Heterogeneity, inherent and acquired drug resistance in patient-derived organoid models of primary liver cancer

Abstract

Purpose

We aimed to elucidate the applicability of tumor organoids for inherent drug resistance of primary liver cancer (PLC) and mechanisms of acquired drug resistance.

Methods

PLC tissues were used to establish organoids, organoid-derived xenograft (ODX) and patient-derived xenograft (PDX) models. Acquired drug resistance was induced in hepatocellular carcinoma (HCC) organoids. Gene expression profiling was performed by RNA-sequencing.

Results

Fifty-two organoids were established from 153 PLC patients. Compared with establishing PDX models, establishing organoids of HCC showed a trend toward a higher success rate (29.0% vs. 23.7%) and took less time (13.0 ± 4.7 vs. 25.1 ± 5.4 days, p = 2.28 × 10⁻¹³). Larger tumors, vascular invasion, higher serum AFP levels, advanced stages and upregulation of stemness- and proliferation-related genes were significantly associated with the successful establishment of HCC organoids and PDX. Organoids and ODX recapitulated PLC histopathological features, but were enriched in more aggressive cell types. PLC organoids were mostly resistant to lenvatinib in vitro but sensitive to lenvatinib in ODX models. Stemness– and epithelial–mesenchymal transition (EMT)–related gene sets were found to be upregulated, whereas liver development– and liver specific molecule–related gene sets were downregulated in acquired sorafenib-resistant organoids. Targeting the mTOR signaling pathway was effective in treating acquired sorafenib-resistant HCC organoids, possibly via inducing phosphorylated S6 kinase. Genes upregulated in acquired sorafenib-resistant HCC organoids were associated with an unfavorable prognosis.

Conclusions

HCC organoids perform better than PDX for drug screening. Acquired sorafenib resistance in organoids promotes HCC aggressiveness via facilitating stemness, retro-differentiation and EMT. Phosphorylated S6 kinase may be predictive for drug resistance in HCC.

Supplementary Information

The online version contains supplementary material available at 10.1007/s13402-022-00707-3.

Introduction

Primary liver cancer (PLC) is the sixth most commonly diagnosed cancer and the third leading cause of cancer death worldwide [1]. In China, PLC remains the second cause of cancer death. The age-standardized mortality rate is higher in the 40–64-year-old population than in the ≥65-year-old population [2]. The major histotypes of PLC are hepatocellular carcinoma (HCC; comprising 94.6%), intrahepatic cholangiocarcinoma (ICC; 3.7%), and combined hepatocellular cholangiocarcinoma (CHC; 1.7%) in Eastern China [3]. Surgical resection is the mainstay of curative treatment. Frequent postoperative recurrence is the major obstacle to improving the prognosis. Approximately 70% of HCC patients relapse within five years after hepatectomy and the incidence of post-liver transplantation recurrence remains 10–20% [4, 5]. As yet, effective treatment options for postoperative recurrence are scarce.

Effective therapeutic options for PLC recurrence are hindered by a shortage of reproducible and reliable human models. Tumor-derived cell lines have long been applied to study the underlying biologic processes and as platforms for discovering and evaluating anticancer therapeutics. Although tumor cell lines have allowed pioneering advances in cancer biology, they may not recapitulate critical characteristics and the heterogeneity of PLC in situ. Patient-derived xenograft (PDX) models are immunodeficient mice engrafted with patient tissues. Implantation of small pieces of tumors into immunodeficient mice allows tumor growth and subsequent transplantation into secondary recipient mice [6]. PDXs maintain the proportions of cancer cells and stromal cells, histopathological structures, major mutations, gene expression profiles and inflammatory microenvironments of original tumor tissues [6, 7]. These characteristics indicate that PDXs may be more useful as predictive experimental models of therapeutic responses. However, PDX models have shortcomings. PDXs are expensive with a low transplantation success rate and require long culture times [6]. Moreover, the original human stromal cells in the tumors are gradually replaced by the stromal cells of the mice as the xenografts grow [8]. Recently, patient-derived liver organoids including PLC organoids have been developed [9–14]. An organoid is a multi-cell mass constructed from a 3D culture in vitro. Tumor organoids have been proven to retain the morphology, genetic heterogeneity and expression pattern of the original PLC tumor markers. Patient-derived organoids bridge the gaps between cancer cell lines cultured in vitro and PDX models cultured in vivo. These organoids not only help to elucidate the molecular mechanism of oncogenesis, but also provide an opportunity of drug screening for precision treatment.

However, several key issues remain to be addressed to construct a platform of PLC organoids for precision medicine. First, the factors affecting the success rate of establishing organoids are not established, due to the small sample sizes of surgically removed tissues applied in each study. The numbers of patients from whom the organoids developed were eight [10], eleven [11], five [12], four [13] and seven [14] in previous studies. Second, the success rate of establishing organoids is low, 26% in PLC [11]. Third, acquired drug resistance reflecting the evolution of PLC has not been evaluated in tumor organoids, in which the heterogeneity should be distinct from that in cell lines [15]. Here, we successfully established 52 tumor organoids from 153 patients and evaluated the factors affecting the development of organoids in HCC. We show that PLC organoids recapitulate the histological features of the original tumors, but enriched aggressive cell types in vitro and in vivo. We also generated acquired sorafenib-resistant HCC organoids and found that targeting mTOR signaling was effective in treating HCC with sorafenib resistance. Our study not only provides comprehensive technology to establish organoid platforms for drug selection, but also elucidates the mechanism by which PLC evolves under sorafenib pressure.

Materials and methods

Patients and surgically removed cancer specimens

Patients who received surgical resection of their tumors by Prof. Zhao’s group at the Eastern Hepatobiliary Surgery Hospital (Shanghai, China) were included. Demographic information, pathological examinations including tumor nodule number, tumor capsule integrity and cancer embolus including microvascular invasion (MVI), and the results of preoperative laboratory examinations including serum α-fetoprotein (AFP), carbohydrate antigen 19–9 (CA19–9) and HBV parameters, routine blood tests and liver function tests were extracted from electronic medical records. The clinical information of the included patients is listed in Table S1. This study was approved by the Ethics Committee of Eastern Hepatobiliary Surgery Hospital. Each patient provided informed consent before the enrollment. Each surgically removed tumor specimen was divided into three parts. One part was processed to generate organoids and PDX. Another part was snap-frozen in liquid nitrogen for RNA extraction. The remaining tissue was fixed in 10% formalin and paraffin-embedded for histological and immunohistochemical analyses. For the culture of organoids, the samples were delivered to our laboratory at 4 °C and processed immediately after excision. The pathological examination was performed thereafter.

Culture of PLC organoids

Tumor-derived organoids were cultured as follows: tumor specimens were minced into pieces of roughly 1–2 mm³ and incubated with a digestion solution at 37 °C for 0.5–1 hr. The digestion solution contained 0.125 mg/ml collagenase IV (Sigma, St. Louis, MO, USA) and 0.1 mg/ml DNase (Sigma). The suspension was filtered through a 100 μm nylon cell strainer and centrifugally separated at 450 rpm for 3 min. The pellet was washed with pre-cooled advanced DMEM/F12 (GIBCO, Billings, MT, USA) and then mixed with Matrigel (CORNING, Corning, NY, USA). A total of 3 × 10³ - 5 × 10³ cells were then seeded into pre-warmed 24-well plates. After solidification of Matrigel (approximately 15–20 min), organoid culture medium was added to the cells. Organoid medium contained advanced DMEM/F12 supplemented with 1% penicillin/streptomycin (GIBCO), 1% glutamax (GIBCO), 10 mM HEPES (GIBCO), 1:50 B27 supplement (GIBCO), 1:100 N2 supplement (GIBCO), 1.25 mM N-acetyl-L-cysteine (Sigma, St. Louis, MO, USA), 10 mM nicotinamide (Sigma), 50 ng/ml recombinant human epidermal growth factor (Peprotech, Cranbury, NJ, USA), 100 ng/ml recombinant human fibroblast growth factor 10 (Peprotech), 25 ng/ml recombinant human hepatic growth factor (HGF) (Peprotech), 10 μM forskolin (Selleckchemicals, Houston, TX, USA), and 5 μM A8301 (Selleckchemicals). The culture medium was changed twice a week. Organoids were passaged at a 1:3 split ratio upon the attainment of dense cultures and then frozen in liquid nitrogen. All organoid cultures were tested for mycoplasma contamination every three months using a Lookout Mycoplasma PCR Detection Kit (Sigma). For each randomly selected case of HCC, ICC, and CHC organoids, three biological replicates were set up.

Xenografts of organoids and PDX models in nude mice

All animal experiments including organoid-derived xenografts (ODX) were conducted in accordance with guidelines of animal welfare and were approved by the Institutional Animal Care and Use Committee of Second Military Medical University. For developing ODX, 1 × 10⁶ cells were released from Matrigel by incubation in Cell Recovery Solution (CORNING), resuspended in 100 μl 50:50 (v/v) matrigel/culture medium, and then injected subcutaneously into 6–8 week old nude mice (Shanghai Jihui Laboratory, Shanghai, China) to generate ODX models. PDX models were generated as follows: surgically removed tumor samples were collected in serum-free DMEM media within 30 min after resection and kept on ice. The time from collection to implantation in nude mice was less than 3 hours. Tumors were cut into fragments (2–3 mm³) and implanted subcutaneously into the flanks of three week old male nude mice. The nude mice were anesthetized using 3% isoflurane. A dorsal midline incision (< 10 mm) was made at the level of the flank. Tumor fragments were placed in bilateral subcutaneous pockets after which the incisions were closed. The tumors were measured every two days using a caliper. When the tumor diameter reached 2 cm, the mice were sacrificed. Tumors were weighed and excised into small pieces and then implanted into another group of nude mice. The remaining tumors were processed for histology and immunohistochemistry.

To assess the anti-tumor efficiency of sorafenib (LC Laboratories, Woburn, MA, USA), regorafenib (LC Laboratories), lenvatinib (LC Laboratories) and RAD001 (LC Laboratories)/TAK228 (Selleck, Houston, TX, USA)/phenformin (Selleck) (RTP) in vivo, mice were randomized by splitting size-matched tumors into five groups. The sorafenib/regorafenib/lenvatinib group received 30 mg/kg by gavage per day. In the RTP group, RAD001 (2.5 mg/kg) and TAK228 (0.3 mg/kg) were orally administrated by gavage and phenformin (100 mg/kg) was administrated by intraperitoneal injection every day as previously described [16]. The mice of vehicle group were gavaged with ddH₂O at equal volumes to the oral drugs and injected with saline at an equal volume to phenformin. The treatment took 30 days. For animal welfare, the HCC-118 experiment was prematurely stopped on day 18 due to excessive tumor growth. The tumor sizes were measured and recorded every three days with a caliper. The tumor volumes were calculated using the formula Vol = (L × W²)/2, where L and W are the long and short diameters of the tumor, respectively. The tumor weights were measured when the mice were sacrificed.

Histological analysis and immunohistochemistry

Tumor tissues, organoids, ODX, and PDX were fixed in 4% paraformaldehyde and embedded into paraffin blocks. Sections were subjected to H&E staining and immunohistochemical examinations. Immunoreactive scores for each antibody were evaluated as previously described [17]. The primary antibodies are listed in Table S2.

Inherent drug response of tumor organoids to targeted drugs

The sensitivity of tumor organoids to each targeted drug was evaluated in vitro. Tumor organoids were digested and seeded at the density of 2–5 × 10³ cells in 8 μl Matrigel droplets. On Day 6, each drug compound with a series of experimental concentrations was added to the cultures. The concentration ranges of the drugs were based on previously reported data [10, 16, 18]. Six days after treatment, cell viability was measured using Thiazolyl Blue Tetrazolium Bromide reagent (MTT, Sigma) in a Synergy H1 Multi-Mode Reader (BioTek Instruments, Winooski, VT, USA). Dimethyl sulfoxide (DMSO) (Wak Chemie Medical Gmbh, Steinbach/Ts, Germany)-treated tumor organoids were defined as 100% viability, and the results were normalized to the DMSO control. All experiments were performed in triplicate.

Establishment of acquired sorafenib-resistant HCC organoids

Tumor organoids from four HCC patients were randomly selected and cultured with sorafenib with an increasing concentration ranging 1.2–10 μM (20% per passage) for about 3–5 months. The half-maximal inhibitory concentration (IC₅₀) of sorafenib-resistant organoids was determined by MTT assay [8]. Sorafenib-resistant tumor organoids were then maintained in medium with 4 μM of sorafenib.

RNA sequencing (RNA-Seq)

Total RNA was extracted from HCC tissues or organoids using TRIzol (Invitrogen, Carlsbad, CA, USA). RNA concentration and integrity were evaluated using a NanoDrop 2000 Spectrophotometer (Thermo Fisher Scientific, Waltham, MA, USA) and an Agilent 2100 Bioanalyzer (Agilent Technologies, Santa Clara, CA, USA), respectively. Purified RNA was preserved at −80 °C. The libraries were constructed using TruSeq Stranded Total RNA with Ribo-Zero Gold (Illumina, San Diego, CA, USA) and then sequenced on an Illumina NovaSeq 6000 sequencing platform (Illumina, San Diego, CA, USA). Reads with 150 bp paired-ends were generated. The adapters were removed using Trimmomatic software [19]. A NGSQCToolkit was applied to perform quality control [20]. The read pairs were mapped to the reference genome of NCBI (GRCh38) by HISAT2 [21]. The transcriptomes were assembled using StringTie and compared with reference gene annotation using Cuffcompare software [22, 23]. Finally, the read pairs were mapped to the pool of assembled transcripts by bowtie2 and the abundances of transcripts were estimated by performing eXpress [24, 25]. Hierarchical clustering analysis (Pearson correlation-based distance, Ward method) was performed to investigate the variation among samples.

Differential gene expression was analyzed using the edgeR package [26]. The genes with a fold change (FC) ≥ 2 and a false discovery rate (FDR) < 0.05 (obtained by Benjamini–Hochberg method) were considered as differentially expressed. The fold changes were collected to perform the “Prerank” method in gene set enrichment analysis (GSEA) [27]. Gene sets of chemical and genetic perturbations in MsigDB v7.4 (http://software.broadinstitute.org/gsea/index.jsp) were imported into the GSEA and randomized for 1000 times. Gene sets with a FDR < 0.01 were considered significantly enriched. The original RNA-Seq data of tumor tissues were deposited in the Sequence Read Archive (SRA) database with accession PRJNA841062. The original RNA-Seq data of organoids were deposited in the Gene Expression Omnibus (GEO) database with accession GSE182593. The Cancer Genome Atlas (TCGA) HCC (LIHC) dataset and the corresponding patients’ clinical information were retrieved from the Xena website (https://xenabrowser.net).

Western blotting

Tumor organoids were released from the Matrigel through incubation with cell recovery solution and then lysed with ice-cold Pierce RIPA Buffer with Halt Protease Inhibitor Cocktail (Thermo Fisher, Waltham, MA, USA). The remaining steps were performed as previously described [13]. All experiments were conducted in triplicate. All antibodies used are listed in Table S2.

Statistical analysis

All statistical tests were performed using R platform (v 4.0.2) and GraphPad Prism v7 (GraphPad Software, San Diego, CA, USA). The differences between two or multiple groups were determined by t-test or ANOVA. All tests were two-sided, and the threshold of α was 0.05 except when explicitly stated. The false discovery rate was calculated by Benjamini–Hochberg method. Curve fitting was performed using GraphPad Prism software and nonlinear regression equation.

Results

Establishment of tumor organoids and PDX models

Tumor organoids were cultured using surgically removed primary tumor tissues of 153 patients (131 with HCC, 17 with ICC, and five with CHC) from January 2018 to January 2019, from which we successfully constructed organoids from the tissues of 38 patients with HCC (29.0%), nine with ICC (52.9%) and five with CHC (Fig. S1a). The organoids from HCC and ICC were frozen in liquid nitrogen and then thawed for in vitro culture for several (3–5) cycles. This process lasted for up to 3.5 years and the first established organoids have been passaged up to 15 times. The 15th passage organoids of three randomly selected cases were subjected to H&E staining together with the first passage of these organoids. We found that for each case, the first and 15th passage organoids showed the same histological features (Fig. S2). We also successfully developed PDX models using tissues from 31 patients with HCC (23.7%), seven with ICC (41.2%) and three with CHC (Fig. S1b). The duration of HCC organoid culture from tissue separation to the first passage was significantly shorter than that of PDX development from the first tumor implantation to the second implantation (13.0 ± 4.7 vs. 25.1 ± 5.4 days, p = .28 × 10⁻¹³) (Fig. S1c, d).

Factors affecting the success rate of developing organoids and PDX models

To identify factors influencing the success rate of constructing HCC organoids, we compared the clinical data of patients whose tissues were successfully developed to organoids with those whose tissues failed. We found that larger tumor size (tumor diameter more than 5 cm, p = 0.003), MVI (p = 0.014), macrovascular invasion (p < 0.001), advanced TNM stage (p = 0.006), advanced Barcelona Clinic Liver Cancer (BCLC) stage (p = 0.010), and elevated AFP serum level (more than 20 μg/l, p = 0.010) were associated with a successful development of HCC organoids (Table 1). Larger tumor size (p = 0.005), MVI (p = 0.014), macrovascular invasion (p = 0.001), advanced TNM stage (p = 0.004), and advanced BCLC stage (p = 0.010) were also associated with a successful PDX establishment (Table S3).

Table 1.

Baseline characteristics of 131 HCC patients from whom tumor organoids were developed

Variable	No. of Patients		P value
	Successfully developing organoid	Unsuccessfully developing organoid
All cases	38	93
Age(year) >50 vs. ≤50	17:21	63:30	0.014*
Gender male vs. female	31:7	84:9	0.274
HBsAg + vs. -	34:4	79:14	0.495
HBeAg + vs. -	10:28	22:71	0.748
Liver cirrhosis + vs. -	20:18	41:52	0.374
Liver function Child A vs. B	36:2	89:4	1.000
AFP (μg/L) >20:≤20	30:8	51:42	0.010*
Tumor size(cm) >5:≤5	30:8	47:46	0.003*
No. tumor nodule: Solitary vs. multiple	32:6	78:15	0.962
Differentiation I + II vs. III + IV	0:38	5:88	0.340
Micro vascular invasion + vs. -	31:7	55:38	0.014*
Macro vascular invasion + vs. -	11:27	4:89	<0.001*
TNM stage I vs. II + III + IV	5:33	35:58	0.006*
BCLC stage 0-A vs. B-C	2:36	23:70	0.010*

RNA-Seq analysis was applied to investigate molecular features of HCCs from which organoids were successfully established. Four HCC samples that successfully developed organoids and another four that failed were subjected to RNA-Seq. To exclude the influence of clinical characteristics, we matched HCC samples with BCLC stage C, advanced TNM stage, a tumor diameter more than 5 cm, serum AFP more than 20 μg/l, and MVI between HCC tissues that successfully developed organoids and those that failed (Table S4). The median number (interquartile range, IQR) of the read pairs generated from organoid samples was 32.7 × 10⁶ (32.6 × 10⁶–33.5 × 10⁶). The rates of high-quality reads and alignment rates were close to 100% (Fig. S3). Hierarchical clustering of the data indicated that there were big differences in the expression patterns between the two groups of HCC samples (Fig. 1a). A total of 132 genes were significantly up-regulated (fold change >2 and p < 5 × 10⁻³) in HCC samples that successfully developed tumor organoids, compared to the failed ones (Table S5). The top 5 significantly up-regulated genes were MMP12, SLC1A7, TREM1, CLEC5A, and POSTN. GSEA demonstrated that 119 cancer-related gene sets were significantly enriched in HCC samples that successfully developed tumor organoids (familywise-error rate < 1 × 10⁻³) (Table S6). Notably, a stemness-related gene set and a proliferation-related gene set were enriched with the highest normalized enrichment scores (Fig. 1b).

Fig. 1.

RNA-seq analysis of four HCC samples successfully developing organoids and four failing samples. (a) Hierarchical clustering analysis of eight HCCs. (b) Gene sets of stemness (upper panel) and proliferation (lower panel) significantly enriched in the HCCs successfully developing organoids. FWER, familywise-error rate

PLC organoids and PDX recapitulate the histopathological features of the original PLC types in vitro and in vivo in animal models

Next, we set out to confirm that tumor organoids recapitulate histological features of the original tumors. The HCC organoids mostly displayed solid structures. ICC organoids formed compact spheroids and irregularly shaped cyst-like structures. To assess whether the tumor organoids retained the histological characteristics of the corresponding PLC subtypes in vivo, we subcutaneously transplanted the organoids of HCC, ICC and CHC into nude mice to establish ODX. We found that the ODX retained the histological characteristics of the corresponding PLC subtypes in nude mice. Moreover, PDX in nude mice also well recapitulated the histology of the corresponding PLC subtypes (Fig. 2).

Fig. 2.

Histology of primary tumor tissues, tumor organoids, ODX and PDX from HCC-25, HCC-118, ICC-6, and CHC-3 patients. H&E staining of the tumor tissues, bright-field and H&E staining of tumor organoids originating from the same corresponding tissues, H&E staining of ODX, and H&E staining of PDX are shown from the top row to the bottom row. Scale bars, 50 μm

Tumor organoids enrich the aggressive cell types of the original PLC subtypes, but retain their heterogeneity

Using immunohistochemistry, we found that AFP was highly expressed in HCC organoids, rather than in ICC and CHC organoids, which was consistent with the expression patterns of the original tumors (Fig. S4a). The expression of cytokeratin 19 (CK19), a marker of HCC stem cells, was higher in tumor organoids than in the corresponding tumor tissues of HCC-25 (immunoreactive score: 12.0 vs. 1.2, p = 8.1 × 10⁻⁹), ICC-6 (12 vs. 7.5, p = 0.015) and CHC-3 (10.2 vs. 2.8, p = 3.0 × 10⁻⁵), rather than in that from HCC-118. The expression of CK19 was higher in ODX than in the corresponding tumor tissues of HCC-25 (9.2 vs. 1.2, p = 4.0 × 10⁻⁴) and CHC-3 (8.2 vs. 2.8, p = 0.001), rather than in those from HCC-118 and ICC-6. The expression of CK19 was not statistically different in the primary tumors and PDX of HCC-25, HCC-118, ICC-6, and CHC-3 (Fig. S4b). The expression of epithelial cell adhesion molecule (EpCAM), another epithelial stem marker, was higher in tumor organoids of HCC-25 (8.8 vs. 2.8, p = 2.8 × 10⁻⁶) and CHC-3 (3.7 vs. 1.7, p = 0.023) as well as in the ODX from HCC-25 (8.3 vs. 2.8, p = 9.8 × 10⁻⁶) than in the corresponding tumor tissues. No differences were found between the organoids and primary tumors of HCC-118 and ICC-6 as well as between PDX and the corresponding tumor tissues from HCC-25, HCC-118, ICC-6, and CHC-3 (Fig. S4c). These results indicate that tumor organoids enriched the aggressive cell types of the corresponding PLC subtypes; however, the heterogeneity is evident.

Heterogeneity of inherent drug response in tumor organoids

To evaluate if tumor organoids are suitable for drug selection in vitro, we treated PLC organoids with different concentrations of sorafenib, regorafenib, lenvatinib, and mTOR inhibitor compound RTP and then examined cell viability. We found that sorafenib and regorafenib inhibited the growth of HCC organoids, whereas this effect was not observed in all the ICC and CHC organoids. RTP decreased the growth of PLC organoids in a dose-dependent manner. However, the majority of the PLC organoids showed resistance to lenvatinib treatment, with an IC₅₀ value higher than the maximum screening concentration (Fig. 3a). In organoids of HCC, ICC, and CHC, the average IC₅₀ value of RTP was significantly lower than that of sorafenib and regorafenib, which suggests that RTP has a stronger antitumor effect (Fig. 3b, c).

Fig. 3.

Sensitivities of PLC organoids to targeted drugs in vitro and in vivo. (a) Response of the HCC, ICC and CHC organoids to sorafenib, regorafenib, lenvatinib and RTP treatment for six days, respectively. The data shown are the average values of three technical and three biological repeats. (b) Average IC₅₀ values of sorafenib, regorafenib, lenvatinib, and RTP for the organoids of HCC (left), ICC (middle), and CHC (right). (c) Representative bright-field images of HCC, ICC, and CHC organoids after being in vitro treated with sorafenib (5 μM), regorafenib (5 μM), lenvatinib (20 μM) or RTP (5 nM) on day 6. Scale bars, 50 μm. (d) Tumor volumes of HCC-17, HCC-118, ICC-3, and CHC-3 ODX during treatment with sorafenib, regorafenib, lenvatinib, and RTP, respectively. (e) Tumor weights of HCC-17, HCC-118, ICC-3, and CHC-3 ODX at the end of the treatment with sorafenib, regorafenib, lenvatinib, and RTP, respectively. * p < 0.05; ** p < 0.01; *** p < 0.005; **** p < 0.001 (Mann–Whitney test, two-tailed)

To explore the therapeutic potential of targeted therapeutics including RTP on PLC of each histotype, we examined their effects on the growth of tumor ODX models. We found that the anti-tumor effect of RTP was comparable to that of sorafenib in HCC, while RTP inhibited ICC and CHC more effectively than sorafenib. In contrast to the resistance in the majority of PLC organoids, lenvatinib was effective in all the tested ODX models (Fig. 3d, e).

Acquired sorafenib-resistant HCC organoids and gene expression profiles

To explore the evolution of resistance to sorafenib in HCC organoids, four HCC organoids (HCC-10, HCC-25, HCC-52, and HCC-118) were randomly selected. HCC-10 was derived from a female HBV-positive HCC (HBV-HCC) patient with lymph node metastases. HCC-25 was an invasive tumor organoid derived from a female HBV-positive HCC patient. HCC-52 was an invasive tumor organoid derived from a male HBsAg-negative HCC patient. HCC-118 was a non-invasive tumor organoid derived from a male HBV-positive HCC patient who had received long-term antiviral treatment before surgery. After approximately 3–5 months of culture, the IC₅₀ values of drug-resistant HCC-10, HCC-25, HCC-52, and HCC-118 strains increased 1.59-, 4.11-, 2.01-, and 2.31-fold, respectively (Table S7).

Next, RNA sequencing was applied to assess the RNA expression profiles between sorafenib-resistant and parental tumor organoids from four HCC patients (HCC-10, HCC-25, HCC-52, and HCC-118). The median number (IQR) of the read pairs generated from the organoid samples was 47.7 × 10⁶ (42.2 × 10⁶–49.4 × 10⁶) (Fig. 4a). The rates of high-quality reads were generally close to 100%. Moreover, the alignment rates were above 90% for all of the samples (median, 96.9%; IQR, 95.4%–98.2%). Hierarchical clustering analysis revealed a strong heterogeneity among the four patients, whereas the parental and sorafenib-resistant organoids of each patient fell into the same cluster (Fig. 4b). Meanwhile, for each patient, parental and sorafenib-resistant organoids also showed evidently different expression patterns, as they formed distinct clusters. Combining the data from the four organoids, we identified 244 differentially expressed genes (Fig. 4c). Among them, 37 genes including MCM6 and RRS1 were upregulated while 207 genes including TP53INP2 and MYH14 were downregulated in acquired sorafenib-resistant HCC organoids (Table S8). GSEA identified a group of gene sets enriched in acquired sorafenib-resistant organoids (Table S9). Importantly, cancer stemness-related gene sets including Myc- and EGFR-related gene sets were enriched in the sorafenib-resistant organoids. Epithelial–mesenchymal transition (EMT)-related gene sets including TGFβ1- and E2F-related gene sets were also enriched in the sorafenib-resistant organoids (Fig. 5a–f). Importantly, liver development- and liver specific gene-related gene sets were often downregulated, whereas undifferentiated cancer- and proliferation-related gene sets were upregulated in the sorafenib-resistant organoids (Fig. 5g–j). Although the sorafenib-resistant organoids were enriched with gene signatures of stemness and EMT, the heterogeneity among the four HCC organoids was still evident. Sorafenib-resistant HCC-25, HCC-52, and HCC-10 were associated with upregulation of stemness-gene sets, whereas sorafenib-resistant HCC-118 was associated with downregulation of stemness-related gene sets (Fig. S5). Sorafenib-resistant HCC-118 and HCC-10 were associated with upregulation of EMT-related gene sets, whereas sorafenib-resistant HCC-52 and HCC-25 were associated with downregulation of EMT-related gene sets (Fig. S5).

Fig. 4.

RNA-Seq analysis of organoids. (a) Quality control of RNA-Seq data. The number of read pairs (marked on left Y-axis), the rate of high quality (HQ) reads (calculated by NGSQCToolkit) (marked on right Y-axis) and the alignment rate (calculated by HISAT2) (marked on right Y-axis) for each sample are shown. (b) Hierarchical clustering analysis of RNA-Seq data. (c) Volcano plot of 244 differentially expressed genes

Fig. 5.

Representative gene sets enriched in acquired sorafenib-resistant HCC organoids. (a) Gene sets representing the signature of cancer stemness, (b) Gene sets representing the signature of EMT, (c) Gene sets representing the signature of liver development and (d) Gene sets representing the signature of tumor de-differentiation. EMT, epithelial–mesenchymal transition; NES, normalized enrichment score; FDR, false discovery rate

We then evaluated the effect of drug resistance on the expression of stemness-related genes in our sorafenib-resistant HCC organoids by either Western blotting or immunohistochemistry. Epithelial markers E-cadherin and ZO-1 were downregulated in sorafenib-resistant HCC-25, HCC-10, and HCC-52 compared to their parental counterparts, in contrast to HCC-118, while β-catenin was upregulated in sorafenib-resistant HCC-25 and HCC-10 (Fig. 6a). Stem cell markers CD133, CK19, and β-catenin were upregulated in sorafenib-resistant HCC-25 compared to its parental HCC-25, and this was not apparent in HCC-118. Consistent with the Western blot result, immunohistochemistry showed that E-cadherin was upregulated in sorafenib-resistant HCC-118 compared to its parental counterparts, in contrast to HCC-25 (Fig. 6b). The expression of ABCG2 was upregulated in sorafenib-resistant HCC-118 and HCC-52 compared to their parental counterparts and the expression of EpCAM was upregulated in sorafenib-resistant HCC-118 and HCC-10 compared to their parental counterparts. Claudin-1 and N-cadherin were upregulated in sorafenib-resistant HCC-10 compared to its parental counterpart, while CD44 was upregulated in sorafenib-resistant HCC-25 compared to its parental counterpart using both Western blotting and immunohistochemistry assays (Fig. S6). Immunohistochemistry for other organoids was technically unsuccessful. Thus, induction of sorafenib resistance is related to the processes of cancer stemness and EMT/partial EMT and the heterogeneity is apparent.

Fig. 6.

Expression of cancer stemness- and EMT-related genes between parental and sorafenib-resistant HCC organoids. (a) Western blotting showed the expression patterns of E-cadherin, β-catenin, and ZO-1 in four sorafenib-resistant HCC organoids and their parental ones. GAPDH was used as loading control. (b) Immunohistochemistry showing the expression patterns of CD133, CK-19, β-catenin, and E-cadherin in parental and acquired sorafenib-resistant HCC-25 and HCC-118 organoids. Scale bars, 50 μm

Targeting the mTOR signaling pathway is effective in treating acquired sorafenib-resistant HCC organoids

After establishing acquired sorafenib-resistant HCC organoids, we sought to investigate if inhibitors targeting the sorafenib-induced signaling pathways could inhibit the growth of sorafenib-resistant HCC organoids. We tested the effects of available inhibitors on the cell viability of sorafenib-resistant HCC organoids and found that RTP was effective in inhibiting sorafenib-resistant HCC-25 and HCC-118 organoids, with IC₅₀ values of 1.918 and 1.828 nM, respectively (Fig. 7a, b). However, the expression levels of mTOR signaling molecules were not significantly upregulated in sorafenib-resistant HCC-25 and HCC-118 organoids (Table S8). In addition, we found that phosphorylation of S6, a key downstream target of PI3K/AKT/mTOR pathway [28], was apparently upregulated in sorafenib-resistant HCC-25 and HCC-118 organoids (Fig. 7c). Thus, phospho-S6 should be predictive in sorafenib-resistant HCC.

Fig. 7.

Effects of mTOR signaling pathway inhibitor on the viability of acquired sorafenib-resistant HCC organoids and its possible mechanisms. (a) Response of acquired sorafenib-resistant HCC-25 organoids to treatment with RTP for six days. (b) Response of acquired sorafenib-resistant HCC-118 organoids to treatment with RTP for six days. (c) Expression of S6, one of the downstream effectors of the mTOR pathway, and phosphorylation of S6 (P-S6) in parental and acquired sorafenib-resistant organoids HCC-118 and HCC-25 determined by Western blotting. (d) Western blot showing downregulation of β-actin expression in acquired sorafenib-resistant organoid HCC-25. (e) Recovery of β-actin expression in HCC-25 organoids after withdrawal of sorafenib. (f) The expression of β-actin was not affected in sorafenib-resistant HCC-118 organoids. GAPDH was used as loading control

As acquired sorafenib-resistant HCC-25 organoids expressed more stemness markers than did acquired sorafenib-resistant HCC-118 (Fig. 6b) and the HCC-118 organoids responded to RTP more efficiently than did HCC-25 (Fig. 7a, b), we sought to identify factors related to the difference. Interestingly, we found that the expression of β-actin was apparently downregulated in acquired sorafenib-resistant HCC-25 (Fig. 7d). We then removed sorafenib and cultured sorafenib-resistant HCC-25 organoids for an additional two weeks, and found that the expression of β-actin was restored after sorafenib withdrawal (Fig. 7e). However, the expression of β-actin did not differ between parental and sorafenib-resistant HCC-118 (Fig. 7f). These data imply that β-actin may be related to efficient cell migration [29] and may have an opposite effect on maintaining cancer stemness. Further study is needed to address this issue.

Molecules upregulated in sorafenib-resistant organoids are often associated with an unfavorable prognosis in HCC

The association of differentially expressed genes related to sorafenib-resistance with postoperative prognosis was evaluated using data from the TCGA database. The genes whose expression levels were dysregulated in the sorafenib-resistant HCC organoids are listed in Table S10. Of the 26 overall survival (OS)-related genes upregulated in the sorafenib-resistant HCC organoids, 21 were significantly associated with an unfavorable prognosis. The genes upregulated in the sorafenib-resistant HCC organoids were more related to an unfavorable OS than were the downregulated genes (21/26 (86.8%) vs. 61/119 (51.3%), p = 0.010); the genes upregulated in the sorafenib-resistant HCC organoids were more related to an unfavorable disease-free survival (DFS) than were the downregulated genes (12/16 (75.0%) vs. 36/95 (37.9%), p = 0.012) (Fig. S7). These genes may reflect HCC evolution and development during the generation of drug resistance against sorafenib.

Discussion

In this study, we established tumor organoids and PDX of each PLC histotype from 153 patients. The mean duration of establishing tumor organoids was significantly shorter than that of establishing PDX. Furthermore, the establishment of tumor organoids showed a trend toward a higher success rate than the establishment of PDX (HCC: 29.0% vs. 23.7%; ICC: 52.9% vs. 41.2%). These data indicate that tumor organoids perform better than PDX as a platform to select drugs for the treatment of PLC recurrence. To decrease the cost, we only included essential growth factors in the culture conditions to establish PLC tumor organoids. Two factors, Wnt3a and Rspo-1, were not included in our culture medium compared to other studies [11, 12]. As a result, our tumor organoids and PDX exactly recapitulate the histopathological features of the original PLC types both in vitro and in vivo. However, more cancer stem cell markers were identified in tumor organoids than in the PDX models. We found that a larger tumor size, the presence of vascular invasion, as well as advanced TNM and BCLC stages were associated with the successful development of tumor organoids and PDX models. These factors are also significantly associated with an unfavorable postoperative prognosis [3–5, 30], indicating that aggressive PLC tumor cell subpopulations may have a growth advantage in tumor organoids. However, some HCC tissues with the above clinical characteristics failed to develop organoids. Our RNA-Seq analysis confirmed significant differences in expression profiles among HCCs with similar clinical characteristics, but different outcomes of organoid establishment. Gene sets of stemness and proliferation were significantly enriched in the HCCs that successfully developed tumor organoids, compared to those that failed to develop organoids. Cancer stemness and EMT are often associated with the aggressiveness of HCC [31]. Our tumor organoids appear to enrich aggressive PLC subpopulations, which may be quite suitable for screening sensitive drugs. Our RNA-Seq analysis also revealed that upregulation of MMP12, SLC1A7, TREM1, CLEC5A, and POSTN may be important in developing HCC organoids. POSTN plays an important role in the maintenance of stemness in many cancers including HCC and colorectal carcinoma [17, 32]. SLC1A7 is a member of the Solute Carrier 1A family that functions as important amino acid transporters. Another member of this family, SLC1A5, has been reported to promote HCC development through affecting activation of the mTOR signaling pathway [33]. However, reports on SLC1A7 are rare. Here, we report an association between SLC1A7 and HCC evolution. Interestingly, the other three significantly upregulated genes, MMP12, TREM1 and CLEC5A, all participate in the regulation of inflammatory microenvironments [34–36]. All of these three molecules can activate the NF-κB pathway while TREM1 and CLEC5A also enhance immunosuppression. Thus, characteristics of the immune microenvironment may also significantly affect the success rate of organoid establishment.

Our data indicate that sorafenib and regorafenib can inhibit the growth of HCC organoids, which is consistent with clinical findings [37]. Our ICC and CHC organoids are resistant to regorafenib, which is not consistent with a previous PDX model [38]. We also found that RTP decreased the growth of all PLC organoids both in vitro and in ODX models. Thus, targeting the mTOR signaling pathway may be important in treating PLC of each histotype, especially ICC and CHC. Interestingly, PLC organoids were found to be resistant to lenvatinib in vitro, but it was effective in all the tested ODX models. The antitumor efficacy of lenvatinib is mainly attributed to its anti-angiogenic activity. In addition, athymic nude mice have normal or even compensated NK activity [39]. A previous study has demonstrated that lenvatinib inhibits melanoma and renal cancer by enhancing the tumor infiltration and activation of natural killer (NK) cells [40]. We speculate that the remaining innate immunity function in athymic nude mice may also influence the antitumor effect of lenvatinib. Our findings indicate that tumor organoids and ODX models may be complementary in testing PLC drug sensitivity. PLC organoids are more suitable for evaluating the efficacy of drugs directly targeting tumor cells while ODX models are more suitable for investigating the effects of therapeutic agents that inhibit the formation of a cancer-promoting microenvironment.

Sorafenib is the first-line targeted agent that has been shown to prolong the median survival time of patients with advanced HCC by nearly three months. Unfortunately, however, most patients develop resistance to this drug within six months [41]. The mechanisms by which HCC develops resistance to sorafenib are largely unknown, although some molecular events such as Hedgehog signaling, Jak-STAT pathway activation, oncogenic KIF14 and IGF/FGF signaling have been suggested to be related to sorafenib resistance in HCC [13, 14, 42, 43]. Here, we generated acquired sorafenib-resistant organoids from four HCC patients following 3–5 months of culture. RNA-sequencing revealed that EMT-promoting molecules including MCM6 and RRS1 [44, 45] were upregulated, while EMT-suppressing molecules including TP53INP2 and MYH14 [46, 47] were downregulated in sorafenib-resistant HCC organoids. Our GSEA revealed that cancer stemness-related gene sets including Myc- and EGFR-related gene sets [48, 49] as well as EMT-related gene sets including TGFβ1- and E2F-related gene sets [50, 51] were significantly enriched in acquired sorafenib-resistant HCC organoids. Thus, cancer stemness and EMT may contribute to the development of sorafenib resistance in HCC.

Heterogeneity in expression patterns of the EMT-related and stemness-related genes were evident among HCC-118 and the other three organoids. The expression of E-cadherin and ZO-1, well-established suppressors of EMT [52], was downregulated in sorafenib-resistant HCC organoids except HCC-118. CD133, CK19, β-catenin, ABCG2, claudin-1, N-cadherin, and CD44, the cancer stem/EMT markers [14, 52–55], were upregulated only in some rather than all of the sorafenib-resistant HCC organoids. This heterogeneity was confirmed by Western blotting. These data supports the notion that partial/hybrid EMT programs, defined by incomplete activation of EMT transcription factors [51, 56, 57], may play an important role in generating sorafenib resistance in HCC organoids.

After a series of tests to identify available inhibitors that can be applied to treat sorafenib-resistant HCC, we found that the mTOR inhibitor compound RTP was effective for the sorafenib-resistant HCC organoids. Further study revealed that phosphorylated S6 was markedly upregulated, which contributes to active mTOR signaling in the sorafenib-resistant HCC organoids. Sorafenib resistance has been correlated with the upregulation of several signaling pathways including the mTOR pathway as assessed by ribosomal protein S6 kinase phosphorylation in tumor biopsies [58]. Interestingly, we found that the expression of β-actin was downregulated in acquired sorafenib-resistant HCC-25. It has been reported that the translation of human β-actin mRNA can be inhibited in the presence of a mTOR kinase inhibitor [59]. Newly synthesized β-actin protein is indispensable for efficient cell migration [29]. The absence of β-actin protein may facilitate the retro-differentiation of HCC organoids under therapeutic pressure. Thus, phosphorylated ribosomal S6 kinase may be predictive for unresponsiveness of HCC to sorafenib and responsiveness of HCC to RTP.

Lastly, we found that the genes upregulated in the sorafenib-resistant HCC organoids were more associated with an unfavorable postoperative prognosis in the TCGA databases than were the downregulated genes. These results are consistent with the data showing that cancer stemness- and EMT-related genes contribute to sorafenib resistance in HCC organoids, while stemness- and EMT-related genes are prognostic in HCC [31, 54–56]. Our GSEA data also indicated that liver development- and liver specific gene-related gene sets were often downregulated, whereas undifferentiated cancer- and proliferation-related gene sets were upregulated in sorafenib-resistant HCC organoids (Fig. 3). These data indicate that a percentage of cancer cells retro-differentiated into stem-like cell types and that EMT features are actively selected in a sorafenib-presenting microenvironment and become more aggressive, sorafenib-resistant PLC. This process reflects the evolutionary and developmental nature of HCC during the generation of drug resistance against sorafenib.

This study has some limitations. Since HCC is the major histological type of primary liver cancer, we focused on HCC cases during our RNA-sequencing analyses and assays for drug resistance. The role of functional molecules identified in HCC organoids should be further investigated in ICC and CHC organoids. This study suggests an association of phosphorylated ribosomal S6 kinase with therapeutic effects of sorafenib and RTP in organoid and ODX models. The predictive value of phosphorylated ribosomal S6 kinase on the responsiveness of HCC to sorafenib and mTOR inhibitors should be further investigated in epidemiological studies.

Conclusively, HCC organoids need a shorter time to construct and enrich more aggressive cancer cell types than does PDX. The majority of PLC organoids was resistant to lenvatinib in vitro, but sensitive to lenvatinib in the ODX models. Upregulation of stemness and EMT/partial EMT contributed to the acquired sorafenib resistance in HCC organoids, and targeting of the mTOR signaling pathway using mTOR inhibitor compound RTP was effective in treating acquired sorafenib-resistant HCC. Furthermore, phosphorylated ribosomal protein S6 may serve as a candidate predictive marker for the treatment of acquired sorafenib-resistant HCC with mTOR inhibitor(s). Our data may not only help to elucidate the mechanisms by which PLC evolves under distinct pressures, but may also provide qualified PLC organoids for drug screening and targeted treatment for PLC aggressiveness.

Author contributions

Cao G conceived and supervised the study. Xian L, Zhao P, and Wei Z were responsible for the culture and maintenance of tumor organoids. Chen X performed the bioinformatics analyses. Ji H and Zhao J did surgical treatments and provided suitable clinical specimens. Liu D, Li Z, Liu W, Zhou X, Fan J, Zhu X, Yin J, and Tan X took part in the cell experiments and animal care. Qian Y and Dong H took part in the histology and immunohistochemistry assays. Chen X, Han X, and Yu H conducted the statistical analyses. Cao G interpreted the data and drafted the manuscript. All authors read and approved the final manuscript.

Funding

This work was supported by the National Key Basic Research Program of China [grant number 2015CB554006 (GC)], the National Natural Science Foundation of China [grant numbers 91529305 (GC), 81520108021 (GC), 81673250 (GC), 81521091 (GC), 82003538 (WL), and 81502882 (XC)], the Shanghai Yangfan Program [grant numbers 20YF1458800 (WL)] and the “3-year public health promotion” program of Shanghai Municipal Health Commission [grant numbers GWV-10.1-XK17 (GC)].

Data availability

The original RNA-Seq data of tumor tissues are deposited in the Sequence Read Archive (SRA) database with accession PRJNA841062. The original RNA-Seq data in this study are deposited in the Gene Expression Omnibus (GEO) database with accession GSE182593. Other data are available from the corresponding author upon reasonable request.

Declarations

Ethical approval and consent to participate

This study protocol was reviewed and approved by Ethics Committee of Eastern Hepatobiliary Surgery Hospital, approval number 2019UE-023. All procedures performed in studies involving human participants were in accordance with the 1964 Helsinki Declaration and its later amendments. Informed consent was obtained from all participants included in the study. All animal experiments in this study were conducted in accordance with the guidelines of the animal ethical committee for animal experimentation in China. The experimental design was approved by the Institutional Animal Care and Use Committee of Second Military Medical University.

Consent for publication

All patients signed informed consent regarding publishing their data.

Competing interests

The authors declare that they have no competing interests.