LGR5 marks targetable tumor-initiating cells in mouse liver cancer

Cancer stem cells (CSCs) or tumor-initiating cells (TICs) are thought to be the main drivers for disease progression and treatment resistance across various cancer types. Identifying and targeting these rare cancer cells, however, remains challenging with respect to therapeutic benefit. Here, we report the enrichment of LGR5 expressing cells, a well-recognized stem cell marker, in mouse liver tumors, and the upregulation of LGR5 expression in human hepatocellular carcinoma. Isolated LGR5 expressing cells from mouse liver tumors are superior in initiating organoids and forming tumors upon engraftment, featuring candidate TICs. These cells are resistant to conventional treatment including sorafenib and 5-FU. Importantly, LGR5 lineage ablation significantly inhibits organoid initiation and tumor growth. The combination of LGR5 ablation with 5-FU, but not sorafenib, further augments the therapeutic efficacy in vivo. Thus, we have identified the LGR5+ compartment as an important TIC population, representing a viable therapeutic target for combating liver cancer. Cancer stem cells (CSCs) are thought to be the main drivers for disease progression and treatment resistance in liver cancer. This study identifies the LGR5+ compartment as an important CSC population, representing a viable therapeutic target for combating liver cancer.

T he key concept underlying the cancer stem cell (CSC) or tumor-initiating cell (TIC) theory is that tumors are maintained through a hierarchical structure, in which different cell populations have different functionalities in pathophysiology 1 . The bulk of a tumor is thought to consist of CSCs/ TICs as well as rapidly proliferating cells. CSCs/TICs are responsible for tumor initiation, resistance to conventional treatment, and distant metastasis. Rapidly proliferating cancer cells, thought to be derived from the tumor stem cell pool, are responsible for volume increment of the tumor 2 . A prediction based on this model is that ablation of the relatively small CSC compartment would ultimately result in cessation of tumor growth and metastasis, and provoke sensitization of the tumor to conventional treatment as well.
Within the framework of this theory, CSCs/TICs would be characterized by a large capacity for self-renewal, a potential for differentiation into different cell types that constitute the tumor, and a resistance to conventional treatment 1 . These key features largely overlap with those of normal stem cells, making it extremely difficult to specifically identify CSCs/TICs, but on the other hand would allow techniques traditionally used for identifying normal stem cells also to be applied for CSCs/TICs 3 .
LGR5 (leucine-rich repeat-containing G protein-coupled receptor 5) evokes special interest as a potential marker for the CSC/TIC compartment in this respect.
LGR5 is a wellcharacterized stem cell marker in several tissues/organs, including the small intestine, colon, and liver [4][5][6] . In the colon and intestine, the LGR5 stem cell pool constantly proliferates and differentiates into mature cell types to compensate for the loss of functional epithelial cells. Interestingly, these LGR5 stem cells also participate in the process of oncogenesis, acting as the cells-of-origin of intestinal cancer 7 . Importantly, LGR5 marks CSCs in colon cancer [8][9][10] , intestinal cancer 11 , and basal cell carcinoma 12 . In intestinal adenoma as well as malignant carcinoma, LGR5 cells account consistently for a ratio of 5-10% of tumor cells and fuel tumor growth 8,13 . Proof-of-concept showing that specific elimination of LGR5 cells delays tumor growth in colon cancer has been provided 9 . Given the essential role of CSCs/TICs, these cells are attractive targets for anticancer treatment, whereas their resistance to conventional therapies impedes the therapeutic development.
In contrast to the colon and intestine, LGR5 stem cells are absent in the homeostatic liver, but emerge upon tissue injury 4,14 . These liver LGR5 cells are likely to be an intermediate stem/progenitor cell population that responds to injury, but they may have a limited contribution to tissue repair 14 . Whether an LGR5 + compartment exists in liver cancer remains obscure, and the possible importance of such a compartment in this disease is unexplored. Nevertheless, research into this possibility is urgently needed as liver cancer is one of the most common forms of malignancy worldwide, with nearly 800,000 cases reported yearly, and it is characterized by a depressing lack of treatment options 15 . Hepatocellular carcinoma (HCC) and cholangiocarcinoma (CC) are the two main types of primary liver cancer. Currently, surgery remains the only potentially curative therapeutic strategy available, but is well-known for its high recurrence rate following tumor resection. Chemotherapy and targeted treatment are generally ineffective, with sorafenib providing some extension of life expectancy to HCC patients. The unusual treatment resistance of liver cancer is thought to be associated with the presence of CSCs/TICs, but this notion remains largely unproven 16 . Thus, we aimed to investigate whether LGR5 marks CSCs/TICs in liver cancer, and to explore the potential for therapeutic targeting of these cells. Our results show that in liver cancer, an LGR5 + compartment exists that is superior in tumor initiation and mediates therapy resistance. Targeting this compartment constitutes a rational avenue for combating this disease.

Enrichment of LGR5-expressing cells in primary liver tumors.
Homeostatic livers are reported to be devoid of LGR5 + cells, but injury does induce such cells 14 . Whether LGR5 + cells are present in liver cancer is largely unknown. By adopting Lgr5-DTR-GFP knock-in mice (Fig. 1a), we first investigated the presence of LGR5 + cells (GFP-co-expressing cells) in the healthy and injured liver, and during carcinogenesis. Carbon tetrachloride (CCl 4 ) was used to trigger liver injury. Diethylnitrosamine (DEN) was used to induce primary liver tumor formation ( Fig. 1b; Supplementary  Fig. 1). Although LGR5 cells are absent in the homeostatic liver (Fig. 1c), either a single course or repeated administration of DEN can rapidly trigger the emergence of LGR5-GFP + cells (post DEN induction day 7; relative size of the LGR5-GFP + compartment following 1 × DEN: 0.025 ± 0.05%, n = 3 [mean ± SEM]; Supplementary Fig. 2a, b). Animals were monitored for liver tumor formation from 4 to 14 months post DEN induction (Supplementary Data 1). Analysis of the resulting hepatic neoplasms revealed the stable presence of an LGR5 + compartment in these liver tumors (Fig. 1c). The relative abundance of LGR5 cells in the tumors (Supplementary Data 1, Supplementary Fig. 2c, d) is significantly higher as compared with those in the tumorsurrounding tissues (Fig. 1d) or as detected in CCl 4 -injured livers (Fig. 1c). The LGR5 expression levels in the tumor cells show substantial variation, but are substantially and significantly higher compared with that in injured liver (Fig. 1e). Immunohistochemistry (IHC) and immunofluorescence (IF) staining of GFP expression further confirms the presence of an LGR5 + compartment and enables detailed analysis of spatial distribution of LGR5-GFP + cells in the liver (IF: Fig. 1f; IHC: Supplementary  Fig. 2e, f). Co-staining with hepatocyte marker (HNF4α) or cholangiocyte marker (CK19) revealed that a proportion of LGR5 cells in the tumor express HNF4α or CK19 (Fig. 1g, h), suggesting that LGR5 + cells may give rise to both a HCC-like and a CC-like phenotype, the two main types of primary liver cancer. Thus, these data have demonstrated the presence of an LGR5 + compartment in primary murine liver cancer.
To examine the clinical relevance, we investigated LGR5 expression in human HCC tumors from our patient cohort (Erasmus MC cohort). We found that LGR5 expression is significantly elevated in tumor tissues compared with the paired tumor-free liver tissues (Fig. 2a), and also in some subpopulations of patients with specific etiologies of HCC (Fig. 2b). Survival analysis by predicting Kaplan-Meier curves revealed a tendency toward worse clinical outcome in patients with higher LGR5 expression (Fig. 2c). Further analysis of online publically available datasets confirmed the upregulation of LGR5 expression in HCC ( Supplementary Fig. 3a), and possible association with clinical outcome, especially in subpopulations of specific patients ( Supplementary Fig. 3b). Interestingly, with data from the TCGA database and International Cancer Genome Consortium-France (LICA-FR) and International Cancer Genome Consortium-Japan (LIRI-JP), we found that the upregulation of LGR5 expression is more pronounced in HCC tumors with β-catenin mutation ( Supplementary Fig. 4). This is in line with LGR5 being a βcatenin target gene both in the intestine and liver 5,17 . Taken together, LGR5 cells are enriched in both mouse and human liver tumors, and bear substantial clinical relevance. established routine procedures 21 for creating organoid cultures from primary liver tumors of DEN-induced mice ( Supplementary  Fig. 1 LGR5-expressing cells (%) LGR5-expressing cells (%) *** HCC-like phenotype (Fig. 3a, b). Importantly, these cultured organoids maintain a population of LGR5-positive cells (Fig. 3c).
To investigate whether these murine organoid lines are malignant, we transplanted all the 63 strains into immunodeficient NOG mice (Fig. 3d). One to 6 months after allografting, 11 out of 63 organoid strains formed palpable tumors in the immune-deficient mice (17.5%). All contained an LGR5-GFP + compartment as determined by FACS analysis of the tumors (Fig. 3e).
The ultimate measure of potential functionality of LGR5 + cells in pathophysiology is their capacity to form allograft tumors in vivo (Fig. 4g). Hence, identical numbers of FACS-sorted LGR5-GFP + and LGR5-GFP − cells derived from primary liver tumors were subcutaneously engrafted into NOG mice, and tumor formation was monitored (Supplementary Data 6). As expected, LGR5 + cells display a stronger capacity for tumor initiation as compared with LGR5 − cells (LGR5 + vs. LGR5 − : 33.3% vs. 11.1%) ( Fig. 4h; Supplementary Data 6). Moreover, tumors initiated from LGR5 + cells contain both LGR5-positive and -negative populations ( Fig. 4i-n). In addition, we have performed a tumor formation assay for the cells that were derived from the allograft tumors ( Fig. 4o). Again, the LGR5 + compartment proved markedly more potent in this respect relative to the LGR5-GFP − compartment (Fig. 4p, q; Supplementary Table 1). Collectively, our results are best interpreted that liver tumor-derived LGR5 + cells constitute a bona fide TIC compartment.
Anticancer treatment enriches LGR5-expressing cells. CSCs or TICs are presumed to be relatively resistant to conventional anticancer treatment. A prediction would thus be that in liver cancer, the LGR5 + cells would be more resistant to anticancer treatment as compared with the LGR5 − cells. Hence, we challenged tumor organoids with sorafenib, the FDA-approved drug for treating advanced HCC, and compared the relative potential of LGR5-GFP + and LGR5-GFP − cells to withstand such treatment (Fig. 5a). Treatment of tumor organoids with sorafenib significantly increased the percentage of LGR5-positive cells in the population (Fig. 5b-d). This effect became even more profound when the organoids were treated with the chemotherapeutic agent, 5-fluorouracil (5-FU) (Fig. 5a-d).
Subsequently, the relative size of the LGR5 + compartment following in vivo treatment with these therapeutic agents was assessed (Fig. 5e). Treatment with either sorafenib or 5-FU to mice-bearing allograft tumors, formed by engrafting tumor organoids, substantially increased the fraction of the LGR5-GFP + cells in the tumors (Fig. 5f). Also when LGR5-GFP + and LGR5-GFP − cells were isolated from tumor organoids and used for organoid re-initiation, while subsequently being treated with 5-FU, the resulting cultures were dominated by LGR5-GFP + -expressing cells, independent from whether LGR5-GFP + or LGR5-GFP − were used as starting material (Fig. 5g). Of note, LGR5 + cells isolated from 5-FU-treated tumors retained the ability of organoid and tumor initiation (Supplementary Fig. 8). Interestingly, 5-FU treatment effectively rewired the transcriptome of LGR5 + cells ( Fig. 5h; Supplementary Figs. 6 and 9). Gene enrichment analysis of the 1464 differentially expressed genes between 5-FU treated compared with untreated LGR5 + cells revealed the involvement of stem cell-related pathways, including "Wnt Signaling", "Notch Signaling Pathway", "ErbB signaling pathway", "Hedgehog Signaling Pathway", and "BMP Signaling Pathway" (Fig. 5i; Supplementary Data 3). These pathways are commonly activated in many types of solid tumors, associated with cancer initiation, progression, and metastasis 23 . Interestingly, there are several pathways, including "TGF Beta Signaling Pathway", "EGFR1 Signaling Pathway", "PPAR signaling pathway", "G1 to S cell cycle control", "Mismatch repair", "p53 signaling", and "Apoptosis Modulation/Apoptosis pathway", which are known to be implicated in anticancer treatment response and DNA damage response 24 . These results may partially explain the enrichment of LGR5-expressing cells upon 5-FU treatment. In conclusion, besides resistance, conventional anticancer treatment also triggers the generation and propagation of LGR5-expressing cells.
LGR5 lineage ablation inhibits organoid and tumor growth. From the results described above, we inferred that ablation of the LGR5 + compartment should impair liver cancer growth. To experimentally test this notion, we exploited the co-expression of the diphtheria toxin receptor (DTR) in the Lgr5-DTR-GFP mice. This would allow us to specifically deplete the Lgr5-DTR-GFP + compartment through diphtheria toxin (DT) administration (Fig. 1a). We have optimized the concentrations of DT treatment (1-10 ng/ml) for LGR5 depletion, with organoids derived from healthy Lgr5-DTR-GFP mice 14 . Accordingly, we evaluated the effects on organoid initiation and proliferation (Fig. 6a, b), and sorafenib treatment served as a positive control. DT treatment inhibited the growth of tumor organoids in an effect that showed close correlation as to the effects on the numbers of LGR5-GFP + cells (Fig. 6c-e). DT treatment did not influence the growth of tumor organoids of genetically wild-type ( Fig. 6c: left panel).
We further assessed therapeutic targeting of LGR5 liver cancer cells in vivo. We first evaluated the effect of DT treatment after formation of visible tumors, following transplantation of tumor organoids into immunodeficient mice ( Supplementary Fig. 10a) LGR5- LGR5-Tumor weight (g)

Primary
Allograft LGR5 percentage (%) LGR5 Tumor weight (g)  (Fig. 6g). Using absolute tumor size as a measure, DT-mediated depletion of the LGR5 + compartment impaired tumor growth (Fig. 6i). The enrichment of stem cell markers also differed in control and DT-treated LGR5 +/− cells ( Supplementary Fig. 11). Interestingly, there was an inverse correlation between tumor and the relative size of the LGR5-GFP + compartment at the end of the experiment ( Supplementary Fig. 10e, f), indicating that LGR5-expressing cells are probably more active in the tumor initiation period. As control, DT treatment did not influence initiation and growth of tumors formed from the wild-type tumor organoids ( Supplementary Fig. 12a, b). Thus, LGR5 lineage ablation impedes organoid and tumor initiation and further growth.
Combination therapy enhances the anticancer efficacy. As LGR5 + cells appear to mediate resistance against conventional anticancer treatment, it is rational to evaluate the combination of LGR5 + lineage ablation with conventional anticancer treatment.
To experimentally test this notion, we first combined DT with sorafenib treatment. However, with different strategies of combination therapy, no enhanced antitumor activity was observed on allografted tumors (Fig. 7). We next tested the combination of 5-FU and DT. Allograft tumor-bearing mice were first subjected to 5-FU (which increases the relative size of the LGR5 + compartment) followed by cessation of 5-FU therapy and start of DT treatment as to kill the LGR5 + cells (Fig. 8a;  Supplementary Fig. 13a-c). Indeed, this approach is effective in combating allograft tumor formation (Fig. 8b) and is substantially superior to monotherapy with 5-FU, stand-alone DT treatment (Fig. 8c, d; Supplementary Fig. 14a) or initial treatment with DT followed by 5-FU therapy ( Supplementary Fig. 13d-h). Simultaneous administration of 5-FU and DT (Fig. 8e) also resulted in anticancer effects (Fig. 8f-h; Supplementary Fig. 14b). Thus, targeting the LGR5 + compartment markedly enhances the efficacy of conventional treatment aimed at combating liver cancer.

Discussion
This study has demonstrated that liver cancer contains an LGR5 + compartment that has various hallmarks of TICs/CSCs, including an increased capacity for tumor organoid formation in culture and allograft formation in mice, as well as resistance against conventional anticancer therapy. Functionally, these cells seem more important in tumor initiation, whereas the LGR5 − compartment appears to bear the proliferative burden. Simultaneously targeting both compartments as demonstrated in this study by 5-FU treatment in combination with DT-mediated ablation of the LGR5 + compartment was effective in combating liver cancer in experimental models. Although our study primarily focused on mouse models, the relevance of the LGR5 + compartment deserves to be further investigated in human liver cancer. Conceivably, the role of these cells could be more prominent for a subset of patients with high levels of LGR5 expression, such as CTNNB1-mutated or alcohol-related HCC patients 25,26 . Although LGR5 + -targeting therapies are still largely in their infancy, the analogy with neuroendocrine tumors, which are successfully combated by radioactive somatostatin analogs (e.g., 177 Lu-Dotatate) that target receptors with homology to LGR5 27 , suggests that radioactive drugs (e.g., R-spondin) may be explored for developing CSC-targeted therapeutics against liver cancer 28 . Our results, however, indicate that such therapy may require combination with particular conventional anticancer therapies to enhance the efficacy.
Overexpression of LGR5 has been previously reported in patient HCC 17 , and we confirmed that this is more pronounced in β-catenin-mutated liver tumors. Although the elevation of LGR5 expression and potential association with clinical outcome have been observed in HCC patients, whether it can serve as an independent prognostic biomarker remains to be further investigated in specifically designed tumor marker prognostic studies in patients 29 .
Of note, these early observations are based on mRNA expression, due to the lack of a reliable anti-LGR5 antibody. We now used transgenic mice in which LGR5-expressing cells coexpress GFP, and we can conditionally ablate these cells with the DT-DTR system 30 . This model allows for the identification and direct visualization of LGR5-expressing cells based on GFP expression, as well as isolation of LGR5-GFP + cells for further functional analyses and detailed characterization.
In intestinal adenomas, LGR5 marks 5-10% of the cells, which keep fueling the growth of established mouse adenomas 13 . We found that the percentages of LGR5-GFP + cells in murine liver tumors vary from 0.1% up to 55% (7.3 ± 1.8%, n = 55). Over 32% of the primary liver tumors harbor relatively high percentages (over 5%) (Supplementary Fig. 2c). In colon cancer, the percentage of LGR5-expressing cells has been reported to be associated with different background of the tumors, especially the accumulation of certain oncogenic mutations 8 . Thus, we speculate that the large variation of the Fig. 4 Single LGR5 + cells from liver tumors are superior in organoid and tumor initiation. a An outline of the experimental strategy for studying ex vivo organoid initiation of cells derived from primary murine liver tumors. b A representative picture of organoids derived from single LGR5 + cells. Scale bar = 50 µm. c Representative confocal micrograph of a single LGR5 + cell-initiated organoid dominated by LGR5-expressing cells.
abundance of LGR5 cells in murine liver tumors may also be related to different types of oncogenic mutations, although this hypothesis requires further investigation. Of note, DEN was used to induce primary liver cancer in this study, and this compound is associated with the accumulation of liver βcatenin mutations 31 . Therefore, it is possible that our results are mainly relevant to liver cancers with deregulated Wnt/β-catenin signaling, and their importance requires further investigation in LGR5- LGR5- LGR5+ LGR5+ LGR5+ e f ×200 ×100 ×200 ×200 ×200  other mutational backgrounds (e.g., deregulated TSC/mTOR signaling) 32 as well. Through a series of functional assays, in particular in vitro organoid initiation and in vivo tumor formation, we have demonstrated the importance of these LGR5 TICs in liver cancer. To define the potential for therapeutic targeting, we have performed LGR5 lineage ablation in organoids in vitro and in the tumor-bearing mouse model. Of note, the presence of LGR5 cells in tumors is likely dynamic. We observed large variations of their percentages among different primary murine liver tumors, and allograft tumors generally contain lower numbers of LGR5 cells (less than 1%). In colorectal cancer, advanced stages compared with the early stages contain less LGR5 cells 33 . A speculative explanation could be that the tumors are derived from LGR5-positive stem cells, yet these cells are suppressed thereafter during tumor progression 33 . The dynamics could be essential for determining at which stage to target LGR5 TICs. When we performed LGR5 cell ablation in established tumors, we only observed a minor effect, probably due to a low percentage of LGR5 cells as well as their dispensable function at that stage, while depletion at the early stage yielded optimal antitumor effects. This result is in line with previous findings showing that LGR5 cells play distinct roles in primary and metastatic colon cancer 8 . In addition, the percentage of LGR5-expressing cells in primary murine tumors seems not strictly correlated with the ability of initiating allograft (Fig. 3e). Because primary tumors were first cultured into tumor organoids, these organoids were engrafted into mice to initiate the allograft. Thus, this indirect assay may not fully recapitulate the initial status of the primary tumor.

5-FU
Although we have demonstrated the feasibility and value of targeting LGR5 TICs in murine liver cancer, therapeutic ablation of these cells remains challenging. Resistance to conventional therapy is a common feature of CSCs 2 . We found that treatment with sorafenib or 5-FU enriches LGR5 cells, consistent with the findings in gastric 34 and colorectal cancer 35 . Different mechanisms may contribute to treatment resistance. Although LGR5 stem cells are generally fast-cycling in the intestine 36 , the existence of quiescent LGR5 cells has been reported in basal cell carcinoma, which mediates relapse after treatment 12 . Cell plasticity could be one of the potential mechanisms of treatment resistance. The loss of LGR5 stem cells in the intestine can be compensated by transdifferentiation from other stem cell pools 30 , or through plasticity of their enterocyte-lineage daughters 37 . Cancer cell plasticity, shifting dynamically between a differentiated and a stemness state, has also been proposed as an important feature contributing to tumor progression, metastasis, and therapeutic response 38 . We have now observed the induction of LGR5 + from LGR5 − liver cancer cells. This may implicate cell plasticity of LGR5 CSCs, but there could also be other mechanisms regulating the origin and expansion of LGR5 cells, as for example, changes in the culture environment. Eventually, these LGR5 liver cancer cells may partially contribute to treatment resistance.
Currently, several innovative scenarios are being explored to therapeutically target CSCs, including antibody-drug conjugates 9 , targeting quiescent CSCs 39 , and inhibiting CSCrelated pathways 2 . However, as discussed, different mechanisms could lead to treatment resistance 8 . Thus, combined therapies are likely necessary in this respect. With the intention to fully expand the stem cell pool, cetuximab has been used to first trigger the LGR5 population, followed by the ablation of these CSCs. This combined therapy has resulted in potent efficacy against colorectal cancer 10 . Similarly, we have observed that the combination of LGR5 lineage ablation with 5-FU chemotherapy can also lead to enhanced anti-liver cancer activity in mouse organoid. However, a combination of LGR5 lineage ablation with sorafenib did not yield enhanced antitumor activity. This is probably related to the mild effect of sorafenib in triggering the LGR5 CSC pool.
Last, a potential concern of such strategies is the possible harmful effects on normal LGR5 stem cells. In the intestine, colon, and skin, although LGR5 stem cells essentially contribute to tissue renewal at a daily basis 5,6 , their loss can be compensated by transdifferentiation from other reserve stem cell pools 30,40 or through plasticity of their daughter cells 37 . Importantly, antibody-conjugated drug targeting LGR5 CSCs in colon cancer has no major impact on the function of normal LGR5 stem cells 9 . In the liver, LGR5 stem cells are absent during homeostasis, but only transiently activated upon injury likely without major contribution toward tissue repair 4,14 . Thus, we envision that our identification of targetable LGR5 TICs in murine liver cancer bears important implications for future therapeutic development.

Methods
Primary liver tumor model. Lgr5-DTR-GFP transgenic mice (kindly provided by Genentech) specifically co-express the diphtheria toxin (DT) receptor (DTR) and green florescent protein (GFP) under the control of the Lgr5 promotor 30 . Thus, LGR5 + cells can be identified by GFP expression, and LGR5-GFP + cells can be specifically depleted by DT administration. Lgr5-DTR-GFP transgenic mice (3-4 weeks) were administered with diethylnitrosamine (DEN) by intraperitoneal injection (Sigma-Aldrich, i.p., 100 mg/kg) weekly for 6-17 weeks 41 . DEN is used to induce liver tumor in Lgr5-DTR-GFP transgenic and wild-type mice, which could cause liver disease from basophilic foci, hyperplasic nodules, hepatocellular adenoma, and finally lead to HCC 31,42,43 . Mice were killed 3-16 months after the last DEN injection, and livers/tumors were collected for further experiments (Supplementary Data 1: in total, 41 mice were monitored; 80.5%, 33 out of 41, mice formed liver tumors; the expression of LGR5 in each tumor/tumor-surrounding tissues was also listed; Supplementary Fig. 2g, h). For each liver, biopsies were taken from the tumor and tumor-surrounding tissue. If livers contain more than one tumor, individual tumors were collected and analyzed separately. For CCl 4 -induced liver injury, Lgr5-DTR-GFP transgenic mice were weekly repeated administered with (6 or 17 weeks) intraperitoneal CCl 4 injection (10 µl/20 g body weight of 10% CCl 4 solution in corn oil or corn oil as control). All animal experiments were approved by the Committee on the Ethics of Animal Experiments of the Erasmus Medical Center. Fig. 5 Anticancer treatment selects for LGR5 + cells. a Outline of the ex vivo experimental strategy used to assess the effects of drug treatment on the size of the LGR5 + compartment. b The fraction of LGR5-GFP + cells is significantly increased upon treatment with sorafenib or 5-FU (vehicle control vs. 10 µM sorafenib vs. 10 µM 5-FU: 2.6 ± 0.5% vs. 4.6 ± 0.4% vs. 21.3 ± 1.9%). c, d Representative FACS plots (c) and confocal pictures (d) demonstrating that the fraction of LGR5-GFP + cells is increased upon treatment with sorafenib or 5-FU, scale bar = 50 µm. e An outline of the experimental strategy used for testing the effects of drug administration in vivo. f The percentages of LGR5-GFP + cells is increased upon administration of sorafenib or 5-FU to allografted animals (vehicle control vs. sorafenib vs. 5-FU: 0.13 ± 0.04%, n = 6 vs. 0.42 ± 0.13 %, n = 8 vs. 0.66 ± 0.17%, n = 7). g Representative confocal pictures showing that both single LGR5-GFP + and LGR5-GFP − cell-initiated organoids contain LGR5-expressing cells, and the relative fraction of LGR5-expressing cells is increased in treatment-resistant organoids. LGR5-driven GFP: green, scale bar = 50 µm. h A volcano plot showing the most significantly differentially expressed genes between 5-FU treated/untreated LGR5 + cells. i Gene enrichment analysis (with the library of Wiki2019) within the differentially expressed genes. Mean ± SEM. Source data are provided as a Source Data file.
HCC specimens and patient information. HCC specimens (paired tumor tissue and adjacent tumor-free liver tissue) were collected from HCC patients undergoing tumor resection at the Erasmus Medical Center, The Netherlands. Samples were stored at −80°C and then used for RNA extraction. In total, 74 specimens were obtained from HCC patients, and the corresponding clinicopathological data are summarized in Supplementary Table 2. HCC-specific survival was assessed in all patients and they were stratified according to relative LGR5 expression (below and above median -0.047). The Kaplan-Meier method was used to estimate survival outcome curves, and the log-rank test was used to compare the survival between the two groups. The hazard ratio (HR) for HCC-specific survival was estimated LGR5+ LGR5+ LGR5+ LGR5+  Histology, immunohistochemistry, and immunofluorescence. The liver or tumor was fixed in 4% paraformaldehyde (PFA) overnight at 4°C. For immunofluorescence, samples were further dehydrated with 30% sucrose (Sigma-Aldrich, 4°C, overnight), stored at −80°C, and then sectioned at 8 μm for further analysis. Images were acquired with a Zeiss LSM510META confocal microscope. For histology and immunohistochemistry, materials were dehydrated with 70% ethanol, embedded with paraffin, and sectioned at 4 μm for staining. Images were acquired with a Zeiss Axioskop 20 microscope. All antibodies are listed in Supplementary  Table 3.
Organoid-based allograft tumor model. Cold advanced DMEM/F12 medium was used to collect the organoids. After centrifuging, the supernatant was discarded and organoid pellets (organoid fragmentation size: range from 5 to 150 µm) were mixed directly with matrigel in a ratio of 1:1 with a total volume of 200 µl. In total, 4-6week-old female NOD.Cg-PrkdcSCIDIl2rgtm1Wjl/SzJ (NSG) mice, NOG/JicTac (CIEA NOD.Cg-Prkdc-scid Il2rg-tm1Sug) mice, or nude mice (NMRI:BomTac-Nude) were purchased from Taconic, and subcutaneously injected with the collected tumor organoids. The characterization of phenotypes for murine allograft tumor is based on the expression of EpCAM/CK19 for CC-like and HNF4ɑ/AFP for HCC-like phenotype (Supplementary Data 2). Tumor dimensions were measured using calipers, and tumor volume was calculated as 0.523 × length × width × width 9 . Tumor formation was monitored every other day, and mice were killed to harvest tumors after the tumor diameter reached~2 cm. Tumor tissues were stored or cultured as described above.
Cell ablation by DT and treatment of 5-FU/sorafenib. To ablate LGR5 + cells in organoids, DT (Calbiochem, 1-10 ng/ml) was added to organoid expansion/ initiation medium, followed by further analysis 14  Metabolic activity analysis of organoids. Different organoid lines were seeded separately in a 24-/48-well plate. Sorafenib (10 µM) or 5-FU (10 µM) was added to the organoid culture since the initial day or post initiation day 3, respectively. Drugs were refreshed every other day. At days 6-7, organoids were incubated with Alamar Blue (Invitrogen, 1:20 in DMEM) for 4 h (37°C), and then the medium was collected for analysis of the metabolic activity of the cells. Absorbance was determined by using fluorescence plate reader (CytoFluor ® Series 4000, Perseptive Biosystems) at the excitation of 530/25 nm and emission of 590/35. Each treatment condition was repeated four times and matrigel with medium only was used as a blank control.
Statistical analysis. Prism software (GraphPad Software) was used for all statistical analyses. For statistical significance of the differences between the means of groups, we used Mann-Whitney U test; for statistical significance of the differences between groups with inequivalent sample sizes, we used Welch test; for statistical significance of the differences between paired samples, we used paired T test; for statistical significance of the differences between multiple independent groups, we used two-way ANOVA (except for Mann-Whitney U test, the use of other statistical methods is indicated in the legends). Differences were considered significant at a P value less than 0.05.
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article. Fig. 6 LGR5 lineage ablation inhibits organoid and tumor growth. a, b The outlines of the ex vivo experimental strategy to assess the effects of anticancer drug treatment on organoid initiation, and delineate its temporal aspect (a) or during organoid expansion (b

Data availability
The RNA-seq data have been deposited in the GEO database under the accession code GSE137517. The human LGR5 expression and HCC mutation data referenced during the study are available in a public repository from the TCGA and LICA databases. The source data underlying Figs. 1c-d, 2, 3e, 4d, f, h, i, p, 5b, f, h-i, 6c, h, I, 7b, c, f, g, 8b, c, f, g and Supplementary Figs. 2a, g, 3, 4, 6b-c, 7b-e, 8a-b, 9, 10c-d, 11, 12a-b and 13g-h are provided as a Source Data file. All the other data supporting the findings of this study are available within the article and its supplementary information files and from the corresponding author upon reasonable request. A reporting summary for this article is available as a Supplementary Information file.