Isolation and identification of patient-derived liver cancer stem cells and development of personalized treatment strategies

Abstract

Background

Liver cancer stem cells (LCSCs) are thought to drive the metastasis and recurrence, however, the heterogeneity of molecular markers of LCSCs has hindered the development of effective methods to isolate them.

Methods

This study introduced an effective approach to isolate and culture LCSCs from human primary liver cancer (HPLC), leveraging mouse embryonic fibroblasts (MEFs) as feeder cells in conjunction with using defined medium. Isolated LCSCs were further characterized by multiple approaches. Transcriptome sequencing data analysis was conducted to identify highly expressed genes in LCSCs and classify different subtypes of liver cancers.

Results

Total sixteen cell strains were directly isolated from 24 tissues of three types of HPLC without sorting, seven of which could be maintained long-term culture as colony growth on MEFs, which is unique characteristics of stem cells. Even 10 of cloned cells formed the tumors in immunodeficient mice, indicating that those cloned cells were tumorgenic. The histologies and gene expression pattern of human xenografts were very similar to those of HPLC where these cloned cells were isolated. Moreover, putative markers of LCSCs were further verified to all express in cloned cells, confirming that these cells were LCSCs. These cloned LCSCs could be cryopreserved, and still maintained the feature of colony growth on MEFs after the recovery. Compared to suspension culture as conventional approach to culture LCSCs, our approach much better maintained stemness of LCSCs for a long time. To date, these cloned cells could be cultured on MEFs over 12 passages. Moreover, bioinformatics analysis of sequencing data revealed the gene expression profiles in LCSCs, and liver cancers were classified into two subtypes C1 and C2 based on genes associated with the prognosis of LCSCs. Patients of the C2 subtype, which is closely related to the extracellular matrix, were found to be sensitive to treatments such as Cisplatin, Axitinib, JAK1 inhibitors, WNT-c59, Sorafenib, and RO-3306.

Conclusion

In summary, this effective approach offers new insights into the molecular landscape of human liver cancers, and the identification of the C2 subtype and its unique response to the treatment pave the way for the creation of more effective, personalized therapeutic strategies.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12967-024-05870-9.

Introduction

Human primary liver cancer (HPLC) is one of the most prevalent malignant tumors globally, ranking the fifth in terms of incidence worldwide [1]. HPLC is mainly classified into three subtypes, the most common subtype of liver cancer is hepatocellular carcinoma (85^~95%) [1], followed by cholangiocarcinoma (10^~15%) [2]and mixed liver cancer accounts for less than 1% of all liver cancers [3]. At present, the treatment of liver cancer is still a great challenge in the world. Currently this extremely heterogeneous cancer is treated by many ways, including surgical resection, local ablation, trans-arterial chemoembolization (TACE), systemic treatment [4]. Despite these treatments, the 5-year recurrence and metastasis rates for liver cancers remain significantly high, ranging from 40–70% [5]. Recent research indicates that cancer stem cells (CSCs) are a key factor leading to tumor metastasis and recurrence [6]. Therefore, investigating LCSCs is instrumental in uncovering new mechanisms that can aid in finding appropriate approaches to treat the metastasis and recurrence of liver cancers, thereby improving its prognosis.

CSCs are widely recognized as a population of cancer cells with the capacity to initiate tumors, exhibit drug resistance, and possess a high metastatic potential, playing a significant role in the onset and progression of cancers [7]. To date, numerous putative markers for liver cancer stem cells (LCSCs) have been identified, such as CD133 [8], CD90 [9], CD44 [10], CD13 [11, 12], EPCAM [13], ALDH [14], OV6 [15, 16], CK19 [17], CD34 [18], CD47 [19], CD24 [20], NCAM1 [21], SOX2 [22], SOX9 [23]. Owing to the heterogeneity of these markers, no individual biomarker has yet been identified that can consistently detect the presence of all cancer stem cells. At present, the most commonly used culture method of CSCs is suspension culture [24]. However, there are two ways of CSCs division, one is asymmetrical (producing into a CSC and a non-CSC), the other is symmetric (producing into two CSCs or two non-CSCs) [25]. The division characteristics of stem cells in the spheroids formed through suspension culture led to being mixed aggregates rather than pure CSCs. For instance, research has found that even CD24⁺ cells sorted under hypoxic conditions can experience a drop of positivity rate to approximately 20% within just 96 h [26]. CD24⁺ cells were associated with CSCs, and widely used as a marker for CSCs in liver cancers [20], pancreatic cancers [27], and colon cancers [28], cervical cancers [29].

Human embryonic stem cells (hESCs) are primarily cultured in vitro using mouse embryonic fibroblasts (MEFs) to provide a stem cell niche and growth factors that maintain the pluripotency and stemness of hESCs. This cultivation method has led to the establishment of multiple hESC lines [30]. Utilizing MEFs for stem cell culture has thus become a vital technique in stem cell biology. However, culturing CSCs using MEFs is relatively scarce. In 2013, PARK et al. successfully cultured CD34⁺ PLC/PRF/5 CSCs using MEFs, maintaining stemness over 22 passages [18].

In this study, we employed MEFs as feeder cells, digested patient-derived tumor tissues without sorting, and seeded them with a specific culture medium to cultivate CSCs on MEFs. We conducted the identification of these CSCs regarding the morphology, surface markers, differentiation potential, tumorigenicity. Additionally, we compared this method with the traditional suspension culture of CSCs, our results indicated that we successfully established a method for culturing liver cancer stem cells without relying on putative CSC markers. Considering LCSCs are the origin of the recurrence and metastasis of liver cancers, we performed transcriptome sequencing analysis, and identified markers that were highly expressed in isolated LCSCs and were correlated with liver cancer prognosis. Based on these results, we utilized TCGA-LIHC database, ICGC database and single-cell data from GEO to perform cluster analysis on these genes, aiming to further elucidate the role of LCSC-related genes in the development and occurrence of liver cancers, and to develop novel treatment strategies.

Materials and methods

Clinical specimens

All liver cancer specimens were obtained from hepatobiliary surgery in Guangzhou First People’s Hospital, and the use of these samples was approved by Ethics Committee of Guangzhou First People’s Hospital. The samples we collected were tested and excluded for conditions such as syphilis, HIV, hepatitis C, and positive for HBsAg, HBeAg, and HBcAb of hepatitis B. In addition, the s the specimens were collected within 1–2 h post-surgical resection and transported to the laboratory on ice. During the transportation, the specimens were kept in DMEM medium, and the entire process was maintained under sterile conditions. All cases were diagnosed as primary liver cancers by both hepatobiliary surgeons and pathologists, rather than metastatic liver cancers from other cancers.

Preparation of feeder cells

Feeder layer cells were prepared from embryonic fibroblasts (MEF) of CF1 strain mice which were purchased from Slrc Laboratory Animal, Shanghai. Briefly, when the growth of MEFs reached approximately 90%, MEFs were treated with DMEM (Gibco) complete medium containing 10 µg/ml mitomycin C for 2.5 h at 37 °C. Subsequently, MEFs were washed five times with PBS, digested with 0.05% trypsin (Procell), centrifuged for 5 min at 350×g, and then resuspended in DMEM medium and cryopreserved in freezing medium containing 10% dimethyl sulfoxide (DMSO), 50% complete culture medium and 40% Fetal Bovine Serum (FBS).

A six-well plate was coated with 0.01% gelatin (Sigma) for at least 2 h, then MEFs were seeded into coated-plates at a density of 50,000 cells/cm². The feeder cells plate could then be used within 48 h.

Isolation and cultivation of LCSCs derived from human liver cancer tissues

Fresh tissues were loaded into DMEM medium and transported back to laboratory from hospital within 1–2 h. The fresh tissues were then minced into 1 mm³ pieces and digested with Type IV Collagenase (1 mg/mL, Thermo), incubated for 1 h with shaking at 80 RPM rotating speed in a 37 °C constant temperature shaker. Subsequently, 70 μm cell strainers were used to filter out undigested tissues, and the suspensions were centrifuged for 5 minutes at 1000–1500 rpm. The cell pellets were then resuspended in DMEM/F12 (Gibco) after lysing red blood cells. Next, 1.5 × 10⁵ to 2.0 × 10⁵ cells were seeded on MEFs per well in a six-well plate with complete culture medium containing DMEM/F12 medium supplemented with epidermal growth factor (20 ng/mL, PeproTech), basic fibroblast growth factor (4 ng/mL, PeproTech), 1×ITSE (BioGems), nicotinamide (1.08 mg/mL, Sigma), L-ascorbic acid-2 phosphate (0.29 mg/mL, Sigma), L-proline (30 µg/mL, Sigma), hydrocortisone (10 nM, Sigma), 1×antibiotic/antimycotic (Gibco), and 0.1% bovine serum albumin (Sigma). The medium was changed daily.

Cells on MEFs were passaged in 7–14 days according to the sizes and numbers of the colonies and the conditions of the feeder cells. Type IV Collagenase (Thermo) was used to digest the colonies in an incubator at 37 °C, and the colonies were pipetted into small colonies 30 min after the digestion. After the neutralization with FBS, the digested colonies were collected and centrifuged, and then resuspended in complete culture medium and seeded back into new feeder cells at the ratio of 1:4. For the cryopreservation of cloned cells, the digested colonies’ suspension was mixed with freezing solution (20% DMSO, 20% complete culture medium, 60% FBS) at 1:1 ratio. For the recovery, the cryopreserved cells were thawed immediately in a 37 °C water bath upon the removal from liquid nitrogen, then centrifuged to remove DMSO, and the cloned cells were seeded onto new six-well plate with freshly prepared feeder cells.

ALDEFLUOR assay

ALDEFLUOR kit (Stem cell Technologies) was used to identify and isolate ALDH positive cells. Briefly, cloned cells were dissociated into single cells and then suspended in ALDEFLUOR buffer containing activated ALDEFLUOR™ with or without the addition of DEAB, and incubated at 37 °C for 35–45 min. To determine ALDH positive rates or the isolation of ALDH positive cells in cloned cells, PE-anti-mouse feeder cell antibody (Miltenyi Biotec) was used to remove MEFs.

Quantitative RT-PCR

Total RNAs from cells were extracted using Universal RNA Extraction Kit (TaKaRa), and cDNAs were subsequently synthesized from total RNAs using the 5X PrimeScript RT Master Mix (TaKaRa), following the manufacturer’s instructions. The qPCR was conducted with PowerUp™ SYBR™ Green Master Mix (Applied Biosystems) on a real-time PCR system (Applied Biosystems), according to the manufacturer’s recommendations. Primer information was listed in Supplementary Table 1.

In vitro differentiation

ALDH⁺ cells sorted from cloned cells were plated at a density of 20,000 cells per well of six-well plates, and cultured in DMEM supplemented with 10% FBS on the plate without MEF cells. Microscopic images were captured and recorded using a Nikon microscope twice a week. To assess the stemness of the cells under differentiated and undifferentiated culture conditions, we collected differentiated cells and cloned cells for further analysis using for flow cytometry, qPCR, and immunofluorescence staining at day 7 after the differentiation or the culture of cloned cells.

Tumorigenicity by cloned ALDH+ LCSC

ALDH⁺ cells sorted from cloned cells and primary cancer cells were injected into Female immunodeficient NCG mice (GemPharmatech Co., Ltd)). ALDH⁺ cells from each cloned cell line were divided into three groups (n = 4 per group), with each group being inoculated subcutaneously with 10, 100, and 1,000 cells per injection, respectively. Similarly, primary cancer cells from each patient were also divided into three groups (n = 4 per group), and each group was inoculated subcutaneously with 100, 1000, and 100,000 cells per injection, respectively. Tumor formations were observed every two days. When the xenografts reached a diameter of 1–1.5 cm, they were removed for further analysis.

Hematoxylin and Eosin (H&E) staining

The xenografts and primary tumor tissues were fixed overnight in 4% paraformaldehyde, then dehydrated, and subsequently embedded in paraffin. The tissues were sectioned at the thickness of 4 μm. Next, the sections were treated by dewaxing and rehydration. They were stained with hematoxylin for a few minutes, followed by differentiation and turning blue. After that, they were dehydrated, counterstained with alcohol-eosin dye for 2–3 min, and finally dehydrated through a graded ethanol series for a few minutes before being cleared with xylene. The images were scanned using Aperio CS2 Digital Pathology Slide Scanners (Leica).

Immunofluorescence staining

The cloned cells and their differentiated counterparts were collected for immunofluorescent staining at day 7 after the differentiation or culture. The primary cancer tissues and xenografts were embedded in OCT, and then the frozen tissues were sectioned at the thickness of 5 μm. After collecting the samples, they were first fixed with 4% paraformaldehyde for 15 min. Following the fixation, the samples were permeabilized with 0.5% Triton-X for another 15 min. Subsequently, they were washed three times with PBS, with each wash for 5 min. Finally, the samples were blocked in commercial goat serum (Boster Biological Technology) for 50 min at room temperature. Next, the samples were incubated with primary antibodies at 4 °C overnight targeting AFP (MAB1368, R&D system), CD44 (3570 S, Cell Signaling Technology), CD133 (64326 S, Cell Signaling Technology), CD31 (AF23628-SP, R&D system), ALB (ab106582, abcam), NCAM (3576 S, Cell Signaling Technology), SOX2 (4900, Cell Signaling Technology), SOX9 (82630 S, Cell Signaling Technology), CD24 (ab202073, abcam), KI67 (9449, Cell Signaling Technology), OV6 (MAB2020, R&D system), and EPCAM (2929 S, Cell Signaling Technology). The following day, all samples were washed three times with PBS for 5 min per each wash, and then stained with the appropriate secondary antibodies anti-rabbit Alexa 488 (4412 S, Cell Signaling Technology), anti-mouse Alexa 488 (4408 S, Cell Signaling Technology), anti-rabbit Alexa 594 (8889 S, Cell Signaling Technology), anti-mouse Alexa 594 (8890 S, Cell Signaling Technology) and Anti-Chicken IgY (Alexa Fluor^® 647) (ab150171, abcam) for 1 h at room temperature. Afterward, all samples were washed three times with PBS for 10 min per each wash. Finally, cell nuclei were stained with DAPI (C1006, Beyotime) for 10 min. Images were captured using a Nikon Ti-E A1 confocal laser-scanning microscope (Nikon). Antibodies information was listed in Supplementary Table 2.

The clone formation rate of single cell

First, the cloned cells were digested into small clones with type IV collagenase for 30 min, then 40 μm cell strainers (NEST) were used to filter the cloned cells to remove MEFs. The cloned cells were further digested into single cells and plated on a 96-well plate with feeder cells using the limiting dilution method. The clone formation rate was calculated 7 after the culture.

Suspension spheroid culture of LCSCs

ALDH⁺ LCSCs were sorted and seeded into a low-adhesion six-well plate at 2 × 10⁵ cells per well in six-well plates with 2 mL of aforementioned complete culture medium, which was replenished every 2–3 days. Concurrently, 2 × 10⁵ cells were plated onto MEF per well in six-well plates, followed by the analysis of qPCR and ALDH positivity rate at day 7 after the culture.

RNA sequencing and analysis

Total RNAs were isolated from LCSCs cultured under various conditions—on MEF with CSC’s complete culture medium, in suspension with CSCs’ complete culture medium, or in DMEM medium with 10% FBS using TaKaRa RNA Extraction Kit. Subsequently, RNA-Seq libraries were generated with NEBNext^® Ultra™ RNA Library Prep Kit for Illumina^® (NEB, USA) and sequenced by Novogene on an Illumina HiSeq X-Ten platform using 150 bp paired-end reads. For data analysis, reads from each sample in FASTQ format were aligned to the genome using Hisat2 to generate SAM files, which were then converted to BAM format using Samtools. Gene quantification was performed using FeatureCounts to obtain a raw count matrix. Subsequently, differential gene expression analysis was performed using the R package DESeq2, and heatmap visualizations were created with the Heatmap R package. To explore the functions and pathways of LCSCs under various conditions, we conducted comprehensive bioinformatics analyses. We used Gene Ontology (GO) for gene function analysis and Kyoto Encyclopedia of Genes and Genomes (KEGG) for pathway analysis, both aided by the ‘clusterProfiler’ package. Additionally, Gene Set Enrichment Analysis (GSEA) was applied for gene set analysis based on MSigDB. Transcription factor (TF) enrichment analysis was conducted using TRRUST database, and bar charts along with transcriptional regulatory network diagrams were created using tools available at Bioinformatics.com.cn, an online platform for data analysis and visualization. The raw data are available in NCBI’s GEO under accession number (GSE274830).

TCGA data analysis

RNA sequencing data (level 3) and corresponding clinical information for liver cancer were retrieved from The Cancer Genome Atlas (TCGA) dataset (https://portal.gdc.cancer.gov). Differential expression analysis of CSC-associated genes between tumors and adjacent non-tumor tissues was conducted using the ‘limma’ R package (version 3.48.3), with a threshold of log2 fold change (log2FC) greater than 1 for high expression in liver cancer. Significant genes impacting prognosis were identified through univariate Cox proportional hazards analysis of the high-expression CSC-related genes in liver cancer. The R package ‘ConsensusClusterPlus’ (version 1.54.0) was used for consensus clustering to determine the optimal number of clusters, with a maximum number of clusters set to 9, coxPfilter = 0.01, clusterAlg = ‘pam’, innerLinkage=‘ward.D2’. Survival analyses for the identified subtypes were performed using the ‘survival’ package in R (version 3.2.11). The tumor microenvironment was characterized using the ‘Estimate’ (version 1.0.13) and ‘MCPcounter’ R packages (version 1.2.0). Heatmaps for clustering were generated with the ‘pheatmap’ R package (version 1.0.12), and all boxplots were constructed using the boxplot function from the ‘ggplot2’ R package. GO enrichment analysis was carried out with the ‘clusterProfiler’ R package (version 4.0.5). Gene Set Variation Analysis (GSVA) of C1 and C2 type cancer was applied for gene set analysis based on MSigDB. Immunotherapy prediction analysis was conducted using scores from two databases, TCIA (https://tcia.at/home) and TIDE (http://tide.dfci.harvard.edu/), followed by statistical analysis and plotting in R software. The immunophenotype score (IPS) in the TCIA database can predict patients’ response to anti-cyotoxic T lymphocyte antigen 4 (CTLA-4) and anti-programmed cell death protein 1 (anti-PD-1) therapies [31]. TIDE uses a set of gene expression markers to assess two distinct mechanisms of tumor immune evasion, including the dysfunction of tumor-infiltrating cytotoxic T lymphocytes (CTLs) and the exclusion of CTLs by immunosuppressive factors. A high TIDE score correlates with poor efficacy of immune checkpoint blockade (ICB) therapy and shorter survival after receiving ICB treatment [32]. The assessment of correlation between genes was conducted using the Spearman rank correlation coefficient method, the correlation plot for multiple genes was displayed using the ‘pheatmap’ package in R software.

ICGC data analysis

RNA sequencing data for liver cancer (hepatocellular carcinoma, HCC) was successfully retrieved from International Cancer Genome Consortium (ICGC) database (https://dcc.icgc.org/). In this study, data from a total of 203 normal samples and 243 tumor samples were collected. The R package ‘ConsensusClusterPlus’ (version 1.54.0) was used for consensus clustering to determine the optimal number of clusters, with a maximum number of clusters set to 9, coxPfilter = 0.01, clusterAlg = ‘km’, innerLinkage=‘ward.D2’. Boxplots for stemness-related genes were generated using the ‘boxplot’ function with Wilcoxon rank tests, differential LCSC-related genes of heatmaps were created with the ‘pheatmap’ R package (version 1.0.12) (log2FC > 1, P < 0.05), differential analysis was performed using the ‘limma’ package (version 3.48.3) (log2FC > 2, P < 0.05), GO enrichment analysis was conducted with the ‘clusterProfiler’ package (version 4.0.5), and bubble charts were plotted using the ‘dotplot’ function.

Single-cell data processing and analysis

Single-cell RNA-seq data from ten liver cancer samples were obtained from GEO database under accession number GSE149614, excluding normal and metastatic samples. Initial normalization was performed using the `NormalizeData` function of the Seurat package, adopting default parameters. The dataset was then subjected to unsupervised clustering and visualization by employing the `RunUMAP` function from the same package, effectively grouping cells into six distinct cell types. To define cell populations, we utilized established marker genes for T cells (CD3, CD8A, CD8B), B cells (CD19, CD79A, MS4A1), cancer-associated fibroblasts (CAF) (ACTA2, FAP, THY1), macrophages (CD14, CD68, CD163), endothelial cells (PECAM1, KDR, CDH5), and liver tumor cells (ALB, AFP, ASGR1). Determination of stemness scores across cells in reduced-dimensional space utilized the UCell algorithm. This was further integrated with pseudotime analysis conducted using the Monocle 2 package, linking stemness scores with developmental trajectory. EMT scores for individual tumor cells were also computed using the UCell algorithm. Expression patterns of stemness-related genes in different liver cancer cell clusters were depicted with Seurat’s ‘Dotplot’ function. The “subset” function facilitated isolation of the CAF subpopulation, with the ‘Vlnplot’ function providing visualization for the expression levels of fibroblast-specific and collagen-related genes.

Chemotherapy response prediction

The R package “oncoPredict” was used to predict chemotherapy responses of liver cancer samples from TCGA using GDSC database. The IC50 values were calculated by ridge regression with default parameters.

Statistical analysis

In this study, statistical analysis between two groups for qPCR and flow cytometry data was conducted using Student’s t-test with GraphPad 8.0 software (La Jolla, California, United States). For RNA sequencing data, the DESeq2 package was utilized to perform differential analysis between the two sequencing groups. The limma package was employed to analyze the expression differences of CSC highly-expressed genes between cancerous and para-cancerous tissues. Univariate COX regression was applied to define the prognostic relevance of highly-expressed genes in LCSCs. The Cox proportional hazards model was used for survival analysis of liver cancer typing. The UCell algorithm was applied to analyze LCSC-related genes and EMT scores in single-cell data. The oncoPredict package was used for drug sensitivity analysis. The Wilcoxon test was conducted for statistical difference analysis between two sample groups. All sequencing data analysis was performed using R version 4.1.1. A p value < 0.05 was considered statistically significant.

Results

Culturing and identification of LCSCs derived from patients

Cancer stem cells were a subpopulation of cancer cells that drove the development of tumors [6]. Recently, we established a method for long-term culture of LCSCs derived from patients with liver cancers, the culture system was adopted and modified from those to culture CD34⁺ LCSCs isolated from PLC/PRF/5 cells as described in our early study [18]. MEFs as feeder layers were prepared one day before the isolation of liver cancer specimens, next day the specimen was treated into single cells and seeded at 1.5 × 10⁵ -2.0 × 10⁵ cells on MEFs per well. Then the growth of cloned cells could be observed within 3–14 days. We had successfully isolated 16 cloned cell strains from 24 patients, including three subtypes of liver cancers (Supplementary Table 3). The morphologies of CSCs from five patients were displayed with round and packed features (Fig. 1A), and those isolated CSCs could be maintained continuous culture over 12 passages (to date) and recovered from frozen cells (Fig. 1A). The morphological characteristics of all cloned LCSCs were shown in Supplementary materials (Fig. 1A, Figure S1A). We observed the proliferation of individual clones at days 3, 5, and 7 and performed KI67 staining to show the cell numbers of proliferating cloned cells (Fig. 1B). The cloned cells were dissociated into single cells and seeded a single cell into one well in 96-well plates using the limited dilution method to determine the clone formation rate of single stem cells. As expected, the clonogenicity of single stem cells could reach more than 80% (Fig. 1B), indicating unique characteristics of symmetric division of stem cells. Then, we evaluated putative CSC markers in these cloned cells by immunofluorescence, and the results showed that CD44, EPCAM, NCAM, CD133, AFP, SOX2, SOX9, OV6, ALDH1A1, and CD24, were expressed (Fig. 1C). Additionally, the cancer-associated fibroblasts marker (FAP) was not expressed in these cloned cells as determined by immunofluorescence (Figure S1B). Consistently, we found that the percentages of ALDH, as functioning marker of progenitor and stem cells, were significantly higher in cloned cells than that in primary tumor cells (Fig. 1D and E). Taken together, these results indicated that we successfully established a novel method for isolating and culturing patient-derived CSCs from primary liver cancers.

Fig. 1.

Clonogenically culturing and identification of liver cancer stem cells (LCSCs). (A) Morphologies of three subtypes of LCSCs cultured on MEFs after continuous passages (Top panel) and recoveries (Bottom panel) after the cryopreservation. Scale bar 100 μm. (B) The proliferation of ALDH⁺ singe LCSCs cultured on MEFs at days 3, 5 and 7 (Left panel), scale bar 50 μm, and ki67 staining at days 3, 5 and 7 (Left panel). Scale bar 25 μm. The colony formation rate of single LCSCs. n = 3 biological replicates (Right panel). (C) Immunofluorescence staining showed that the cultured cloned cells expressed putative LCSC markers (CD44, NCAM, CD133, AFP, SOX2, SOX9, OV6, ALDH1A1, CD24). Scale bar 100 μm. (D, E) Determination of the percentages of ALDH positive cells between primary cancer cells and cloned cells. n = 3 biological replicates. Values were represented as mean ± SD. ****p < 0.0001

Cloned cells could be differentiated in vitro

Previous studies have implicated that cloned cells possessed stem cell characteristics. To further validate these findings, we performed the differentiation of cloned cells in vitro. The cloned cells were cultured in DMEM medium supplemented with 10% fetal bovine serum to induce the differentiation, and then the stemness characteristics of the cells under the two different culture conditions were assessed 7 days after the differentiation. After the differentiation, the morphology of the cloned cells changed from round colonies to polygons, resembling the morphology of ordinary liver tumor cells (Fig. 2A). Next, we assessed positive rate of ALDH in the differentiated cells. As expected, the percentages of ALDH positive cells were only about 20% (Fig. 2B and C). In addition, we found that previously reported liver stem cell markers (CD133, SOX2, NCAM1, CD90, SOX9) were downregulated after the differentiation in three different cloned cell lines (Fig. 2D). In contrast, the liver specific marker (ALB) was upregulated after the differentiation. Furthermore, two stem cell markers were further confirmed at the protein level, the results showed that SOX2 and SOX9 was highly expressed in the cloned cells, as determined by immunofluorescence (Fig. 2E). Collectively, all these results revealed that the stemness of cloned cells was decreased after the differentiation.

Fig. 2.

Stemness analysis between LCSCs and differentiated cells. (A) Morphologies of LCSCs and differentiated cells. Scale bar 100 μm. (B, C) Determination of the percentages of ALDH positive cells between cloned cells and differentiated cells. n = 3 biological replicates. (D) The relative expression levels of putative LCSC marker genes (SOX2, SOX9, CD133, NCAM1, CD90) and differentiated marker gene (ALB) were quantified by qPCR in cloned and differentiated cells. n = 3 biological replicates. (E, F) Immunofluorescence staining was performed to detect the expressions of putative LCSC markers SOX2 (green) (E), and SOX9 (red) (F) in cloned and differentiated cells. Nucleus were counterstained by DAPI (blue), Scale bar 30 μm. Values were represented as mean ± SD. *P < 0.05 and **P < 0.01 and ***P < 0.001

Identification of tumorigenicity of liver cancer stem cells

To evaluate the tumorigenic ability of cloned cells, we sorted ALDH ⁺ cells from the cloned population and injected them subcutaneously into immunodeficient mice (Fig. 3A). Simultaneously, we injected parental tumor cells as controls, and the observation period for all mice was six months. Tumorigenic potential was observed within two months using 100 and 1,000 ALDH⁺ clone cells from three cell strains (HCC2, HCC12, HCC14). However, tumorigenicity with only 10 cells was observed exclusively in HCC12 cells. Then, we observed that the inoculation of 100 or 1,000 ALDH⁺ cloned cells resulted in almost 50% or more mice with the tumorigenesis. More importantly, the inoculation of 1,000 of HCC12 cloned cells resulted in 100% mice (8/8) with the tumorigenesis in about one month. In contrast, it was challenging for the primary tumor cells from patients to form xenografts even with the injection of 100,000 cells subcutaneously. The exception was the case from the cells of HCC14 patient, from which tumors were observed 111 days after the transplantation (Fig. 3B). Subsequently, we conducted hematoxylin and eosin (HE) staining of the xenografts produced by the cloned cells and compared it with the corresponding primary tumor tissues. The results indicated that the histologies of the xenografts was very similar to that of the primary liver cancers (Fig. 3C). Notably, the immunofluorescence staining results showed that the tissue of the xenografts also expressed liver-specific markers (ALB) and liver cancer markers (AFP) as the primary liver cancers did, further confirming that the cloned cells could generate tumors resembling the primary liver cancers (Fig. 3D). Furthermore, the statistical analysis of the immunofluorescence data also indicated that there were no significant differences between the two groups (Fig. 3E). It is extensively acknowledged that CSCs have a stronger ability to promote angiogenesis than ordinary tumor cells [33]. As expected, the tumors produced by cloned cells expressed higher level of CD31 (vascular marker) than those produced by primary tumor cells, which is consistent with previous reports (Fig. 3F, G and Figure S2). In conclusion, these in vivo experimental results demonstrated the characterization of cloned cells with tumorigenicity as CSCs. However, it appeared that clonal cells from different patients exhibited significant differences in tumor-forming ability. For example, HCC12 cells were able to form a greater number of tumors in a shorter period of time when compared to those from HCC2 cells, indicating higher expression levels of genes associated with proliferation to promote rapid tumor growth in HCC12 cells. These individual differences further confirmed the tumorigenic capacity of CSCs across patients.

Fig. 3.

Tumorigenicity of LCSCs and primary liver cancer cells. (A) ALDH⁺ cells from cloned growth cells were sorted after the removal of MEFs and injected into NCG mice. (B) The ability of tumorigenicity in vivo between primary liver cancer cells and LCSCs. (C) Hematoxylin and eosin staining of primary liver cancer tissues from patients and LCSC-derived xenografts formd by the injection of LSCSs into mice. Scale bar 50 μm. (D) Human liver-specific proteins, AFP (green), ALB (purple) were expressed in primary liver cancer tissue (PHCC2) and LCSC-derived xenografts (MHCC2), as determined by immunofluorescence staining. Nucleus were stained by DAPI. Scale bar 50 μm. (E) Bar graph showed the statistical immunofluorescent intensity of AFP and ALB immunostaining in PHCC2 and MHCC2, n = 3. Values were represented as mean ± SD, ns indicates no statistical significance. (F) Tumor formations by primary liver cancer cells and LCSCs (Left panel). Vascular marker CD31(red) was detected by immunofluorescence staining in LCSC-derived xenograft and primary liver cancer cells (Right). Scale bar 50 μm. (G) The bar graph displayed the statistical immunofluorescent intensity of CD31 in tumor tissue formed by primary cells and LCSCs, n = 6. Values were represented as mean ± SD, *P < 0.05

Comparison of CSCs between suspension culture and culture with feeder cells

CSCs are commonly cultured using serum-free suspension culture to maintain their self-renewal ability. To compare the effects of two different culture methods on CSCs and to determine which is more effective in preserving the stemness and proliferative capabilities of CSCs, we sorted ALDH⁺ CSCs and cultured them on MEFs as feeder cells (MEF group) and in low-attachment plates for suspension culture (SP group). CSCs cultured on MEFs exhibited uniform clonal growth, whereas those in suspension culture formed spheroids with uneven sizes and shapes (Fig. 4A). The average diameters of clones on MEFs were larger than those of spheroids from suspension culture 5 to 7 days after culturing (Fig. 4B, Figure S3A). At day 7, we assessed the rate of ALDH-positive cells under two distinct culture conditions. About 87% of HCC2 cells cultured on MEFs were tested positive for ALDH, whereas ALDH positivity rate in suspended cells was only 51.9%. HCC14 cells showed a similar trend. These results indicated that CSCs cultured on MEFs possessed more pronounced stem cell characteristics (Fig. 4C, Figure S3B). To elucidate the differences between MEF-cultured cells and those cultured under suspension culture, we performed transcriptome sequencing, and found that the expression of stemness-associated genes in MEF-cultured cells, including LGR5, PROM1, SOX2, KLF4, ITGA6, NCAM1, LIF, and SOX9, were elevated. In contrast, the expression of liver differentiated gene ALB, hepatocyte marker, was increased in suspension-cultured spheroids (Fig. 4D), indicating the occurrence of partial differentiation. To validate results of the transcriptome sequencing, qPCR was performed to verify the expression of selected stemness genes, confirming that these genes were more highly expressed in MEF-cultured cells. In contrast, the expression of ALB gene was increased in suspended culture cells. Especially, HCC2 CSCs reduced the expression of CD133, CTNNB1, NCAM1, and SOX2, were significantly reduced to 18.68%±6.11%, 6.03%±1.06%, 0.12 ± 0.16%%, and 34.42%±3.84%, and ALB expression was 9.31 ± 1.71 times higher in suspended CSCs when compared to MEF-cultured cells (Fig. 4E). Gene Set Enrichment Analysis (GSEA) further revealed significant enrichment of upregulated genes of MEF-cultured cells in pathways related to DNA repair, hypoxia and MYC target (Fig. 4F). These pathways are crucial for enhancing CSCs’ genetic damage repair capabilities, adapting to hypoxic tumor microenvironment, and regulating cell proliferation. In conclusion, clonal cells cultured on MEFs appeared to better preserve the undifferentiated state and stemness of CSCs, whereas spheroids under suspension culture condition showed signs of partial differentiation which could impact the purity, quantity, and growth microenvironment of CSCs, eventually diminishing the conditions favorable for stemness maintenance.

Fig. 4.

Comparison of LCSCs cultured under two different conditions. (A) Morphological comparison of LCSCs cultured on MEFs and in suspension condition respectively. Scale bar 100 μm (B) Comparison of the diameters of LCSCs cultured under two conditions. Values were represented as mean ± SD. *P < 0.05, ***P < 0.001. (C) Comparison of ALDH positive rates of LCSCs cultured under two conditions. (D) Heatmap showing the differences in stemness-related genes and differentiated gene (ALB) in LCSCs cultured under two conditions. The statistical method employed is the default analysis within the DESeq2 R package, and the adjusted P-values were used to indicate statistical significance. *P < 0.05, **P < 0.01, ***P < 0.001, ns indicates no statistical significance. (E) qPCR was performed to verify the expression levels of stemness genes and differentiated gene ALB in LCSCs cultured under two conditions. Values were represented as mean ± SD. *P < 0.05, **P < 0.01, ***P < 0.001, and **** P < 0.0001. (F) GSEA pathway enrichment analysis of LCSCs cultured under two conditions. MEF denoted LCSCs that were cloned and cultivated on MEFs as feeder cells, while Spheroid referred to LCSCs that were cultured under suspension condition

Gene expression analysis of LCSCs

To investigate the differences between LCSCs and their differentiated counterparts, we conducted transcriptome sequencing on LCSCs and cells collected 10 days after the differentiation of LCSCs. Our objective was to identify the changes in gene expression patterns between LCSCs and differentiated cells. Volcano plot revealed significant transcriptional differences with the upregulation of 638 genes and downregulation of 523 genes in differentiated cells compared to those in LCSCs (Fig. 5A). To further refine our analysis, we focused on the expression patterns of stemness-related genes in these two cell populations. Heatmap analysis demonstrated significant upregulation of stemness genes such as ASCL2, SEMA3F, SOX2, LIF, ITGA6, EPCAM, and KLF4 in LCSCs cultured on MEFs (Fig. 5B), indicating a higher state of maintenance in stemness, while differentiated cells exhibited a notable decrease in the expression of stemness-related genes. Subsequently, we performed GO function enrichment and KEGG pathway enrichment analyses on genes highly expressed in LCSCs, and the results showed predominant enrichment in key biological processes such as cellular modified amino acid metabolic process, response to toxic substance, fatty acid derivative metabolic process, tissue homeostasis, and extracellular matrix organization (Fig. 5C). These findings underscored the link between specific gene expression patterns and biological functions in LCSCs. The KEGG analysis showed that highly expressed genes were mainly involved in metabolic pathways related to protein digestion and absorption, drug metabolism-cytochrome P450, drug metabolism-other enzymes, PPAR signaling, and ECM-receptor interactions (Fig. 5D). Furthermore, transcription factor enrichment analysis of highly expressed genes in LCSCs revealed potential regulations by transcription factors including AHR, ARNT, SREBF1, SP1, and KLF4 among others (Fig. 5E). Based on these results, we selected transcription factors AHR, SREBF1, ESR1, SP1, CDX2, and SP3, along with their target genes, to construct a regulatory network of highly expressed genes and transcription factors in LCSCs (Fig. 5F), which was helpful to elucidate the regulatory mechanisms that underpinned the maintenance of stemness in LCSCs.

Fig. 5.

Transcriptome analysis of LCSCs and differentiated cells. (A) The volcano plot displayed the upregulated and downregulated genes in LCSCs on MEFs compared to those in differentiated cells (log2 Fold Change > |2|); (B) The heatmap illustrated the differences in stemness-related genes between LCSCs and differentiated cells. (C) GO functional enrichment results for different biological process in LCSCs; (D) KEGG pathway enrichment results for different pathways in LCSCs; (E) Enrichment results for highly expressed transcription factor genes in LCSCs. (F) The network diagram showed the regulatory relationships between transcription factors and target genes in LCSCs

Analysis of liver cancer classification based on highly expressed genes in LCSCs

To further investigate the impact of genes highly expressed in LCSCs in patients, we conducted a differential analysis of these genes between HCC tissues and adjacent non-cancerous tissues using TCGA-LIHC database. The analysis revealed that 67 genes were significantly upregulated in liver cancers (log2FC > 1), with a heatmap illustrating the top 30 most differentially expressed genes in LCSCs (Fig. 6A). Subsequently, we performed univariate Cox regression analysis on these 67 LCSC-related genes. The results suggested that the majority of highly expressed genes associated with LCSCs could serve as potential risk factors for the development of liver cancers. In contrast, P2RY8 and SLC16A11 genes with low hazard ratio indicated a protective effect against liver cancers. At present, the functions of these two liver cancer protective genes in solid tumors have not been widely reported. It has been shown that P2RY8 is a Gα₁₃-coupled receptor that mediates migration inhibition and regulates the growth of B cells in lymphoid tissues, primarily associated with immune modulation [34]. SLC16A11 is a proton-coupled monocarboxylate transporter, and genetic disruption of SLC16A11 can induce changes in fatty acid and lipid metabolism, which are related to an increased risk of type 2 diabetes. Therefore, enhancing the function of SLC16A11 may have potential therapeutic effects on type 2 diabetes [35]. These research findings provide new research directions for future investigation on these two genes.

Fig. 6.

Classification of liver cancer subtypes based on highly expressed genes in LCSCs within the TCGA-LIHC Database. (A) The heatmap displayed the top 30 differentially expressed LCSC genes in cancerous and para-cancerous tissues; (B) The line chart presented genes highly expressed in LCSCs that are associated with liver cancer prognosis; (C) The clustering results of liver cancer subtypes; (D) The boxplot verified the expression levels of LCSCs-related genes in two subtypes of liver cancers; (E) Survival analysis for patients with two liver cancer subtypes; (F) The heatmap showed the correlation between two liver cancer subtypes and clinical characteristics; (G) The expression levels of putative LCSC markers in two subtypes of liver cancer; (H) GO functional enrichment of genes highly expressed in C2 subtype of liver cancers; (I) The heatmap illustrated the relationship between two liver cancer subtypes and the tumor microenvironment

These results suggested that most genes highly expressed by LCSCs promoted the development of liver cancers (Fig. 6B). In order to identify cancer subtypes associated with LCSCs, considering the close relationship between LCSCs and malignant characteristics of cancer, such as the metastasis, drug resistance, the recurrence, and poor prognosis, we excluded genes related to protective factors and prioritized genes (KCNE3, ALDOA, PRTFDC1, SRXN1, LOX, ENTPD2, MFAP2, CREB3L1, RIBC2, CTHRC1, FAM133A, and MEP1A) associated with poor prognosis in LCSCs for consensus clustering analysis. The aim of this strategy was to enhance the precision of cancer classification for achieving more accurately personalized targeted drug treatments in the future.

To further deepen our understanding of these core genes, we used bar graphs to visually display their expression differences and levels between LCSCs and differentiated cells (Figure S4A). In addition, we conducted a correlation analysis on these core genes and found that most of them showed positive correlations with each other (Figure S4B). Moreover, we explored the correlation between these core genes and stemness-related genes in LCSCs and discovered that the majority of them were positively correlated with stemness-related genes, particularly MFAP2, CREB3L1, and CTHRC1 (Figure S4C). These findings not only highlighted the critical role of these genes in defining the characteristics of LCSCs but also provided valuable insights for future research directions and the selection of therapeutic targets.

Based on the results of the heatmap from the consensus clustering matrix, cumulative distribution function plot, Delta Area Plot, and Tracking Plot for k = 2 to 9, the K = 2 group analysis was ultimately selected (Figure S5A-D). Therefore, we classified liver cancer patients into two subtypes, with 219 patients in group C1 and 151 patients in group C2 (Fig. 6C). Consistently, HCC is also classified into two subtypes based on the genotyping of LCSC-related genes in ICGC database (Figure S6A-C). In this categorization, subtype C2 represented liver cancers with high expression of key genes related to LCSCs (Fig. 6D, Figure S6D). Further survival analysis revealed significant prognostic differences based on the patient grouping constructed using LCSC-related genes (Fig. 6E), with a better prognosis for subtype C1 and a poorer prognosis for subtype C2. We also used a heatmap to display the differential genes between these two subtypes of liver cancers and their correlation with clinical characteristics (Fig. 6F). The analysis found that subtype C2 of liver cancers was significantly associated with clinical stage, grade, and gender, with a higher prevalence of late-stage, high-grade, and female patients in subtype C2. Additionally, genes known to be related to stemness, such as SOX9, EPCAM, KRT9, LIF, CD44, DLK1, SOX4, KLF4, PROM1, CD24, POU5F1, and AFP, were more highly expressed in subtype C2(Fig. 6G, Figure S6E). These results further confirmed that subtype C2 of liver cancers possessed more stemness and malignant characteristics. Furthermore, we have observed that the expression levels of these key genes are higher in liver cancer tissues compared to the adjacent non-cancerous tissues within the ICGC database (Figure S6G).

To identify biological processes enriched in subtype C2, GO enrichment analysis was conducted on genes that were highly expressed in patients with this subtype. The results showed significant enrichment in processes such as extracellular matrix organization and immune regulation (Fig. 6H, Figure S6F). To corroborate these findings, tumor microenvironment-related analyses were performed, and the results indicated that subtype C2 of liver cancers exhibited elevated stromal and immune scores, strongly correlating with the infiltration levels of various immune cells, including fibroblasts, natural killer (NK) cells, CD8⁺ T cells, B cells, monocytes, and myeloid-derived suppressor cells (MDSCs). Notably, neutrophil infiltration levels did not show this correlation (Fig. 6I). Thus, these results highlighted the close association between subtype C2 of liver cancers and the tumor microenvironments.

To investigate whether there is a difference in immunotherapy response between the two subtypes of hepatocellular carcinoma, we applied two predictive models, TCIA and TIDE, to evaluate the potential effect on the immunotherapy. However, the results showed that there was no significant difference in immunotherapy response between patients with hepatocellular carcinoma of C1 and C2 subtypes (Figure S7A, B). Notably, the expression levels of CTLA4 and CD274 (PD-L1) genes were significantly higher in C2 subtype compared to C1 subtype (Figure S7C). Theoretically, C2 subtype might respond better to the immunotherapy, but this was not the case in practice. More interestingly, in TIDE score analysis, we observed a higher expression level of MDSCs in C2 subtype and a positive correlation with the key marker genes of MDSCs, ITGAM and CD33 (Figure S7D, E). MDSCs are capable of suppressing the immune microenvironment [36], which may explain the reason, despite the higher expression of immune checkpoint genes in C2 subtype, it was not responsive to the immunotherapy. These findings revealed the complexity of the tumor immune microenvironment and indicated that co-targeting of genes associated with LCSCs might be necessary to improve the efficacy of the immunotherapy.

Stratification of key genes in LCSCs through single-cell transcriptomic analysis

To investigate the association between the expression of high-risk LCSC-related genes (KCNE3, ALDOA, PRTFDC1, SRXN1, LOX, ENTPD2, MFAP2, CREB3L1, RIBC2, CTHRC1, FAM133A, and MEP1A) and the extracellular matrix (ECM) composition, we acquired and analyzed single-cell sequencing data of HCC from GEO database. Initially, the dataset underwent rigorous preprocessing followed by dimensionality reduction and clustering via Uniform Manifold Approximation and Projection (UMAP). This analysis delineated the cellular heterogeneity within samples by categorizing cells into six populations, designated as tumor cells, T cells, B cells, macrophages (macro), cancer-associated fibroblasts (CAF), and endothelial cells (EC), using established lineage-specific markers (Fig. 7A, B). After segregating the tumor cells, we applied the Ucell algorithm to compute the distribution of stemness-related gene scores within the dimensionality-reduced space. The results of the analysis revealed varying stemness profiles, with samples HCC03T and HCC0T showing lower scores, while HCC08T, HCC09T, and HCC10T exhibited higher scores (Fig. 7C, D). Pseudotime trajectory analysis using Monocle 2 package indicated a differentiation trend where cells with elevated stemness scores progressed towards a state with reduced stemness (Fig. 7E-H). The epithelial-mesenchymal transition (EMT) scores calculated for these tumor cells correlated with stemness, with higher EMT indices observed in patients with higher stemness scores (Fig. 7I). By analyzing the expression of classic LCSC-related genes, we found that the expression levels of these genes (PROM1, CD24, CD44, EPCAM, POU5F1, KLF4, LIF, AFP, DLK1, KER19 and SOX9) were correspondingly elevated in patients with high stemness scores (Fig. 7J). Finally, the analysis within fibroblast subset unveiled an upsurge in the expression of collagen-associated genes (COL1A1, COL1A2, and COL3A1) in patients displaying high stemness scores, suggesting a potential link to enhanced collagen deposition within ECM (Fig. 7K, L). In summary, these results further confirmed that patients expressing high-risk LCSC-related genes had stronger EMT indices and stemness characteristics and might be closely related to collagen deposition in ECM.

Fig. 7.

Analysis of prognosis-associated LCSC gene sets in single-cell sequencing data. (A) Umap-based clustering delineated primary liver cancer cell types; (B) The bubble chart represented the expressions of marker genes characteristic of distinct cellular subpopulations; (C) Extraction of tumor cell subsets from the dataset; (D) Ucell algorithm analysis of the scoring for prognosis-associated LCSCs gene sets; (E) Extraction of patient subsets for HCC03, HCC05T, HCC08T, HCC09T, and HCC10T; (F, G, & H) Illustrations of pseudotime analysis for tumor cells across five patients within the dataset; (I) Ucell algorithm analysis of EMT gene sets scoring in tumor cells from 5 patients; (J) The bubble chart displayed the expressions of LCSC markers in these five patients; (K) Extraction of CAF subsets from these five patients; (L) Analysis of the expressions of fibroblast and collagen-related genes among five patients

Predicting drug sensitivity using LCSC-associated gene scores

In an effort to understand the influence of LCSC-related genes on drug sensitivity of LIHC, we carried out additional drug sensitivity predictions. The results were presented in the form of box plots, revealing a significant statistical difference in drug sensitivities between patients with subtype C1 or C2 of liver cancers. Notably, subtype C1 patients showed heightened sensitivity to 5-FU, ABT-737, Gefitinib, ERK inhibitors, Foretinib, and Erlotinib (Fig. 8A-F). In contrast, subtype C2 patients exhibited increased sensitivity to Cisplatin, Axitinib, JAK1 inhibitors, WNT-c59, Sorafenib, and RO-3306 (Fig. 8G-L). These results indicated that stratifying HCC patients based on LCSC-related genes could facilitate the personalized treatment regimens according to their predicted drug sensitivities.

Fig. 8.

Drug sensitivity analysis of LCSC-related genes in LIHC Patients. (A-F) Box plots depicted drug sensitivity predictions for patients with subtype C1 of liver cancers, highlighting increased sensitivity to 5-FU, ABT-737, Gefitinib, ERK inhibitors, Foretinib, and Erlotinib. (G-L) Box plots illustrated drug sensitivity predictions for patients with subtype C2 of liver cancers, indicating heightened sensitivity to Cisplatin, Axitinib, JAK1 inhibitors, WNT-c59, Sorafenib, and RO-3306

Discussion

Liver cancer, one of the top three malignant tumors in terms of global mortality [2], is primarily treated with targeted drugs such as sorafenib. Although these first-line drugs have achieved some success in improving patient survival, the issues of emerging drug resistance and high recurrence rates remain significant challenges [37]. Studies have shown that the presence of CSCs is one of the key factors leading to therapeutic resistance and relapse metastasis [38].

LCSCs, a small subset within the tumors, are characterized by their enhanced self-renewal, differentiation, and tumorigenic capabilities [39]. Furthermore, CSCs are also resistant to chemotherapy or radiotherapy due to their high expression of drug efflux pumps, quiescent properties, and DNA repair mechanisms [40, 41]. Therefore, the traditional treatments of tumors cannot eliminate CSCs, and may even induce the formation of CSCs. Over the past decades, numerous LCSC-related markers have been identified such as CD133, CD44, CD90, CD24, CD47, OV6, CD49f, CD13, ICAM1, CK19, EPCAM, but there is no standardized criterion to define which are unique markers of LCSCs. This is also a problem and challenge for the elimination of CSCs. Therefore, the development of novel approach to isolate and culture CSCs is very important to find specific targets on CSCs for treating tumors. Currently the most common method to isolate and culture patient-derived CSCs were cell sorting using specific markers and culturing them with spheroid culture. However, the stemness of CSCs was difficult to maintain, because spheroid culture did not provide CSC niche [42]. It is well known that the tumor microenvironment (TME) was very important for the occurrence and development of tumors, and the CSC niche was part of the TME provided CSC niche for the formation and maintenance of CSCs [42]. It had been found that CAF [43], tumor-associated macrophage(TAM) [44, 45], and tumor-associated endothelial cells [46, 47] in TME were conducive to maintaining the stemness of CSCs.

Herein, we have established a new method to isolate and culture LCSCs derived from primary tumor tissues. In this method, we used MEFs as feeder cells to provide stem cell niche to maintain the stemness of CSCs as MEFs do for human embryonic stem cells. Thus, we have successfully isolated and cultured 16 CSC strains from 24 patients with three liver cancer types by using defined medium with MEFs as feeder cells. Notably, seven strains could be maintained long-term culture (Supplementary Table 3). Our results indicated that this developed method might be used for the isolation and culture of CSCs of other cancers without using any specific molecular markers of CSCs. Interestingly, when we performed immunofluorescence staining on cloned cells with previously reported LCSC markers (CD133, EPCAM, CD44, CD24, AFP, OV6, NCAM1, SOX2, SOX9) to identify the LCSCs, and found that all aforementioned LCSC markers were expressed in each strain, but not each clone in every strain expressed all these LCSC markers, indicating that LCSCs were heterogeneous.

To further validate the advantages of our stem cell technology-based LCSC isolation method, we compared the growth characteristics and transcriptome profiles of LCSCs grown in clonal colonies on MEFs with those cultured in suspension culture condition. The results revealed that LCSCs grown in clonal colonies on MEFs exhibited higher expression of stemness markers such as CD133, CD90, and NCAM1, as well as higher positive rate of ALDH. In contrast, cells in suspension culture showed higher expression of ALB, a liver differentiation marker, and a significant decrease in ALDH positivity, indicating that LCSCs in suspension culture underwent a certain degree of the differentiation, leading to a reduction in the purity and quantity of LCSCs.

In this study, we conducted HCC subtyping analysis based on genes highly expressed in LCSCs that are associated with poor prognosis (KCNE3, ALDOA, PRTFDC1, SRXN1, LOX, ENTPD2, MFAP2, CREB3L1, RIBC2, CTHRC1, FAM133A, and MEP1A), and found that patients with C2 subtype of liver cancers highly expressed stemness-related genes and are associated with ECM and the tumor immune microenvironment. Additionally, we performed validation analysis using single-cell data from GEO database, and found that patients with high expression of stemness-related genes indeed also had higher expression of ECM-related proteins COL1A1, COL1A2 and COL3A1, further illustrating the link between stemness-related genes and ECM. Although the C2 subtype showed a close correlation with the immune microenvironment, there was no significant difference between the C1 and C2 subtypes in predictive analyses of immunotherapy efficacy, which may be related to the higher MDSCs scores of hepatocellular carcinoma patients with the C2 subtype. MDSCs are a group of myeloid cells with immunosuppressive functions, which can inhibit the activation and proliferation of T cells and reduce cytotoxic T cells (CTLs) and natural killer (NK) cell activity, thus suppressing anti-tumor immune responses [48]. Therefore, even if the C2 subtype highly expresses CD8⁺ T cells, PD-L1 and CTLA4, it is less effective against immune checkpoint inhibitors due to the presence of MDSCs. Recent studies have indicated that ENTPD2 can inhibit the differentiation of MDSCs, promote the maintenance of MDSCs, leading to an immunosuppressive microenvironment [48]. Furthermore, in another study it was found that ENTPD2 is also able to inhibit the function of CD8⁺ T cells thereby accelerating tumor progression in colon cancer [49]. Targeting ENTPD2 can reduce cancer growth and improve the efficacy of immune checkpoint therapy [48, 49]. Moreover, research indicates that MFAP2⁺ CAFs in gastric cancer establish an immunosuppressive microenvironment marked by the impairment of immune effector cell functions, which correlates with a poor response to immunotherapies [50]. Therefore, targeting these LCSC-associated genes may ameliorate the tumor’s immunosuppressive microenvironment, thereby enhancing the efficacy of immunotherapy.

Additionally, we utilized a computational predictive model to analyze the drug sensitivity of patients with different subtypes of HCC. In the predictive analysis of 198 drugs, we found that patients with C1 and C2 subtypes exhibited varying sensitivities to certain medications. Notably, patients with the C2 subtype showed greater sensitivity to drugs such as cisplatin, axitinib, JAK1 inhibitors, WNT-c59, sorafenib, and RO-3306. We speculate that this higher sensitivity to these drugs in C2 subtype patients may be related to the differentially expressed signaling pathways between the two subtypes. Interestingly, our GSVA analysis revealed that C2 subtype tumors are significantly enriched in signaling pathways such as VEGF, cell cycle, JAK, WNT, and mTOR (Figure S8.), which may explain the increased sensitivity of patients with C2 subtype HCC to the aforementioned drugs.

In summary, we successfully developed effective and efficient approach to isolate and culture LCSCs from primary liver cancers without relying on molecular markers. This method has been shown to better preserve the stemness of LCSCs compared to traditional suspension culture. Furthermore, we also could subtype liver cancers and validate the predicted drug sensitivities based on genes highly expressed in LCSCs,

This discovery offers a new strategic direction for the treatment of advanced liver cancer. In the future, we plan to obtain tissue samples through biopsy and subsequently culture LCSCs in vitro. Concurrently, we will utilize genomic sequencing technology to conduct precise personalized analyses for patients. Based on the predictive outcomes of these analyses, as well as data from in vitro drug sensitivity tests, we will be able to tailor personalized treatment plans for patients. We anticipate that this precision medicine-based personalized treatment approach will significantly enhance the specificity and effectiveness of treatments, reduce the side effects of ineffective or suboptimal treatments, and thereby improve the quality of life and clinical outcomes for liver cancer patients. Moreover, this method is expected to be extended to the treatment of other types of cancer, bringing hope to more cancer patients. We look forward to further research and clinical trials to transform this innovative treatment method into practical clinical applications for the benefit of liver cancer patients worldwide.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Abbreviations

MEFs

Mouse embryonic fibroblasts

hESCs

Human embryonic stem cells

LCSCs

Liver cancer stem cells

HCC

Hepatocellular carcinoma

FBS

Fetal Bovine Serum

HPLC

Human primary liver cancer

TACE

Trans-arterial chemoembolization

CSCs

cancer stem cells

MDSCs

myeloid-derived suppressor cells

UMAP

Uniform Manifold Approximation and Projection

CAFs

cancer-associated fibroblasts

EC

Endothelial cells

EMT

Epithelial-mesenchymal transition

TME

Tumor microenvironment

Author contributions

YD, WG, TG and SZ contributed to the conceptualization and methodology, TG designed and implemented the experiments, collected and analyzed the data, and prepared the draft of the manuscript. SZ collected and prepared clinical samples. WZ and YL performed experiments and collected data, JX, SL, WZ, YQ, YF, YL, YO, KM and BW contributed to collecting data. YD revised the manuscript. YD and WG provided financial support, and approved the manuscript.

Funding

This study was supported in part by Research Starting Funding of South China University of Technology (D6181910, D6201880, K5180910, and K5204120), by Research Starting Funding of the Second Affiliated Hospital of South China University of Technology (Grant No. KY09060026), by Research Agreement between South China University of Technology and Guangzhou First People’s Hospital (D9194290), by Key Project of Guangzhou Science and Technology Program (202102010027).

Data availability

The data sets used or analyzed in the current study could be obtained from the corresponding authors upon reasonable request. The RNA-seq datasets have been deposited in Gene Expression Omnibus (GEO) with accession number GSE274830.

Statement

Ethics approval and consent to participate

Human liver tissues used in this study was approved by Ethics Committee of Guangzhou First People’s Hospital (Approval Number: K-2019-167-02). All animal experiments were approved by Ethics Committee of South China University of Technology (SCUT) (Approval Number: 2019073) for the Use and Care of Laboratory Animals.

Competing interests

The authors declare that they have no competing interests.