CRISPRi Gene Knockdown Is Inducible and Reversible

Unlike gene knockout methods, one advantage of using CRISPRi-based knockdown of gene expression is the fact that this perturbation should be reversible. To test whether CRISPRi regulation could be induced and subsequently reversed, we encoded both dCas9 and mRFP-specific sgRNA (NT1) under the control of the aTc-inducible promoter and performed time course measurement of CRISPRi-mediated regulation of mRFP in response to inducers (Figure 2E). At time zero, cell culture that grew to the early exponential phase without inducers was supplemented with 1 μM of aTc. Our data indicated that the system could quickly respond to the presence of inducers—the fluorescent reporter protein signal started to decrease within 10 min of the addition of the inducer molecule. Because the mRFP protein is stable (Campbell et al., 2002), the rate of fluorescence signal decrease is limited by protein dilution due to cell growth, as seen by a similar cell doubling time and loss of fluorescence half-time (both ~36 min). At 240 min, all cells were uniformly repressed to the same level as the negative control. At 370 min, the inducer was washed away from the growth media, and cells were diluted back to a lower OD. After a delay of 50 min, mRFP fluorescence started to increase. It took a total 300 min for single-cell fluorescence to increase to the same level as the positive control. The 50 min delay is most likely determined by the dCas9/sgRNA turnover rate offset by dilution by cell growth and division. In summary, these results demonstrate that the silencing effects of dCas9-sgRNA can be induced and reversed.

Native Elongating Transcript Sequencing Confirms that CRISPRi Functions by Blocking Transcription

We hypothesized that dCas9 was functioning as an RNA-guided DNA-binding complex that could block RNA polymerase (RNAP) binding during transcription elongation. Because the non-template DNA strand shares the same sequence identity as the transcribed mRNA and only sgRNAs that bind to the non-template DNA strand exhibited silencing, it remains a possibility that the dCas9:sgRNA complex interacts with mRNA and alters its translation or stability. To distinguish these possibilities, we applied a recently described native elongating transcript sequencing (NET-seq) approach to E. coli, which could be used to globally profile the positions of elongating RNA polymerases and monitor the effect of the dCas9:sgRNA complex on transcription (Churchman and Weissman, 2011). In this NET-seq method, we transformed the CRISPRi system into an E. coli MG1655-derived strain that contained a FLAG-tagged RNAP. The CRISPRi contained an sgRNA (NT1) that binds to the mRFP-coding region. In vitro immunopurification of the tagged RNAP followed by sequencing of the nascent transcripts associated with elongating RNAPs allowed us to distinguish the pause sites of the RNAP.

Our experiment demonstrated that the sgRNA induced strong transcriptional pausing upstream of the sgRNA target locus (Figure 3A). The distance between the pause site and the target site is 19 bp, which is in perfect accordance with the previously reported ~18 bp distance between the nucleotide incorporation of RNAP and its front edge (Nudler et al., 1994). This finding is consistent with a mechanism of CRISPRi in which the transcription block is due to physical collision between the elongating RNAP and the dCas9:sgRNA complex (Figure 3B). Binding of the dCas9:sgRNA complex to the template strand had little repressive effect, suggesting that RNAP was able to read through the complex in this particular orientation. In this case, the sgRNA faces the RNAP, which might be unzipped by the helicase activity of RNAP. Though more structural studies are required to provide a comprehensive understanding of the system in vivo, we have demonstrated that CRISPRi utilizes RNAs to directly block transcription. This mechanism is distinct from that of RNAi, for which knockdown of gene expression requires the destruction of already transcribed messenger RNAs prior to their translation (Zamore et al., 2000).

Open in a separate window

Figure 3

CRISPRi Functions by Blocking Transcription Elongation

(A) FLAG-tagged RNAP molecules were immunoprecipitated, and the associated nascent mRNA transcripts were sequenced. (Top) Sequencing results of the nascent mRFP transcript in cells without sgRNA. (Bottom) Results in cells with sgRNA. In the presence of sgRNA, a strong transcriptional pause is observed 19 bp upstream of the target site, after which the number of sequencing reads drops precipitously.

(B) A proposed CRISPRi mechanism based on physical collision between RNAP and dCas9-sgRNA. The distance from the center of RNAP to its front edge is ~19 bp, which matches well with our measured distance between the transcription pause site and 3′ of sgRNA base-pairing region. The paused RNAP aborts transcription elongation upon encountering the dCas9-sgRNA roadblock.

CRISPRi sgRNA-Guided Gene Silencing Is Highly Specific

To evaluate the specificity of CRISPRi on a genome-wide scale, we performed whole-transcriptome shotgun sequencing (RNA-seq) of dCas9-transformed cells with and without sgRNA coexpression (Figure 4A) (Mortazavi et al., 2008). In the presence of the sgRNA targeted to mRFP (NT1), the mRFP transcript was the sole gene exhibiting a decrease in abundance. No other genes showed significant change in expression upon addition of the sgRNA, within sequencing errors. We also performed RNA-seq on cells with different sgRNAs that target different genes. None of these experiments showed significant changes of genes besides the targeted gene (Figure S4). Thus, sgRNA-guided gene targeting and regulation are highly specific and do not have significant off-target effects.

Open in a separate window

Figure 4

Targeting Specificity of the CRISPRi System

(A) Genome-scale mRNA sequencing (RNA-seq) confirms that CRISPRi targeting has no off-target effects. The sgRNA NT1 that binds to the mRFP coding region is used. The dCas9, mRFP, and sfGFP genes are highlighted.

(B) Multiple sgRNAs can independently silence two fluorescent protein reporters in the same cell. Each sgRNA specifically represses its cognate gene, but not the other gene. When both sgRNAs are present, both genes are silenced. Error bars represent SEM from at least three biological replicates.

(C) Microscopic images for using two sgRNAs to control two fluorescent proteins. (Top) Bright-field images of the E. coli cells; (middle) RFP channel; (bottom) GFP channel. Coexpression of one sgRNA and dCas9 only silences the cognate fluorescent protein, but not the other. The knockdown effect is strong, as almost no fluorescence is observed from cells with certain fluorescent protein silenced. Scale bar, 10 μm. Control shows cells without any fluorescent protein reporters.

Fluorescence results represent average and SEM of at least three biological replicates. See also Figure S4.

CRISPRi Can Be Used to Simultaneously Regulate Multiple Genes

We next asked whether the CRISPRi system could allow control of multiple genes independently without crosstalk. We devised a dual-color fluorescence reporter system based on mRFP and sfGFP (Pédelacq et al., 2006). Two sgRNAs with distinct complementary regions to each gene were designed. Expression of each sgRNA only silenced the cognate gene and had no effect on the other. Coexpression of two sgRNAs knocked down both genes (Figures 4B and 4C). These results suggest that the sgRNA-guided targeting is specific, with the specificity dictated by its sequence identity, and is not impacted by the presence of other sgRNAs. This behavior should enable multiplex control of multiple genes simultaneously by CRISPRi.

Factors that Determine CRISPRi Silencing Efficiency

To find determinants of CRISPRi targeting efficiency, we investigated the roles of length, sequence complementarity, and position on silencing efficiency (Figure 5A). As suggested in Figure 2C, the location of the sgRNA target sequence along the gene was important for efficiency. We further designed sgRNAs to cover the full length of the coding regions for both mRFP and sfGFP (see Data S1 for sgRNA sequences). In all cases, repression was inversely correlated with the target distance from the transcription start site (Figure 5B). A strong linear correlation was observed for mRFP. A similar but slightly weaker correlation was observed when sfGFP was used as the target, perhaps indicating varying kinetics of the RNA polymerase during different points in elongation of this gene.

Open in a separate window

Figure 5

Characterization of Factors that Affect Silencing Efficiency

(A) We measured the silencing effects of sgRNAs with different targeting loci on the same gene (distance from the translation start codon) and sgRNAs with different lengths of the base-pairing region to the same target locus (based on NT1).

(B) The silencing efficiency is inversely correlated with the target distance from the translation start codon (orange, mRFP; green, sfGFP). The relative repression activity is calculated by normalizing repression of each sgRNA to that of the sgRNA with the highest repression fold change. Error bars represent SEM from three biological replicates.

(C) The length of the Watson-Crick base-pairing region between the sgRNA and the target DNA affects repression efficiency. Extensions of the base-pairing region all exhibit strong silencing effect, and truncations dramatically decrease repression. The minimal length of the base-pairing region for detectable repression is 12 bp. Error bars represent SEM from three biological replicates.

(D) We introduced single mismatches into every nucleotide on sgRNA (NT1; Figure 2B) and measured how single mismatches affect repression efficiency. Three subregions with distinct importance to the overall silencing can be discerned. They show a step function. The first 7 nt region is critical for silencing and likely constitutes a “seed” region for probing sgRNAs binding to the DNA target. The PAM sequence (NGG) is indispensible for silencing. Error bars represent SEM from three biological replicates.

(E) Silencing effects of sgRNAs with adjacent double mismatches. The relative repression activity of single-mismatched sgRNAs is shown in gray, with the mismatch position labeled on the bottom. Experimentally measured activity of double-mismatched sgRNAs is shown in blue. Calculated activity by multiplying the effects of two single-mismatched sgRNAs is shown in white and labeled with “Com.” In most cases, the silencing activity of a double-mismatched sgRNA is simply a multiplication of the activities of single-mismatched sgRNAs (except Figure S5B), suggesting an independent relationship between single mismatches. Error bars represent SEM from three biological replicates.

(F) Combinatorial silencing effects of using double sgRNAs to target a single mRFP gene. Using two sgRNAs that target the same gene, the overall knockdown effect can be improved to almost 1,000-fold. When two sgRNAs bind to nonoverlapping sequences of the same gene, repression is augmented. When two sgRNAs target overlapping regions, repression is suppressed. Error bars represent SEM from three biological replicates.

Fluorescence results represent average and SEM of at least three biological replicates. See also Figures S5 and S6.

The sgRNA contains a 20 bp region that is complementary to the target. To identify the importance of this base-pairing region, we altered the length of sgRNA NT1 (Figure 5C). Whereas extension of the region from the 5′ end did not affect silencing, truncation of the region severely decreased repression. The minimal length of the base-pairing region needed for gene silencing was 12 bp, with further truncation leading to complete loss of function. We also introduced single mutations into the base-pairing region of sgRNA NT1 and tested how mismatches affected overall silencing. From our results, three subregions could be discerned, each with a distinct contribution to the overall binding and silencing (Figure 5D). Importantly, any single mutation of the first 7 nt dramatically decreased repression, suggesting that this sequence constitutes a “seed region” for binding, as noted previously for both the type I and type II CRISPR systems (Wiedenheft et al., 2011; Jinek et al., 2012; Gasiunas et al., 2012). We further mutated adjacent nucleotides in pairs (Figures 5E and S5). In most cases, the relative repression activity due to a double mutation was multiplicative relative to the effects of the single mutants, suggesting an independent relationship between the mismatches. Furthermore, in agreement with previous results on the importance of the PAM sequence, an incorrect PAM totally abolished silencing even with a 20 bp perfect binding region (Figure 5E). Thus, we conclude that the specificity of the CRISPRi system is determined jointly by the PAM (2 bp) and at least a 12 bp sgRNA-DNA stretch, the space of which is large enough to cover most bacterial genomes for unique target sites.

We tested whether the efficiency of CRISPRi could be enhanced by using two sgRNAs both targeted against the same gene (Figures 5F and S6). Depending on the relative positioning of multiple sgRNAs, we observed distinct combinatorial effects. Combining two sgRNAs—each with about 300-fold repression—allowed us to increase the overall silencing to up to 1,000-fold. Combining two weaker sgRNAs (~5-fold) showed multiplicative effects when used together. We also observed suppressive combinatorial effects using two sgRNAs whose targets overlapped. This was probably due to competition of both sgRNAs for binding to the same region.

Interrogating an Endogenous Regulatory Network Using CRISPRi Gene Knockdown

We next tested whether the CRISPRi system could be used as a gene knockdown platform to interrogate endogenous gene networks. Previous methods to interrogate microbial gene networks have mostly relied on laborious and costly genomic engineering and knockout procedures. By contrast, gene knockdown with CRISPRi requires only the design and synthesis of a small sgRNA bearing a 20 bp complementary region to the desired genes. To demonstrate this, we applied CRISPRi to create E. coli knockdown strains by designing sgRNAs (see Data S1 for sgRNA sequences) to systematically perturb genes that were part of the well-characterized E. coli lactose regulatory pathway (Figure 6A). We performed β-galactosidase assays to measure LacZ expression from the knockdown strains, with and without isopropyl β-D-1-thiogalactopyranoside (IPTG), a chemical that inhibits the lac repressor (LacI) (Lewis, 2005). In wild-type cells, addition of IPTG induced LacZ expression. Our results showed that a lacZ-specific sgRNA could strongly repress LacZ expression (Figure 6B). Conversely, an sgRNA targeting the lacI gene led to activation of LacZ expression even in the absence of IPTG, as would be expected for silencing a direct repressor of LacZ expression.

Open in a separate window

Figure 6

Functional Profiling of a Complex Regulatory Network Using CRISPRi Gene Knockdown

(A) sgRNAs are designed and used to knock down genes (cya, crp, lacI, lacZ, lacY, and lacA) in the lac regulatory pathway or block transcriptional operator sites (A/P/O). LacI is a repressor of the lacZYA operon by binding to a transcription operator site (O site). The lacZ gene encodes an enzyme that catalyzes lactose into glucose. A few trans-acting host genes such as cya and crp are involved in the activation of the lacZYA system. The cAMP-CRP complex binds to a transcription operator site (A site) and recruits RNA polymerase binding to the P site, which initiates transcription of lacZYA. IPTG, a chemical that inhibits LacI function, induces LacZ expression.

(B) β-galactosidase assay of the knockdown strains without (white) and with (gray) IPTG. Control shows that the wild-type cells without CRISPRi perturbation can be induced by addition of IPTG. The sgRNA that targets LacZ strongly represses LacZ expression even in the presence of IPTG. When LacI is targeted, LacZ expression is high even without IPTG. Targeting cya and crp genes leads to decreased LacZ expression level in the presence of IPTG. Presence of 1 mM cAMP rescues cya knockdown, but not crp knockdown. Blocking the transcription operator sites results in LacZ repression, suggesting that these are important cis-acting regulatory sites for LacZ. Upon perturbation, decreased (red arrows) and increased (green arrows) expression of LacZ are indicated. Error bars represent SEM from three biological replicates.

(C) The knockdown experiments allow us to profile the roles of regulators in the lac regulatory circuit. The data is shown on a two-dimensional (2D) graph, with x axis showing LacZ activity without IPTG and y axis showing its activity with IPTG. The spreading of ovals along each axis shows the SEM of three biological replicates.

The β-galactosidase assay results represent average and SEM of three biological replicates. For RNA-seq data on LacI and LacZ targeting, see also Figure S4.

It is known that cAMP-CRP is an essential activator of LacZ expression by binding to a cis regulatory site upstream of the promoter (A site). Consistently, the sgRNA that was targeted to the crp gene or to the A site in the LacZ promoter led to repression, demonstrating a means to link a regulator to its cis-regulatory sequence using CRISPRi experiments. Targeting the adenylate cylase gene (cya), which is necessary to produce the cAMP that makes CRP more effective at the LacZ promoter, only led to partial repression. Addition of 1 mM cAMP to the growth media complemented the effects for cya knockdown, but not for crp knockdown, suggesting that cya is an indirect regulator of LacZ. Furthermore, targeting the LacI cis-regulatory site (O site) with an sgRNA led to inhibition, presumably because Cas9 complex binding at this site sterically blocks RNA polymerase, mimicking the behavior of the LacI transcription repressor. Targeting the known RNAP-binding site (P site) also blocked expression. In summary, these studies demonstrate that the CRISPRi-based gene knockdown method provides a rapid and effective approach for interrogating the regulatory functions (activating or repressing) of genes and cis elements in a complex regulatory network (Figure 6C).

CRISPRi Is Amenable to Genome-Scale Analysis and Regulation

The CRISPRi method is based on the use of sgRNAs, and only the 20 bp matching region needs to be designed for specific gene targets. With the advances of large-scale DNA oligonucleotide synthesis technology, generating large sets of oligonucleotides that contain unique 20 bp regions for genome targeting is fast and inexpensive. These oligonucleotide libraries could potentially allow us to target large numbers of individual genes to infer gene function or to target gene pairs to map genetic interactions. Furthermore, CRISPRi could be used to simultaneously modulate the expression of large sets of genes, as the small size of sgRNAs allows one to concatenate multiple elements into the same expression vector (Qi et al., 2012).

CRISPRi Provides Tools for Manipulating Microbial Genomes

Because the CRISPRi platform is compact and self-contained, it has the potential to be adapted for different organisms. CRISPRi could thus be a powerful tool for studying nonmodel organisms for which genetic engineering methods are not well developed, including pathogens or industrially useful organisms. Unlike most eukaryotes, most bacteria lack the RNAi machinery. As a consequence, regulation of endogenous genes using designed synthetic RNAs is currently limited. CRISPRi could provide an RNAi-like method for gene perturbation in microbes.

Plasmid Construction and E. coli Genome Cloning

The Cas9 and dCas9 genes were cloned from the previously described vectors pMJ806 and pMJ841, respectively (Jinek et al., 2012). The genes were PCR amplified and inserted into a vector containing an aTc-inducible promoter P_LtetO-1 (Lutz and Bujard, 1997), a chloramphenicol-selectable marker, and a p15A replication origin. The sgRNA template was cloned into a vector containing a minimal synthetic promoter (J23119) with an annotated transcription start site, an ampicillin-selectable marker, and a ColE1 replication origin (see Data S1 for sgRNA sequences). Inverse PCR was used to generate sgRNA cassettes with new 20 bp complementary regions. To insert fluorescent reporter genes into E. coli genomes, the fluorescent gene was first cloned onto an entry vector, which was then PCR amplified to generate linearized DNA fragments that contained nsfA 5′/3′ UTR sequences, the fluorescent gene, and a KanR-selectable marker. The E. coli MG1655 strain was transformed with a temperature-sensitive plasmid pKD46 that contained λ-Red recombination proteins (Exo, Beta, and Gam) (Wang et al., 2009). Cell cultures were grown at 30°C to an OD (600 nm) of ~0.5, and 0.2% arabinose was added to induce expression of the λ-Red recombination proteins for 1 hr. The cells were harvested at 4°C and were used for transformation of the linearized DNA fragments by electroporation. Cells that contain correct genome insertions were selected by using 50 μg/ml kanamycin.

Flow Cytometry and Analysis

Strains were cultivated in EZ-RDM containing 100 μg/ml carbenicillin and 34 μg/ml chloramphenicol in 2 ml 96-well deep well plates (Costar 3960) overnight at 37°C and 1200 rpm. 1 μL of this overnight culture was then added to 249 μl of fresh EZ-RDM with the same antibiotic concentrations, with 2 μM aTc supplemented to induce production of the dCas9 protein. When cells were grown to midlog phase (~4 hr), the levels of fluorescence protein were determined using the LSRII flow cytometer (BD Biosciences) equipped with a high-throughput sampler. Cells were sampled with a low flow rate until at least 20,000 cells had been collected. Data were analyzed using FCS Express (De Novo Software) by gating on a polygonal region containing 60% cell population in the forward scatter-side scatter plot. For each experiment, triplicate cultures were measured, and their standard deviation was indicated as the error bar.

β-Galactosidase Assay

To perform β-galactosidase assay, we added 1 μl of overnight culture prepared as above to 249 μl of fresh EZ-RDM with the same antibiotic concentrations with 2 μM aTc, with or without 1 mM IPTG. Cells were grown to midlog phase. The LacZ activity of 100 ul of this culture was measured using the yeast β-galactosidase assay kit (Pierce) following the instructions.

Extraction and Purification of Total RNA

For each sample, a monoclonal culture of E. coli was grown at 37°C from an OD (600 nm) 0.1 in 500 ml of EZ-RDM to early logphase (OD 0.45 ± 0.05), at which point the cells were harvested by filtration over 0.22 μm nitrocellulose filters (GE) and frozen in liquid nitrogen to simultaneously halt all transcriptional progress. Frozen cells (100 μg) were pulverized on a QIAGEN TissueLyser II mixer mill six times at 15 Hz for 3 min in the presence of 500 μl frozen lysis buffer (20 mM Tris [pH 8], 0.4% Triton X-100, 0.1% NP-40, 100 mM NH₄Cl, 50 U/ml SUPERase·In (Ambion), and 1×protease inhibitor cocktail (Complete, EDTA-free, Roche), supplemented with 10 mM MnCl₂ and 15 μM Tagetin transcriptional inhibitor (Epicenter).

The lysate was resuspended on ice by pipetting. RQ1 DNase I (110 U total, Promega) was added and incubated for 20 min on ice. The reaction was quenched with EDTA (25 mM final) and the lysate clarified at 4°C by centrifugation at 20,000 × g for 10 min. The lysate was loaded onto a PD MiniTrap G-25 column (GE Healthcare) and eluted with lysis buffer supplemented with 1 mM EDTA.

Total mRNA Purification

Total RNA was purified from the clarified lysate using the miRNeasy kit (QIAGEN). 1 μg of RNA in 20 μl of 10 mM Tris (pH 7) was mixed with an equal volume of 2× alkaline fragmentation solution (2 mM EDTA, 10 mM Na₂CO₃, and 90 mM NaHCO₃ [pH 9.3]) and incubated for ~25 min at 95°C to generate fragments ranging from 30 to 100 nt. The fragmentation reaction was stopped by adding 0.56 ml of ice-cold precipitation solution (300 mM NaOAc [pH 5.5] plus GlycoBlue [Ambion]), and the RNA was purified by a standard isopropanol precipitation. The fragmented mRNA was then dephosphorylated in a 50 μl reaction with 25 U T4 PNK (NEB) in 1× PNK buffer (without ATP) plus 0.5 U SUPERase·In, and precipitated with GlycoBlue via standard isopropanol precipitation methods.

Nascent RNA Purification

For nascent RNA purification, we added the clarified lysate to 0.5 ml anti-FLAG M2 affinity gel (Sigma Aldrich) as described previously (Churchman and Weissman, 2011). The affinity gel was washed twice with lysis buffer supplemented with 1 mM EDTA before incubation with the clarified lysate at 4°C for 2.5 hr with nutation. The immunoprecipitation was washed 4 × 10 ml with lysis buffer supplemented with 300 mM KCl, and bound RNAP was eluted twice with lysis buffer supplemented with 1 mM EDTA and 2 mg/ml 3× FLAG peptide (Sigma Aldrich). Nascent RNA was purified from the eluate using the miRNeasy kit (QIAGEN) and was converted to DNA using a previously established library generation protocol (Churchman and Weissman, 2011).

DNA Library Preparation and DNA Sequencing

The DNA library was sequenced on an Illumina HiSeq 2000. Reads were processed using the HTSeq Python package, and other custom software written in Python. The 3′ end of the sequenced transcript was aligned to the reference genome using Bowtie (http://bowtie-bio.sourceforge.net/) and the RNAP profiles were generated in MochiView (http://johnsonlab.ucsf.edu/mochi.html).

Plasmid Design and Construction for CRISPRi in Human Cells

The sequence encoding mammalian-codon-optimized Streptococcus pyogenes Cas9 (DNA 2.0) was fused with three C-terminal SV40 nuclear localization sequences (NLS) or to tagBFP flanked by two NLS. Using standard ligation-independent cloning, we cloned these two fusion proteins into MSCV-Puro (Clontech). Guide sgRNAs were expressed using a lentiviral U6-based expression vector derived from pSico, which coexpresses mCherry from a CMV promoter. The sgRNA expression plasmids were cloned by inserting annealed primers into the lentiviral U6-based expression vector that was digested by BstXI and XhoI.

The authors thank Martin Jinek for discussion and distribution of Cas9 and dCas9 genes and Connie Lee for discussion and critical reading of the manuscript. The authors also thank Leonardo Morsut and Esteban Toro for technical advice and help. L.S.Q. acknowledges support from the UCSF Center for Systems and Synthetic Biology. This work was supported by NIH P50 GM081879 (L.S.Q. and W.A.L.), Howard Hughes Medical Institute (L.A.G., J.A.D., J.S.W., and W.A.L.), NSF SynBERC EEC-0540879 (A.P.A. and W.A.L.), a Howard Hughes Collaborative Initiative Award (J.S.W.), and a Ruth L. Kirschstein National Research Service Award (M.H.L.). J.A.D. is a founder of Caribou Biosciences and a member of its scientific advisory board. None of the other authors have a financial interest related to this work. The authors have filed a patent related to this work.