Lenvatinib, Toripalimab plus FOLFOX Chemotherapy in Hepatocellular Carcinoma Patients with Extrahepatic Metastasis: A Biomolecular Exploratory, Phase II Trial (LTSC)

This article is featured in Selected Articles from This Issue, p. 4991

Liver cancer, specifically hepatocellular carcinoma (HCC), is the third leading cause of cancer-related death, and its incidence is increasing (1–3). Despite improvement in HCC surveillance, most patients are diagnosed at advanced disease stages, accompanied by major vascular invasion or extrahepatic metastasis (3, 4). Atezolizumab plus bevacizumab have recently been approved as first-line treatment for patients with advanced HCC (5–7). However, approximately 30% of patients with advanced HCC present extensive extrahepatic metastasis and high intrahepatic tumor burden (8), which are closely associated with poor prognosis after systemic treatments (9). Recently, an updated IMbrave150 study showed that atezolizumab and bevacizumab exhibited limited benefits in patients with high-risk features, such as Vp-4, bile duct invasion, and tumor occupancy ≥50% of the liver, with a median overall survival (OS) of only 7.6 months (9).

In contrast, our previous studies showed that hepatic arterial infusion chemotherapy with oxaliplatin, 5-fluorouracil, and leucovorin (FOLFOX-HAIC) in combination with systemic therapy yielded significantly longer OS and acceptable safety profiles in high-risk patients with unresectable HCC (10–13). However, FOLFOX-HAIC is a locoregional therapy, which is not easy to popularize and has lower OS in patients with extrahepatic metastasis (9.8 months) than in patients without extrahepatic metastasis (14.8 months; ref. 14). As an alternative treatment to FOLFOX-HAIC, systemic chemotherapy with FOLFOX regimen is recommended for patients with extrahepatic metastasis according to the EACH study in China (15–17). However, the EACH study did not fulfill the primary endpoint and is limited to Asian patients, so that FOLFOX is not recommended in Western countries.

Preclinical studies have provided rationale for the therapeutic potential of combining tyrosine kinase inhibitors (TKI) and immune checkpoint inhibitors (ICI) with traditional chemotherapy for the treatment of advanced HCC (18–22). A recent phase II study showed that sintilimab combined with apatinib and capecitabine had good safety and antitumor activity as a first-line treatment for unresectable HCC (23). In this study, we performed a single-arm, phase II (LTSC) study to prospectively investigate the efficacy, safety, and predictive biomarkers of lenvatinib (an oral TKI that targets VEGFR 1–3, FGFR 1–4, PDGFR α, RET, and KIT), toripalimab [a humanized IgG4K monoclonal antibody specific for programmed cell death protein 1 (PD-1)], plus FOLFOX chemotherapy as the first-line treatment in patients with high-risk HCC with extrahepatic metastasis. Single-nucleus RNA sequencing (snRNA-seq) of tumor tissues for biomarker exploration was also performed to promote precision medicine for the treatment of advanced HCC.

The LTSC was performed at the Sun Yet-sen University Cancer Center in China. The trial protocol is presented in Supplement S1. This study was approved by the Institutional Review Board of Sun Yat-sen University Cancer Center and performed in accordance with the Declaration of Helsinki. The LTSC trial was registered with ClinicalTrials.gov (NCT04170179), and all patients provided written informed consent before enrollment.

Eligible patients were 18 years or older and diagnosed with advanced HCC (24). High-risk features were allowed in this study and identified according to IMbrave150 study (9). Other key inclusion criteria included the following: presence of extrahepatic metastasis, no previous treatment for HCC (including systemic and locoregional therapies), and at least one measurable tumor lesions based on RECIST version 1.1 (25). The detailed inclusion and exclusion criteria are available in the eMethods in Supplement S2. The study participant demographics were considered to be representative for the general Chinese population (Supplementary Table S1).

Patients received 21-day treatment cycles of lenvatinib, toripalimab, plus FOLFOX chemotherapy. Once chemotherapy intolerance occurred, or at the end of six cycles, the other study treatments continued as maintenance therapy until radiographic progression, death, intolerable toxicities, patient withdrawal, or investigator's decision. The criteria information for dose reduction, interruption, and discontinuation of the study treatments, follow-up procedures, and tumor response assessment are available in the eMethods in Supplement S2.

The primary endpoint was the progression-free survival (PFS) rate at 6 months, which was defined as the proportion of patients alive, assessable, and free from disease progression at 6 months. Secondary endpoints included PFS, OS, objective response rate (ORR), disease control rate (DCR), and safety. The definitions of clinical outcomes are provided in the eMethods in Supplement S2. Adverse events (AE) were evaluated according to the NCI CTCAE Version 4.03.

To identify potential biomarkers, biopsy samples of tumor tissues were collected and frozen before the initiation of study treatments. Whole-genome transcriptome profiling of tumor biopsies from training and validation cohorts was performed by snRNA-seq. The detailed information is available in the eMethods in Supplement S2.

We used Simon's two-stage design with a one-sided α error of 5% and a power of 80% (26), and the sample size for this study was 30. The detail was written in the eMethods in Supplement S2. The primary efficacy analysis was performed in the intention-to-treat population, and safety analysis included all enrolled patients who received at least one dose of the study treatment. OS and PFS with 95% confidence intervals (95% CI) were analyzed using the Kaplan–Meier method. SPSS (RRID: SCR_002865) and GraphPad Prism (RRID: SCR_002798) were utilized for statistical analyses, and bioinformatics analyses of snRNA-seq and bulk RNA-seq data were performed using R version 4.2.0.

The data generated in this study are not publicly available due to the patient privacy requirements but are available upon reasonable request from the corresponding author.

Between November 19, 2019 and July 4, 2021, 30 patients who fulfilled the inclusion and exclusion criteria were enrolled and received study treatments (Supplementary Fig. S1), among whom 27 (90.0%) were male. The main etiology of HCC was hepatitis B virus (HBV) infection (96.7%), with a median age of 45.5 years [interquartile range (IQR): 38–52.5]. The baseline demographic and clinical characteristics of the population are summarized in Table 1. All patients exhibited extrahepatic metastasis and a high intrahepatic tumor burden at diagnosis, and 25 (83.3%) participants exhibited major vascular invasion. The median size of the maximum tumor lesion measured by RECIST criteria was 10.2 cm (IQR, 8.3–12.6 cm). In general, 24 (80.0%) participants exhibited high-risk features, including Vp4, bile duct invasion, or tumor involvement ≥50% of the liver.

Treatment administration is listed in Supplementary Table S2. A total of 30 patients received at least one treatment cycle, and 11 (36.7%) patients completed six cycles of chemotherapy. The median treatment duration of lenvatinib was 8.5 months. At the date of the clinical data cutoff (July 8, 2022), 3 (10.0%) participants continued maintenance therapy with lenvatinib and/or toripalimab and were still free from disease progression. After termination of the study treatments, 26 (86.7%) participants received second-line treatment, including FOLFOX-HAIC (8), liver resection (1), ablation (1), transarterial chemoembolization (TACE; 2), radiotherapy (2), and other TKIs or TKIs plus PD-1 antibodies (12).

In the first stage, 10 participants were enrolled and the PFS of 7 participants was longer than 6 months, which allowed the LTSC study to continue to the second stage. Overall, 20 patients with PFS higher than 6 months were observed, and the LTSC study met the primary endpoint. Furthermore, 25 patients showed disease progression or died, and the primary endpoint showed a 6-month PFS rate of 66.7%, with a median PFS of 9.73 months (95% CI, 2.89–16.58, Fig. 1A). The intrahepatic and extrahepatic PFS were 9.73 months (95% CI, 2.89–16.58) and 11.67 months (95% CI, 9.48–13.85; Fig. 1B and C), respectively. The median OS was 14.63 months (95% CI, 11.77–17.50; Fig. 1D), with a 12-month OS rate of 60%. After stratification by the absence or presence of high-risk features, the median PFS was 8.3 months (95% CI, 3.02–13.58; Fig. 1E) and OS was 13.7 months (95% CI, 9.24–18.16; Fig. 1F) in patients with high-risk HCC.

The antitumor activity of the triple combination therapy is summarized in Fig. 2A–C, and the tumor response are shown in Supplementary Table S3. Among the 29 patients who underwent tumor response evaluation with the radiologic imaging, 13 (43.3%) achieved a partial response according to RECIST v1.1. Furthermore, according to the mRECIST criteria, 3 patients achieved a complete response, and the ORR and DCR were 56.7% and 83.3%, respectively. In addition, contrast-enhanced CT and MRI scans of 8 representative patients who received the triple combination therapy are shown in the Supplementary Figs. S2 to S9.

All patients who received at least one dose of the study treatments (safety analysis) were monitored for the development of treatment-related AEs and all AEs (Table 2). Treatment-related AEs of any grade, regardless of causality, were reported by all participants in the LTSC study. The main treatment-related AEs (≥30%) were neutropenia, anemia, thrombocytopenia, abdominal pain, diarrhea, vomiting, and mild liver dysfunction, including increased alanine aminotransferase (ALT) and aspartate aminotransferase (AST) levels. The most common grades 3 to 4 treatment-related AEs were increased ALT (13.3%), increased AST (10.0%) and neutropenia (10.0%). The serious immune-related AEs were observed in 3 (10.0%) patients, including dermatitis (n = 2) and hepatitis (n = 1). After discontinuation of toripalimab and treatment with prednisone, these 3 patients recovered.

The toxicities that led to trial discontinuation in 5 patients were thrombocytopenia, fatigue, increased AST, ascites, and infection. Treatment-related AEs led to treatment interruption, dose reduction, and discontinuation of lenvatinib in 13 (43.3%), 11 (36.7%), and 4 (13.3%) participants, respectively. Treatment-related AEs led to treatment interruption and discontinuation of toripalimab in 10 (33.3%) and 8 (26.7%) participants, respectively. Treatment-related AEs led to interruption, dose reduction, and discontinuation of FOLFOX chemotherapy in 8 (26.7%), 3 (10.0%), and 4 (13.3%) participants, respectively. The special AEs that led to treatment interruption and discontinuation are shown in the eResults in Supplement S2.

To determine potential cell subpopulations related to clinical responses or resistance to the triple combination treatment, we performed snRNA-seq of tumor biopsy tissues derived from 6 participants recruited in the LTSC study (3 responders defined as having a complete and partial response, and 3 nonresponders defined as having a stable and progressive disease). The detailed baseline information for this population is summarized in Fig. 3A and Supplementary Table S4.

In this training cohort, we obtained the transcriptomic profiles of 21,601 high-quality single nuclei after removing low-quality cells and doublets. Unsupervised clustering of single-nucleus profiles identified 13 cell clusters, including hepatocytes, immune cells, and stromal cells (Fig. 3B; Supplementary Figs. S10A and S10B). To further identify these heterogeneous cell populations, we reanalyzed malignant hepatocytes by identifying marker genes for cell clusters and evaluating their correlation with clinical response (Fig. 3C; Supplementary Fig. S10C). We identified three cell subtypes that delineated responsive or nonresponsive tumors, including one cluster (C06) enriched in responders and two clusters (C04 and C05) enriched in nonresponders (Fig. 3D; Supplementary Fig. S10D).

Malignant hepatocytes in the C04 cluster were enriched in nonresponsive tumors and expressed higher levels of complement system-related genes, such as CFHR4 and CFH (Fig. 3E), which were involved in the maintenance of stemness in liver cancer stem cells (27), suggesting that CFHR4⁺ malignant hepatocytes might promote resistance to triple combination therapy by maintaining the stemness features of HCC. Malignant hepatocytes in the C05 cluster were enriched in nonresponders and highly expressed AKR1C2 (Fig. 3E), which stimulated cell migration and metastasis in liver cancer (28) and was associated with cisplatin resistance (29), suggesting that AKR1C2⁺ malignant cells induced chemo-resistance by promoting HCC progression.

To generate a landscape of the global cellular microenvironment of advanced HCCs and draw more robust conclusions, snRNA-seq of adequate tumor biopsy samples was performed from 10 additional patients who received the same combination therapy, but were excluded from the LTSC study, as a validation cohort (Fig. 3A).

In this validation cohort, Louvain clustering of single-nucleus profiles identified 20 distinct major cell clusters representing epithelial, immune, endothelial, and fibroblast populations. We annotated these 20 cell clusters with canonical marker genes for major cell types, which consisted of B cells (BANK1; MS4A1), T cells (CD247; CD2; NKG7), plasma B cells (MZB1; IGHG1), myeloid cells (CD163; MRC1; FCN1), fibroblasts (ACTA2; COL1A2), endothelial cells (CDH5; VWF), and hepatocytes (HNF4A; GPC3; Fig. 3F; Supplementary Fig. S10E and S10F). Although all major cell types were shared among the patients, the infiltration levels were different, possibly reflecting substantial heterogeneity and different responses to the triple combination therapy (Supplementary Fig. S10G). We then identified 42,558 malignant hepatocytes that were separated into 12 clusters (Fig. 3G), enabling a systematic exploration of the heterogeneity in responsive or nonresponsive tumors. High intertumoral heterogeneity of hepatocytes was also observed, as three hepatocyte subclusters (C08, C09, and C12) were specifically associated with individual patients (Supplementary Fig. S10H). Hepatocytes in the Val_C01_HCC cluster highly expressed CFHR4, whereas hepatocytes in the Val_C02_HCC cluster expressed higher levels of AKR1C2, and these malignant hepatocyte subtypes shared many common marker genes (Fig. 3E; Supplementary Fig. S11A). The detailed differential expression genes are listed in Supplementary Tables S5 and S6. Next, we integrated the training and validation cohort data to validate the malignant hepatocytes. Cluster similarity analyses exhibited that CFHR4⁺ hepatocytes shared higher similarity with Val_C01_HCC cluster, and AKR1C2⁺ hepatocytes shared higher similarity with Val_C02_HCC cluster in the validation cohort. However, RPS12⁺ hepatocytes enriched in responders, did not share high similarity with any other cluster in the validation cohort (Fig. 3H; Supplementary Fig. S11B). Consistent with the clinical relevance of CFHR4⁺ and AKR1C2⁺ malignant hepatocytes, we observed an increased proportion of hepatocytes in Val_C01_HCC and Val_C02_HCC clusters in nonresponders (Fig. 3I), even though the Val_C01_HCC cluster did not reach statistical significance.

To resolve the substantial differences in malignant cells between responders and nonresponders, we performed pathway analysis of malignant hepatocytes from the training cohort. We observed an enrichment of cell-cycle–related pathways in responders (e.g., WNT-β-catenin signaling, E2F-targets, and G2M-checkpoint pathway), whereas the metabolism-related and immune response pathways were enriched in nonresponders (e.g., bile acid metabolism, xenobiotic metabolism, and TNFA-signaling-via-NF-κβ; Fig. 3J), and we obtained similar results in the validation cohort (Fig. 3K). We also observed that responders expressed higher proliferative and lower hepatic function signals, which were validated in the validation cohort (Fig. 3L; Supplementary Fig. S11C). Given that lenvatinib was one of the study treatments for the combination therapy, we explored the expression profile of major targets of lenvatinib in malignant cells. In both training and validation cohorts, the expression scores of lenvatinib targets were significantly higher in responders than in nonresponders (Fig. 3L; Supplementary Fig. S11C). To further investigate the potential functions of these two malignant subtypes identified in the training cohort, we utilized the hallmark gene sets of the Molecular Signatures Database (MsigDB; ref. 30) to analyze the alterations in pathways. Both CFHR4⁺ and AKR1C2⁺ hepatocytes exhibited a greater enrichment of metabolism-related pathways, including xenobiotic metabolism, bile acid metabolism, glycolysis, fatty acid metabolism, and cholesterol homeostasis. These observations were confirmed in the Val_C01_HCC and Val_C02_HCC clusters in the validation cohort (Fig. 4A). In addition, KEGG pathway enrichment analysis clearly demonstrated distinct metabolic features among these hepatocytes. CFHR4⁺ and AKR1C2⁺ hepatocytes exhibited greater enrichment of glutathione metabolism, steroid hormone biosynthesis, and TCA cycle (Supplementary Fig. S11D), indicating that these hepatocytes might be driven by elevated metabolic processes. These results suggested the metabolic landscape of malignant hepatocytes was closely associated with clinical response to this combination therapy. We next evaluated the association between these two hepatocyte subtypes and previously reported molecular subclasses of HCC (31, 32). And we observed that CFHR4⁺ and AKR1C2⁺ malignant hepatocytes were mainly enriched in S3 subclass of the Hoshida classification (Fig. 4B).

As the sample size of our snRNA-seq dataset was limited, we utilized a deconvolution algorithm Bayesprism (33) to simulate the cell subtype-specific gene expression profile to predict the abundance of each cell type quantified by scRNA-seq in large-scale datasets from TCGA and Fudan HCC cohorts (34, 35). To assess the clinical relevance of infiltration of each hepatocyte subtype, we investigated the correlation between hepatocyte subtype infiltration and OS in patients with HCC. These results revealed that a higher abundance of CFHR4⁺ or AKR1C2⁺ hepatocytes in tumors was strongly associated with worse OS (Fig. 4C). Notably, we observed similar survival trends for Val_C02_HCC cluster in the validation cohort, and patients with higher infiltration of Val_C02_HCC exhibited shorter OS (Supplementary Figs. S11E to S11F). Furthermore, we evaluated the expression levels of several gene signatures that were previously reported to be associated with clinical response to immunotherapy. Notably, the presence of CFHR4⁺ and AKR1C2⁺ hepatocytes exhibited significantly lower expression scores of the immune-related signatures (e.g., cytotoxic activity, TLS, and CD8 T-cell signature), suggesting that these hepatocytes might induce “immune-excluded” phenotypes and lack of responses to immunotherapy (Supplementary Figs. S12A to S12C).

Given that one of the treatments in this combination was toripalimab, the composition and infiltration levels of immune cells in the tumor microenvironment might be the key factors affecting the therapeutic effect. In the validation cohort (n = 10), we identified a total of 23,203 immune cells that were then separated into 17 populations, including six clusters for macrophages [C01-Mph-F13A1, C02-Mph-GPNMB, C03-Mph-Undefined, C05-Mph-MARCO, C08-Mph-ICAM1, C12-Mph-MKI67), five subtypes of T cells (C04-T-TTN, C06-T-TNIP3, C07-T-IL7R, C10-Treg-IL2RA, C13-T-MKI67), two for dendritic cells (C16-cDC-FLT3, C17-pDC-CLEC4C), and four other cell clusters (C09-NK-KLRD1, C11-Mono-FCN1, C14-B-BANK1, C15-Plasma-IGHG4); Fig. 4D–E; Supplementary Figs. S13A and S13B]. The infiltration levels of B cells, plasma B cells, and dendritic cells were relatively low in all patients. Nonetheless, the infiltration levels of intratumoral T and myeloid cells showed substantial heterogeneity across all patients (Supplementary Figs. S13C and S13D). As expected, the proportion of total immune cells was significantly higher in responders than in nonresponders (P = 0.032). The proportions of total T cells and macrophages displayed a similar trend, although the difference was not statistically significant (P = 0.056, Fig. 4F; Supplementary Fig. S13E). Four immune subtypes (C02-Mph-GPNMB, C08-Mph-ICAM1, C10-Treg-IL2RA, C13-T-MKI67) were significantly enriched in responders, and three subtypes (C06-T-TNIP3, C11-Mono-FCN1, C16-cDC-FLT3) exhibited a rising trend of proportion (Fig. 4G; Supplementary Fig. S13F). Specifically, macrophages in the C02-Mph-GPNMB cluster were characterized by higher expression levels of TAM-related genes, including GPNMB and SPP1. C08-Mph-ICAM1 cluster highly expressed IL4I1, GBP1, and ANKRD22, which have been reported as inflammatory regulators expressed in IFNγ-induced macrophages (36, 37), indicating that they were a set of pro-inflammatory macrophages (Fig. 4H). Concerning T cells, C06-T-TNIP3 cells highly expressed genes associated with cytotoxicity (GZMK, IFNG, GRP174) and moderately expressed genes related with exhaustion (PDCD1, HAVCR2, LAG3). Because C10-Treg-IL2RA cluster showed the highest expression of FOXP3, IL2RA, IKZF2, TIGIT, and CTLA4, we designated it as regulatory T cells. C13-T-MKI67 cells were designated as proliferative T cells, due to overexpressing cell-cycle-related genes (MKI67, TOP2A, and STMN1; Fig. 4I).

Taken together, a comprehensive analysis of our snRNA-seq profiles identified two subtypes of malignant hepatocytes (CFHR4⁺ and AKR1C2⁺), which exhibited potential predictive values for resistance to this triple combination therapy. In addition, patients with higher abundance of tumor-infiltrating immune cells were associated with superior clinical outcomes.

The clinical data of patients with advanced HCC having extrahepatic metastasis and a high intrahepatic tumor burden are limited, especially for high-risk patients, because they are usually excluded from other contemporary phase III trials (38, 39). Unlike previous or ongoing clinical trials, the LTSC study mainly focused on patients with advanced HCC having both extensive extrahepatic metastasis and high intrahepatic tumor burden. The baseline characteristics of patients in this study differed from those in trials of systemic therapies (5, 40). In the LTSC study, enrolled participants were diagnosed with extrahepatic spread, with a median diameter of intrahepatic lesions of 10.2 cm (IQR: 8.3–12.6 cm), and 80.0% of the subjects were at high risk. However, in the IMbrave150 trial, only 20.2% of the enrolled participants exhibited high-risk features. The clinical benefits of atezolizumab plus bevacizumab in high-risk populations are limited, with a median OS of only 7.6 months (9).

This study demonstrates the potential synergistic effects and promising preliminary efficacy of this combination strategy, with a 6-month PFS rate of 66.7%. Furthermore, we found a DCR of 83.3% and an ORR of 43.3%, with a median PFS of 9.73 months, and median OS of 14.63 months, suggesting that this combination therapy exerts better antitumor activity than systemic chemotherapy, TKI, and ICI monotherapy in patients with advanced HCC. Post-hoc analyses also demonstrated a superior OS of 13.7 months in high-risk populations receiving lenvatinib, toripalimab, plus FOLFOX chemotherapy, which was higher than the 7.6 months of IMbrave150. Moreover, the primary endpoints of the LEAP-002 trail did not meet the pre-specified statistical significance (OS: 21.2 months vs. 19.0 months; ref. 41), suggesting that combination therapy might not be superior to monotherapy in patients with a low tumor burden. Notably, most patients in this study presented with high-risk features (Vp4, bile duct invasion, or tumor involvement ≥50% of the liver). Therefore, the results reported herein are unlikely to be fully translatable to Western countries.

The potent antitumor activity of this combination therapy may be mainly attributed to the synergistic effects of lenvatinib, toripalimab, and traditional chemotherapy. First, the anti-angiogenic effects of TKIs, such as lenvatinib, can reprogram the structure and function of the tumor vasculature by promoting tumor vascular normalization, thereby increasing the drug delivery of cytotoxic agents and enhancing antitumor responses (18, 42). Second, FOLFOX chemotherapy induces the production and release of tumor antigens, thereby increasing tumor immunogenicity, activating the adaptive immune system, and sensitizing the response to toripalimab (22). Furthermore, the synergistic effects of this combination strategy were implicated in our previous study on lenvatinib, toripalimab, plus FOLFOX-HAIC, which yielded infusive results in high-risk patients (13). Notably, 8 patients received subsequent FOLFOX-HAIC due to progression of intrahepatic lesions and stabilization of extrahepatic lesions, 2 received subsequent TACE due to the presence of new small intrahepatic lesions, and 2 received radiotherapy due to painful bone metastases. These locoregional therapies are allowed for advanced HCC according to the Chinese guideline (43). However, FOLFOX-HAIC is a highly specialized therapeutic modality, which limits its generalizability. Thus, systemic chemotherapy with FOLFOX has the potential to be widely administered.

The spectrum, incidence, and severity of treatment-related AEs observed with this combination therapy were consistent with those observed in previous studies and seemed acceptable and manageable. These AEs were not unexpected and could be alleviated or eliminated by treatment interruption or dose modification. Severe immune-related AEs were observed in 10.0% of the participants, which were associated with toripalimab, as reported in the previous studies (44, 45).

The identification of biomarkers to evaluate tumor response represents an important secondary aim of the LTSC study. Here, we optimized snRNA-seq for frozen biopsy tissues, which has substantial advantages over scRNA-seq, including reduced dissociation bias, compatibility with frozen samples, and improved detection of malignant and stromal cells (46, 47). Biomarker analyses revealed that CFHR4⁺ and AKR1C2⁺ malignant subtypes have the potential to discriminate between responsive and nonresponsive tumors, and may be predictive biomarkers for triple combination treatment. In the validation cohort, we also identified two cell subtypes (Val_C01_HCC, Val_C02_HCC) with the highest similarity to CFHR4⁺ and AKR1C2⁺ malignant hepatocytes, and further validated their similar clinical relevance. In a previous study, Gao and colleagues identified three HCC subtypes with distinct clinical and molecular features, denoted as proliferation, metabolism, and microenvironment dysregulated subgroup (34). Our analyses clearly demonstrated distinct metabolic features among these hepatocyte subtypes. Notably, CFHR4⁺ and AKR1C2⁺ malignant subtypes were characterized by enrichment of metabolism-related pathways, including glycolysis, TCA cycle, steroid hormone, and glutathione metabolism. These results suggest that CFHR4⁺ and AKR1C2⁺ malignant subtypes are driven by elevated metabolic processes and are derived from the metabolism subgroup, consistent with a previous study (34). More importantly, the metabolic landscape of malignant hepatocytes appears to be associated with the clinical response to this combination therapy, but the mechanisms underlying tumor biological metabolism remain to be elucidated. We also observed that CFHR4⁺ and AKR1C2⁺ hepatocytes were primarily enriched in Hoshida S3 subclass, which was denoted as retained hepatocyte-like phenotype, and expressed high levels of hepatocyte function-related genes involved in metabolism, detoxification, coagulation, and oxygen radical scavenging (31). Hoshida S1 subclass reflected aberrant activation of the WNT signaling pathway, whereas Hoshida S2 subclass was characterized by proliferation and MYC activation. Chemotherapeutic drugs such as oxaliplatin or 5-fluorouracil mainly interfere with the synthesis of nucleotides and DNA replication in the process of cell proliferation, therefore the proliferative tumors are more sensitive to chemotherapy (48, 49). However, CFHR4⁺ and AKR1C2⁺ malignant hepatocyte subtypes expressed low levels of WNT-β-catenin signaling and MYC target genes (Fig. 4A), suggesting that CFHR4⁺ and AKR1C2⁺ hepatocytes reflected inactivation of the WNT and MYC signaling pathway and may be less responsive to chemotherapy. In addition, the presence of CFHR4⁺ or AKR1C2⁺ hepatocyte subtypes slightly expressed immune-related signatures and lacked immune cells, which were predominant characteristics of immune exclusion phenotype in HCC (50). Patients with a higher abundance of CFHR4⁺ or AKR1C2⁺ hepatocytes are more likely to develop resistance to immunotherapy.

This study revealed significant heterogeneity in composition and infiltration of immune subtypes in responsive and nonresponsive HCC, characterized by increased proportions of total immune cells, T cells, and macrophages in responders. Consistent with a previous trial evaluating atezolizumab and bevacizumab (51), we identified several specific macrophage and T-cell subtypes that were significantly enriched in responders and associated with better clinical outcomes. These results indicated that responders in our study exhibited an “immune-inflamed” microenvironment, whereas nonresponders exhibited an “immune-excluded” phenotype, characterized by the absence or exclusion of immune cells in the tumor parenchyma. Moreover, the expression score of lenvatinib targets in malignant hepatocytes was significantly higher in responders, in comparison to nonresponders.

This study had several limitations. First, this was a single-arm, phase II study with a relatively small number of enrolled patients and without a standard treatment group as a control, which may have led to participant and selection biases. These findings should be validated in larger phase III trials. Second, this trial was performed in patients with disease characteristics relevant to Asian countries (e.g., 96.7% of patients were HBV-HCC) and preserved liver function. The results reported herein are unlikely to be fully generalizable to Western countries. Moreover, the PFS rate at 6 months is proposed as a potential surrogate endpoint for OS in patients with advanced HCC. PFS or time to tumor progression, is also commonly used in other single-arm trials (13, 52) and not affected by subsequent treatments. Finally, considering the relatively small sample size and lack of a control arm in this biomarker study, further biomarker exploration in a larger cohort is required.

To conclude, the LTSC trial provided clinical evidence that FOLFOX chemotherapy in combination with lenvatinib and toripalimab exhibits promising antitumor activities and manageable safety profiles in patients with advanced HCC with extrahepatic metastasis, including those with high-risk disease. This combination strategy may be suitable as a first-line therapy for advanced HCC with extrahepatic metastasis. Exploratory analyses of snRNA-seq profiles identified that AKR1C2⁺ and CFHR4⁺ hepatocyte subtypes were associated with reduced clinical benefits and may be predictive biomarkers of resistance to this triple combination therapy. Patients with higher abundance of tumor-infiltrating immune cells were more likely to benefit from this combination. These findings warrant further validation in large randomized clinical trials.

No author disclosures were reported.

M.K. He: Conceptualization, data curation, formal analysis, investigation, writing–original draft. Y.X. Huang: Data curation, formal analysis, investigation, writing–original draft. Z.F. Du: Data curation, formal analysis, investigation, visualization. Z.C. Lai: Data curation, formal analysis, investigation. H. Ouyang: Formal analysis, investigation. J.X. Shen: Formal analysis, validation. D.S. Wen: Formal analysis, investigation. Q.J. Li: Formal analysis, investigation. Y.J. Zhang: Validation, investigation. W. Wei: Validation, investigation. M.S. Chen: Validation, investigation. L. Xu: Formal analysis, supervision, validation, investigation. A. Kan: Formal analysis, investigation, methodology, writing–review and editing. M. Shi: Conceptualization, supervision, investigation, project administration, writing–review and editing.

This study was supported by the National Natural Science Foundation of China (Nos. 82203126, 82102985, 82072610, 81902473), Development Planned Project in Key Areas of Guangdong Province (No. 2019B110233002), and China Postdoctoral Science Foundation (No. 2021TQ0383).

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.