This article is mentioned in:

Introduction

Loss-of-function analysis is a crucial issue in reverse genetic studies. In the past decade, RNA interference (RNAi) has been widely used to knockdown gene expression (1,2). RNAi is based on the binding of small interfering (si)RNAs to target transcripts leading to either its degradation or its inhibition at the translational level. siRNAs are either transiently delivered to cells or stably produced from small hairpin (sh)RNAs that are transcribed by polymerase III promoters. Albeit both siRNAs and shRNAs allow specific silencing of target genes (3), knockdowns by RNAi are mostly incomplete, vary between experiments and have unpredictable off-target effects (4,5).

New technologies have recently become available to generate knockouts rather than knockdowns of target genes in cultured cells. Among those techniques are transcription activator-like effector nucleases (TALENs) that use a pair of artificial DNA-binding domains fused to the catalytic domain of restriction endonuclease FokI which causes a double-strand break (DSB) at the targeted genomic locus stimulating DNA repair (6,7). Yet, TALEN is labor-intensive and works with low efficiency (8).

A more recently discovered technique of genomic engineering is clustered regulatory interspaced short palindromic repeats (CRISPR)/CRISPR associated 9 (Cas9), which is a unique mechanism of bacteria and archea to protect themselves against foreign DNA penetration (9). This prokaryotic system has been adapted using a Cas9 endonuclease from Streptococcus pyrogenes that is guided to the target sequence by a guide RNA (gRNA) chimera that includes a protospacer adjacent motif. To reduce off-target effects, a mutant Cas9 termed nickase can be used which requires a pair of gRNAs to introduce site-specific single strand breaks, called nicks, that are together equivalent to a DSB (10). Of note, the use of two gRNAs and the nickase doubles the number of bases that need to be specifically recognized at the target locus and thereby significantly increases specificity.

DSBs introduced by TALEN or CRISPR/Cas9 at the targeted genomic locus are either repaired by the error prone non-homologous end joining (NHEJ) or by homology-directed repair (HDR). NHEJ leads to small insertions or deletions (InDels) that can result in a knockout of gene function due to frameshift mutations (11). The co-delivery of locus-specific homology arms with the site-specific nuclease triggers HDR-mediated genetic alterations and allows efficient integration of transgenes into an endogenous gene locus. First proof-of-principle studies showed that Cas9 can be successfully targeted to endogenous genes in bacteria (12), human pluripotent stem cells (13), as well as in whole organisms such as zebrafish (14), yeast (15), fruit flies (16), mice (17), rats (18) and rabbits (19). In addition, a haploid human cell line named engineered-HAPloid cells has been generated by megabase-scale deletion using CRISPR/Cas9 (20).

An important step in the use of genomic editing techniques is the confirmation of the knockout events. To analyze the targeted genomic locus, the target sequence is amplified by PCR, subcloned into a plasmid vector and subjected to sequencing (21). Another approach uses direct sequencing of the PCR products and analysis by ‘Tracking InDels by Decomposition’ (TIDE) which quantifies the editing efficacy and identifies predominant types of InDels in the targeted pool of cells (22). Other methods analyzing the efficiency of the Cas9-mediated DNA cleavage include heteroduplex formation that is examined either by high resolution melting analysis, heteroduplex mobility assay or T7 endonuclease I cutting. Using these methods, the ratio of homo- to heteroduplexes can be determined in order to estimate the nuclease efficiency. However, the latter method fails to accurately detect InDels (23).

Contrary to applications of CRISPR/Cas9 in haploid or diploid cells, genomic editing is more challenging when applied to hyperdiploid genomes as in the case of most cancer cells. In particular, all functional copies of the target gene must be edited in cancer cell lines to accomplish a complete knockout situation (24). As NHEJ works in a random fashion, there may arise altered structures without gene inactivation along NHEJ repair events. These insufficient knockout events, often combined with cellular heterogeneity, enhance the probability to generate partial knockouts that still harbor alleles coding for functional gene products or gene products with altered functionality (24). Hence, the determination of target gene copy number and cellular heterogeneity is essential in cancer cell populations to allow generation of solid CRISPR/Cas9-mediated knockouts and to correctly interpret the subsequent confirmation of knockout events.

The increase in aberrant ploidy levels and karyotypic complexity correlates with the progression of tumor cells from a benign neoplasm to malignant cancer. Chromosomal abnormalities occur in 75% of blood cancers and in more than 90% of solid tumors including hepatocellular carcinoma (HCC) (25,26). The overexpression of the receptor tyrosine kinase Axl along with the release of soluble Axl (sAxl) was detected in HCC and correlated with poor survival of HCC patients (27,28). In this study, we have generated knockouts of Axl in hepatoma cell lines using NHEJ-mediated CRISPR/Cas9 knockout technology. Our data show the dynamics of CRISPR/Cas9-mediated editing of the AXL locus in hyperdiploid hepatoma cell lines and reveal the drawbacks of the CRISPR/Cas9 technique in cancer cells.

Materials and methods

Cell culture

The human hepatoma cell lines SNU449 and HLF were cultured in RPMI-1640 supplemented with 10% fetal calf serum (FCS; Sigma, St. Louis, MO, USA) and DMEM plus 10% FCS, respectively. All cells were kept at 37°C and 5% CO2 and routinely screened for the absence of mycoplasma. All hepatoma cell lines were validated by short tandem repeat analysis.

CRISPR/Cas9-mediated genomic editing of the AXL locus

The AXL gene was disrupted in hepatoma cell lines using the human Axl gRNA CRISPR lentivirus set (K0161411, ABM, Milton, Ontario, Canada). HLF and SNU449 cells were infected with ready-to-use lentiviral stocks. Two or 6 µl of the gRNA pair and 2 or 6 µl of Cas9 nickase-encoding virus were added in a ratio of 1:2 to the cells. The medium was exchanged with standard culture medium after overnight incubation and cells were selected for stable expression of Cas9 for at least one week with 1 µg/ml puromycin. Single cell clones were generated by limiting dilution in 96-well plates and expanded to 100 mm culture plates. The single cell clones were further processed for sequence analysis of the AXL locus.

Sequence analysis of the genomic AXL locus

The gRNA binding region was amplified by polymerase chain reaction (PCR), cloned into a pGEM-T easy vector (A1360; Promega Corp., Madison, WI, USA) and transformed into E. coli DH5α competent bacteria for blue-white screening. The plasmid DNA of 20 bacterial colonies was sequenced by Sanger method and analyzed for sequence variations in the gRNA binding region. Single cell clones showing InDels in all 20 colonies were considered to be Axl-deficient (Axl−).

Western blot analysis

Immunoblotting was done as described previously (29). The antibodies used were anti-Axl, 1:1,000 (AF154; R&D Systems, Inc., Minneapolis, MN, USA) and anti-Actin, 1:2,500 (A2066; Sigma-Aldrich, St. Louis, MO, USA).

Enzyme-linked immunosorbent assay (ELISA)

Soluble Axl (sAxl) levels were assessed by ELISA using the DuoSet ELISA Development System Human Axl (DY154; R&D Systems, Inc.) from supernatants (SNs) of HLF and SNU449 cells. 1.5×106 HLF or 0.8×106 SNU449 cells were each seeded on 60 mm tissue culture plates. 24 h after plating, the cells were incubated in serum-free medium for further 24 h. Whole-cell lysates of the same cells were measured by ELISA using a protein concentration of 0.1 mg/ml. The samples of parental cells were measured in 1:10 and 1:20 dilutions. To ensure the detection of Axl in the knockout situation, the SNs and protein lysates of Axl− cells were measured undiluted and at a dilution of 1:5. The concentrations of sAxl and Axl were normalized to 1.0×106 cells. All values below the standard curve were considered as noise and were therefore excluded. A seven-point standard curve was generated for each plate and quantification was performed using GraphPad Prism 5.01 software (GraphPad Software, Inc., La Jolla, CA, USA).

Cell migration

Cells (1×106)were seeded in 6-well tissue culture plates. Cells were scratched with a sterile pipette tip to generate artificial wounds. Phase contrast images were taken using the microscope Nikon-Eclipse Ti-S (Nikon Corporation, Tokyo, Japan) to monitor wound closure. The second set of phase contrast images was taken after 24 h of migration. The area of migration into the artificial wound was analyzed using ImageJ 1.44p.

Array comparative genomic hybridization (aCGH)

Genomic DNA was isolated from SNU449 and HLF hepatoma cells including parental cells, Axl− single cell clones and a control clones expressing nickase without gRNAs using the QIAamp DNA Blood Mini kit (Qiagen, Hilden, Germany) according to the manufacturer's protocol. Direct and indirect aCGH analyses were performed using human whole genome oligonucleotide-based microarrays (SNU449: Cancer Research Array + single nucleotide polymorphism, 2×400K; HLF: 4×44K; both arrays from Agilent Technologies, Inc., Santa Clara, CA, USA). Normal human male DNA (Agilent Technologies, Inc.) or parental SNU449-derived DNA were used as reference samples in case of direct and indirect aCGH experiments, respectively. Labeling and hybridization was carried out following the protocols provided by the manufacturer. Slides were scanned with a G2600D Microarray Scanner (Agilent Technologies, Inc.). Feature extraction and data analysis were performed using the Feature Extraction and Agilent Genomic Workbench software (Agilent Technologies, Inc.), respectively.

Statistics

Data were expressed as means ± standard deviations. The statistical significance of differences was evaluated using a one-way analysis of variance followed by a Tukey's post hoc test. *P<0.05, **P<0.01 and ***P<0.005 were considered to indicate a statistically significant difference.

Results

Sequence analysis of the AXL locus in CRISPR/Cas9-edited hepatoma cells

To address the role of Axl signaling in HCC cells, we aimed at generating Axl-deficient HLF and SNU449 hepatoma cells. Both HLF and SNU449 cells were previously shown to exhibit a hyperdiploid DNA status (30,31). To accomplish this task, we performed CRISPR/Cas9-mediated genomic editing using a pair of gRNAs targeting the genomic AXL locus together with the expression of nickase. Genomic editing was induced in exon 7 or 8 depending on the pair of gRNAs used (Fig. 1A). The genomic editing events of two single cell clones each from HLF and SNU449 cells were analyzed after PCR amplification of the AXL locus by sequencing of 20 bacterial colonies each. In HLF cells, a gRNA pair was used that binds to DNA in Exon 7 (Fig. 1A and B). By analyzing the two HLF single cell clones, the first supposed Axl-negative clone, designated HLF-Axl−-1, showed one genomic editing event with one defined deletion (Fig. 1B). Sequencing of the second Axl− clone (HLF-Axl−-2) revealed two genomic editing events corresponding to deletions of various lengths (Fig. 1B). In SNU449 cells, the used gRNA pair bound to exon 8 of the AXL gene (Fig. 1A and B). Notably, one Axl− clone (SNU449-Axl−-1) showed five genomic editing events, where two events were deletions and the other three ones resulted in insertions (Fig. 1B). The second Axl− clone (SNU449-Axl−-2) exhibited six different genomic editing events consisting of two deletions, two insertions and two events resulting in a combination of InDels (Fig. 1B). From these data we conclude that multiple genomic editing events occurred at the AXL loci that were differentially affected by independent gRNA/Cas9 complexes.

Figure 1. Genomic editing by CRISPR/Cas9 in HLF and SNU449 cells. (A) Grey boxes represent exons of the AXL gene locus. The CDS is indicated. (B) The genomic editing events in two clones of HLF (left panel; Axl−-1, Axl−-2) and SNU449 cells (right panel; Axl−-1, Axl−-2) are shown. The gRNA binding sites are underlined. All genomic editing events in the single cell clones are listed below the original sequence. CDS, coding sequence.

Mutations of the AXL locus in CRISPR/Cas9-edited cells

As a next step we addressed the question of the consequences of these genomic editing events on Axl protein expression. All InDels that are not a multiple of three resulted in frameshift mutations, and as a consequence in premature stops of translation (Table I). However, the genomic editing event not resulting in a frameshift and premature stop was editing event 3 of SNU449-Axl−-1, where six basepairs were inserted into the genomic AXL locus, which resulted in an insertion of two amino acids in the extracellular domain of Axl (Fig. 1B, Table I). In summary, these data suggest that CRISPR/Cas9-mediated genomic editing may result in the expected protein deficiency (HLF-Axl−-1, HLF-Axl−-2 and SNU449-Axl−-2) or in the incomplete knockout of the Axl protein expression based on a single allele editing event (SNU449-Axl−-1).

Table I. Genomic editing by CRISPR/Cas9 in HLF and SNU449 cells.

Table I.

Genomic editing by CRISPR/Cas9 in HLF and SNU449 cells.

Cell type	Editing event	Insertion (bp)	Deletion (bp)	Effect on protein
HLF-Axl−-1	Editing 1		10	Frameshift-premature stop
HLF-Axl−-2	Editing 1		73	Frameshift-premature stop
	Editing 2		29	Frameshift-premature stop
SNU449-Axl−-1	Editing 1		14	Frameshift-premature stop
	Editing 2		24+5	Frameshift-premature stop
	Editing 3	6		2 AA insertion
	Editing 4	37		Frameshift-premature stop
	Editing 5	95		Frameshift-premature stop
SNU449-Axl−-2	Editing 1		19	Frameshift-premature stop
	Editing 2	7		Frameshift-premature stop
	Editing 3		7	Frameshift-premature stop
	Editing 4	77	2	Frameshift-premature stop
	Editing 5	49+3	1	Frameshift-premature stop
	Editing 6	176+9		Frameshift-premature stop

[i] AA, amino acid; Axl⁻, Axl knockout; bp, base pair; CRISPR, clustered regularly interspaced short palindromic repeat.

Expression of Axl in CRISPR/Cas9-edited cells

We next performed western blot analysis to detect the effects of the genomic editing events on Axl protein expression. Interestingly, no protein expression was observed in each of the HLF-Axl− and SNU449-Axl− single cell clones as compared to the HLF and SNU449 parental cells (p) or control HLF and SNU449 cells (n) expressing nickase without gRNAs (Fig. 2A). Furthermore, enzyme-linked immunosorbent assays (ELISAs) were performed to detect very low levels of Axl protein expression that could not be detected by the less sensitive western blot analysis. Thereby, we measured both the cleaved extracellular domain of Axl, termed soluble Axl (sAxl) in supernatants (Fig. 2B) as well as the full length Axl receptor in protein lysates (total Axl; Fig. 2C). Most notably, while no protein expression was observed in HLF-Axl−-1, Axl−-2 and SNU449-Axl−-2 cells, significant expression of sAxl and total Axl was detected in SNU449-Axl−-1 cells. Axl expression in this single cell clone confirmed the incomplete knockout (Fig. 1B, Table I), resulting in a 4-fold reduced expression to 24.7% of combined Axl values (sAxl + total Axl) as compared to parental SNU449-p cells (Table II). Noteworthy, all other CRISPR/Cas9-edited HLF and SNU449 single cell clones displayed no Axl expression at all (HLF-Axl−-2, and SNU449-Axl−-2) or showed very low Axl levels that were below the detection limit and thus considered as signal noise (HLF-Axl−-1; Table II). Together, these data demonstrate that the genomic editing is incomplete in multi-allelic hepatoma cell lines, as three of four clones show deficiency of Axl and one clone maintains Axl expression despite genomic editing.

Figure 2. Expression of Axl protein in CRISPR/Cas9-edited hepatoma cells. (A) Western blot analysis of parental HLF and SNU449 cells as well as those expressing nickase without gRNAs (n) or nickase plus a pair of gRNAs as described in Fig. 1 (Axl−-1, Axl−-2). Actin is shown as a loading control. (B) Release of sAxl into cell supernatants of the same cells shown in (A) as determined by ELISA. sAxl protein was normalized to cell numbers. (C) Expression of total Axl as analyzed by ELISA of whole cell lysates. Total Axl protein was normalized to cell numbers. Two biological replicates were analyzed in triplicates by ELISA. Data are expressed as means ± standard deviations. ***P<0.0005. P, parental.

Table II. Axl protein levels in CRISPR/Cas9 edited HLF and SNU449 cells.

Table II.

Axl protein levels in CRISPR/Cas9 edited HLF and SNU449 cells.

Cell type	sAxl (%)	Total Axl (%)	Combined Axl (%)
HLF-p	100	100	100
HLF-n	96.18	98.83	97.58
HLF-Axl−-1	0	1.93	1.02
HLF-Axl−-2	0	0	0
SNU449-p	100	100	100
SNU449-n	109.56	113.93	112.08
SNU449-Axl−-1	17.66	29.88	24.7
SNU449-Axl−-2	0	0	0

[i] Axl, tyrosine-protein kinase receptor UFO; Axl⁻, Axl knockout; n, nickase; p, parental; sAxl, soluble Axl; CRISPR, clustered regularly interspaced short palindromic repeat. Values given for HLF-n, HLF-Axl⁻-1, HLF-Axl⁻-2, SNU449-n, SNU449-Axl⁻-1 and SNU449-Axl⁻-2 cells are relative to parental cell lines.

Migration of CRISPR/Cas9-edited hepatoma cells

Our recent data demonstrated that the siRNA-mediated loss of Axl expression caused a decrease of cell motility of human hepatoma cells (27). Therefore, we next analyzed the effects of the genomic editing events on the migratory cell behavior by performing a wound healing assay. A significant difference in migration was observed between the parental cells (p) or cells expressing nickase without gRNAs (n) and the Axl− clones in both, HLF and SNU449 cells (Fig. 3A and B). HLF-Axl−-1 and HLF-Axl−-2 cells showed comparable levels of migration which is around 40% reduced when compared to HLF-n or HLF-p cells. Interestingly, migration of SNU449-Axl−-1 cells was reduced by approximately 20%, whereas SNU449-Axl−-2 cells displayed half of the migration level of SNU449-n or SNU-449-p cells. In conclusion, the CRISPR/Cas9-edited hepatoma cells show a clear phenotype of diminished migration, thus confirming recent data obtained with siRNA. In addition, the incompletely edited SNU449-Axl−-1 cells display an inhibition of migration with a lesser extent than those cells with a complete knockout (HLF-Axl−, SNU449-Axl−-2), indicating a close correlation between the genotype and the phenotype.

Figure 3. Migration of CRISPR/Cas9-edited hepatoma cells. (A) Migration of HLF parental (p), nickase (n), Axl−-1 and Axl−-2 cells. (B) Migration of SNU449 parental (p), nickase (n), Axl−-1 and Axl−-2 cells. Migration levels were normalized to migration of parental cells (100%). Data are expressed as means ± standard deviations. ***P-value <0.005.

Genomic changes in CRISPR/Cas9-edited hepatoma cells

We further addressed the question on the gain and loss of genomic DNA in parental SNU449-p, control SNU449-n cells and SNU449-Axl− single cell clones by employing direct (SNU449-p) and indirect (control SNU449-n, SNU449-Axl−-1, SNU449-Axl−-2) array comparative genomic hybridization (aCGH) analyses. As expected, SNU449-p cells showed vast changes in genomic DNA as compared to the diploid reference DNA by multiple gains and losses (Fig. 4A). In more detail, these alterations included on the one hand focal deletions/amplifications such as e.g. deletions at chromosome 9p21 (CDKN2A) and 16q21 (CDH8, CDH11) and amplifications at 17p11.2 (MAPK7, MFAP4). On the other hand, these genomic changes involved gains and losses of whole chromosomes/chromosome arms such as loss of chromosome 3, loss of chromosome 7q, and gain of chromosome 10. Importantly, SNU449-n cells showed multiple differences in genomic DNA alterations as compared to SNU449-p cells (Fig. 4B, red line), suggesting that the constitutive expression of nickase induced changes in the genomic DNA despite the absence of gRNAs. A very similar pattern of genomic alterations was observed in both SNU449-Axl−-1 and SNU449-Axl−-2 cells expressing gRNAs (Fig. 4B, turquoise and blue lines, respectively), suggesting that a multitude of genomic changes in control SNU449-n, SNU449-Axl−-1 and SNU449-Axl−-2 cells is primarily caused by nickase activity.

Figure 4. Changes in genomic DNA of SNU449 HCC cells subjected to CRISPR/Cas9-mediated editing of the AXL locus. aCGH analysis of parental SNU449 cells (SNU449-p) and those expressing either nickase without gRNAs (SNU449-n) or nickase plus gRNAs for the knockout of AXL (SNU449-Axl−-1 and SNU449-Axl−-2). (A) Gain and loss of genomic DNA in SNU449-p cells compared to diploid cells. (B) Comparison of genomic changes between control SNU449-n cells as well as SNU449-Axl−-1 and SNU449-Axl−-2 cells to SNU449-p cells. (C) Magnification of genomic changes in chromosome 19. The loss of the AXL locus in SNU449-Axl−-1 and SNU449-Axl−-2 cells at 19q is indicated by the green circle plus an arrow. The arrow in (A) and (B) indicates chromosome 19. Red line, SNU449-n; turquoise line, SNU449-Axl−-1; blue line, SNU449-Axl−-2 cells.

SNU449-p cells showed a gain at the AXL locus with a mean log2 ratio of 0.3 (data not shown). SNU449-n, SNU449-Axl−-1 and SNU449-Axl−-2 cells each displayed an additional low level gain of chromosome 19q at the AXL locus (Fig. 4C). However, both SNU449-Axl−-1 and SNU449-Axl−-2 cells displayed the expected deletion of the AXL locus as a decrease in AXL-specific genomic DNA could be detected (Fig. 4C), thus confirming CRISPR/Cas9-specific genomic editing events. In particular, five and four AXL-specific oligonucleotides indicated a loss of genomic DNA in SNU449-Axl−-1 and SNU449-Axl−-2 cells between nucleotide 20198 and 21255 as well as 21179 and 22539 of the AXL gene, respectively. Both regions encompassing the AXL-specific oligonucleotides were located downstream of the gRNA binding sites.

We performed additional aCGH analyses of parental HLF cells, one HLF clone derived from nickase-treated cells and two Axl knockout clones. Comparable to the results obtained in SNU449 cells, parental HLF cells showed a multitude of gains and losses which is characteristic for cancer cells, thus mirroring a highly aberrant and re-arranged genome (Fig. 5A). In these cells as well, the chromosomal region 19q13.2 containing the AXL locus showed a low level gain (about 1.2-fold) as compared to normal diploid cells (data not shown). Interestingly, both Axl knockout clones as well as the clone derived from the nickase treated cells displayed various new, additional changes as compared to parental HLF cells (Fig. 5B). Together, these data provide evidence that high resolution aCGH detects CRISPR/Cas9-mediated genomic editing of AXL. Furthermore, the expression of nickase alone rather than the CRISPR/Cas9-dependent editing of the AXL locus causes pleiotropic gene-dose changes in the genome of established HCC cell lines.

Figure 5. Changes in genomic DNA of HLF HCC cells subjected to CRISPR/Cas9-mediated editing of the AXL locus. aCGH analysis of parental HLF cells (HLF-p) and those expressing either nickase without gRNAs (HLF-n) or nickase plus gRNAs for the knockout of AXL (HLF-Axl−-1 and HLF-Axl−-2). (A) Gain and loss of genomic DNA in HLF-p cells compared to diploid cells. (B) Comparison of genomic changes between control HLF-n cells as well as HLF-Axl−-1 and HLF-Axl−-2 cells to HLF-p cells. The arrow in (A) and (B) indicates chromosome 19. Red line, HLF-n; turquoise line, HLF-Axl−-1; blue line, HLF-Axl−-2 cells.

Discussion

Despite the extraordinary achievements of the CRISPR/Cas9 technology in genome engineering, its application still poses a challenging task in hyperdiploid cancer cell models. The data obtained in this study strongly suggest that ploidy and gene copy number play an important role in the efficacy of the CRISPR-dependent genomic editing. Two recent studies demonstrated that CRISPR/Cas9 is not suitable in cellular cancer models showing gene amplifications, as overrepresented genomic regions cause false positive results. Aguirre et al (32) provided evidence that DNA breaks generated by the CRISPR/Cas9 system lead to gene-independent cell toxicity and the number of DNA cuts rather than the gene knockout predicts the cellular response. Munoz et al (33) found a higher number of lethal genes when using CRISPR/Cas9 as compared to RNAi. In aneuploid cancer cell models, the genes located in amplified regions scored as lethal, indicating false positive results as already described by Aguirre et al (32). Thus, the number of loci representing the gene of interest in a cancer cell line under investigation is of crucial importance for the physiological response to CRISPR/Cas9 and for the experimental interpretation.

Established cancer cell lines frequently represent a pool of genetically heterogeneous subclones with varying numbers of chromosomes and different physiological behavior. These progressive genomic alterations and aneuploidy might drive a high degree of genomic instability leading to new mutations and gene-dose alterations over time and thus give rise to new subclones (34). The predominance of such subclones can change depending on time and conditions of cultivation. Upon selection of CRISPR/Cas9-positive cells, expansion of single cells harbor the risk of developing a phenotype not specific for the knockout as cells go through multiple cell divisions in which compensatory mechanisms could arise due to the loss of the target protein. Even without any genetic manipulation, single cell clones can behave differently from the original cancer cell population and might not be representative for a certain cancer cell model. Thus, the use of CRISPR/Cas9 might have drawbacks in cancer cell lines as the reliable evaluation of experiments is less predictable in the absence of data about target gene copies and the dynamic modulation of subclone heterogeneity also under the stress of clonal selection.

Notably, the expression of nickase in the absence of gRNAs in SNU449 and HLF cells causes multiple genomic changes comparable to those in CRISPR/Cas9-processed cells in the presence of gRNAs (Figs. 4B and 5B). These data show that gRNA expression results in specific editing of the targeted AXL locus and further suggest that cells expressing nickase without gRNAs represent the most suitable control for CRISPR/Cas9-edited cells. In this context, however, it is an open issue whether the genomic changes observed in nickase-only expressing cells are due to the pleiotropic activity of nickase or due to the selection of a favorable subclone from heterogenous SNU449-p or HLF-p cells. Thus, further aCGH profiling of nickase-only expressing cells vs. representative subclones of parental hepatoma cells could clarify the impact of clonal selection vs. nickase activity and the evolution of genomic alterations as compared to the parental cell pool.

In order to reduce unspecific genomic changes by the nickase, CRISPR-Cas9 inhibitory proteins could be expressed to inhibit the activity of Cas9 after genomic editing (35,36). The prophage-encoded inhibitor proteins AcrIIA2 and AcrIIA4, which allow phages to evade the bacterial host's CRISPR/Cas immune system, were found to inhibit Cas9-based targeting in their native host Listeria monocytogenes, as well as Cas9 of Streptococcus pyogenes in bacteria and human cells. Another approach for minimizing off-target effects employs the delivery of Cas9 protein/gRNA ribonucleoprotein complexes (37). When delivering Cas9 protein directly, cleavage occurs only temporarily, as the Cas9 protein is rapidly degraded in cells (38).

The confirmation of the desired gene knockout requires the determination of target protein levels in a large number of single cell clones using immunoblotting and the more sensitive ELISA. Interestingly, this study showed different levels of Axl protein in SNU449-Axl−-1 cells when detected either by immunoblotting or ELISA (Fig. 2), albeit the same polyclonal Axl-specific antibody was employed in both methods. We speculate that the insertion of two additional amino acids into the AXL locus of CRISPR/Cas9-edited SNU449-Axl−-1 cells (Fig. 1B, Table I) accounts for changes in the protein structure or the stability of the Axl protein allowing detection in its native form by ELISA but no detection of the reduced form by immunoblotting. Obviously one incompletely edited copy of AXL was enough to produce considerable amount of protein that is detectable by ELISA. In accordance with these data, both SNU449-Axl− clones exhibited a significant reduction of the migratory phenotype, which was less pronounced in incompletely edited SNU449-Axl−-1 cells (Fig. 3).

Furthermore, 20 bacterial colonies of the AXL-specific PCR-amplified region were each analyzed by sequencing in order to verify the complete knockout. As 20 analyses might not cover all genomic changes, we suggest applying next generation sequencing (NGS) methods to identify all InDels in the targeted gene locus. NGS analysis could be even employed at various passage numbers during selection in order to examine the dynamics of subclonal evolution. In case of many alleles and an incomplete knockout, one intact and functional target gene copy, even if not detected in the beginning, could provide a particular growth advantage in culture over time.

Loss-of-function studies employing CRISPR/Cas9, TALEN or stable RNAi involve selection of edited cells through multiple cell divisions in which compensatory mechanisms could arise due to the loss/knockdown of the target protein, resulting in altered cellular phenotypes. Notably, transient RNAi by siRNA transfection could overcome such drawbacks as cells are not subjected to selection and might not adapt to environmental conditions. Thus, siRNA-induced changes are more likely due to the downregulation of the target protein rather than to an artifact derived from clonal selection. Since transient RNAi is only used for short-term analysis, we suggest to compare e.g. the CRISPR/Cas9 phenotype with the one obtained by transient RNAi, both of which must coincide in phenotypical characteristics in vitro.

In conclusion, the CRISPR/Cas9 system has the advantage of generating a complete and stable knockout which can be successfully used in cancer cell lines by considering the above mentioned aspects. Once the knockout is confirmed and selection artefacts can be excluded, a robust and trustworthy system has been established.

Acknowledgements

The present study was supported by the Austrian Science Fund, FWF, P25356 and the Herzfelder Family Foundation.

Glossary

Abbreviations

Abbreviations:

References