Research Article

RNA-guided DNA insertion with CRISPR-associated transposases

See allHide authors and affiliations

Science  05 Jul 2019:
Vol. 365, Issue 6448, pp. 48-53
DOI: 10.1126/science.aax9181

Beyond adaptive immunity

Prokaryotic CRISPR-Cas systems defend bacterial cells from phage and plasmid infection. Strecker et al. characterized a CRISPR-Cas system that functions beyond adaptive immunity (see the Perspective by Hou and Zhang). Type V-K CRISPR-Cas from cyanobacteria was associated with a Tn7-like transposon and a natural nuclease–deficient effector Cas12k. Cas12k directed the insertion of Tn7-like transposons into target sites via RNA-guided Tn7 transposition. This system was reprogrammed to efficiently and specifically insert DNA both in vitro and into the Escherichia coli genome.

Science, this issue p. 48; see also p. 25


CRISPR-Cas nucleases are powerful tools for manipulating nucleic acids; however, targeted insertion of DNA remains a challenge, as it requires host cell repair machinery. Here we characterize a CRISPR-associated transposase from cyanobacteria Scytonema hofmanni (ShCAST) that consists of Tn7-like transposase subunits and the type V-K CRISPR effector (Cas12k). ShCAST catalyzes RNA-guided DNA transposition by unidirectionally inserting segments of DNA 60 to 66 base pairs downstream of the protospacer. ShCAST integrates DNA into targeted sites in the Escherichia coli genome with frequencies of up to 80% without positive selection. This work expands our understanding of the functional diversity of CRISPR-Cas systems and establishes a paradigm for precision DNA insertion.

Prokaryotic clustered regularly interspaced short palindromic repeats (CRISPR) and CRISPR-associated proteins (Cas) systems provide adaptive immunity against foreign genetic elements via guide RNA–dependent DNA or RNA nuclease activity (13). CRISPR effectors, such as Cas9 and Cas12, have been harnessed for genome editing (49) and create targeted DNA double-strand breaks in the genome, which are then repaired by endogenous DNA damage repair pathways. Although it is possible to achieve precise integration of new DNA after Cas9 cleavage either through homologous recombination (10) or nonhomologous end-joining (11, 12), these processes are inefficient and vary greatly depending on cell type. Homologous recombination repair is also tied to active cell division, making it unsuitable for postmitotic cells. Recently, an alternative approach for making point mutations on DNA has been developed that relies on using dead Cas9 (13) to recruit cytidine or adenine deaminases to achieve base editing of genomic DNA (1416). However, base editing is restricted to nucleotide substitutions, and thus efficient and targeted integration of DNA into the genome remains a major challenge.

To overcome these limitations, we sought to leverage self-sufficient DNA insertion mechanisms, such as transposons. We explored bioengineering approaches of CRISPR-Cas effectors to facilitate DNA transposition (fig. S1). Cas9 binding to DNA generates an R-loop structure, exposing a substrate for enzymes that act on single-stranded DNA (ssDNA). By tethering nickase Cas9(D10A) to the ssDNA transposase TnpA from Helicobacter pylori IS608 (17, 18), we observe targeted DNA insertions in vitro and in Escherichia coli that are dependent on TnpA transposase activity, Cas9 single guide RNA (sgRNA), and the presence of an insertion site within the ssDNA. However, the requirement of an ssDNA donor will necessitate continued development for efficient synthesis and delivery to cells.

A number of CRISPR-Cas systems lacking active nuclease domains have been identified previously, including minimal type I loci lacking the Cas3 helicase-nuclease (19) and type V loci containing a Cas12 effector with a naturally inactivated RuvC-like nuclease domain (20). The absence of nuclease domains raises questions about the biological function of these CRISPR-Cas systems that can bind but not cleave DNA. Recently, an association between Tn7-like transposons and subtype I-F, subtype I-B, or subtype V-K (formerly, V-U5) CRISPR-Cas systems was reported (21, 22). The CRISPR-Cas–associated Tn7-like transposons contain tnsA, tnsB, tnsC, and tniQ genes (21), similar to the canonical Tn7 heterotrimeric TnsABC complex (23, 24). Tn7 is targeted to DNA via two alternative pathways that are mediated, respectively, by TnsD, a sequence-specific DNA binding protein that recognizes the Tn7 attachment site (25, 26), and TnsE, which facilitates transposition into conjugal plasmids and replicating DNA (27).

The association between Tn7-like transposons and CRISPR-Cas systems suggests that the transposons might have hijacked CRISPR effectors to generate R-loops in target sites and facilitate the spread of transposons via plasmids and phages (21). In the case of subtype V-K, the position of the CRISPR-Cas locus is frequently conserved in predicted transposons, suggesting that CRISPR-Cas is linked with transposition (22). However, because canonical Tn7 transposons often carry cargo genes with defense functions that are beneficial to the host cell (24), it is also possible that CRISPR-Cas may be cargo genes. To date, no functional data on transposon-encoded CRISPR-Cas systems have been reported. Here we show that Tn7-like transposons can be directed to target sites via CRISPR RNA (crRNA)–guided targeting and we elucidate the mechanism of crRNA-guided Tn7 transposition. We further demonstrate that Tn7 transposition can be reprogrammed to insert DNA into the genome of E. coli, highlighting the potential of using RNA-guided Tn7-like transposons for genome editing.

Characterization of a transposon associated with a type V CRISPR system

Among the transposon-encoded CRISPR-Cas variants, the subtype V-K are the simplest because they contain a single-protein CRISPR-Cas effector (20, 21, 28), Cas12k (formerly, C2c5). Subtype V-K systems are thus far limited to cyanobacteria, and the latest nonredundant set includes 63 loci that, in the phylogenetic tree of Cas12k, split into four major branches, covering a broad taxonomic range of Cyanobacteria (22). All V-K systems are embedded within predicted Tn7-like transposable elements with no additional cas genes, suggesting that, if they are active CRISPR-Cas systems, they might rely on adaptation modules supplied in trans. Of the 560 analyzed V-K spacers, only six protospacer matches were identified: three from cyanobacterial plasmids and three from single-stranded transposons of IS200 or IS650 families (22). These findings suggest that V-K systems might provide a biological advantage for the host transposons by directing integration into other mobile genetic elements, to enhance transposon mobility while minimizing damage to the host.

For experimental characterization, we selected two Tn7-like transposons encoding subtype V-K CRISPR-Cas systems [hereafter, CAST (CRISPR-associated transposase)]. The selected CAST loci were 20 to 25 kb in length and contained Tn7-like transposase genes at one end of the transposon with a CRISPR array and Cas12k on the other end, flanking internal cargo genes (Fig. 1A and fig. S2, A and B). We first cultured the cyanobacteria Scytonema hofmanni (UTEX B 2349) (Fig. 1B) and Anabaena cylindrica (PCC 7122) and performed small RNA sequencing to determine whether the CRISPR-Cas systems are expressed and active. For both loci, we identified a long putative trans-activating crRNA (tracrRNA) that mapped to the region between Cas12k and the CRISPR array, and in the case of S. hofmanni (ShCAST) we detected crRNAs 28 to 34 nucleotides (nt) long, consisting of 11 to 14 nt of direct repeat (DR) sequence with 17 to 20 nt of spacer (Fig. 1C and fig. S2C).

Fig. 1 Targeting requirements for CRISPR-associated transposase (CAST) systems.

(A) Schematic of the S. hofmanni CAST locus containing Tn7-like proteins, the CRISPR-Cas effector Cas12k, and a CRISPR array. (B) Fluorescent micrograph of the cyanobacteria S. hofmanni. Scale bars, 40 μm. (C) Alignment of small RNA sequencing reads from S. hofmanni. The location of the putative tracrRNA is marked. (D) Schematic of experiment to test CAST system activity in E. coli. kanR, kanamycin resistance gene; F, forward primer; R, reverse primer. (E) PAM motifs for insertions mediated by ShCAST and AcCAST. (F) ShCAST and AcCAST insertion positions identified by deep sequencing. (G) Insertion frequency of ShCAST system in E. coli with pTarget substrates as determined by ddPCR. Error bars represent SD from n = 3 replicates. n.t., nontargeting.

To investigate whether ShCAST and AcCAST function as RNA-guided transposases, we cloned the four CAST genes (tnsB, tnsC, tniQ, and Cas12k) into a helper plasmid (pHelper) along with the endogenous tracrRNA region and a crRNA targeting a synthetic protospacer (PSP1). We predicted the ends of the transposons by searching for TGTACA-like terminal repeats surrounded by a duplicated insertion site (21) and constructed donor plasmids (pDonor) containing the kanamycin resistance gene flanked by the transposon left end (LE) and right end (RE). Given that CRISPR-Cas effectors require a protospacer adjacent motif (PAM) to recognize target DNA (29), we generated a target plasmid (pTarget) library containing the PSP1 sequence flanked by a 6N motif upstream of the protospacer. We coelectroporated pHelper, pDonor, and pTarget into E. coli and extracted plasmid DNA after 16 hours (Fig. 1D). We detected insertions into the target plasmid (pInsert) by polymerase chain reaction (PCR) for both ShCAST and AcCAST, and deep sequencing confirmed the insertion of the LE into pTarget. Analysis of PAM sequences in pInsert plasmids revealed a preference for GTN PAMs for both ShCAST and AcCAST systems, suggesting that these insertions result from Cas12k targeting (Fig. 1E and fig. S3, A and B). We next examined the position of the donor in pInsert products relative to the protospacer. Insertions were detected within a small window 60 to 66 base pairs (bp) downstream from the PAM for ShCAST and 49 to 56 bp downstream from the PAM for AcCAST (Fig. 1F). No insertions were detected in the opposite orientation for either system, indicating that CAST functions unidirectionally. Although DNA insertions could potentially arise from genetic recombination in E. coli, the discovery of an associated PAM sequence and the constrained position of insertions argues against this possibility.

To validate these findings, we transformed E. coli with ShCAST pHelper and pDonor plasmids along with target plasmids containing a GGTT PAM, an AACC PAM, and a scrambled nontarget sequence. We assessed insertion events by quantitative droplet digital PCR (ddPCR), which revealed insertions of the donor only in the presence of pHelper and a pTarget containing a GGTT PAM and crRNA-matching protospacer sequence (Fig. 1G). Additional experiments with 16 PAM sequences confirmed a preference for NGTN motifs (fig. S3C). As further validation, we recovered pInsert products and performed Sanger sequencing of both LE and RE junctions. All sequenced insertions were located 60 to 66 bp from the PAM and contained a 5-bp duplicated insertion motif flanking the inserted DNA (fig. S4), consistent with the staggered DNA breaks generated by Tn7 (30). Because Tn7 inserts into a CCCGC motif downstream of its attachment site, we hypothesized that the sequence within the insertion window might also be important for CAST function. We generated a second target library with an 8N motif located 55 bp from the PAM and again cotransformed the library into E. coli with ShCAST pHelper and pDonor followed by deep sequencing (fig. S5A). We observed only a minor sequence preference upstream of the LE in pInsert, with a slight T or A preference three bases upstream of the insertion site (fig. S5, B to D). ShCAST can therefore target a wide range of DNA sequences with minimal targeting rules. Together, these results indicate that AcCAST and ShCAST catalyze DNA insertion in a heterologous host and that these insertions are dependent on a targeting protospacer and a distinct PAM sequence.

Genetic requirements for RNA-guided insertions

We next sought to determine the genetic requirements for ShCAST insertions in E. coli and constructed a series of pHelper plasmids with deletions of each element. Insertions into pTarget required all four CAST proteins and the tracrRNA region (Fig. 2A). To better characterize the tracrRNA sequence, we complemented pHelperΔtracrRNA with various tracrRNA driven by the pJ23119 promoter. Expression of the 216-nt tracrRNA variant 6 alone was sufficient to restore DNA transposition (Fig. 2B). The 3′ end of the tracrRNA is predicted to hybridize with a crRNA containing 14 nt of the DR sequence, and we designed sgRNA testing two linkers between the tracrRNA and crRNA sequences. Both designs supported insertion activity in the context of the tracrRNA variant 6 (Fig. 2C). We observed that expression of tracrRNA or sgRNA with the pJ23119 promoter resulted in a fivefold increase in the insertion activity compared with the natural locus, suggesting that RNA was rate limiting during heterologous expression.

Fig. 2 Genetic requirements for RNA-guided insertions.

(A) Genetic requirement of tnsB, tnsC, tniQ, Cas12k, and tracrRNA on insertion activity. Deleted components are indicated by a dashed outline. (B) Insertion activity of six tracrRNA variants expressed with the pJ23119 promoter. (C) Schematic of tracrRNA and crRNA base pairing and two sgRNA designs highlighting the linker sequence (blue). (D) Insertion activity into pTarget containing ShCAST transposon ends relative to activity into pTarget without previous insertion. bars represent SD from n = 3 replicates.

As ShCAST does not destroy the protospacer upon DNA insertion, we asked whether multiple insertions could occur in pTarget, or if these are inhibited, as with canonical Tn7 (31, 32). We generated target plasmids containing LE + RE, or LE alone, and measured ShCAST transposition activity at six nearby protospacers. We observed a strong inhibitory effect on transposition at a protospacer 62 bp from the LE (<1% of relative activity to pTarget), and only 5.7% relative activity 542 bp from the LE (Fig. 2D), indicating that CAST transposon ends act in cis to prevent multiple insertions. The presence of LE alone resulted in a weaker inhibitory effect, and we observed 61.1% of activity at 542 bp away from the transposon end (fig. S6, A and B).

Our original pDonor contained 2.2 kb of cargo DNA, and we next tested the effect of donor length on ShCAST activity ranging from 500 bp to 10 kb. We observed a twofold higher insertion rate with a 500-bp donor and a similar rate of insertions with 10 kb of payload compared with the original pDonor (fig. S6C). We were unable to detect rejoined pDonor backbone during transposition in E. coli (fig. S6, D and E), suggesting that a linear donor backbone is formed, not a rejoined product, consistent with the known reaction products of canonical Tn7 (30, 33). Finally, we investigated the requirement of the LE and RE transposon end sequences contained in pDonor for transposition. Removal of all flanking genomic sequences or the 5bp duplicated target sites had little effect on insertion frequency, and ShCAST tolerated truncations of LE and RE to 113 and 155 bp, respectively (fig. S7A). Removal of additional donor sequence completely abolished transposase activity, which is consistent with the loss of predicted Tn7 TnsB-like binding motifs (fig. S7, B and C).

In vitro reconstitution of ShCAST

Although our data strongly suggested that ShCAST mediates RNA-guided DNA insertion, to exclude the requirement of additional host factors, we next sought to reconstitute the reaction in vitro. We purified all four ShCAST proteins (fig. S8A) and performed in vitro reactions using pDonor, pTarget, and purified RNA (Fig. 3A). Addition of all four protein components, crRNA, and tracrRNA resulted in DNA insertions detected by both LE and RE junction PCRs, as did reactions containing the four protein components and sgRNA (Fig. 3B). The truncated tracrRNA variant 5 was also able to support DNA insertion in vitro, in contrast with the activity observed in E. coli. ShCAST-catalyzed transposition in vitro occurred between 37° and 50°C and depended on adenosine triphosphate and Mg2+ (fig. S8, B and C). To confirm that in vitro insertions are in fact targeted, we performed reactions with target plasmids containing a GGTT PAM, an AACC PAM, and a scrambled nontarget sequence, and we could only detect DNA insertions into the GGTT PAM substrate with the target sequence (Fig. 3C). In vitro DNA transposition depended on all four CAST proteins, although we identified weak but detectable insertions in the absence of tniQ (Fig. 3D).

Fig. 3 In vitro reconstitution of an RNA-guided transposase.

(A) Schematic of in vitro transposition reactions with purified ShCAST proteins and plasmid donor and targets. (B) RNA requirements for in vitro transposition. pInsert was detected by PCR for LE and RE junctions. All reactions contained pDonor and pTarget. Schematics indicate the location of primers and the expected product sizes for all reactions. (C) Targeting specificity of ShCAST in vitro. All reactions contained ShCAST proteins and sgRNA. (D) Protein requirements for in vitro transposition. All reactions contained pDonor, pTarget, and sgRNA. (E) CRISPR-Cas effector requirements for in vitro transposition. All reactions contained ShCAST proteins, pDonor, and pTarget. (F) Chromatograms of pInsert reaction products after transformation and extraction from E. coli. LE and RE elements are highlighted and the duplicated insertion sites denoted. For all panels, ShCAST proteins were used at a final concentration of 50 nM, and n = 3 replicates for all reactions were performed with a representative image shown.

We were unable to detect DNA cleavage in the presence of Cas12k and sgRNA across a range of buffer conditions (fig. S8D), which is consistent with the predicted lack of nuclease activity of Cas12k. To determine whether other CRISPR-Cas effectors could also stimulate DNA transposition, we performed reactions with tnsB, tnsC, and tniQ, along with dCas9 and a sgRNA targeting the same GGTT PAM substrate. We were unable to detect any insertions after dCas9 incubation (Fig. 3E), which indicates that the function of Cas12k is not merely DNA binding, and that DNA transposition by CAST does not simply occur at R-loop structures. As final validation, we transformed in vitro reaction products into E. coli and performed Sanger sequencing to determine the LE and RE junctions. All sequenced donors were located in pTarget 60 to 66 bp from the PAM and contained duplicated 5-bp insertion sites, demonstrating complete reconstitution of ShCAST with purified components.

ShCAST mediates efficient and precise genome insertions in E. coli

To test whether ShCAST could be reprogrammed as a DNA insertion tool, we selected 48 targets in the E. coli genome and cotransformed pDonor and pHelper plasmids expressing targeting sgRNAs (Fig. 4A). We detected insertions by PCR at 29 of the 48 sites (60.4%) and selected 10 sites for additional validation (fig. S9A). We performed ddPCR to quantitate insertion frequency after 16 hours and measured rates up to 80% at PSP42 and PSP49 (Fig. 4B). This high efficiency of insertion was surprising given that insertion events were not selected for by antibiotic resistance, so we performed PCR of target sites to confirm. We detected the 2.5-kb insertion product in the transformed population (Fig. 4C). Restreaking transformed E. coli yielded pure single colonies, the majority of which contained the targeted insertion (fig. S9B), and the high efficiency of integration was maintained with a variety of donor DNA lengths (fig. S9C). We analyzed the position of genome insertions by targeted deep sequencing of the LE and RE junctions and observed insertions within the 60- to 66-bp window at all 10 sites (Fig. 4D and fig. S10A).

Fig. 4 ShCAST mediates genome insertions in E. coli.

(A) Schematic of experiment to test for genome insertions in E. coli. (B) Insertion frequency at 10 tested protospacers after ShCAST transformation. Insertion frequency was determined by ddPCR on extracted genomic DNA. Error bars represent SD from n = 3 replicates. (C) Flanking PCR of three tested protospacers in a population of E. coli after ShCAST transformation. Schematics indicate the location of primers and the expected product sizes. (D) Insertion site position as determined by deep sequencing after ShCAST transformation. (E) Insertion positions determined by unbiased donor detection. The location of each protospacer is annotated along with the percent of total donor reads that map to the target.

We next assayed the specificity of RNA-guided DNA transposition. We performed unbiased sequencing of donor insertion sites after Tn5 tagmentation of genomic DNA. We observed one prominent insertion site in each sample, which mapped to the target site and contained >50% of the total insertion reads (Fig. 4E). The remaining off-target reads were scattered across the genome, and analysis of the top off-target sites revealed strong overlap between samples, demonstrating that these events are independent of the guide sequence (fig. S10B and table S5). Top off-target sites were located near highly expressed loci such as ribosomal genes, serine-tRNA ligase, and enolase, although insertion frequency in these regions were all <1% of the on-target site (table S5). We identified one potential RNA-guided off-target after targeting of PSP42, which contains four mismatches to the guide sequence (fig. S10C). Together, these results indicate that ShCAST robustly and precisely inserts DNA into the target site.


Here we demonstrate that CRISPR-Cas systems associated with Tn7-like transposon mediate RNA-guided DNA transposition and elucidate its mechanism. ShCAST mediates unidirectional insertions in a narrow window downstream of the target and inhibits repeated insertions into a single target site (Fig. 5). Although ShCAST and AcCAST exhibit similar PAM preferences, one notable difference is that their respective positions of insertion, relative to the PAM, differ by 10 to 11 bp, which roughly corresponds to one turn of DNA. Deeper exploration of microbial genomes is expected to uncover CAST systems with a range of diverse properties, including targeting preference and activity across different conditions.

Fig. 5 Model for RNA-guided DNA transposition.

The ShCAST complex that consists of Cas12k, TnsB, TnsC, and TniQ mediates insertion of DNA 60 to 66 bp downstream of the PAM. Transposon LE and RE sequences, along with any additional cargo genes, are inserted into DNA resulting in the duplication of 5-bp insertion sites.

Targeted DNA insertion by ShCAST results in the incorporation of LE and RE elements and is therefore not a scarless integration method. One potential generalizable strategy for the use of CAST in the therapeutic context would be to insert corrected exons into the intron before the mutated exon (fig. S11). CAST could also be used to insert transgenes into “safe harbor” loci (34) or downstream of endogenous promoters so that the expression of transgenes of interest can benefit from endogenous gene regulation.

Further studies should improve our understanding of the function of each transposase subunit in the CAST complex, notably, TniQ, which contains a predicted DNA binding domain. We originally hypothesized that TniQ is analogous to the site-specific DNA binding protein TnsD of Tn7 and, therefore, might be dispensable for RNA-guided insertions; however, we observed that TniQ is required for RNA-guided insertions in E. coli. The observation that in vitro transposition can occur to a limited extent in the absence of TniQ is compatible with a model in which TniQ facilitates the formation of the CAST complex and is not essential for catalytic function. Therefore, it might be possible to engineer simplified versions of CAST systems without TniQ or with fragments of TniQ.

Our analysis indicated that ShCAST is specific but, under overexpression conditions, can integrate at nontargeted sites in the E. coli genome via Cas12k-independent mechanisms, and this guide-independent integration seems to favor highly expressed genes. We also observed nontargeted insertions into pHelper in E. coli that were independent of Cas12k (fig. S12) and reminiscent of TnsE-mediated Tn7 insertions into conjugal plasmids and replicating DNA (27). Future protein engineering of the transposase components could improve the targeting specificity of CAST systems.

This work identifies a function for CRISPR-Cas systems beyond adaptive immunity that does not require Cas nuclease activity and provides a strategy for targeted insertion of DNA without engaging homologous recombination pathways, with a particularly exciting potential for genome editing in eukaryotic cells.

Supplementary Materials

Materials and Methods

Figs. S1 to S12

Tables S1 to S6

References (3538)

References and Notes

Acknowledgments: We thank R. Macrae for critical reading of the manuscript and the entire Zhang laboratory for support and advice. We thank F. Chen for imaging S. hofmanni. Funding: J.S. is supported by the Human Frontier Science Program. F.Z. is a New York Stem Cell Foundation–Robertson Investigator. F.Z. is supported by NIH grants (1R01-HG009761, 1R01-MH110049, and 1DP1-HL141201); the Howard Hughes Medical Institute; the New York Stem Cell and Mathers Foundations; the Poitras Center for Psychiatric Disorders Research at MIT; the Hock E. Tan and K. Lisa Yang Center for Autism Research at MIT; J. and P. Poitras; and the Phillips family. Author contributions: J.S. and F.Z. conceived of the project. J.S., A.L., and Z.G. performed bacterial experiments. J.S. purified CAST proteins and performed in vitro reactions. A.L. performed genome targeting experiments. A.L. performed insertion specificity analysis with help from J.L.S.-B. K.S.M. and E.V.K. identified CAST loci and performed bioinformatics analysis. F.Z. supervised the research and experimental design. J.S. and F.Z. wrote and revised the manuscript with input from all authors. Competing interests: J.S. and F.Z. are coinventors on U.S. provisional patent application no. 62/780,658 filed by the Broad Institute, relating to CRISPR-associated transposases. F.Z. is a cofounder of Editas Medicine, Beam Therapeutics, Pairwise Plants, Arbor Biotechnologies, and Sherlock Biosciences. Data and materials availability: Expression plasmids are available from Addgene under a uniform biological material transfer agreement; support forums and computational tools are available via the Zhang lab website (

Correction (5 December 2019): The Phillips family was omitted from the list of funding sources in the Acknowledgments. This error has been corrected.

View Abstract

Stay Connected to Science

Navigate This Article