab27 suppressed tumor growth in liver and colon cancer xenograft models, including sorafenib-resistant HCC, which occurred concomitantly with a reduction in the phosphorylation levels of FAK, p27 Kip1 , and STAT3. Combination treatment with Ab27 and sorafenib or doxorubicin exerted higher antitumor activity than monotherapy. No general toxicities of Ab27 were observed in vivo. Furthermore, we developed a humanized antibody Ab27-hz9 via the CDR grafting technique. Ab27-hz9 exhibited specific target recognition and antitumor activity comparable with those of Ab27, indicating that Ab27-hz9 could serve as a starting platform for further development of clinical treatments for liver cancer and CRC.

Production of scFv-hFc
The anti-TM4SF5 chimeric monoclonal antibody Ab27 and its humanized monoclonal antibody Ab27-hz9 were produced using the Expi293 or the ExpiCHO Expression system (Thermo Fisher Scientific, Ab27-hz9 Waltham, MA, USA) and purified using protein A affinity chromatography with MabSelect SuRe Column (GE Healthcare, Uppsala, Sweden) or as previously described. 14

Internalization analysis
Cell-surface binding and subsequent internalization of Ab27 and Ab27-hz9 were analyzed using flow cytometry. Briefly, cells were incubated with Ab27 or Ab27-hz9 for 45 min at 4 C, washed to remove unbound antibodies, and then either warmed to 37 C to allow internalization or maintained at 4 C. Cells were transferred to icecold buffer to stop the occurrence of internalization and stained with FITC-conjugated anti-human IgG. PI À cells were analyzed by flow cytometry. The relative levels of Ab27 or Ab27-hz9 bound to the cell surface were calculated as the shift in the fluorescence signal of the antibody occupancy relative to that detected at the beginning of the internalization period.
To detect lysosomal localization of Ab27, SNU-449Tp cells were seeded on coverslips in a 6-well plate and incubated for 48 h at 37 C. Cells were incubated with Ab27 conjugated to DyLight 755 for 3 h at 37 C and stained with 200 nM LysoTracker red DND-99 (Thermo Fisher Scientific) for 2 h at 37 C. The cells were counterstained with DAPI before observation with a confocal microscope.

ELISAs (antigen-binding and competition ELISAs)
The recombinant human EC2-Fc fusion protein (EC2 [amino acid residues 113-157] fused to the Fc of human IgG1) was previously described. 14 Recombinant human EC2-mouse Fc (derived from mouse IgG2a) fusion protein and mouse EC2 (amino acid residues 112-156)-mouse Fc fusion protein were produced in a similar way. A DNA fragment encoding human EC2 followed by glutathione S-transferase (GST) was subcloned into the pET21 vector (Novagen, Darmstadt, Germany). The recombinant human EC2 domain fused with GST was expressed in E. coli BL21 (DE3) and affinity purified using a glutathioneSepharose 4B column (GE Healthcare).
Ab27 scFv sequence was subcloned into the pComb3X vector digested with SfiI. Ab27 scFv tagged with 6ÂHis and HA was expressed in Escherichia coli BL21 by isopropyl b-D-1-thiogalactopyranoside (IPTG) induction (final concentration of 0.1 mM), and a periplasmic extract was obtained using the osmotic shock method. 34 Ab27 of the scFv-6ÂHis-HA format was purified from the periplasmic extract by Ni-NTA affinity chromatography.
For antigen-binding ELISA, 96-well immunoplates (eBiosciences, San Diego, CA, USA) were coated with the purified hEC2-GST protein (100 ng/well) diluted in 50 mM sodium carbonate buffer (pH 9.6) at 4 C overnight and then blocked with 2% BSA in PBS. The plates were washed 3 times with PBS containing 0.05% Tween 20 between all of the steps. Ab27 or Ab27-hz9 (amounts ranging from 0 to 90 ng/well) was added into each well, and then horseradish peroxidase (HRP)-conjugated anti-human Fc was added. All of the incubations were carried out at 37 C for 1 to 2 h. Color was developed with 3,3 0 ,5,5 0 -tetramethylbenzidine (TMB) substrate solution, and the absorbance was measured at 450 nm using a microplate reader (BMG LABTECH GmbH, Ortenber, Germany). Binding of Ab27 scFv-6ÂHis-HA to hEC2-mFc or mEC2-mFc protein was also analyzed using HRP-conjugated anti-HA in a similar manner.
For competition ELISA to determine affinity (K D value), 96-well plates were coated with the hEC2-GST protein. Ab27 or Ab27-hz9 (amount reaching absorbance 1, which was determined from the above antigen-binding ELISA) were incubated with hEC2-GST protein ranging from 0 to 1 mM for 2 h and then the mixtures were put onto the plates for 1 h. HRP-conjugated anti-human Fc incubation, color development, and absorbance measurement were performed as described above.

Construction of humanized antibody Ab27-hz9
Six CDRs determined by combined Kabat/IMGT/Paratome numbering 35 and some framework residues (H71 and H73) of the Ab27 were grafted into the human IGHV1-2 and IGKV3-20 frameworks. Humanized heavy-chain variable domain (VH) and kappa chain variable domain (Vk)-encoding genes were synthesized as an www.moleculartherapy.org scFv format and then inserted into the pDR-OriP-Fc1 mammalian expression vector as described previously. 14

SPR analysis
The kinetic parameters of the interaction between the scFv-Fc forms of Ab27 and Ab27-hz9 and hEC2-GST were determined at 25 C using the Biacore T200 (Cytiva, Marlborough, MA, USA). Briefly, antihuman Fc antibody was aminecoupled on a CM5 sensor chip using the Human Antibody Capture Kit (Cytiva) and then scFv-Fc was injected for 30 s at a flow rate of 10 mL/min according to the manufacturer's instructions. hEC2-GST in HBS-EP buffer with 350 mM NaCl at a concentration range of 12.5-100 nM was then injected over 3 min at a flow rate of 30 mL/min. After each binding cycle, a regeneration solution (3 M MgCl 2 ) was injected for 1 min to remove any non-covalently bound protein. The association rate (k a ), dissociation rate (k d ), and equilibrium dissociation constant (K D , k d /k a ) were determined using a 1:1 binding model and Biacore BIAevaluation software version 1.0.

Cell proliferation assay
Cell proliferation was determined using the colorimetric WST-1 Cell Proliferation Assay Kit (Takara Bio, Otsu, Shiga, Japan). Briefly, cells were seeded into 96-well plates at a density of 5 Â 10 3 cells/well and incubated for 48 to 72 h in the presence of antibodies. The cells were then incubated with the WST-1 reagent (1/10th of the medium volume), and formazan dye formation was determined by measuring absorbance at 450 nm using a microplate reader.

Soft agar anchorage-independent growth assay
Cells were seeded at a density of 1 Â 10 3 cells/well in 6-well tissue culture plates in 0.4% agar (Sigma) over a 0.6% agar feed layer. Cells were allowed to grow at 37 C in 5% CO 2 for 13 days, and the number of resulting colonies with diameter >0.3 or 0.5 mm was counted per well.

Anoikis assay
Cells (5 Â 10 5 ) were seeded in the presence of antibodies into 6-well plates with an Ultra-Low Attachment Surface (Corning, Corning, NY, USA) to induce anoikis. Cells were washed and stained with 5 mL of annexin V and 5 mL of PI per 1 Â 10 5 cells for 15 min at room temperature (RT) in the dark, and the percentage of apoptotic cells was analyzed using flow cytometry. The cells were harvested after the induction of anoikis, washed with PBS, and lysed for immunoblot analysis. HA6, a scFv-Fc recognizing the hepatitis A virus, 14 was used as a negative control antibody. old) were obtained from Japan SLC (Nishi-ku, Hamamatsu, Shizuoka, Japan) or Nara Biotech (Seoul, Korea). SNU-449T 7 -luc (stably overexpressing TM4SF5 and luciferase) cells 10 were injected orthotopically into the liver (5 Â 10 5 cells) after minimal incision or subcutaneously into the back (1 Â 10 6 cells) of mice. On day 7 or 14, the mice were randomized into control and treatment groups. Ab27 (100 mg/mouse) was i.p. injected at 2-or 3-day intervals for 2 or 3 weeks (total 7 or 8 times). Bioluminescence from SNU-449T7 cells were acquired in an IVIS Luminar imaging system (PerkinElmer, Santa Clara, CA, USA), as previously described. 10 SNU-449T 7 cells were treated with sorafenib (1-10 mM) for 2 weeks, and the surviving cells (sorafenib-resistant SNU-449T 7 ; 1 Â 10 6 cells) were mixed with Matrigel on ice and then injected subcutaneously into the backs of mice. After 11 days, when tumor volumes reached $100 mm 3 , the tumor-bearing mice were randomized into control and treatment groups (n = 7 per group). Ab27 (250 mg/mouse) or sorafenib (400 mg/mouse) was i.p. injected at 2-or 3-day intervals (total 8 times).
SNU-449T 7 (3 Â 10 6 ) cells were subcutaneously injected into the flanks of mice. When volumes reached $100 mm 3 , the tumor-bearing mice were randomized into control and treatment groups (n = 6 per group). Ab27, Ab27-hz9, or cetuximab (300 mg/mouse) was i.p. injected at 2-or 3-day intervals (total 12 times). Normal human IgG was injected as a negative control. Body weight and tumor volume were measured before injection of the antibody or drug. Tumor volumes were calculated using the formula width 2 Â length/2. Tumor masses were lysed as previously described 7 for immunoblot analysis, or tumor masses were fixed in 10% formalin and embedded in paraffin.

Immunohistochemistry
Formalin-fixed and paraffin-embedded 6-mm-thick tissue sections from xenograft tumors were processed for immunohistochemistry analysis as per the standard protocol. Sections were stained with anti-Ki67 antibody (SP6; Abcam) using the peroxidase technique. The proliferative index (%) was determined by calculating the number of Ki67 + cells relative to the total number of cells, which consisted of at least 1,000 cells per field. Five randomly selected fields from tumor sections per mouse were analyzed using ImageJ software.

In vivo toxicity of Ab27
In vivo toxicity testing of Ab27 was performed by KBIOhealth (Osong, Korea). Six-week-old ICR mice (female, n = 5) were i.v. injected with 48 mg/kg of Ab27. General clinical signs, including body weight, were monitored. Twenty-eight days after injection, the mice were sacrificed, and urine and blood samples were collected. Standard hematological analysis and urinalysis were performed. Blood samples were centrifuged at 1,700 Â g for 10 min, and biochemical parameters of serum were analyzed using a Konelab 60i (Thermo Fisher Scientific, Vantaa, Finland).

In vivo tumor targeting
Ab27, Ab27-hz9, and normal human IgG were conjugated with Dy-Light 755 and purified using a DyLight 755 Antibody Labeling Kit (Thermo Fisher Scientific). Seventy micrograms of dye-labeled antibody were injected into the tail vein of nude mice bearing SNU-449T 7 or SNU-398 cell-derived tumors at a tumor size of $100 mm 3 . The distribution profiles of the antibody were quantified by in vivo fluorescence using an in vivo imaging system (IVIS) at 0, 24, 48, 72, and 96 h after antibody injection. The tumor was removed at 96 h after injection to determine the distribution of the DyLight755labeled antibody in tumor tissues.
Analysis of TCGA data cBioPortal (www.cbioportal.org) 36,37 was used to analyze TCGAgenerated human liver HCC data (TCGA, PanCancer Atlas). All of the patient samples with available mRNA and protein expression profiles were included in the correlation analysis.

Statistical analysis
For the statistical analysis, all of the data points were tested for normality using either the D'Agostino-Pearson or Shapiro-Wilk normality test. The Student's t test or one-way ANOVA was performed for those data points that passed normality with Dunnett's multiple comparison post-test. The Mann-Whitney test or the Kruskal-Wallis test was performed for those data points that did not pass normality. All of the statistical calculations were done using Prism software (GraphPad, San Diego, CA, USA). Pearson's test was performed for correlation analysis. p < 0.05 was considered statistically significant.