Induction of primary mouse liver cancer by multiplexed CRISPR/Cas9 targeting 34 TSGs
To induce mouse liver tumors using multiplexed CRISPR/Cas9, we chose 34 TSGs along with a 5’-upstream site of the Setd5 locus as a negative control for CRISPR/Cas9-mediated gene inactivation (Fig. 1A and Fig. S1). These 34 TSGs are involved in at least 11 cancer-related signaling pathways (Fig. 1A). We designed and constructed at least 3 sgRNAs for each target gene except p53, and analyzed these sgRNAs for their efficiency of frame-shift editing using our BGN reporter established previously (Fig. S2A)33. The 23-bp target sequence containing the PAM for each sgRNA was inserted into the I-SceI-EcoRI site of the BGN reporter (Fig. S2A). To test each sgRNA, the expression plasmids for Cas9 and sgRNA and the BGN reporter plasmid containing the Cas9-sgRNA target were transfected into mouse embryonic fibroblast cell line NIH-3T3 cells and the frequencies of “3n+1”-bp frame-shift measured by flow cytometry for GFP+ cells (Fig. S2B). After testing of each sgRNA, we selected the most effective sgRNA for each target gene to establish a plasmid library of 35 sgRNAs. We then used hydrodynamic tail vein injection (HTVI) to deliver this sgRNA plasmid library mixed with the Streptococcus pyogenes Cas9 (SpCas9) expression plasmid into mouse liver cells to induce liver tumors19,34. Total volume injected was 2mL with the amount of the SpCas9 plasmid fixed at 200 μg for each mouse. The amounts of each sgRNA in the sgRNA library injected ranged from 0.005 μg to 5 μg (Fig. 1B). Liver tumors were formed and visible in mice within 30-60 days (Fig. 1B-C). Dilution of sgRNA delayed development of Cas9/sgRNA-induced mouse liver cancer and reduced cancer induction (Fig. 1B). After each sgRNA was reduced to 0.2 μg, tumor development rarely occurred (Fig. 1B). Histological analysis of tumor nodules indicated that the hepatic lobule structures were destructed in tumor nodules (Fig. 1D). Staining for the HCC and ICC biomarkers alpha-fetoprotein (AFP), cytokeratin 19 (CK19) and Golgi glycoprotein 73 (GP73) indicated induction of HCC, ICC and mixed HCC-ICC type in CRISPR/Cas9-induced mouse liver tumors (Fig. 1D). In one section of a single tumor nodule, three selected regions exhibited distinct histologic features (Fig. 1E), indicating strong ITH in CRISPR/Cas9-induced liver cancer in mice.
Heterogeneity of targeted TSG mutations in CRISPR/Cas9-induced mouse liver tumors
We performed next generation sequencing (NGS) of PCR-amplified target sites in tumor nodules to identify targeted TSG mutations and determine the frequencies and spectra of these mutations. We define an allelic mutation with a frequency at no less than 5% as a true mutation to reduce the interference of sequencing errors. While different targets showed different mutational spectra, we also found that many same target sites carried different allelic mutations with varying frequencies. For example, in one analysis, in addition to their respective wild-type (WT) allele, the p53 target site had 3 types of mutations, 28.12% for 1-bp deletion on the left side of the break (termed Del1|0), 16.13% for 7-bp deletion on the right side of the break (Del0|7) and 13.87% for insertion of 1A at the break (Ins1A); Atm had two (23.81% for D6|0 and 20.07% for Ins1C) and Rb1 had none in addition to WT alleles (Fig. 2A). In each tumor, majority of target genes were mutated at intended sites and the mutation frequencies (MF) were high at some of these sites after the frequencies of all allelic mutations were combined for each site (Fig. 2B-C, and Table S1). The mutation profiles differed between tumors within a same mouse or from different mice, indicating intertumor heterogeneity in CRISPR/Cas9-induced liver cancer with respect to targeted mutations of TSGs.
In addition, a single tumor nodule was microdissected into four regions, and each region analyzed for targeted TSG mutations by NGS of PCR-amplified target sites. Significant variations existed in mutation profiles between these four regions (Fig. 2C and Table S1), suggesting ITH of targeted TSG mutations. These variations may be also attributable to varying healthy stromal components and/or residual normal tissue in individual tumor nodules.
To further analyze ITH of targeted TSG mutations, we established single-cell clones from tumor nodules and analyzed targeted mutations in these tumor cells by targeted PCR amplicon deep sequencing. While about half of 34 TSGs were highly mutated between individual clones from the same nodules (in red in Fig. 2C), the frequencies of targeted mutations varied widely between 0% and 100% in the remaining genes (from Smad4 to Hnf1a; Fig. 2C). Between different clones from the same nodules, some targets sites carried identical mutations with similar frequencies but some harbored identical mutations with different frequencies or even different mutations (Fig. 2C and Table S2). These results indicated strong ITH of targeted TSG mutations in CRISPR/Cas9-induced mouse liver cancer.
As observed previously22, individual target sites often carried more than two different allelic mutations within a tumor nodule (Table S1). This could be explained at least by the following three possibilities: 1) Several transfected founder cells happened to start together with a different subset of mutations induced by multiplexed CRISPR/Cas9 genome editing and developed into a single tumor nodule. 2) Some mutations were first induced in a single transfected founder cell by multiplexed CRISPR/Cas9 genome editing and others occurred only after the first cell division in subsequent daughter or granddaughter cells22. 3) More than two copies of target sites exist in the genome of a single transfected founder cell and are differently mutated by CRISPR/Cas9. However, in some single-cell clones (e.g., 1C3 and 6C7), a few target sites harbored more than two different allelic mutations and even up to 9 different allelic mutations, some of which occurred with varying frequencies (Table S2). This was unexpected for a single-cell clone where only two copies of a target site are the most likely for CRISPR/Cas9-induced mutations unless the cells are hyperploid or the target sites have repeat sequences. Nevertheless, this type of copy number variations (CNVs) at the single-cell level represents a new source for heterogeneity of targeted TSG mutations in CRISPR/Cas9-induced mouse liver cancer.
Type and frequency of targeted mutation alterations from parental clones to subclones
Given the presence of more than two allelic mutations with significant different frequencies in a single-cell clone, we wondered whether this mutation pattern is inheritable. We thus selected two single-cell clones (i.e., 1C3 and 6C7) that carried more than two copies of allelic mutations at some target sites, isolated 4 subclones from 1C3 and 8 subclones from 6C7, and analyzed targeted mutations in these clones and subclones. The profile of target site mutations was different between parental clones and their respective subclones and between subclones (Fig. 3A and Table S3). For instance, 6 Rb1 target site variants, 8 Lkb1 target site variants, 4 Arid1a target site variants and 5 Smad4 target site variants were detected across parental 1C3 clone and its 4 subclones, but each type displayed different frequencies within individual clones and the mutation profiles were also different between clones (Fig. 3B-E). The frequency of the wide type Rb1 was less than 2% in paternal clone 1C3, but over 30% in 4 subclones (Fig. 3B). In contrast, the frequency of the Rb1 Del22|5 mutation was nearly 30% in 1C3, but hardly detectable in subclone 1C3-1 or 1C3-2, less than 3% in 1C3-3 and about 5% in 1C3-4 (Fig. 3B). The Lkb1 Del4|0 mutation was barely detected in 1C3, but the frequencies of this mutation were 20% or more in 4 subclones (Fig. 3C). The Lkb1 Del10|0 mutation was a dominant mutation with the frequency at over 30% in 1C3 and its subclone 1C3-4, but less than 1% in subclone 1C3-1 and 1C3-2 and around 16% in subclone 1C3-3 (Fig. 3C). The Arid1a Del11|16 mutation was dominant at more than 70% in 1C3-1 but negligible at less than 2% in 1C3-4 (Fig. 3D). In contrast to Del11|16, the Arid1a Del10|9 mutation was nearly undetectable in 1C3-1 but highly frequent at more than 50% in 1C3-4 (Fig. 3D). Similarly, while the Smad4 WT allele was detected at over 60% in 1C3 but less than 6% in 1C3-4, the frequency of the Smad4 Del2|0 was 2% in 1C3 but nearly 35% in 1C3-4 (Fig. 3E). These results indicate continuing oscillation in the frequencies of target site mutations between these parental clones and subclones. In particular, predominant mutations in parental clones could disappear in subclones whereas negligible mutations in parental clones could appear in subclones.
As spontaneous mutation at a given site normally occurs with extremely low probability during cell proliferation35, it is surprising that the frequency of a specific targeted mutation oscillates significantly from parental single-cell clones to daughter single-cell subclones. We speculated that Cas9-sgRNA might be stably expressed in clones isolated and continue to edit the WT target sites, thus altering the frequency of the targeted mutation. However, we did not detect SpCas9 proteins in the parental clones 1C3 and 6C7 and their subclones (Fig. S3A). It remains possible that a small amount of SpCas9 proteins stably synthesized in the cells could actively mutate their target sites, even if the protein level is hardly detectable by Western blot. We thus transfected these parental clones or subclones with the expression plasmids for two Col1a1 sgRNAs and one Rosa26 sgRNA with or without the SpCas9 expression plasmid and measured the editing frequencies at the target sites by PCR amplicon deep sequencing. While the editing is efficient with transfection of both SpCas9 and sgRNA, no editing was detected with sgRNA transfection alone or with neither SpCas9 nor sgRNA (Fig. S3B-C). This excludes the possibility that continuing CRISPR/Cas9 genome editing alters the types and frequencies of targeted TSG mutations from parental clones to subclones.
Continuing type and frequency oscillation of targeted mutations during clonal expansion
Next, we asked whether the types and frequencies of targeted mutations change over time during expansion of a subclone. We continuously cultured 4 subclones (i.e., 1C3-1, 1C3-2, 1C3-3 and 1C3-4) and analyzed target site mutations of each subclone on Day 0, Day 15 and Day 30 (Fig. 4A and Fig. S4). The frequencies of some targeted mutation changed over time during clonal expansion, e.g., from 2.47% at Day 0 to 35.09% at Day 15 and 34.09% at Day 30 for Rb1 Ins1T in 1C3-1 (Fig. S4 and Table S4). After storage in liquid nitrogen for a year, we thawed and continued to culture the 1C3-1 clone. We repeated analysis of target site mutations in the cultured cells on day 0 (i.e., Mon12), day 90 (i.e., Mon15) and day 180 (i.e., Mon18) (Fig. 4A). The frequencies of some targeted mutations continued to change over time during cell culturing of 1C3-1 (Fig. 4B and Table S4). For example, the Rb1 Ins1T mutation was negligible on Day 0 but appeared with the frequencies at over 30% on Day 15 and Day 30. Del1|0 mutation was significant at over 60% on Day 0 and decreased to about 30% on Day 15, Day 30 and Mon 12 and even 10% on Mon 15. On Mon 18, this mutation was slightly increased to over 20% (Fig. 4C). Both Lkb1 Del5|2 and Ins1G mutations remained frequent at over 40% from Day 0 to Mon 12, but decreased to about 20% on Mon 15 and increased again to 30% on Mon 18 (Fig. 4D). Arid1a Del11|16 and Del10|9 mutations changed in opposite direction on Mon 15 and Mon 18 (Fig. 4E). Similarly, Smad4 WT and Del16|0 mutations oscillated in opposite direction on Mon 15 and Mon 18 (Fig. 4F). This again confirms that some targeted mutations are not stable during proliferation of single tumor cells derived from CRISPR/Cas9-induced liver cancer in mice.
Type and frequency alteration of targeted mutations from single-cell clones to subcutaneous grafts
In order to evaluate this instability of targeted TSG mutations in vivo, we implanted the subclone 1C3-1 cells into 4 immunodeficient SCID mice subcutaneously (Fig. 5A). All mice grew a visible tumor in about 5 days. We harvested the tumor tissues from these 4 mice at the 14th day and established 2 single tumor cell clones from each tumor tissue (Fig. 5A). Analysis of targeted mutations revealed that tumor tissues harbored more frequent WT allele than both 1C3-1 and tumor cell clones derived from tumor tissues, likely due to the presence of normal cells in subcutaneous tumor samples (Table S5). Therefore, in order to exclude the interference of normal cells from tumor tissues on the frequency of targeted mutations in tumor cells, we compared the frequencies of targeted mutations only between parental 1C3-1 clone and tumor cell clones (Fig. 5B). The frequencies of some targeted mutations changed significantly among these tumor cells (Fig. 5B and Table S5). For instance, 1bp deletion (Del1|0) of Rb1 was dominant with the frequencies at over 50% in SG1-1, SG1-2, SG2-5 and SG2-6, but infrequent at less than 5% in SG4-1, SG4-3, SG5-1 and SG5-2 (Fig. 5C). Differently, the parental subclone 1C3-1 harbored this Rb1 Del1|0 mutation at 40% (Fig. 5C). Furthermore, while the Del5|2 mutation of Lkb1 changed little with the frequency at about 40% among all tumor cells tested, the Lkb1 Ins1G mutation oscillated significantly among these tumor cells, with the frequency at about 40% in SG1-1, SG1-2, SG2-5, SG2-6 as well as 1C3-1, but nearly undetectable in SG4-1, SG4-3, SG5-1 and SG5-2 (Fig. 5D). In addition, two Col1a1 mutations, i.e., Del1|0 and Ins1T, oscillated in an opposite direction (Fig. 5E). While the Col1a1 Del1|0 mutation dominant in 1C3-1, SG1-1, SG1-2, SG2-5 and SG2-6 was negligible in SG4-1, SG4-3, SG5-1 and SG5-2, the Col1a1 Ins1T mutation infrequent in 1C3-1, SG1-1, SG1-2, SG2-5 and SG2-6 occurred frequently in SG4-1, SG4-3, SG5-1 and SG5-2 (Fig. 5E). Similarly, the Del1|0 and Ins1A mutations of Rasa1 started with the frequencies at around 50% in parental 1C3-1 clone and then oscillated in a reverse pattern among single-cell clones from tumor grafts (Fig. 5F). These results indicate that targeted mutations in tumor cells derived from CRISPR/Cas9-induced liver cancer in mice are also unstable in subcutaneous grafts derived from tumor cells with the frequencies of some targeted mutations oscillating in vivo. Of note, the WT alleles of Rb1, LKb1, Col1a1 and Rasa1 not shown in Fig. 5B were all below 5% in frequency among the clones analyzed, but indicated in Fig. 5C-F as a negative comparison.
Increased genomic instability in CRISPR/Cas9-induced mouse liver cancer cells
As mentioned previously35, spontaneous mutations at a given site normally occurs at an extremely low rate and are impossible to cause significant alterations in the types and frequencies of targeted mutations detected in the study. Continuing targeted editing by Cas9-sgRNA was also excluded as a causal factor because neither stable expression nor the editing activity of Cas9-sgRNA was detected in these single-cell clones. As cancer with genomic instability has an increased tendency for constant genomic alteration, we wondered whether these primary liver tumor cells are associated with strong genomic instability. We first examined spontaneous γ-H2AX and 53BP1 foci formation in 10 primary tumor cell lines derived from CRISPR/Cas9-induced mouse liver cancer, NIH3T3 and the mouse liver cancer cell line Hepa1-6 (Fig. 6A). All of these liver cancer cell lines except 1C3-2 showed a higher level of spontaneous γH2AX focus formation than NIH3T3 (Fig. 6A-B), indicating strong induction of spontaneous DNA DSBs and activation of DNA damage response in these cancer cells. It was however unclear why 53BP1 focus formation was much less frequent than γH2AX focus formation in nearly all of these cell lines as both γH2AX and 53BP1 foci indicated the site of DSBs (Fig. 6A-B).
We also analyzed these CRISPR/Cas9-induced primary liver tumor cells for micronucleus formation, which is frequently involved in chromosomal aberrations and genomic instability in cancer36. Micronucleus formation was readily detected in these tumor cells (Fig. 6C). The percentages of micronucleated cells were significantly higher in tumor cells than in NIH3T3 cells (Fig. 6D). In particular, while about 1% of NIH3T3 cells contained micronuclei, over 10% of 1C3 or 1C3-1 cells were micronucleated (Fig. 6D). Metaphase spread analysis further identified significant chromosomal aberrations in primary liver tumor cells (Fig. 6E). The number of chromosomes in primary liver tumor cells varied in average from 57.4 in 1C3 to 105.7 in 6C7-2 and was much more than 40 in a diploid mouse cell (Fig. 6E-F). This indicates that these tumor cells are hyperploid, at least in part contributing to more than two allelic variations at a target site of TSGs in single-cell clones. Biarmed chromosomes appeared in all Hepa1-6 cells as reported previously37, but only existed in 2.0-6.0% of the primary liver tumor cell lines 1C3, all 1C3 subclones and 6C7-5 as well as in 8.9% of NIH-3T3 (Fig. 6E-F). In contrast, we did not detect any biarmed chromosomes in 6C7 or its subclones 6C7-2, 6C7-3 and 6C7-4. The difference in the generation of biarmed chromosomes between 6C7-5 and its parental clone 6C7 or other 6C7 subclones indicates potential variations in genomic instability in these cells (Fig. 6E-F). In addition, primary liver tumor cells exhibited more frequent chromosomal fragments than NIH-3T3 and Hepa1-6 (Fig. 6E-F). As frequent micronucleus formation, chromosomal aberrations and chromosomal fragments in primary tumor cells manifested the genomic instability of CRISPR/Cas9-induced mouse liver tumors, these findings indicated a possible connection between the genomic instability and the oscillation in the frequencies of targeted TSG mutations in tumor cells. Variations in genomic instability among parental clones (i.e., 1C3 and 6C7) and their respective subclones again implied that an intrinsic genetic force drive the oscillation of target site mutations in these single-cell clones.
Detectable eccDNA harboring targeted site mutations
Due to unstable nature of chromosomal fragments, the frequent presence of small chromosomal fragments observed in our study could drive significant alterations in the frequencies of targeted mutations in tumor cells derived from CRISPR/Cas9-induced mouse liver cancer. However, because it was difficult to separate chromosomal fragments from the regular chromosomes, we could not determine the contribution of chromosomal fragments to the oscillation in the frequencies of targeted mutations in tumor cells. Instead, like chromosomal fragments, eccDNA could be formed in CRISPR/Cas9-induced mouse liver cancer; this form of ecDNA could however be separated from genomic DNA (gDNA) and analyzed for targeted mutations by targeted PCR amplicon deep sequencing (Fig. S5A). Indeed, using eccDNA isolated from 1C3-1, 1C3-2, 1C3-3, 6C7-2 and 6C7-4 as the template for targeted PCR, we found that many of 35 target sites were poorly amplified as compared to gDNA and the numbers of deep sequencing reads for these sites were low, indicating the absence of eccDNA harboring these sites (Fig. S5B, and Table S6). Only 13 targets (i.e., Atm, Cdkn2a, Col1a1, Fbxw7, Kmt2c, Nf1, Notch3, p53, Pik3r1, Rb1, Rps6ka3, Setd5 and Tet2) consistently exhibited sufficient reads in eccDNAs (Fig. S5B and Fig. S6). Deep sequencing of targeted amplicons further confirmed that eccDNA in 1C3-1, 1C3-2, 1C3-3, 6C7-2 and 6C7-4 clones contain intact TSG target sites and targeted TSG mutations (Table S4 and Fig. S6).
To compare the distribution of targeted mutations across 34 target sites between eccDNA and gDNA, we calculated percentages of combined reads from all 34 target sites for reads of each target site respectively in both eccDNA and gDNA and derived relative read ratio of eccDNA to gDNA for each target site in 1C3-1 by dividing percentage values for eccDNA to respective ones for gDNA (Fig. S5C-D). Because PCR variations of eccDNA were normalized by PCR variations of gDNA in this relative read ratio of eccDNA to gDNA, the interference of PCR variations on sequencing reads in eccDNA across target sites were partly excluded. While the profiles of targeted mutations were different between eccDNA and gDNA as expected, it was surprising that some mutations undetectable or with a low frequency in gDNA increased significantly in frequency in eccDNAs (Fig. 7A, Fig. S5C-D and Fig. S6). For example, the frequency of Atm Del3|0 mutation was less than 2.35% in gDNA, but was elevated to 17.88% in eccDNA (Fig. 7B). Similarly, the frequency of the Rb1 Del22|5 mutation is 9.95% in gDNA, but increased to 28.53% in eccDNA (Fig. 7C). In addition, some targeted mutations detected at a high rate in gDNA were undetectable or much less frequent in eccDNA harboring these targets (Fig. 7A, Fig. S5D, Fig. S6 and Table S6). For instance, the frequency of the Rb1 Del1|0 mutation was 23.82% in gDNA, but was lowered to 7.53% in eccDNA (Fig. 7C).
To directly identify eccDNA harboring targeted mutations, we designed outward PCR primer pairs to determine circularization junctions of eccDNA carrying the respective targets whereas inward primer pairs were usually used to identify targeted mutations (Fig. S7A). Because 13 targets were frequently found in eccDNA isolated from 1C3-1, we followed these 13 sites to design outward PCR primer pairs and identify circularization junctions of eccDNA. Because the outward PCR primer pairs are positioned at one side to targeted sites, target sites with or without any variations could also be identified (Fig. S7A). Some of these 13 targets generated PCR bands with outward primer pairs, and PCR products with outward primer pairs were purified and cloned into a pUC19 vector for Sanger sequencing. We detected the circularization junctions of eccDNA harboring either intact target site or targeted mutations of Atm, Rb1 and Kmt2c (i.e., Atm, Rb1 and Kmt2c eccDNA) (Fig. 7D-F and Fig. S7B). Based on positions of circularization junctions and targeted mutations, we determined that the sizes of Atm, Rb1 and Kmt2c eccDNA were about 200 bp, 211 bp and 213 bp, respectively (Fig. 7D-F). Among 10 clones sequenced for Atm on eccDNA, 1 contained the WT sequence at the target site and Ins1G at the circularization junction, 1 contained Del3|0 at the target site and Del1T Ins1C at the circularization junction, and 8 contained Del8|0 Ins4CCCG at the target site and no deletion/insertion at the circularization junction (Fig. 7D). Among 4 clones sequenced for Rb1 on eccDNA, all had Del22|5 at the target site and were joined accurately at the circularization junction, whereas 1 clone from Kmt2c eccDNA had WT target site and no deletion/insertion at the circularization junction (Fig. 7E-F). Most of these variations such as Atm WT, Atm Del3|0, Rb1 Del22|5 and Kmt2c WT were also present in gDNA (Fig. 7B-C and Table S6). Taken together, these results demonstrated the existence of eccDNA harboring some targeted mutations of TSGs in CRISPR/Cas9-induced mouse liver cancer. As eccDNA is unstable in varying copies and could be degraded or integrated back to regular chromosomes38, eccDNA harboring targeted mutations may provide genetic materials to alter the frequencies of targeted mutations in the genome, driving ITH in CRISPR/Cas9-induced mouse liver cancer.