Genome editing inside living organisms using a small Cas9 protein from Campylobacter jejuni

First, we determined the PAM sequences recognized by CjCas9 in vitro (Fig. 1b). A PCR amplicon library containing a CjCas9 target sequence followed by randomized 10-base pair (bp) sequences was cleaved by CjCas9 (and SpCas9 as a control) and a sgRNA specific to the target sequence. Cleaved duplexes were subjected to deep sequencing to identify PAM sequences recognized by CjCas9. This assay revealed that CjCas9 recognized 5′-NNNNACAC-3′ or 5′-NNNNRYAC-3′ (where R and Y stands for purines and pyrimidines, respectively) as PAMs in vitro. Using a different in vitro assay, Fonfara et al.18 reported that the optimal PAM for CjCas9 was 5′-NNNNACA-3′.

To confirm the PAM specificity in human cells, we performed cell-based reporter assays by co-transfecting plasmids encoding CjCas9 and its sgRNA and reporter plasmids containing the target site with variable PAM sequences between RFP and GFP sequences19 (Fig. 1c). Because the RFP sequence is fused to the GFP sequence out of frame in these reporter plasmids, cells express RFP but not GFP (RFP+ GFP−) in the absence of CjCas9. When CjCas9 cleaves the target site and induces small insertions or deletions (indels), the GFP sequence can be fused to the RFP sequence in frame. These reporter assays showed that CjCas9 cleaved target sites containing 5′-NNNNRYAC-3′ PAM sequences in HEK 293 cells. Importantly, the cytosine nucleotide at the 3′-end was essential.

We next optimized the sgRNA length for CjCas9-mediated genome editing in human and mouse cells (Fig. 2; Supplementary Fig. 2; Supplementary Table 1). We co-transfected CjCas9 and a series of sgRNAs with variable lengths into HEK 293 cells or NIH 3T3 cells and measured indel frequencies using targeted amplicon sequencing. The first nucleotide in the sgRNAs was fixed to an extra guanine (G) because sgRNAs are transcribed under the control of the U6 promoter, which requires a guanine at the 5′ end. Unlike SpCas9, which is most active with GX₂₀ sgRNAs that hybridize with a 20-nucleotide target DNA sequence upstream of a PAM, CjCas9 was most active with GX₂₂ sgRNAs that hybridize with a 22 nucleotides target sequence. GX₂₀ and GX₁₉ sgRNAs failed to induce indels at 8 out of 19 sites and 3 out of 4 sites, respectively in human or mouse cells.

We tested GX₂₂ sgRNAs at 12 sites containing the optimal 5′-NNNNACAC-3′ PAM in human cells. All of these sgRNAs co-transfected with CjCas9 were able to induce indels at frequencies that ranged from 1.0 to 64% (21±5%, on average). CjCas9 also induced indels at sites with 5′-NNNNGCAC-3′ PAMs, 5′-NNNNGTAC-3′ PAMs, and 5′-NNNNATAC-3′ PAMs albeit less efficiently (10±3%, 10±4%, and 16±5%, respectively).

We then determined genome-wide specificities of CjCas9 using nuclease-digested whole-genome sequencing (WGS; Digenome-seq)20,21,22. Cell-free genomic DNA digested with CjCas9 in vitro was subjected to WGS. Uniform cleavage patterns corresponding to on-target and off-target cleavage sites were computationally identified. CjCas9 designed to target 6 different DNA regions cleaved between one and 27 sites (7±4 sites, on average) in the human or mouse genome (Fig. 3a–d). In parallel, we used Digenome-seq to test three SpCas9 nucleases designed to cleave sites that overlapped with CjCas9 target sites. SpCas9 cleaved 15 to 147 sites (70±40 sites, on average) in the human or mouse genome, in line with our previous results showing that SpCas9 targeted to 11 different sites cleaved 90±30 sites in the human genome21. One target site in the human genome contained a 5′-NGGNACAC-3′ PAM recognized by both CjCas9 and SpCas9. CjCas9 and SpCas9 cleaved human genomic DNA at 5 sites and 45 sites, respectively, although the two Cas9 orthologues showed comparable indel frequencies at this particular site. Thus, the higher specificity of CjCas9 was not gained at the expense of its lower editing efficiency. Strikingly, two CjCas9 nucleases targeting the Rosa26 locus or the Vegfa gene in the mouse genome cleaved genomic DNA only at the single on-target site, reminiscent of the remarkable specificity of Cpf1 nucleases22,23. Sequence logos obtained computationally by comparing in vitro cleavage sites with each other unambiguously showed that these sites contained 5′-NNNNACAC-3′ or 5′-NNNNRYAC-3′ PAMs.

We chose the most promiscuous CjCas9 nuclease, which cleaved genomic DNA at 27 sites, and performed targeted amplicon sequencing to measure indel frequencies in human cells (Supplementary Table 2). Indels occurred at the on-target site but were not detectably induced at the other in vitro cleavage sites. Taken together, these results show that CjCas9 nucleases are highly specific in human cells. The remarkable specificity of CjCas9 can be at least partially attributed to its extended, 22 nucleotides target and 4 nucleotides PAM sequences, compared to the 20 nucleotides target and 2 nucleotides PAM sequences recognized by SpCas9.

We compared genome editing efficiencies and specificities of CjCas9 with those of SaCas9. We chose 3 overlapping sites in the human genome with 5′-NNGRRTAC-3′ PAM sequences, which can be targeted by both CjCas9 and SaCas9. CjCas9 and SaCas9 generated indels at these sites in HEK 293 cells with comparable frequencies of 35±7% and 36±13%, respectively (Fig. 3e). We then determined genome-wide specificities of CjCas9 and SaCas9 nucleases targeted to two of these overlapping sites using Digenome-seq (Fig. 3f). CjCas9 and SaCas9 targeted to the AAVS1-TS16 site cleaved human genomic DNA at 59 and 118 sites, respectively. CjCas9 and SaCas9 targeted to the AAVS1-TS39 site were more specific than were those targeted to the TS16 site, cleaving human genomic DNA at 2 and 15 sites, respectively. Note that CjCas9 targeted to the TS16 site was more efficient than SaCas9 in terms of on-target indel frequency (46% versus 28%), suggesting that the higher specificity of CjCas9 was not gained at the expense of a lower on-target efficiency. These results demonstrate that CjCas9 is as efficient as SaCas9 but is more specific than SaCas9 at least at the sites we tested in human cells.

With its small size, the CjCas9 gene plus a sgRNA sequence can be packaged into an all-in-one AAV vector (Fig. 4a). CjCas9 directed to the Rosa26 locus was expressed in C2C12 mouse myotubes using an AAV serotype DJ (AAVDJ) vector. Indels accumulated at the target site in a time- and dose-dependent manner with a frequency of up to 79±7% (Fig. 4b). Because no sites other than the on-target site were cleaved by this particular CjCas9 nuclease in the mouse genome, as shown above using Digenome-seq, we identified, using Cas-OFFinder24, potential off-target sites that differed from the on-target site by up to 4 nucleotides in the genome. No indels were detected at the resulting 20 homologous sites by targeted deep sequencing even at day 14 post-infection (Supplementary Fig. 3; Supplementary Table 3), confirming the high specificity of this CjCas9 nuclease.

We next packaged the muscle-specific Spc512 promoter-driven CjCas9 and the U6 promoter-driven Rosa26-specific sgRNA into a muscle-tropic AAV serotype 9 (AAV9) vector25 (Fig. 4c). The resulting virus was administered via intramuscular injection into tibialis anterior (TA) muscles of C57BL6J mice (n=3 per group). CjCas9-induced indels were observed at the target site in TA muscles with a frequency of 17±1% and 13±2%, 8 weeks and 32 weeks, respectively, after injection. No indels were detectably induced at the 20 possible off-target sites in TA muscles even 32 weeks after injection (Fig. 4d and Supplementary Table 3), although CjCas9 was still expressed (Supplementary Fig. 4). This result suggests that a long-term expression of CjCas9 via AAV in vivo does not necessarily aggravate off-target effects.

To show that CjCas9 can be expressed via AAV in other tissues such as retina in mice and to investigate the therapeutic potential of CjCas9-mediated gene surgery for the treatment of age-related macular degeneration (AMD), a leading cause of blindness in adults, we prepared an AAV9 vector encoding CjCas9 under the control of the elongation factor-1 short (EFS) promoter, enhanced green fluorescent protein (eGFP) linked to the C terminus of CjCas9 with the self-cleaving T2A peptide, and a U6 promoter-driven sgRNA specific to the Vegfa or Hif1a gene, whose expression in the retina is associated with choroidal neovascularization (CNV)26,27 (Fig. 5a,b). We monitored the expression of CjCas9 in the eye and measured indel frequencies using targeted deep sequencing and VEGFA protein levels using ELISA, 6 weeks after the resulting viruses were administered into the eye via intravitreal injection.

In retinal pigment epithelium (RPE) cells, primary target cells for the treatment of AMD, AAV encoding the Vegfa-specific CjCas9 (AAV-CjCas9: Vegfa) achieved indels with frequencies that ranged from 22 to 30% at day 14, 28, and 42 post injection (Fig. 5c). As expected, CjCas9-linked eGFP was expressed in RPE cells (Fig. 5d and Supplementary Fig. 5). At day 42 post injection, CjCas9-induced indels were observed at Rosa26, Vegfa, and Hif1a target sites in the retina with a frequency of 44±18%, 20±5, 58±12, respectively (Fig. 5e) and in RPE cells with a frequency of 14±5%, 22±3%, and 31±2%, respectively (Fig. 5f). As expected, the VEGFA protein level measured using ELISA was decreased in the retina treated with AAV encoding the Vegfa- or Hif1a-specific CjCas9 (AAV-CjCas9: Vegfa and Hif1a) but not in those treated with AAV-CjCas9: Rosa26 (Fig. 5g). The VEGFA protein level was also decreased in RPE cells treated with the AAV-CjCas9: Vegfa. However, the protein level was not decreased in those with the AAV-CjCas9: Hif1a or Rosa26 (Fig. 5h), suggesting that VEGFA expression is differentially regulated in the retina and in RPE cells. We did not measure HIF-1α protein levels because HIF-1α is degraded under normoxia conditions. No off-target indels were detectably induced in these cells at one or two in vitro cleavage sites captured by Digenome-seq using the Vegfa-specific or the Hif1a specific sgRNA, respectively (Fig. 5i), ruling out the possibility that the partial suppression of VEGFA expression in the retina and RPE cells were caused by CjCas9 off-target effects.

Next, we induced CNV in the eye by laser treatment 6 weeks after injection of AAV and measured the area of CNV 1 week later (Fig. 6a). Both AAV-CjCas9: Vegfa and AAV-CjCas9: Hif1a reduced the area of CNV by 24±4% and 20±4%, respectively, compared to the AAV-uninjected negative control (Fig. 6b–d). The Rosa26-specific CjCas9, used as another negative control, did not show any therapeutic effect. We noted that CjCas9 linked eGFP was expressed in RPE cells surrounding CNV, which are a major source of VEGFA in laser-induced CNV (Fig. 6b). This result shows that the reduction of CNV area coincides with targeted Vegfa mutagenesis and partial suppression of VEGFA expression in RPE cells (Fig. 5h).

We then investigated whether the partial gene knockout of Vegfa and Hif1a in RPE cells using AAV caused any side effect. It was reported that a conditional knockout of the Vegfa gene but not that of the Hif1a gene in mouse RPE cells leads to cone dysfunction28. We measured cone function using full-field electroretinography (ERG) in laser-untreated CNV-free mice 8 weeks after injection of AAV-CjCas9: Vegfa and AAV-CjCas9: Hif1a (Fig. 6e,f). No significant decrease in photopic response (Fig. 6e) or 30 Hz flicker response (Fig. 6f) was observed in these mice, compared to AAV-uninjected control mice. We also measured the size of the opsin-positive area, which is closely related to cone function, in contact with RPE cells expressing CjCas9. The AAV-CjCas9: Vegfa reduced the size by 30±10%, compared with the AAV-uninjected control (Fig. 6g,h; Supplementary Fig. 6), suggesting that the partial Vegfa gene knockout using AAV can still cause local opsin dysfunction near Vegfa-edited RPE cells. As expected, however, the AAV-CjCas9: Rosa26 or AAV-CjCas9: Hif1a did not cause any such cone dysfunction. Taken together, these results raise a concern about targeted inactivation of Vegfa in RPE cells using AAV and suggest that Hif1a could be inactivated without causing cone dysfunction to avoid neovascularization for the treatment of AMD.

The care, use, and treatment of all animals in this study were in strict agreement with the ARVO statement for the Use of Animals in Ophthalmic and Vision Research and College of Veterinary Medicine and the guidelines established by the Seoul National University Institutional Animal Care and Use Committee, which granted permission to perform animal experiments. Eight-week-old, male, specific pathogen-free C57BL/6J mice (n=3–9) were used in this study. Mice were maintained under a 12 h dark–light cycle.

After mice were anaesthetized, pupils were dilated with an eye drop containing phenylephrine (0.5%) and tropicamide (0.5%). Laser photocoagulation was performed using an indirect head set delivery system (Iridex) and laser system (Ilooda). Laser parameters were 810 nm wave length, 200 μm spot size, 800 mW power and 70 ms exposure time. Laser burn was induced three to four times around the optic disc. Only burns that produced a bubble without vitreous haemorrhage were included in the study. Seven days later, the eyes were fixed in 4% paraformaldehyde for 1 h at room temperature. RPE complexes (RPE/choroid/sclera) were prepared for immunostaining and then incubated with isolectin-B4 (Thermo Fisher Scientific, cat. no. I21413↗, 1:100) and anti-GFP antibody (Abcam, ab6556, 1:100) overnight at 4 °C. The RPE complex was flat-mounted and viewed with a fluorescent microscope (Eclipse 90i, Nikon) or a confocal microscope (LSM 710, Carl Zeiss) at a magnification of × 100. The CNV area was measured using Image J software (1.47v, NIH) by blinded observers. An average of 3–4 CNV areas per eye were analysed. Each group consisted of 17–18 eyes.

A human codon-optimized CjCas9-coding sequence, derived from Campylobacter jejuni subsp. Jejuni NCTC 11168, was synthesized with a nuclear localization signal and an HA epitope at its C-terminal end (GeneArt Gene Synthesis, Thermo Fisher Scientific) and cloned into the p3s plasmid4. The trans-activating crRNA (tracrRNA) sequence and the precursor CRISPR RNA (pre-crRNA) sequence were fused with a GAAA or TGAA linker to form a sgRNA sequence (Supplementary Fig. 1). sgRNAs were transcribed under the control of the U6 promoter. We used addgene plasmid (# 61591) for SaCas9 expression.

The recombinant Cas9 protein was expressed in E. coli and purified as described previously3. sgRNAs were transcribed using T7 RNA polymerase as described3. To make a randomized PAM (N₁₀) library, the DNA sequence including the AAVS1-TS1 target site cloned in a plasmid was amplified with a randomized primer. After gel purification, the amplicon library (1 μg) was digested with the SpCas9 or CjCas9 protein and in vitro transcribed sgRNA for 30 min at 37 °C. Digested library was purified by column filtration and subjected to deep sequencing using Miseq (Illumina). Miseq reads that perfectly matched the reference sequence were sorted. The randomized PAM region was extracted and analysed with WebLogo.

The AAVS1-TS1 target sites with variable PAM sequences, which were randomized at position X (5′-NNNNXCAC-3′, 5′-NNNNAXAC-3′, 5′-NNNNACXC-3′ and 5′-NNNNACAX-3′), were synthesized (Macrogen, Inc.) and cloned in a surrogate reporter plasmid encoding RFP and GFP19. To determine optimal PAM sequences, each of the resulting reporter plasmids (100 ng) and plasmids encoding CjCas9 and its sgRNA (225 and 675 ng, respectively) were co-transfected into HEK293 cells (1 × 10⁵) using lipofectamine 2000 (Invitrogen). At day 2 post-transfection, the fraction of GFP and RFP double-positive cells was determined by flow cytometry (BD Accuri C6, BD).

HEK 293 (ATCC, CRL-1573) cells and mouse NIH 3T3 (ATCC, CRL-1658) cells were maintained in DMEM supplemented with 100 units per ml penicillin, 100 mg ml⁻¹ streptomycin, and 10% fetal bovine serum (FBS). sgRNA plasmid (750 ng) and CjCas9 plasmid (250 ng) were transfected into cells (0.5∼1 × 10⁵) with lipofectamine 2000 (Invitrogen). After 48 h of transfection, genomic DNA was isolated using a DNeasy Blood & Tissue kit (Qiagen) and on-target or off-target loci were amplified using specific primers (Supplementary Table 4) for targeted deep sequencing. Deep-sequencing libraries were generated by PCR. TruSeq HT Dual Index primers were used to label each sample. Pooled libraries were subjected to paired-end sequencing (LAS, Inc.). Indel frequencies were calculated as described previously20.

Digenome-seq was performed as described previously20,21. Genomic DNA was isolated using a DNeasy Tissue kit (Qiagen) according to the manufacturer's instructions. Genomic DNA (8 μg) with CjCas9 or SaCas9 protein (300 nM) and sgRNA (900 nM) in a 400 μl reaction volume (100 mM NaCl, 50 mM Tris-HCl, 10 mM MgCl₂, and 100 μg ml⁻¹ BSA) and incubated the mixture for 8 h at 37 °C. Digested genomic DNA was incubated with RNase A (50 μg ml⁻¹) for 30 min at 37 °C and purified again with a DNeasy Tissue kit (Qiagen). Digested DNA was fragmented using the Covaris system and ligated with adaptors for library formation. DNA libraries were subjected to WGS using an Illumina HiSeq X Ten Sequencer at Macrogen. We used the Isaac aligner to generate a Bam file using the following parameters: ver. 01.14.03.12; Human genome reference, hg19 from UCSC (original GRCh37 from NCBI, Feb. 2009), Mouse genome reference, mm10 from UCSC; Base quality cutoff, 15; Keep duplicate reads, yes; Variable read length support, yes; Realign gaps, no; and Adaptor clipping, yes (adaptor: 5′-AGATCGGAAGAGC*-3′, 5′-*GCTCTTCCGATCT-3′)33.

AAV inverted terminal repeat-based vector plasmids carrying a sgRNA sequence and the CjCas9 gene with a nuclear localization signal and an HA tag at the C terminus were constructed. sgRNA transcription was driven by the U6 promoter and CjCas9 expression was controlled by the EFS promoter in C2C12 myoblast cells or by the Spc512 promoter in TA muscles of C57BL6J mice mice. For retinal delivery, an AAV vector encoding CjCas9 under the control of the EFS promoter, enhanced green fluorescent protein (eGFP) linked to the C terminus of CjCas9 with the self-cleaving T2A peptide, and a U6 promoter-driven sgRNA specific to the Vegfa or Hif1a gene was constructed.

To produce AAV vectors, they were pseudotyped in AAVDJ or AAV9 capsids. HEK293T cells were transfected with pAAV-ITR-CjCas9-sgRNA, pAAVED2/9 and helper plasmid. HEK293T cells were cultured in DMEM with 2% FBS. Recombinant pseudotyped AAV vector stocks were generated using PEI coprecipitation with PEIpro (Polyplus-transfection) and triple-transfection with plasmids at a molar ratio of 1:1:1 in HEK293T cells. After 72 h of incubation, cells were lysed and particles were purified by iodixanol (Sigma-Aldrich) step-gradient ultracentrifugation. The number of vector genomes was determined by quantitative PCR.

Mouse myoblast cells were infected with AAVDJ-CjCas9 at different viral doses (multiplicity of infection (MOI): 1, 5, 10, 50, and 100 determined by quantitative PCR) and maintained in DMEM with 2% FBS. At different time points, cells were collected for targeted deep sequencing. An MOI of 1 was estimated with one infectious virus particle in 100 total viral particles determined by quantitative PCR.

AAV was administered to 8-week-old young adult male C57BL/6J mice anaesthetized with 2–4% isoflurane. The mice were injected intramuscularly with AAV9-CjCas9 (1 × 10¹¹ viral genome) in physiological saline (40 μl) using an ultra-fine insulin syringe with a 31 G needle (BD). As a negative control, C57BL/6J mice were injected with physiological saline (40 μl) only.

8-week-old mice were anaesthetized with an intraperitoneal injection of a mixture of tiletamine and zolazepam (1:1, 2.25 mg per kg body weight) and xylazine hydrochloride (0.7 mg per kg body weight). AAV9-CjCas9 (2 × 10¹⁰ viral genome in 2 μl) was intravitreally injected using a Nanofil syringe with a 33 G blunt needle (World Precision Instruments Inc.) under an operating microscope (Leica Microsystems Ltd.).

For the analysis of opsin-positive area, formalin-fixed paraffin-embedded samples were prepared at day 42 post injection (n=4). Cross-section samples were immunostained with anti-HA antibody (Roche, 3F10, 1:1,000), anti-opsin antibody (Millipore, AB5405, 1:1,000), and Alexa Fluor 488 or 594 antibodies (Thermo Fisher Scientific, 1:500). The opsin-positive area corresponding to RPE cells expressing HA-tagged CjCas9 was measured using Image J software (1.47v, NIH) by blinded observers. For the distribution of CjCas9 and eGFP, the eyes were fixed in 4% paraformaldehyde for 1 h at room temperature. RPE complexes (RPE/choroid/sclera) were prepared for immunostaining and then incubated with anti-GFP antibody (Abcam, ab6556, 1:100) overnight at 4 °C. After stain with Alexa Fluor 488 antibodies (1:500), the RPE flat-mounts was imaged using a confocal microscope (LSM 710, Carl Zeiss). The scanning parameters were as follows: scaling (x=0.042 μm per pixel, y=0.042 μm per pixel, z=0.603 μm per pixel), dimensions (x=1,024, y=1,024, channels: 2, 8-bit) with objective C-Apochromat × 40 per 1.20 W Korr M27. ZEN 2 software was used to process the images.

For DNA extraction from muscle, muscle tissue was homogenized using tungsten carbide beads (3 mm; Qiagen) and a TissueLyser II (Qiagen). For extraction from RPE, after imaging of RPE flat-mounts, tissue samples were washed in PBS. RPE cells were mechanically isolated from choroid/sclera by vortexing for 30 s in lysis buffer (NucleoSpin Tissue, Macherey-Nagel), as described34. Genomic DNA from the remnant choroid/sclera tissues was analysed to confirm complete isolation of RPE cells. Genomic DNA was analysed by targeted deep sequencing.

At day 42 post injection, whole RPE complexes were separated from neural retina tissue and frozen for further analysis. Sample tissues were lysed with Cell Lysis Buffer (120 μl) (CST #9803) and Vegfa protein levels were measured using a mouse VEGF Quantikine ELISA kit (MMV00, R&D systems) according to the manufacturer's instructions.

Mice were dark-adapted over 16 h. Mice were anaesthetized with an intraperitoneal injection of a mixture of tiletamine and zolazepam (1:1, 2.25 mg per kg body weight) and xylazine hydrochloride (0.7 mg per kg body weight). Pupils were dilated with an eye drop containing phenylephrine (0.5%) and tropicamide (0.5%). Contact lens electrodes were placed on both eyes with a drop of methylcellose. Full-field ERGs were recorded as described35 by using the universal testing and electrophysiologic system 2000 (UTAS E-2000, LKC Technologies, Gaithersburg, MD). The responses were recorded at a gain of 2 k using a notch filter at 60 Hz, and were bandpass filtered between 0.1 and 1,500 Hz. In the light-adapted state (photopic), with a 30 cd/m² background light to desensitize the rods and isolate cones, cone responses were recorded in response to single flashed of 0 dB for photopic response, and a flicker sequence of 30 Hz, averaging 20 responses. The amplitude of the a-wave was measured from the baseline to the lowest negative-going voltage, whereas peak b-wave amplitudes were measured from the trough of the a-wave to the highest peak of the positive b-wave.

The CjCas9 protein expressed in TA muscles of C57BL/6J mice at 8 months after injection of AAV was detected using western blotting. Samples containing equal amounts of protein (20 μg) were analysed; Cas9 and GAPDH were detected with an anti-HA high affinity antibody (Abcam, ab9110, 1:5,000) and an anti-GAPDH antibody (Abcam, ab9485, 1:2,500), respectively. Goat anti-rabbit IgG-HRP antibody (Abcam, ab6721, 1:5,000) was used for signal detection. ImageQuant LAS4000 (GE healthcare) was used for digital imaging. Uncropped scans of blots in main figures are presented in. Supplementary Fig. 4

No statistical methods were used to predetermine sample size for in vitro or in vivo experiments. All group results are expressed as mean±s.e.m., if not stated otherwise. Comparisons between groups were made using the two-tailed Student's t-test or one-way analysis of variance (ANOVA) and Tukey's post hoc tests for multiple groups. Statistical significance as compared with untreated controls was denoted with *P<0.05, **P<0.01, ***P<0.001 in the figures and figure legends. Statistical analysis was performed in Graph Pad PRISM 5.

The deep-sequencing data from this study have been submitted to the NCBI Sequence Read Archive under accession number SRP095501 and SRP095507. The data that support the findings of this study are available from the corresponding author upon reasonable request.

The deep-sequencing data from this study have been submitted to the NCBI Sequence Read Archive under accession number SRP095501 and SRP095507. The data that support the findings of this study are available from the corresponding author upon reasonable request.