Brefeldin A, 3-methyladenine (3-MA), MG132, N-acetyl-l-cysteine, puromycin, and DAPI stain solution were purchased from Sigma-Aldrich (St. Louis, MO, United States). 2′,7′-dichlorodihydrofluorescein diacetate (H2DCFDA) and Lipofectamine^® 3000 transfection reagents were purchased from Thermo Fisher Scientific Inc., (Waltham, MA, United States). Small interfering RNA of Stx17 and scramble were purchased from Thermo Fisher Scientific Inc., (Waltham, MA, United States). Small interfering RNA of CD63 and Rab11 were purchased from Santa Cruz Biotechnology, Inc., (Dallas, TX, United States). Recombinant mouse macrophage colony-stimulating factor (M-CSF) was purchased from PeproTech (Cranbury, NJ, United States). Antibodies against Gal-1, Stx17, CD63, Rab11, VAMP7, calreticulin, heat shock protein 60 (HSP60), and β-actin were purchased from Abcam (Cambridge, United Kingdom). Gold nanoparticles (20 nm) conjugated secondary antibodies were also purchased from Abcam (Cambridge, United Kingdom). Antibodies against LC3, Atg5, and p62 were purchased from MBL (Nagoya, Japan). Antibodies against SOCS3 were purchased from the Proteintech group (Rosemont, IL, United States). Antibodies against CD68 were purchased from Agilent (Santa Clara, CA, United States). Mouse IL-10 was detected by ELISA (R&D Systems, Minneapolis, MN, United States).

Mouse hepatoma cell line ML-1_4a was adapted from ML-1 cells in BALB/c mice (Chang et al., 2013). Mouse macrophage cell line RAW 264.7 and human embryonic kidney cells 293T were purchased from the American Type Culture Collection Cell bank. Wild type and Atg5-/- mouse embryonic fibroblasts (MEF) were kindly provided by Dr. Tamotsu Yoshimori (Osaka University, Japan). MEF, ML-1_4a, and 293T cells were cultured in Dulbecco’s modified Eagle medium (DMEM) while RAW 264.7 cells were cultured in RPMI 1640 medium supplemented with 10% heat-inactivated fetal bovine serum (FBS), 0.05 mg/ml streptomycin, and 50 U/mL penicillin. To generate mouse bone marrow-derived macrophages (BMDMs), bone marrow cells were collected from femurs and tibias of 6- to 10-week-old mice. These cells were cultured for 5 to 6 days in RPMI 1640 medium supplemented with 10% heat-inactivated FBS and 10 ng/ml recombinant mouse M-CSF to differentiate into BMDMs.

C57BL/6 mice and athymic nude mice (8–10 weeks old) were purchased from the National Laboratory Animal Center, Taiwan. Gal-1 knockout mice with C57BL/6 background were obtained from the Mutant Mouse Regional Resource Center and maintained in the Animal Laboratory of National Cheng Kung University. TLR2 knockout mice with C57BL/6 background were kindly provided by Dr. John T. Kung (Institute of Molecular Biology, Academia Sinica, Taiwan). TLR4 knockout mice with C57BL/6 background were kindly provided by Dr. Akira Shizuo (Laboratory of Host Defense, WPI Immunology Frontier Research Center, Osaka University). All mice were raised and cared for according to the guidelines set up by the Institutional Animal Care and Use Committee (IACUC) of National Cheng Kung University, and the Committee also approved this study (Ethics of Animal Experiments of National Cheng Kung University, Permit Number:107130). To examine the Gal-1 role in TAMs, 5 × 10⁵ ML-1_4a cells were mixed with 5 × 10⁵ wild type or Gal-1-/- BMDMs and then subcutaneously injected into nude mice. The tumor volume in HCC-bearing nude mice was monitored.

Atg5 and VAMP7 were knocked down in RAW 267.4 cells by stably expressing lentivirus-based shRNA. The clones were obtained from the National RNAi Core Facility (Institute of Molecular Biology/Genomic Research Center, Academia Sinica, Taiwan) and the target sequence for Atg5 was TRCN 0000375819 5′-AGCCGAAGCCTTTGCTCAATG -3′, while for VAMP7 it was TRCN 0000353419 5′-GCACTTCCTTATGCTATGAAT-3′, and for control luciferase it was TRCN0000072247 5′-GAATCGTCGTATGCAGTGAAA-3′. To generate shRNA carrying lentivirus, 293T cells (1 × 10⁶) were seeded into a 6 cm-plate, and then DNA mixture in 10 μL GeneJammer transfection reagent (Agilent, CA, United States) was added (pCMVdeltaR8.91: 0.9 μg/well; pMD.G: 0.1 μg/well; shRNA: 1 μg/well). After 48 h of incubation, the supernatants were collected to concentrate the virus. RAW 267.4 cells were infected with the lentivirus in the presence of polybrene (8 μg/ml) at 37°C for another 48 h. The cells were cultured with puromycin (5 μg/ml) containing medium to select stably expressing lentivirus-based shRNA. Stx17, CD63, and Rab11 were silenced in RAW 267.4 cells by transfecting small interfering RNA. The siRNA was packaged in liposomes by Lipofectamine^® 3000 transfection reagent according to the manufacturer’s instructions. The silencing efficacy of these proteins was determined by Western blotting.

Raw 264.7 cells (2 × 10⁶/per 10 cm dish) were treated with ML-1_4a hepatoma culture medium (MCM) for 6 h. The cells were suspended in 10% sucrose and mixed with 0.5 ml of the buffer (1 M Hepes/0.1 M EDTA), and then cells were physically broken down by using a 25G syringe. This homogenate was diluted with homogenization buffer (0.25 M sucrose, 10 mM Hepes, 1 mM EDTA, pH = 7.3) containing 0.5 mM glycyl-l-phenylalanine 2-naphthylamide (Cayman Chemical, MI, United States) and 1% dimethyl sulfoxide. After incubation for 7 min at 37°C, this homogenate was centrifuged at 4000 rpm for 2 min to collect the supernatant (post-nuclear supernatant). The supernatant was added into a tube with 9.5% Nycodenz (Alere Technologies AS, Norway) (middle layer) and 22.5% Nycodenz (bottom layer). The Nycodenz gradient was centrifuged at 28000 rpm in an SW28 rotor overnight, and then three fractions were observed. The interface band was added to the tube with 33% Percoll (middle layer) and 22.5% Nycodenz (bottom layer) and then centrifuged for 1 h at 20000 rpm in the SW28 rotor to collect the autophagosome-enriched layer.

The cells were collected for homogenization in a cell lysis buffer (Cell Signaling Technology, MA, United States) on ice for 20 min, and then centrifuged at 13000 rpm under 4°C for 20 min to collect the supernatants. Proteins were separated through SDS gel electrophoresis and transferred onto the PVDF membranes. The PVDF membranes were blocked with 5% skimmed milk and incubated with the appropriate primary antibodies at 4°C overnight. The membranes were then washed and incubated with peroxidase-conjugated secondary antibodies. The blots were visualized by enhancing chemiluminescence reagents (PerkinElmer Life Sciences, Hopkinton, MA, United States).

To monitor the protein distribution in macrophages, BMDMs or RAW264.7 cells were seeded into a cover slide, fixed with 4% formaldehyde, and stained with primary antibodies against Gal-1, LC3, CD63, Rab11, and VAMP7 at 4°C overnight. For detection of the expression of Gal-1 and LC3 in TAMs of human HCC, paraffin-embedded HCC tissue sections were deparaffinized with xylene and alcohol, and then antigen retrieval was performed with 10 mM citrate buffer at 95–100°C for 10 min. Next, the tissue sections were stained with primary antibodies against Gal-1, LC3, and CD68 at 4°C overnight. All above slides were stained with fluorescent dye-conjugated secondary antibodies at room temperature for 2 h. The protein distribution was determined by a confocal fluorescence microscope (Olympus FV 3000, Japan). The percentages of LC3-positive or Gal-1-positive cells in CD68-positive TAMs were visually evaluated and digitally documented with merged images. Further, we defined 50% as the cut-off point, e.g., the high Gal-1 expression (>50%) and the low Ga1-1 expression (≦50%). Subsequently, we divided the 93 HCCs into four groups, (1) low Gal-1 and low LC3 (Gal-1^– LC3^–), (2) low Gal-1 and high LC3 (Gal-1^– LC3⁺), (3) high Gal-1 and low LC3 (Gal-1⁺ LC3^–), and (4) high Gal-1 and high LC3 (Gal-1⁺ LC3⁺).

For immunoelectron microscopy analysis, macrophages were fixed in 4% formaldehyde and 1% glutaraldehyde in 0.1 M PBS, followed by dehydration and sectioning. The cell sections were first blocked with 1% BSA at room temperature for 1 h. Next, the primary antibodies against Gal-1 were co-incubated with sections at 4°C overnight. After washing steps, the sections were incubated with gold nanoparticles (20 nM) conjugated secondary antibodies at room temperature for 2 h. Sections were further stained with uranyl acetate for 20 min and lead citrate for 4 min and then examined using a JEM-1400 transmission electron microscope (Japan Electron Optics Laboratory Co., Ltd., Japan).

The institutional review board approved the study design and patient recruitment (CYCH-IRB No. 106-030 and CYCH-IRB No. 104079, Ditmanson Medical Foundation Chia-Yi Christian Hospital, Taiwan). Twenty-one healthy volunteers and 25 HCC patients were recruited to compare the serum level of Gal-1. A total of 93 surgically treated patients with HCCs were retrospectively enrolled at Ditmanson Medical Foundation Chia-Yi Christian Hospital, Taiwan, from January 2007 to December 2012. The inclusion criteria included: (1) a tissue-proved HCC after liver resection; (2) medical histories and therapeutic outcomes were complete and well traceable; (3) at least 1 month of post-operation survival; and (4) no neoadjuvant therapies before surgery. The exclusion criteria included: (1) metachronous or synchronous tumor in other organs; (2) a mixed type of liver tumor. Clinical data, including gender, age, viral status (hepatitis B virus and hepatitis C virus), tumor size, stage, survival time, disease recurrence and progression, image features, and treatment response, were obtained by chart review. The tumor stage was re-evaluated according to the eighth edition of the American Joint Commission on Cancer (AJCC) staging system. Histopathological characteristics, including diagnosis, the modified histology activity index (HAI) score, and the presence of cirrhosis, were re-evaluated by an experienced pathologist blinded to the clinical information. The clinical and pathologic parameters of the patients are shown in Supplementary Table 1. All patients were followed till the death or till February 2017 and the follow-up time was between 1–120 months (median, 68 months).

Gene Set Enrichment Analysis (GSEA) was performed on RNA-Seq gene expression data from The Cancer Genome Atlas (TCGA) liver cancer dataset. GSEA was performed to determine galectin-1 enrichment in gene lists extracted from MSigDB (Mootha et al., 2003; Subramanian et al., 2005) in order to determine the enrichment in gene sets from the immunologic signatures (C7) collection.

All cell and mouse experiments were performed at least two to three times, and the data are expressed as means ± SD. Data were analyzed by One-Way ANOVA using GraphPad Prism 8 software (GraphPad Software Inc., San Diego, CA, United States), and p < 0.05 was considered to represent significant differences between groups. Regarding the clinicopathological characteristics, data were analyzed using one-way ANOVA for continuous variables and the Fisher exact test for categorical variables to evaluate the association between different groups with different Gal-1/LC3 expression levels. In addition, Spearman rank-order correlation analysis was employed to explore the relationship between Gal-1 and LC3 in TAMs. Progression-free survival (PFS) was the interval between tumor resection and the first time of either disease progression recurrence, relapse, or death from any cause. Overall survival (OS) was defined as the time from surgery to the date of death. The follow-up time was defined as the period between surgery and the end of follow-up or death. The OS and PFS rates were calculated using the Kaplan–Meier method with Bonferroni adjustment for multiple pairwise comparisons. Hazard ratios (HRs) and confidence intervals (CIs) were calculated using the Cox regression method. Statistical analyses were performed using SPSS (version 22.0; IBM Inc.) or GraphPad Prism 8 software (GraphPad Software Inc., San Diego, CA, United States), and two-tailed p < 0.05 was considered to represent significant differences between groups.

Upregulated expression of Gal-1 is found in the tumor area and peripheral blood of HCC patients, and this correlates with the poor prognosis of these patients (Wu et al., 2012). Immune cells are known to express and produce Gal-1 in response to stimuli (Thiemann and Baum, 2016), but their roles in upregulated Gal-1 observed in HCC are still a puzzle. To assess Gal-1 expression in HCC-associated immune cells, we analyzed the relationship of Gal-1 mRNA expression to various immune subpopulation markers of human HCC from available RNAseq data in TCGA by GSEA analysis (Figure 1A). We found a high correlation of Gal-1 expression to macrophage-rich HCC tumors, as identified by gene signatures, but not to Th1, Th2, CD8⁺ T, B cells, or dendritic cells (Figure 1A). This indicates that HCC-associated macrophages may contribute to the increase of Gal-1 in TME. To investigate this process, we have introduced a co-culture system in which the bone marrow-derived macrophages (BMDMs) were co-cultured with MCM to differentiate as the M2-like phenotype with increased expression of SOCS3 and IL-10 (Supplementary Figures 1A,B) as we previously reported (Chang et al., 2013). The expression and secretion levels of Gal-1 in MCM-treated BMDMs were then monitored. As the results showed in Figure 1B, MCM treatment stimulated up-regulation of Gal-1 mRNA in BMDMs, which is consistent with findings in Figure 1A. By Western blot analysis, the protein level of Gal-1 was increased at 6 and 12 h in MCM-treated BMDMs; however, the level was gradually reduced at 24 h (Figure 1C). A similar dynamic expression pattern of Gal-1 was observed by immuno-fluorescent analysis (Figure 1D). Since Gal-1 mRNA was not down-regulated at 24 h in MCM-treated BMDMs (Figure 1B), we examined whether the reduction of the protein level of Gal-1 was caused by protein instability. We found no restoration of Gal-1 in MCM-treated macrophages in the presence of proteasome inhibitor MG132 (Figure 1E). These findings suggested that cytosolic Gal-1 may be secreted from MCM-treated macrophages. As expected, a significant increase of supernatant Gal-1 was observed in MCM-treated BMDMs and RAW264.7 cells (Figure 1C and Supplementary Figures 2A,B). To rule out the possibility of passive release of Gal-1, we checked cell death by LDH release analysis in which MCM was not able to induce cell death (Supplementary Figure 2C). All these data collectively indicate that HCC cells are capable of triggering macrophages to secrete Gal-1 actively. Next, we further examined whether Gal-1 produced by HCC-associated macrophages contributes to HCC progression. Consequently, the ML-1_4a hepatoma cells were mixed with wild type (WT) or Gal-1-/- BMDMs and then inoculated subcutaneously into nude mice. As indicated by the results shown in Figures 1F,G, mice carrying ML-1_4a tumors co-injected with Gal-1-/- BMDMs displayed reduced tumor growth, size, and weight compared with co-injection with WT BMDMs. Furthermore, the serum level of Gal-1 was significantly decreased in HCC-bearing mice co-injected with Gal-1-/- BMDMs compared with co-injection of WT BMDMs (Figure 1H). These results indicate that Gal-1 produced by HCC-associated macrophages facilitates HCC progression.

It is known that Gal-1 lacks a specific signal sequence for secretion via the conventional pathway, which suggests the involvement of unconventional secretion pathways, such as secretory autophagy (Martinez-Bosch and Navarro, 2020). To determine the secretory pathway of Gal-1 in MCM-treated macrophages, the autophagy inhibitor, 3-MA, and the classical secretion pathway inhibitor, brefeldin A (BFA), were used. Our data revealed that treatment of 3-MA, but not BFA, significantly inhibited the secretion of Gal-1 in MCM-treated macrophages (Figure 2A). Accordingly, the upregulated expression of LC3 puncta in MCM-treated macrophages was observed by confocal microscopy (Figure 2B). Consistent with the 3-MA results described above, silencing of Atg5 in macrophages reduced MCM-induced Gal-1 secretion as well (Figure 2C). These data indicate that autophagy plays a role in regulating this Gal-1 secretion. To understand whether this autophagy-regulated Gal-1 secretion relies on the process of lysosome fusion, we knocked down syntaxin 17 (Stx17), a SNARE protein that is essential to mediate fusion of lysosomes with autophagosomes (Itakura et al., 2012), by transfection of siRNA targeting Stx17. Knockdown of Stx17 resulted in an increase of accumulation of p62 in macrophages, indicating the blockage of fusion between lysosomes and autophagosomes. However, these Stx17-knockdown macrophages did not suppress Gal-1 secretion after MCM treatment compared with scramble siRNA transfecting control cells, indicating that autolysosomes are not involved in the secretion of Gal-1 (Figure 2D). Next, we asked whether Gal-1 is a cargo protein of autophagosomes. Using immunostaining analysis, we found that Gal-1 was co-localized with LC3, and located inside the double-layer autophagosomes in MCM-treated macrophages, according to confocal microscopy and immuno-electron microscopy observations, respectively, (Figures 2E,F). In addition, the presence of Gal-1 in isolated autophagosomes of MCM-treated macrophages was increased compared to control cells (Figure 2G). For real-time visualization of the effect of autophagy on Gal-1 trafficking, we performed total internal reflection fluorescence (TIRF) microscopy analysis to selectively visualize labeled vesicles which are located close to the plasma membrane. To determine this, we transfected GFP-Gal-1 into WT or Atg5-/- mouse embryonic fibroblasts (MEFs) and tracked the movement of GFP-Gal-1 by TIRF microscopy. The expression of Atg5 in WT or Atg5-/- MEF cells was confirmed by immunoblots (Figure 2H). The GFP-Gal-1 dots moved quickly and diffused laterally in the TIRF evanescence field of WT MEF cells (Upper, Figure 2H and Supplementary Video 1), while the signal remains steady in the Atg5-/- cells (Lower, Figure 2H and Supplementary Video 2). These data apparently indicate that Gal-1 is carried by autophagosomes and is then subsequently released from HCC-stimulated macrophages.

We next went further to address the pathways involved in autophagy-dependent Gal-1 secretion. MVBs are the endosome molecules that can interact with the autophagosomes to form hybrid structures called amphisomes to facilitate protein secretion (Hessvik and Llorente, 2018). Hence, we investigated whether the MVBs are involved in the secretion of Gal-1. To examine this, we first checked the co-localization of CD63, an MVB marker, and Gal-1 by confocal immunofluorescence microscopy. As shown by the results in Figure 3A, Gal-1 was found to co-localize with CD63 in MCM-treated macrophages. Then, the CD63 was knocked down by siRNA to examine MVB contribution on Gal-1 secretion. This was followed by Western blot analysis, where the recovery of cytosolic Gal-1 along with the reduction of supernatant Gal-1 was observed in CD63 silencing MCM-treated macrophages compared with control cells (Figure 3B). These data indicate the involvement of MVB in the secretion of Gal-1. Rab GTPase family proteins are well-studied key regulators of vesicle trafficking and protein secretion. One of these proteins, called Rab11, has been shown to play a pivotal role in amphisome-mediated protein secretion (Hessvik and Llorente, 2018). Thus, we examined whether Rab11 participates in Gal-1 secretion. According to our immunostaining observation, Rab11-positive vesicles were found to co-localize with Gal-1 (Figure 3C). Knockdown of Rab11 by siRNA significantly reduced secretion of Gal-1 from MCM-treated macrophages (Figure 3D), suggesting that Rab11 is essential for Gal-1 secretion. Furthermore, Rab11 is reported to interact with one of the SNARE proteins called VAMP7 to promote vesicle trafficking (Kandachar et al., 2018). Hence, we further investigated the role of VAMP7 in the secretion of Gal-1. Consistent with what we found in Figures 3C,D, the confocal image analysis has revealed a significant co-localization between the VAMP7 and Gal-1 (Figure 3E). Silencing of VAMP7 abrogated the supernatant level of Gal-1 in MCM-treated macrophages (Figure 3F). These results collectively suggest that an MVB/Rab11/VAMP7 axis facilitates the autophagy-mediated secretion of Gal-1.

We have recently demonstrated that TLR2 signal-induced reactive oxygen species (ROS) is essential to trigger autophagy in HCC-associated macrophages (Shiau et al., 2020). Thus, we examined whether TLR2 and ROS are responsible for mediating the autophagy-regulated secretion of Gal-1 from MCM-treated macrophages. Subsequently, the secretion of Gal-1 in MCM-treated wild type (WT), TLR2-/-, and TLR4-/- BMDMs was monitored. According to the results shown in Figure 4A, a significant decrease of Gal-1 secretion in MCM-treated TLR2-/- BMDMs was found compared with WT or TLR4-/- BMDMs. In addition, the LC3 puncta and co-localization of Gal-1 and LC3 were abolished in MCM-treated TLR2-/- BMDMs (Figure 4B). These data suggest that TLR2-dependent autophagy is crucial for Gal-1 secretion. To further address the role of ROS in Gal-1 secretion, we first monitored ROS production from MCM-treated WT, TLR2-/-, and TLR4-/- BMDMs. Our data showed that MCM triggers ROS generation in WT and TLR4-/- BMDMs, which was blocked in TLR2-/- macrophages (Figure 4C). Next, a well-known scavenger of ROS, N-acetyl cysteine (NAC), was used to inhibit MCM-triggered ROS in BMDMs (Figure 4D). Accordingly, the MCM-induced autophagy and Gal-1 secretion were determined. As shown by the results in Figures 4E,F, inhibition of ROS by NAC decreased the formation of LC3 puncta and LC3-II conversion in MCM-treated BMDMs compared with control cells. Furthermore, treatment of NAC was able to effectively suppress the Gal-1 secretion in MCM-treated BMDMs (Figure 4G). Together these findings suggest that TLR2-mediated ROS production is essential to induce autophagy-dependent secretion of Gal-1.

The relationship between autophagy and Gal-1 secretion has never been examined in TME of human HCC patients. Thus, in our pilot study, we first enrolled 25 HCC patients and 21 healthy volunteers to determine the serum level of Gal-1. In line with previous reports, an increased serum level of Gal-1 was found in HCC patients compared with healthy volunteers (Figure 5A). To examine the correlation between serum level of Gal-1 and autophagy activity in TAMs of HCC patients, we performed immunofluorescence staining in HCC patients’ tissues from those with high serum levels (over 250 ng/ml) or low serum levels (below 150 ng/ml) of Gal-1. Interestingly, we found a high proportion of Gal-1^–LC3⁺CD68⁺ cells in HCC tissues with a high level of serum Gal-1 while Gal-1⁺LC3^–CD68⁺ cells were rich in HCC tissues with a low level of serum Gal-1 (Figure 5B). The results suggest that autophagy activity in TAM may correlate to Gal-1 secretion in HCC patients. We next examined whether Gal-1^–LC3⁺CD68⁺ cells correlate to HCC patients’ progression and poor prognosis. A total of 93 HCC patients were retrospectively recruited. The clinicopathological parameters of patients are summarized in Supplementary Table 1. Among 93 HCC patients, 66 (70.97%) were male, and 52 (55.91%) patients had liver cirrhosis with a mean age of 63.06 years. According to immunofluorescence staining of Gal-1 and LC3 on CD68⁺ TAMs, we divided the 93 HCCs into four groups, Gal-1^– LC3^–, Gal-1^–LC3⁺, Gal-1⁺ LC3^–, and Gal-1⁺ LC3⁺. Although most clinicopathological variables of each group were not significantly different, there were differences in tumor size, stage, recurrence, and death between different groups. Groups with low Gal-1 expression (Gal-1^– LC3^– and Gal-1^– LC3⁺) had larger tumor size (p < 0.001), higher stage (p = 0.005), and higher frequencies of tumor recurrence (p = 0.033) and deaths (p < 0.001) than groups with high Gal-1 expression (Gal-1⁺ LC3^– and Gal-1⁺ LC3⁺) (Supplementary Table 1). Consistently, we found that high expressed Gal-1 TAMs (Gal-1⁺CD68⁺) showed a low level of LC3 while low expressed Gal-1 TAMs (Gal-1^–CD68⁺) carried a high level of LC3 (Figure 5C). Notably, a significant negative correlation between Gal-1 and LC3 expression in TAMs of HCC patients was observed. Bivariate correlation analysis using Pearson’s correlation (r = −0.535, p < 0.001) implied a moderate negative correlation, yet a statistically significant linear relationship exists between LC3 and Gal-1. Moreover, Spearman’s correlation also demonstrated a statistically significant negative correlation (r = −0.534, p < 0.001) (Figure 5D). Moreover, Kaplan-Meier survival analyses revealed that the patients carrying Gal-1^– LC3⁺ TAMs had the worst overall survival (OS) and progression-free survival (PFS) rates, and the statistical significance was present in comparison with all three of the other groups (with the Gal-1^– LC3^– group: p < 0.001; with the Gal-1⁺ LC3^– group: p < 0.001; with Gal-1⁺ LC3⁺ group: p < 0.001) (Figure 5E). After adjusting for patients’ age, gender, tumor size, and stage, the Cox proportional Hazard model demonstrated that patients carrying Gal-1^–LC3⁺ TAMs are significantly and independently at high-risk with a poor prognosis (Supplementary Table 2). Collectively, these data indicate that Gal-1^– LC3⁺ CD68⁺ cells serve as a prognostic factor for HCC’s poor progression.