CRISPR-induced exon skipping of β-catenin reveals tumorigenic mutants driving distinct subtypes of liver cancer
No conflicts of interest were declared.
Abstract
CRISPR/Cas9-driven cancer modeling studies are based on the disruption of tumor suppressor genes by small insertions or deletions (indels) that lead to frame-shift mutations. In addition, CRISPR/Cas9 is widely used to define the significance of cancer oncogenes and genetic dependencies in loss-of-function studies. However, how CRISPR/Cas9 influences gain-of-function oncogenic mutations is elusive. Here, we demonstrate that single guide RNA targeting exon 3 of Ctnnb1 (encoding β-catenin) results in exon skipping and generates gain-of-function isoforms in vivo. CRISPR/Cas9-mediated exon skipping of Ctnnb1 induces liver tumor formation in synergy with YAPS127A in mice. We define two distinct exon skipping-induced tumor subtypes with different histological and transcriptional features. Notably, ectopic expression of two exon-skipped β-catenin transcript isoforms together with YAPS127A phenocopies the two distinct subtypes of liver cancer. Moreover, we identify similar CTNNB1 exon-skipping events in patients with hepatocellular carcinoma. Collectively, our findings advance our understanding of β-catenin-related tumorigenesis and reveal that CRISPR/Cas9 can be repurposed, in vivo, to study gain-of-function mutations of oncogenes in cancer. © 2023 The Authors. The Journal of Pathology published by John Wiley & Sons Ltd on behalf of The Pathological Society of Great Britain and Ireland.
Introduction
Liver cancer significantly contributes to cancer-related mortality [1]. The most frequent type of liver cancer in adults is hepatocellular carcinoma (HCC), accounting for ~90% of all liver cancers [1]. Although hepatoblastoma (HB) is a rare form predominant in children [2], aberrant oncogenic activation of Wnt/β-catenin signaling pathway is linked to both HCC and HB [1]. Approximately 26% of human HCC patients present β-catenin mutations [3]. Most hotspot mutations are within exon 3 of CTNNB1 that encodes the domain containing phosphorylated serine/threonine residues targeting β-catenin for degradation in cytoplasm. Such mutations impair β-catenin degradation and enable β-catenin to accumulate in the nucleus and to drive the transcription of various oncogenes [4].
β-catenin mutations correlate with tumor progression [5] and have been used to refine liver cancer classification [6]. In HB, different mutations of β-catenin associate with histologically and transcriptionally distinct subtypes [7], such as HCC patients with a weak S45 mutation. For instance, the weak, intrinsic β-catenin activity related to S45 mutation in some HCC patients is complemented by additional mutations [5]. Thus, better understanding of β-catenin mutations in liver tumorigenesis is important to define disease heterogeneity and patient outcomes.
It has long been believed that mutated β-catenin alone is insufficient to form liver tumor, although Loesch et al recently demonstrated in mice that mutated β-catenin can induce liver tumor per se. However, tumor formation requires approximately a year [8]. Thus, additional oncogenic events, such as Yap1 or c-Met [9], encourage malignant transformation. Mutated β-catenin N90 (ΔN90) and YAPS127A have been used to model liver cancer with features of HB [9]. Depending on its oncogenic partner, β-catenin can also drive HCC [9]. However, it is unknown whether different β-catenin mutations can drive different tumor types in synergy with the same oncogene. Most of the CRISPR/Cas9-based cancer modeling involves the disruption of TSGs by introducing indels that lead to frameshift mutations and loss of function. Whether indels also lead to gain of function of oncogenes is not well known. We previously demonstrated that a single guide RNA targeting Ctnnb1 exon 3 induces exon-skipped in-frame transcripts in lung cancer KrasG12D;Trp53−/− cells (KP cell line) [10]. However, whether CRISPR-induced exon skipping occurs in vivo remains unknown.
Here, we delivered a single guide RNA (sgCtnnb1) targeting Ctnnb1 exon 3 to liver cells that resulted in nuclear localization of active β-catenin. Further simultaneous delivery of sgCtnnb1 and YAPS127A confers growth advantage to exon-skipped hepatocytes, thereby inducing tumor formation in vivo. Intriguingly, we report two distinct subtypes of liver tumor: one is induced by exon 3 deleted β-catenin isoform and YAPS127A, while the other is induced by exons 3 and 4 deleted β-catenin isoform and YAPS127A. Notably, we found three human HCC cases, in The Cancer Genome Atlas (TCGA), revealing exons 3 & 4 skipped or exon 3 skipped β-catenin isoforms. Moreover, our transcriptome analysis provides further evidence for the relevance of our tumor models to human liver cancer.
Materials and methods
Plasmids and cloning
Sleeping Beauty-based β-catenin N90 (ΔN90) (Catalogue No. 31785), Yap (Catalogue No. 86497), and c-Met (known also as Met) (Catalogue No. 31784) plasmids were purchased from Addgene (Watertown, Massachusetts, USA). gBlocks Gene Fragments of skipped β-catenin isoforms for Gibson cloning were synthesized by Integrated DNA Technologies (IDT; Coralville, IA, USA). We assembled the gBlocks Gene Fragments together with PGK promoter into the Sleeping Beauty vector, which retains transposase-mediated integration capacity, by following the instructions from the Gibson cloning kit (NEB; Ipswich, MA, USA; Catalogue No. E2611). All sgRNAs used for hydrodynamic injection were cloned into px330 backbone (Addgene; Catalogue No. 42230). Primers and oligos used for cloning are listed in supplementary material, Table S1.
Cell culture
We cultured KPC cells derived from pancreatic cancer model (KrasG12D; Trp53−/−) with DMEM (Corning, NY, USA; Catalogue No. 10-13-CV) supplemented with 10% FBS (Thermo Fisher Scientific, Waltham, Massachusetts, USA; Catalogue No. 16000044). For transfection we used lipofectamine 2000 reagents (Invitrogen, Waltham, MA, USA; Catalogue No. 11668027) following the manufacturer's instructions. At 2 days after transfection, we collected cells to isolate genomic DNA (Roche Diagnostics, Indianapolis, IN, USA; Catalogue No. 06650767001) and total RNA (Qiagen, Germantown, MD, USA, RNeasy Mini kit; Catalogue No. 74104) by following the manufacturer's instructions.
Hydrodynamic tail vein injection of plasmids
To prepare large quantities of plasmids, we cultured 250 ml Escherichia coli bacteria bearing transformed plasmids and isolated the plasmids using Qiagen Maxi-Prep Endotoxin-free kit (Qiagen; Catalogue No. 12362) following the manufacturer's instructions. Hydrodynamic tail vein injection was performed as previously described [11]. In brief, we mixed 40–60 μg plasmid DNA in 2 ml 0.9% sterile saline solution (w/v) at room temperature and delivered the mixture to mice via tail vein injection within 5–7 s. Mice were then warmed by heating pad for 30 min to recover from the injection shock.
Histology and immunohistochemistry
We euthanized the mice, collected the tissues, and fixed them overnight in 4% neutral buffered formalin (v/v) overnight. Fixed tissues were then switched to 70% ethanol (v/v) and submitted to the Histology Core Facility of Cold Spring Harbor Laboratory for paraffin embedding, sectioning, and hematoxylin and eosin (H&E) staining. We performed immunohistochemistry (IHC) on unstained slides. In brief, we deparaffinized the slides in 100% xylene three times and hydrated the slides in series of ethanol (v/v) 100, 95, 90, and 70%, three times each. We then boiled the slides for 10 min in 1 mM citrate buffer (w/v), pH 6.0, to retrieve the antigens. We used the ImmPRESS Excel Amplified HRP Polymer Staining kit (Vector Laboratories, Newark, CA, USA; Catalogue No. MP-7601) for IHC staining. BLOXALL reagent was used to block endogenous peroxidases, and 2.5% normal horse serum (w/v) was used to block nonspecific antibody binding. Primary antibodies against β-catenin (Santa Cruz Biotechnology, Dallas, TX, USA; Catalogue No. sc-7,199; 1:200 dilution), YAP (Cell Signaling Technology, Danvers, MA, USA; Catalogue No. 14074; 1:200 dilution) or Ki67 (Thermo Fisher Scientific; Catalogue No. MA5-14520; 1:200 dilution) were incubated overnight at 4 °C. We then used the ImmPRESS kit to generate a brown color signal from the 3,3′–diaminobenzidine (DAB) substrate provided in the kit. To visualize a red color signal, we added a biotinylated secondary antibody and used an avidin/biotin complex (ABC) system provided in Vectorstain ABC-AP kit (Vector Laboratories; Catalogue No. AK5200) and Vector Red substrate kit (Vector Laboratories; Catalogue No. SK-5100). The slides were then counterstained with hematoxylin for 2 min and dehydrated in series of ethanol 70, 90, 95, and 100%, three times each. Slides were then immersed in 100% xylene three times and coverslipped for long-term storage. Images were captured using Olympus microscope (Olympus Corporation, Shinjuku-ku, Tokyo, Japan; Catalogue No. DP72).
RT-PCR, TOPO cloning, and Sanger sequencing
We reverse-transcribed isolated mRNA using a high-capacity cDNA reverse transcription kit (Thermo Fisher Scientific; Catalogue No. 4374966). We then performed LaTaq PCR (Takara Bio, San Jose, CA, USA; Catalogue No. RR002A) to amplify the Ctnnb1 transcripts, which were then electrophoresed through 1.5% agarose gels (w/v) to visualize the amplicon's size. Skipping transcripts were gel purified using a QIAquick gel extraction kit (Qiagen; Catalogue No. 28704), cloned into a TOPO vector using a TOPO TA cloning kit (Invitrogen; Catalogue No. 450071) and then Sanger sequenced (Eurofins Company sequencing service; Louisville, KY, USA).
Reverse Transcription-quantitative PCR (RT-qPCR)
Total RNA was extracted from snap-frozen tissue samples using a Quick-RNA Miniprep Plus Kit following the manufacturer's instructions (Zymo Research, Irvine, CA, USA; Catalogue No. R1057). Then 500 ng of total RNA was converted to cDNA using SuperScript™ IV VILO™ Master Mix (Thermo Fisher Scientific; Catalogue No. 11756050). The cDNA was diluted at a 1:10 ratio with UltraPure™ DNase/RNase-Free Distilled Water (Thermo Fisher Scientific; Catalogue No. 10977015). The qPCR reactions were performed using TaqMan™ Fast Advanced Master Mix (Thermo Fisher Scientific; Catalogue No. 4444557). PCR amplification signals were detected using Applied Biosystems™ QuantStudio™ 6 Flex Real-Time PCR System (Thermo Fisher Scientific; Catalogue No. 4485691). The probes utilized for The TaqMan assays included Mm02619580_g1 (Actb, Catalogue No. 4453320), Mm00725701_s1 (Glul, Catalogue No. 4453320), Mm00443610_m1 (Axin2, Catalogue No. 4453320), Mm00521920_m1 (Lect2, Catalogue No. 4448892), and Mm01195726_m1 (Tbx3, Catalogue No. 4453320).
Western blotting
We homogenized the collected tissues in RIPA lysis buffer (Cell Signaling Technology; Catalogue No. 9806) (25 mm Tris–HCl, pH 7.6, 150 mm NaCl, 1% NP-40 [v/v], 1% sodium deoxycholate [w/v], and 0.1% SDS [w/v]) containing 1× E-64 protease inhibitor (Sigma; Rockville, MD, USA; Catalogue No. E3132) and centrifuged at 12,000 rpm to obtain a clear supernatant. The protein concentration of samples was determined using a Pierce Bicinchoninic Acid (BCA) protein assay kit (Thermo Fisher Scientific; Catalogue No. 23225). Samples were boiled in 4X NuPAGE lithium dodecyl sulfate (LDS) sample buffer (Invitrogen, Catalogue No. NP008) containing 100 mm dithiothreitol. We ran 20 μg total proteins on 4–12% NuPAGE Bis-Tris mini protein gels (Invitrogen; Catalogue No. NP0321BOX). Proteins separated by their approximate sizes on the gel were then transferred onto a nitrocellulose membrane (Thermo Fisher Scientific; Catalogue No. 88018). We next blocked the membrane in blocking buffer (Li-Cor, Lincoln, NE, USA; Catalogue No. 927-60001) for 1 h at room temperature before incubation with primary antibodies against β-catenin (Cell Signaling Technology; Catalogue No. 8480 S; 1:2000 dilution), and HSP90 (BD Biosciences; Franklin Lakes, NJ, USA; Catalogue No. 610418; 1:2000 dilution) overnight at 4 °C. The following day, bound primary antibody was detected using dye-conjugated secondary antibodies: IRDye 800CW (Li-Cor; Catalogue No. 926-32213; 1:15,000 dilution) or 680RD (Li-Cor; Catalogue No. 926-68072; 1:15,000 dilution) and visualized using a Li-Cor Odyssey imaging instrument (Li-Cor).
RNA sequencing
We isolated total RNA from tumors using a MirVana miRNA Isolation kit (Invitrogen; Catalogue No. AM1561) and removed ribosomal RNA using 186 oligos anti-sense to rRNAs, which helped degrade rRNAs using RNAse H, as described previously in an established protocol [12-16]. To enrich RNA >200 nt and remove tRNAs, we purified RNA samples using RNA Clean & Concentrator-100 (Zymo; Catalogue No. R1019). We prepared strand-specific libraries for high-throughput RNA sequencing (RNA-seq), as described previously [14-16]. All reagents related to RNA-seq library preparation are listed in supplementary material, Table S2. The prepared RNA libraries were sequenced by 79 + 79 paired-end reading using a NextSeq550 (Illumina, San Diego, CA, USA).
Raw reads were processed using DolphinNext [17] RNA-Seq pipeline Revision 2. STAR (version 2.6.1) [18] and RSEM (version 1.3.1) [19] indices were created based on the mouse genome assembly mm10 and gene annotations downloaded from University of California, Santa Cruz (UCSC) Genome Browser. The RSEM software package [19] was then used to estimate relative gene expressions using STAR [18] aligner. To examine differentially expressed transcripts, we used DEBrowser [20], an interactive differential expression analysis tool. We compared the raw gene counts estimated by RSEM from biological replicates in different conditions using DESeq2 [21]. After normalization by DESeq2's Median Ration Normalization method, we performed Combat [22] to adjust for possible batch effect or conditional biases. Gene ontology (GO) term enrichment analysis was performed using the Database for Annotation, Visualization and Integrated Discovery (DAVID) functional annotation tool [23]. A complete list of GO term categories with significant enrichment was extracted. Gene set enrichment analysis (GSEA) was conducted using the GSEAPreranked [24] software package on results from differential analysis generated by DESeq2.
Exon skipping events in human tissue and cancer samples
The Cancer Genome Atlas (TCGA) samples
Aligned genomic BAM RNA-seq files (GRCh38/hg38 coordinates) for TCGA tissue samples with primary diagnosis hepatocellular carcinoma, NOS (HCC) were downloaded from Genomic Data commons TCGA-LIHC (dbGAP accession phs000178.v11.p8, project ID 26811). As described in the TCGA data processing protocol (https://docs.gdc.cancer.gov/Data/Bioinformatics_Pipelines/Expression_mRNA_Pipeline/), genomic Bam RNA-seq files were generated using STAR [18] in two-pass mode with a splice junction detection step, allowing for downstream splicing analysis. We used samtools [25] to extract all reads mapping to the coding region of CTNNB1: ‘samtools view -o output.bam input.bam chr3:41194741-41260096’.
ExonSkipDB
We queried ExonSkipDB [26] for CTNNB1 exon skipping events in any tissue or cancer as curated based on data from the genotype-tissue expression (GTEx) portal and TCGA. ExonSkipDB reports its exon skipping events in GRCh37/hg19 coordinates. We used the UCSC liftover tool [27] with the following parameters to convert the coordinates of the exon skipping events to GRCh38/hg38: minimum ratio of bases that must remap: 0.95; allow multiple output regions: off; minimum hit size in query: 0; minimum chain size in target: 0; if thickStart/thickEnd is not mapped, use the closest mapped base: off. All originally reported positions were successfully converted.
Visualization of TCGA and ExonSkipDB events
Exon skipping events identified via ExonSkipDB and TCGA HCC RNA-seq reads were visualized using Gviz [28]. As a reference, the exon and intron positions for canonical β-catenin isoforms expressed in liver (ENST00000349496, ENST00000396185, ENST00000396183) were added as a separate track.
Gene expression analysis
Gene expression counts normalized as transcript per million (TPM) were downloaded using TCGAbiolinks (version 2.24.3 [29]) with project = ‘TCGA-LIHC’, data.category = ‘Transcriptome Profiling’, data.type= ‘Gene Expression Quantification’, and workflow.type = ‘STAR – Counts’. TPM values were scaled across samples and visualized using ComplexHeatmap (version 2.12.1 [30]).
Results
Single guide RNA targeting Ctnnb1 exon 3 induces nuclear accumulation of β-catenin in mouse liver cells
To test whether CRISPR-mediated exon skipping events produced gain-of-function β-catenin isoforms in vivo, we delivered Cas9/sgCtnnb1.1 targeting Ctnnb1 exon 3 to liver cells via hydrodynamic tail vein injection (HDI) [10, 11] (Figure 1A and Table S1). To examine CRISPR-mediated exon skipping events, we performed RT-PCR from the liver tissues of injected and noninjected mice at 2 weeks and 2 months after HDI: an ~0.9 kbp PCR band was detected in injected liver compared to noninjected control at 2 weeks after HDI, while the band was under the limit of detection at 2 months after HDI, suggesting that a low number of hepatocytes are edited in vivo that are subsequently replaced by nonedited hepatocytes (Figure 1B and supplementary material, Table S3). Consistent with this, IHC staining for β-catenin revealed that at 2 weeks following HDI, 1.07% of hepatocytes (median = 1.1 ± 0.2%) retained nuclear β-catenin, while at 2 months following HDI, the positive nuclear staining for β-catenin was not detectable (Figure 1C and supplementary material, Figure S1A). Together, these data demonstrate that a single guide RNA targeting Ctnnb1 exon 3 is sufficient to induce a rare but active nuclear exon 3 skipped β-catenin isoform in vivo.
CRISPR-mediated Ctnnb1 exon skipping along with YAPS127A induces liver tumorigenesis
β-catenin mutations are frequently observed in HB patients, especially in exons 3 and 4 (supplementary material, Figure S1B) [6]. Moreover, activation of β-catenin together with YAPS127A (a constitutively active nuclear YAP) induces HB [9]. In line with this, we hypothesized that if CRISPR-mediated exon skipping generates a functional nuclear β-catenin, cells with both exon-skipped β-catenin and YAPS127A will have a growth advantage and form liver tumors.
To directly test this, we delivered Sleeping Beauty-based YAPS127A (SB-YAPS127A) and sgCtnnb1.1/Cas9 together to liver cells (supplementary material, Figure S1C). At 2 months following HDI, we observed gross tumor formation in liver compared to control mice (Figure 1D). Furthermore, RT-PCR using freshly frozen tumors detected two major Ctnnb1 skipping isoforms (supplementary material, Figure S1D). We then Sanger sequenced these two bands: isoform 1 skipped the entire exon 3, which is in frame with exon 4; isoform 2 skipped the entire exons 3 and 4 but with a 16-nucleotide intronic sequence, causing this isoform to be in frame with exon 5 (Figure 1E). To test whether these skipping Ctnnb1 isoforms encode proteins, we performed western blotting and observed that additional β-catenin protein bands appeared in tumor samples with smaller molecular weight compared to control liver tissues (Figure 1F).
To rule out the CRISPR off-target effect, we delivered two different single guide RNAs targeting Ctnnb1 exon 3 with distinct protospacer adjacent motif (PAM) sequences (supplementary material, Table S1; sgCtnnb1.2 and sgCtnnb1.3), along with YapS127A. In accordance with sgCtnnb1.1, both induced tumors with nuclear β-catenin and YAP localization (supplementary material, Figure S2A). To further substantiate our findings, we delivered another oncogene, c-Met (Met), alongside sgCtnnb1.1 to liver cells via HDI (supplementary material, Figure S2B). At 1 month following HDI, we observed liver tumor formation (supplementary material, Figure S2B). Together, these results demonstrate that a single guide RNA targeting Ctnnb1 exon 3 results in exon-skipped isoforms acting synergistically with YAPS127A or MET to drive liver tumor formation.
Exon-skipped β-catenin isoforms induce distinct liver tumor subtypes
We examined the tumor sections and surprisingly encountered two histologically distinct tumor lesions (henceforth, Subtypes A and B) (Figure 2A). Both Subtypes A and B tumors showed Ki67-positive IHC staining, which is a well-established proliferation marker (Subtype A tumors, mean fraction of Ki67[+] cells = 0.30 ± 0.07; Subtype B tumors, mean fraction of Ki67[+] cells = 0.40 ± 0.01; supplementary material, Figure S3A). Subtype A was characterized by pleomorphic neoplastic hepatoblasts with particularly variable sized round to oval nuclei (anisokaryosis) and uniformly condensed chromatin. The neoplastic nuclear/cytoplasmic ratio varied significantly. There was extensive cytoplasmic clearing, especially perinuclear, and occasional mild accumulation of microvesicles. These cytoplasmic changes were reminiscent of glycogen and lipid accumulation associated with altered metabolism. Subtype B tumors were characterized by uniform small neoplastic epithelial hepatoblasts with relatively small, round nuclei with conspicuous peripheral clumping of chromatin and central pallor. There was mild cytologic atypia, and occasional nuclei demonstrated prominent nucleoli. The cells had eosinophilic cytoplasm. For both subtypes, occasional mitotic figures were present, and the neoplastic masses were not encapsulated but were well demarcated with mild compression of the adjacent liver (Figure 2A). Next, to assess lipid accumulation in these histologically distinct tumor subtypes, we performed Oil Red O staining that stains for neutral triglycerides and lipids on frozen sections. We found that the cytosol of Subtype A cells showed clear Oil Red O staining, while the signal was below the limit of detection in the cytosol of Subtype B cells (supplementary material, Figure S3B). Concordant with this, our Periodic Acid-Schiff staining, which detects glycogen, revealed accumulated glycogen in the cytoplasm of Subtype A cells (supplementary material, Figure S3C). Interestingly, while both Subtypes A and B showed similar YAP nuclei-positive signal, the localization pattern for β-catenin was different. Specifically, while Subtype B showed the expected nuclear β-catenin staining, Subtype A surprisingly revealed no clear nuclear β-catenin accumulation. Instead, β-catenin was detected mostly around the cytoplasm and membrane in Subtype A (Figure 2A). These data indicate that CRISPR-mediated exon skipping results in oncogenic isoforms of β-catenin that drive histologically distinct liver tumors in mice and exhibit different cellular localization patterns.
To determine the transcriptional signatures of these histologically distinct tumor subtypes driven by exon-skipped β-catenin isoforms, we randomly collected 13 individual liver tumor samples and three normal liver tissue samples from five different mice and performed RNA-seq. Unsupervised hierarchical clustering of all 13 liver tumor samples revealed 6,027 differentially expressed genes in tumors compared to normal liver (supplementary material, Figure S4A and Table S4; <twofold and >twofold change, false discovery rate [FDR] < 0.01). Expression levels of well-established liver tumor marker genes, such as Gpc3, Mki67, Afp, and Cd34 [31-33], were increased in tumor groups compared to control (supplementary material, Figure S4B). Clustering analysis identified two major groups of tumor samples: T1 to T4 were in one group (Subtype A), while T5 to T13 were in another group (Subtype B) (supplementary material, Figure S4A). Differential gene expression analysis between the two groups of tumor samples revealed 460 differentially expressed genes (<twofold and >twofold change, FDR < 0.01) (Figure 2B and supplementary material, Table S1). To define functional categories enriched in distinct subtypes, we performed GSEA. Consistent with the histological features of Subtype A, metabolic pathways, including lipid metabolism, were among the top functional categories enriched in upregulated genes in T1 to T4 tumor samples (Subtype A) (Figure 2C). Pathways that were enriched in upregulated genes for T5 to T13 (Subtype B) included tissue development and morphogenesis (supplementary material, Figure S4C). Importantly, our GSEA confirmed the existence of liver tumor and YAP activation signatures for all 13 tumor samples compared to normal liver (supplementary material, Figure S4D). Because these subtypes exhibited distinct β-catenin localization patterns, we compared the expression levels of β-catenin target genes between Subtypes A and B. Several notable β-catenin targets were differentially expressed, including Id2, Lgr4, Cd44, S100a6 (supplementary material, Figure S4E). Taken together, we consider that exon-skipped β-catenin isoforms synergize with YAPS127A to drive two histologically and transcriptionally distinct tumor subsets exhibiting differences in lipid metabolism and expression of β-catenin target genes. The ways in which different β-catenin isoforms influence lipid metabolism and β-catenin target gene transcription warrant further investigation.
Ectopic expression of exon-skipped β-catenin isoforms together with YAPS127A drive distinct liver tumor subtypes
To definitively assess whether the phenotypes we observed are driven by the oncogenic effect of CRISPR-induced exon-skipped β-catenin isoforms rather than the off-target activity of Cas9, we delivered two in-frame cDNA of skipped Ctnnb1 isoforms together with YapS127A into liver cells: one with the entire exon 3 skipped (Skip-1), and the other with both exons 3 and 4 skipped entirely but with a 16 nt intron insertion (Skip-2) (supplementary material, Table S5). Introducing both β-catenin isoforms along with YAPS127A into liver cells formed tumors (Figure 3A). Our western blotting from the tumor protein lysates validated the expected molecular weight of Skip-1 and Skip-2 isoforms (Figure 3B). The gross histology of tumor samples revealed that, similarly to Subtype A, Skip-1-induced tumor cells retained larger nuclei and unstained cytosol (Figure 3C), whereas Skip-2-induced tumor cells showed smaller nuclei and more structured and condensed tumor cells phenocopying Subtype B, which is typically the histological hallmark of HB [9] (Figure 3C). In line with what we observed in Subtype A, Skip-1-induced tumors retained less nuclear staining for β-catenin compared to Skip-2 (Figure 3C).
To determine the transcriptional profile of tumors induced by Skip-1 and Skip-2, we performed RNA-seq. Using the most varied 1,000 genes, principal component analysis (PCA) distinguished Skip-2 tumors from Skip-1 tumors and placed Skip-2 tumors near ΔN90-β-catenin/YAPS127A-induced HB tumor samples, an established HB model [9] (supplementary material, Figure S5A). Furthermore, control mice injected with either Skip-1 or Skip-2 isoforms without YAPS127A were located near wild-type animals (supplementary material, Figure S5A). Accordingly, unsupervised hierarchical clustering analysis grouped Skip-2-induced tumors with ΔN90-β-catenin-induced tumors, while Skip-1-induced tumors were separated from them (Figure 3D).
Given the histology of Skip-1-induced tumors resemble Subtype A and Skip-2-induced tumors resemble Subtype B, we further examined whether the transcriptome of Skip-1 and -2-induced tumors was similar to that of Subtype A and B tumors. Our unsupervised hierarchical clustering of all tumor samples using 460 genes differentially expressed in Subtypes A and B revealed that the majority of Skip-2-induced tumors were clustered with Subtype B, while Skip-1-induced tumors were clustered with Subtype A (Figure 3E). Moreover, GSEA using differentially expressed genes in Skip-1 and Skip-2 compared to their controls revealed that only Skip-2-induced tumors retained a HB signature and that both Skip-1 and -2 tumors showed a YAP activation signature (supplementary material, Figure S5B). Of note, both Skip-1 and -2 tumors showed Ki67 IHC-positive staining, indicating active proliferation of tumor cells (Skip-1 tumors, mean fraction of Ki67[+] cells = 0.24 ± 0.04; Skip-2 tumors, mean fraction of Ki67[+] cells = 0.29 ± 0.05; supplementary material, Figure S5C). Furthermore, the mRNA abundance of liver tumor marker genes, Cd34, Gpc3, and Mki67 [31, 33, 34], increased in both Skip-1 and -2 tumors compared to controls (supplementary material, Figure S6A). Our transcriptome data revealed that, while the mRNA abundance of Glul (a direct β-catenin target and well-established HCC marker [35]) increased in Skip-1 tumors, we observed that several β-catenin pathway genes, such as Cps1, Abcd2, Mmp15, and Igfbp2 [36-38], were exclusively upregulated in Skip-2 tumors (supplementary material, Figure S6B).
We further examined the change in the mRNA abundance of several other β-catenin target genes, Axin2, Lect2, and Tbx3, along with Glul in Skip-1 and -2 tumors using RT-qPCR. Among these four genes we assayed, consistent with our RNA-seq analysis, only Glul mRNA was increased in Skip-1 tumors compared to Skip-2 tumors and normal liver tissues (supplementary material, Figure S6C), suggesting that the dysregulation of Axin2, Lect2, and Tbx3 genes might require the activation of other oncogenes. Consistent with the increased mRNA abundance of Glul in Skip-1 tumors, IHC revealed profound staining for glutamine synthetase in the gross section of a Skip-1 tumor compared to that of a Skip-2 tumor and normal liver tissue (supplementary material, Figure S6D).
Finally, because a single guide RNA-driven exon 3-skipped β-catenin isoform acts with YAPS127A to generate tumors, we tested whether using two single guide RNAs (targeting introns before and after exon 3; supplementary material, Table S1) also produced a gain-of-function β-catenin isoform in vivo to form tumors in synergy with YAPS127A (supplementary material, Figure S7A). In KP cells, our newly designed single guide RNAs yielded an ~0.9 kbp PCR band (supplementary material, Figure S7B) encoding a shorter β-catenin skipped isoform (supplementary material, Figure S7C). Delivering these two single guide RNAs along with YAPS127A to liver cells induced tumor formation at 2 months after HDI (supplementary material, Figure S7D). Histological analysis of these tumors revealed similar features with Subtype A and Skip-1-induced tumors that are driven by the removal of exon 3 of β-catenin (supplementary material, Figure S7E). These data confirm the oncogenic activity of loss of β-catenin exon 3 in vivo.
Transcriptome of tumors driven by exon-skipped β-catenin isoforms is similar to tumors from Apc loss of function and exon 3 Ctnnb1 deletion mouse models
Recently, Loesch et al, using either a Cre-lox or CRISPR/Cas9 system, generated two mouse models, one of which had an Apc loss-of-function mutation and the other retained exon 3 deleted Ctnnb1 gene [8]. Both models brought about two distinct phenotypes: differentiated or undifferentiated. Intriguingly, similar to Subtype A and Skip-1 tumors, differentiated tumors revealed extensive cytoplasmic clearing, whereas undifferentiated tumors had uniform small hepatoblasts with small, round nuclei that we found in Subtype B and Skip-2 tumors. Moreover, the transcriptome of differentiated tumors resembles that of HCC [39], while undifferentiated tumors retained a transcriptome signature that was close to HB [8]. We therefore sought to test the idea that the transcriptome of Subtype A and Skip-1 tumors was similar to that of differentiated tumors, while Subtype B and Skip-2 tumors resembled undifferentiated tumors. To test this, we reanalyzed the RNA-seq data of differentiated and undifferentiated tumors and performed unsupervised hierarchical clustering of all our tumors along with differentiated and undifferentiated tumors. Our analysis revealed that Subtype A and Skip-1 tumors were grouped with differentiated tumors, whereas four of the five undifferentiated tumors clustered with Subtype B and Skip-2 tumors (supplementary material, Figure S8). Together, our analysis substantiates the idea that, while Subtype A/Skip-1 tumors resembles HCC, Subtype B/Skip-2 tumors are similar to HB.
β-catenin isoforms are found in human HCC
Given that we identified two different CRISPR-mediated isoforms of β-catenin leading to histologically distinct liver tumors in mice, we next sought to understand whether these β-catenin isoforms might contribute to the etiology of human liver cancer. To this end, we analyzed exon skipping events using the transcriptome data of 371 HCC patients from TCGA. Intriguingly, our analysis identified three HCC patients retaining β-catenin (CTNNB1) exon skipping events: two cases (LG-A9QC and XR-A8TE) revealed CTNNB1 isoform with the exclusion of exons 3 and 4, while one case evidenced a CTNNB1 isoform with only exon 3 skipped (DD-AAD6) (supplementary material, Figure S9A,B). Given that tumors induced by exons 3- and 4-skipped Ctnnb1 isoform (i.e., Subtype B and Skip-2 tumors) histologically resemble HB, we next examined the differential expression of 13 genes featured in HB. Supervised hierarchical clustering of the transcript abundance of these genes grouped two HCC cases retaining the exon 3- and 4-skipped Ctnnb1 isoform together and placed them further from the HCC case with skipped exon 3 (supplementary material, Figure S9C). Together, our data represent the possible relevance of our model in the etiology of human liver cancer, and even the fraction of cases retaining exon 3- or exon 3- and 4-skipped CTNNB1 isoform is low among the cases we examined.
The transcriptome of Subtype B and Skip-2 tumors resembles that of human hepatoblastoma
To further substantiate the relevance of our model to human liver cancer etiology, we examined the association between the transcriptome signature of mouse liver tumors to that of human liver cancer. To do this, we reanalyzed the RNA-seq data of 14 human HB tumors [40]. Our supervised hierarchical clustering analysis based on 460 genes, differentially expressed between Subtypes A and B (Figure 2B), revealed that Subtype B and Skip-2 tumors clustered with human HB (supplementary material, Figure S10). These data suggest that Subtype B and Skip-2 tumors are transcriptionally similar to human HB.
Discussion
Herein we report that a single guide RNA targeting exon 3 of Ctnnb1results in exon-skipped transcripts in vivo. In fact, exon-skipped β-catenin isoforms encode functional proteins that act synergistically with YAPS127A or C-MET to induce liver tumors. More importantly, CRISPR-mediated exon-skipped β-catenin isoforms drive two distinct liver tumor subtypes.
Exon skipping and alternative splicing
Although CRISPR-mediated genome editing driven by single guide RNA has been used to disrupt gene activity, we recently reported that in a lung adenocarcinoma cell line, a single guide RNA can generate large deletions or alternative splicing in exons of oncogenes that trigger aberrant juxtaposition of exons encoding shorter protein isoforms [10]. However, the exact underlying mechanisms leading to CRISPR-induced exon skipping remain elusive. Mutations affecting splice-sites can lead to alternative splicing events [41, 42]. Accordingly, the indels within the Ctnnb1 exon 3 introduced by CRISPR leading to the disruption of exonic splicing enhancer sites could be a potential contributing factor for the complete skipping of exon 3. Alternatively, if splice sites in exons 3 and 4 are weak, the DNA insertion mediated by CRISPR can create stronger cryptic splice sites in the flanking intron, which can lead to the inclusion of intronic fragments, while exons 3 and 4 are removed. A recent study suggests that the disruption of essential splice sites by premature termination codon mutations may contribute to exon skipping [43]. Moreover, altered chromatin modifications [44], posttranslational regulation of the exon skipping machinery [45], or proteins regulating mRNAs, such as polypyrimidine tract-binding proteins 1 (PTBP1) [46], are other potential mechanisms underlying exon skipping.
Complete loss of CTNNB1 exon 3 has been reported in microsatellite stable colorectal cancer [47, 48]. Interstitial deletions involving exon 3 induces oncogenic β-catenin in primary colorectal carcinomas without APC mutations [47]. Skipping events are also observed in other oncogenes, such as BRCA1 and MET [48, 49]. Additionally, altered splicing resulting in in-frame deletion of exons from the COL11A2, FBN1, and FLOT1 genes is demonstrated in other disease settings [50-53]. More importantly, exon skipping or splicing has already been accepted as a therapeutic tool [54, 55]. Therefore, utilizing CRISPR-based exon skipping may provide invaluable tool to model disease in vivo to better understand the underlying molecular mechanisms. We also envision that the CRISPR-mediated exon skipping phenomenon may be used as a therapeutic tool to correct diseases. For example, patients with Duchenne muscular dystrophy who retain exon 44 exhibit more severe disease progression [50].
CRISPR-induced β-catenin isoforms drive distinct liver tumor subtypes
We unexpectedly found two histologically distinct tumors driven by alternatively spliced β-catenin isoforms. Tumor cells induced by Skip-1 isoform with an entire exon 3 skipped retained lipid and glycogen accumulation within their cytosol and revealed an accumulation of β-catenin mostly around the cytoplasm and membrane, while Skip-2-induced tumor cells retained strong nuclear β-catenin staining. How, then, does exon 3-skipped β-catenin isoform accumulate in cytosol and evade cytosolic degradation? Although most HCC cases retain highly active nuclear β-catenin, weak activating mutations in β-catenin were observed in some HCC patients with compromised nuclear β-catenin accumulation. However, weak activating mutations can be complemented by the existence of a second mutation [5], suggesting that hyperactivation of β-catenin is accompanied by additional events. In line with this, one possible explanation for cytosolic and membranous accumulation of β-catenin in Skip-1-induced tumors could be perturbed cascades that translocate or degrade β-catenin. In fact, the steady-state mRNA abundance of the Csnk1a1 gene, whose protein product is involved in β-catenin degradation, was reduced by 11% in Skip-1 tumors compared to Skip-2 tumors (mean decrease = 0.89 ± 0.09).
Mutations in CTNNB1 exon 3 correlate with β-catenin accumulation patterns in HCC [5]. Consistent with this, our transcriptome analysis revealed differentially expressed β-catenin targets. In particular, the mRNA and protein abundance of Glul, a classical β-catenin target gene, were increased in Skip-1-induced tumors, despite the lack of nuclear β-catenin compared to Skip-2-induced tumors. Accordingly, two subsets of human HCC patients, both bearing activated β-catenin, showed different patterns of β-catenin IHC staining: one subset with nuclear β-catenin signal and the other without nuclear signal [56]. Moreover, HCC patients with β-catenin exon 3 in-frame deletion exhibited increased GLUL expression, despite the fact that they retained less nuclear β-catenin staining [5], suggesting that β-catenin target genes were activated by additional events.
Our findings also provide insight into pathways downstream of β-catenin, potentially opening a window for identifying new physically interacting partners of β-catenin and the dynamics of β-catenin interactions with other oncogenes and how they drive distinct tumor subsets. This model can be expanded to study the functional domains of other proteins because skipping transcripts might generate potential gain-of-function mutations.
Acknowledgements
We thank CSHL Cancer Center Shared Resources (Animal, Flow Cytometry and Histology, Sequencing and Bioinformatics Core Facilities) supported by National Cancer Institute (NCI) Cancer Center Support Grant 5P30CA045508. We thank Soren Hough (Cambridge University, UK) and Angela Park (Weill Cornell Medical College, USA) for editing the manuscript. This work was supported by the National Institutes of Health (P30CA045508-33 to SB), Mark Foundation for Cancer Research (20-028-EDV to SB and TJ), the Oliver S. and Jennie R. Donaldson Charitable Trust (SB), and the STARR Cancer Consortium (I13-0052, SB). AK was supported by the National Center for Advancing Translational Sciences (Grant UL1 TR001453-01). DMO was supported by the Swedish Research Council (Grant 2020-03818).
Author contributions statement
DMO and SB conceived the study. HM, OE, KAÖ, MG, JY, ZW and YG performed experiments. AK, TY, HVM and YJ performed bioinformatic analysis. JEW performed histopathological analyses. DMO wrote the manuscript.
Open Research
Data availability statement
Sequencing data are available from the National Center for Biotechnology Information Sequence Read Archive using accession number GSE164109 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE164109).
