CRISPR adaptation biases explain preference for acquisition of foreign DNA

Abstract

CRISPR-Cas (clustered, regularly interspaced short palindromic repeats coupled with CRISPR-associated proteins) is a bacterial immunity system that protects against invading phages or plasmids. In the process of CRISPR adaptation, short pieces of DNA ('spacers') are acquired from foreign elements and integrated into the CRISPR array. So far, it has remained a mystery how spacers are preferentially acquired from the foreign DNA while the self chromosome is avoided. Here we show that spacer acquisition is replication-dependent, and that DNA breaks formed at stalled replication forks promote spacer acquisition. Chromosomal hotspots of spacer acquisition were confined by Chi sites, which are sequence octamers highly enriched on the bacterial chromosome, suggesting that these sites limit spacer acquisition from self DNA. We further show that the avoidance of self is mediated by the RecBCD double-stranded DNA break repair complex. Our results suggest that, in Escherichia coli, acquisition of new spacers largely depends on RecBCD-mediated processing of double-stranded DNA breaks occurring primarily at replication forks, and that the preference for foreign DNA is achieved through the higher density of Chi sites on the self chromosome, in combination with the higher number of forks on the foreign DNA. This model explains the strong preference to acquire spacers both from high copy plasmids and from phages.

Extended Data Figure 1.. Graphic overview of the procedure for characterizing the frequency and sequence of newly acquired spacers

DNA from cultures of either E. coli K-12 (left) or E. coli BL21-AI (right) strains expressing Cas1+2 from two different plasmids were used as templates for PCR. Round 1 was used to determine the frequency of spacer acquisition by comparing occurrences of expanded arrays to WT arrays. Round 2 amplified only the expanded arrays and, followed by deep sequencing, was used to determine the sequence, location, and source of newly acquired spacers.

Extended Data Figure 2.. PAMs and DNA content along the E. coli BL21-AI genome

(A) Distribution of PAM (AAG) sequences. Each data point represents the number of PAMs in a window of 10kb. (B) DNA content of a culture growing in log phase. Genomic DNA was extracted from E. coli BL21-AI cells carrying the pCas plasmid, grown at log phase, and was sequenced using the Illumina technology. The resulting reads were mapped to the sequenced E. coli BL21(DE3) genome (genbank accession NC_012947). Areas where little or no reads map to the genome represent regions that are present in the reference BL21(DE3) genome but are missing from the genome of the sequenced strain (BL21-AI).

Extended Data Figure 3.. Distribution of newly acquired spacers on the genome during synchronized replication

E. coli K-12ΔcasCdnaC2 cells were transferred from 39°C (replication restrictive temperature) to 30°C (replication permissive). Cas1+2 were induced in these cells 30 minutes prior to the transfer to 30°C and during the growth in 30°C. At given time points: (A) following 20 minutes; (B) following 40 minutes; (C) following 60 minutes from replication initiation, newly acquired spacers were sequenced. Shown are the positions of the newly acquired spacers in windows of 100kb, and their fraction out of the total new spacers in the sample.

Extended Data Figure 4.. A model explaining the preference for spacer acquisition near TerC as compared to TerA in E. coli BL21-AI

The DNA manipulation at the CRISPR region forms a replication fork stalling site, and leads to extensive spacer acquisition upstream to the CRISPR. While the clockwise fork is stalled at the CRISPR, the counterclockwise fork reaches the Ter region and is stalled at the respective Ter site, TerC, leading to extensive spacer acquisition upstream to TerC. Another factor that can contribute to the observed TerC/TerA bias may be that the clockwise replichore in E. coli (oriC to TerA) is longer than the counter clockwise one (oriC to TerC), leading the forks to stall at TerC more often than at TerA.

Extended Data Figure 5.. The protein product of T7 gene 5.9 inhibits spacer acquisition activity

E. coli BL21-AI strains harboring pBAD-Cas1+2 and pBAD33-gp5.9 (lane 1) or pBAD33 vector control (lane 2) were grown overnight in the presence of inducers (0.4% L-arabinose). Gel shows PCR products amplified from the indicated cultures using primers annealing to the leader and to the fifth spacer of the CRISPR array. Results represent one of three independent experiments.

Extended Data Figure 6.. Distribution of protospacers across (A) pCtrl-Chi and (B) pChi plasmids

Circular representation of the 4.7kb plasmid is presented, with the inserted 4-Chi cluster present at the top-middle of the circle. Black bars indicate the number of PAM-derived spacers sequenced from each position; green bars represent non-PAM spacers. Scale bar indicates 100k spacers. Pooled protospacers from two replicates are presented for each panel.

Figure 1. Chromosome-scale hotspots for spacer acquisition

(A) Distribution of protospacers across the E. coli BL21-AI genome. Protospacers were deduced from aligning new spacers, acquired into the CRISPR I array after 16 hours growth with no arabinose, to the bacterial genome. Only unique protospacers are presented, to avoid possible biases stemming from PCR amplification of the CRISPR array. Pooled protospacers from two replicates are presented. (B) Protospacer density across a circular representation of the E. coli genome, normalized to the DNA content of the culture. Dark brown, normalized protospacer numbers; orange, PAM density. (C) Protospacer distribution at the Ter region. Protospacer density is shown in 1kb windows. (D) Protospacer density in an E. coli BL21-AI in which the native 23bp-long TerB site was engineered into the pheA locus.

Figure 2. Dependence of spacer acquisition in replication

(A) Spacer acquisition rates in antibiotic-treated E. coli BL21-AI cells. Cells were grown at log phase 16 hours during Cas1+2 induction, with addition of the replication inhibitor nalidixic acid (Nal) or the transcription inhibitor rifampicin (Rif). (B) Spacer acquisition rates of K-12ΔcasCdnaC2 and an isogenic K-12ΔcasC strains during overnight Cas1+2 induction. (C) Spacer acquisition patterns measured following transfer of K-12ΔcasCdnaC2 cells from 39°C to 30°C, during induction of Cas1+2. For all panels, average and error margins for two biological replicates are shown.

Figure 3. Chi sites define boundaries of protospacer hotspots

(A-D) Protospacer hotspot peaks. Each panel shows a 100kb window around a major hotspot for spacer acquisition. Short blue and red ticks mark positive- and negative-strand Chi sites, respectively. Green line mark a replication fork stalling site (TerA, TerC) or putative stalling site (CRISPR array). Dashed line marks the first properly oriented Chi site upstream relative to the fork stalling site. (A) The CRISPR region in E. coli BL21-AI. (B) The CRISPR region in E. coli K-12. (C) The TerC region and (D) the TerA region in E. coli BL21-AI. In panel C, the Chi site drawn at^~2260k represents a cluster of 3 consecutive Chi sites found in the same 1kb window.

Figure 4. Involvement of the dsDNA break repair machinery in defining spacer acquisition patterns

(A) The overall number of protospacers around all Chi sites in E. coli BL21-AI, that are not included in the CRISPR region (950,000-1,050,000) or the Ter region (2M-2.5M), is shown in windows of 0.5 kb. (B) Protospacer hotspot peak resulting from a dsDNA break formed by the homing endonuclease I-SceI.(C) The overall number of protospacers around all Chi sites that are not included in the CRISPR or the Ter regions in a BL21-AIΔrecB strain. (D) The protospacer hotspot at the CRISPR region in the BL21-AIΔrecB strain is not confined by a Chi site (compare to the same hotspot in the WT strain, Fig. 3A). (E) Adaption levels in WT BL21-AI and BL21-AIΔrecB, ΔrecC or ΔrecD strains following overnight growth without arabinose induction of Cas1+2. (F) Percent new spacers derived from the self chromosome in the experiment described in Panel E. (G) Percent new spacers derived from the self chromosome in the presence of a plasmid that contains a cluster of 4 Chi sites (pChi) as compared to an identical plasmid that lacks Chi sites (pCtrl-Chi). For panels E-G, average and error margins for two biological replicates are shown.

Figure 5. A model explaining the preference for foreign DNA in spacer acquisition

(A) RecBCD localizes to double strand DNA breaks (DSBs) and unwinds/degrades the DNA until reaching the nearest properly oriented Chi site. The RecBCD activity generates significant amounts of DNA “debris”, including short and long ssDNA fragments and degraded dsDNA, all of which may serve as substrates for spacer acquisition by Cas1+2. (B) High density of Chi sites on the chromosome reduces spacer acquisition from self DNA. On average, the 8bp-long Chi sites are found every 4.6kb on the E. coli chromosome, 14 times more often than on random DNA. When a DSB occurs on the chromosome, RecBCD DNA degradation activity will quickly be moderated by a nearby Chi site, but a similar DSB on a foreign DNA will lead to much more extensive DNA processing, providing more substrate for spacer acquisition. (C) Preference for spacer acquisition from high copy plasmids. In a replicating cell, most replication forks (blue circles) localize to the multiple copies of the plasmid. Since most DBSs occur during replication , at stalled replication forks ^,, plasmid DNA would become more amenable for spacer acquisition. (D) Most phages inject linear DNA into the infected cell. When such linear DNA is not protected, RecBCD will quickly degrade it, providing an intrinsic preference for spacer acquisition from phage DNA.