CRISPRi is not strand-specific at all loci and redefines the transcriptional landscape

CRISPRi, an adapted CRISPR-Cas9 system, is proposed to act as a strand-specific roadblock to repress transcription in eukaryotic cells using guide RNAs (sgRNAs) to target catalytically inactive Cas9 (dCas9) and offers an alternative to genetic interventions for studying pervasive antisense transcription. Here, we successfully use click chemistry to construct DNA templates for sgRNA expression and show, rather than acting simply as a roadblock, sgRNA/dCas9 binding creates an environment that is permissive for transcription initiation/termination, thus generating novel sense and antisense transcripts. At HMS2 in Saccharomyces cerevisiae, sgRNA/dCas9 targeting to the non-template strand for antisense transcription results in antisense transcription termination, premature termination of a proportion of sense transcripts and initiation of a novel antisense transcript downstream of the sgRNA/dCas9-binding site. This redefinition of the transcriptional landscape by CRISPRi demonstrates that it is not strand-specific and highlights the controls and locus understanding required to properly interpret results from CRISPRi interventions.


Introduction
Eukaryotic genomes are pervasively transcribed but the effect of this inter-and intragenic transcription is not fully understood (Mellor et al., 2016). Of particular interest is the function of antisense transcription, which alters the chromatin in the vicinity of sense promoters (Lavender et al., 2016;Murray et al., 2015;, and is associated with repression, activation or no change in levels of the corresponding sense transcript (Murray et al., 2015). Discerning the mechanism by which antisense transcription functions in gene regulation is confounded by the difficulty in ablating antisense transcription without direct genetic intervention (Bassett et al., 2014). A new approach, CRISPR interference (CRISPRi) , can circumvent these issues by avoiding the need to manipulate endogenous DNA sequences. An endonucleolytically dead version of Cas9 (dCas9) is recruited by single base-pairing guide RNAs (sgRNAs) targeting the non-template (NT) DNA strand, where it blocks transcription strand-specifically at the loci tested (Lenstra et al., 2015;Qi et al., 2013). In a further adaptation, dCas9 fusion with transcriptional repressors or activators allows both negative and positive regulations of transcription, respectively Gilbert et al., 2014), but the strand-specificity is lost and thus is not suitable for strand-specific repression of antisense transcription. The sgRNA compenent of the CRISPRi system consists of two regions: a constant region (82 nt) that binds to dCas9 and a variable region (20 nt) that is responsible for targeting. The modular nature of the sgRNA lends itself to production from DNA templates constructed using synthetic copper(I)-catalysed alkyne-azide cycloaddition (CuAAC) click chemistry (El-Sagheer and Brown, 2010). The stable artificial triazole DNA linker generated is biocompatibile, being read and accurately copied by DNA and RNA polymerases (Birts et al., 2014;. We show that click chemistry is an efficient method for sgRNA template construction and that when combined with dCas9, these sgRNAs are as effective as sgRNAs produced from chemically synthesised full-length DNA templates at reducing levels of transcripts in vivo. However, it is still not completely understood how the CRISPRi system functions strand-specifically or why sgRNAs must target the non-template strand.
Here, we use the CRISPRi system to study the effect of blocking antisense transcription at loci with well-characterised sense:antisense transcript pairs. We have previously used a promoter deletion of the antisense transcript SUT650 at the HMS2 locus to show that SUT650 represses HMS2 sense transcription (Nguyen et al., 2014). Now we use CRISPRi to examine the effects of blocking SUT650 antisense transcription to ask (i) whether SUT650 represses HMS2 sense transcription without a genetic intervention and (ii) whether CRISPRi is strand-specific using, in addition to HMS2, an engineered GAL1 gene with a well-characterised antisense transcript (Murray et al., 2012(Murray et al., , 2015. The main conclusion from this study is that CRISPRi at the HMS2 locus is not fully strand-specific and results in (i) premature termination of the sense transcript and (ii) initiation of a new unstable antisense transcript in the vicinity of the sgRNA binding site. As transcription from this new antisense initiation site extends in to the HMS2 promoter, there is no net change in HMS2 sense transcript levels. Thus, CRISPRi redefines the transcriptional landscape at HMS2. This suggests that routine use of CRISPRi for gene expression analysis will require rigorous analysis of transcript integrity and function before conclusions can be drawn.

Results and discussion
DNA templates for sgRNA production can be made using click chemistry CRISPRi-mediated transcriptional repression requires co-expression of a mature sgRNA and dCas9 ( Figure 1A). A small library of single-stranded DNAs, comprised of templates for sgRNA variable regions, were joined to the constant region using click chemistry ( Figure 1B) and used successfully as templates for PCR, with no significantly different efficiencies when compared to control full-length synthesised oligonucleotides ( Figure 1C). The PCR products were inserted in place of a URA3 selection cassette in the endogenous snR52 locus for expression of a transcript that is then processed to form the mature nuclear-retained sgRNA ( Figure 1A). Levels of dCas9 protein were uniform between strains (Figure 1-figure supplement 1). Neither insertion of URA3 into snR52 nor dCas9 expression in the control strains affected growth rate, although strains expressing some sgRNAs grew more slowly indicating a physiological effect ( Figure 1D).

CRISPRi represses the production of antisense transcripts at HMS2 and GAL1
CRISPRi represses transcription when sgRNAs/dCas9 are targeted to the non-template (NT) strand next to a protospacer adjacent motif (PAM)   (Figure 2A). Firstly, we used an engineered version of GAL1 that has a stable antisense transcript (GAL1 AS) initiating within an ADH1 terminator inserted into the GAL1 coding region (Murray et al., 2012) ( Figure 2B). GAL1 AS is present in cells grown in glucose-containing media when the GAL1 gene is repressed and is reduced as cells are switched into galactose-containing media and GAL1 sense is induced ( Figure 2C lanes 1-3). We designed sgRNAs adjacent to two PAM sequences on the non-template strand near the antisense transcription start site (TSS) (AS+28NT and AS+112NT) and a third strand-specificity control sgRNA on the template strand in this region (AS + 93T) ( Figure 2B). Only sgRNA AS+112NT/dCas9 caused significant (p=0.004) reduction in GAL1 antisense transcript levels, as assessed by Northern blotting ( Figure 2C,D).
Next, we examined the HMS2 locus which has a stable antisense transcript, SUT650, initiating within the 3' coding region of HMS2 (Nguyen et al., 2014;. We designed three SUT650-targeting sgRNAs, located 59, 243 and 276 nucleotides downstream of the major SUT650 TSS on the non-template strand (sgRNAs AS+59NT, AS+243NT, AS+276NT, respectively) A B D S + 7 7 N T S + 1 0 6 N T S + 1 7 9 N T A S + 2 4 3 N T A S + 2 7 6 N T A S + 5 9 N T Clicked product (  DNA templates for sgRNA production were generated using a click chemistry reaction between a single-stranded DNA oligonucleotide with a 3' alkyne group encoding the constant region (dark blue) and a number of single-stranded DNA oligonucleotides with 5' azido groups encoding the different variable regions (green). The resulting single-stranded DNA oligonucleotide was purified, amplified by PCR and transformed into yeast to replace the URA3 selection cassette that had previously been introduced into the endogenous snR52 locus. Correct insertion was confirmed by Sanger sequencing. The expression of the sgRNA is driven from the endogenous RNA polymerase III snR52 promoter and processed to produce a mature sgRNA. The sgRNA couples with enzymatically dead Cas9 (dCas9, red), expressed under the control of the TDH3 promoter off a plasmid, to block transcription. See Materials and methods for more detail. (B) A polyacrylamide gel visualised by UV shadowing reveals a high efficiency of the click reaction for all the HMS2 constructs (see Figure 2E and  Figure 2E). Note that there is considerable heterogeneity in the SUT650 TSS ( Figure 2-figure supplement 1) and thus in some cells, sgRNA AS+59NT may be binding upstream of the SUT650 TSS. AS+243NT/dCas9 significantly (p=0.011) reduced SUT650 ( Figure 2F lanes 1 and 3, 2G) and this repression was comparable for the clicked and control full-length synthetic constructs ( Figure 2G). However, there was no obvious effect of the other two sgRNAs/dCas9 on SUT650 levels ( Figure 2F lanes 1, 4 and 5, 2G). Consistently, we could only detect dCas9 binding to the site of sgRNA AS +243NT binding and not to the other two SUT650-targeting sgRNA binding sites tested (Figure 2figure supplement 2). CRISPR efficiency has been linked to nucleosome occupancy (Horlbeck et al., 2016b;Isaac et al., 2016), but we found no correlation between the level of repression and the nucleosome occupancy (Knight et al., 2014) at each of the SUT650 sgRNA target regions (Figure 2-figure supplement 3). However, other factors such as sequence determinants can also influence repression (Horlbeck et al., 2016a). These results at GAL1 and HMS2 confirm CRISPRi is suitable for reducing levels of antisense transcripts in yeast but controls are needed for each sgRNA designed to ensure that repression has been achieved.

CRISPRi repression of SUT650 induces a new shorter antisense transcript at HMS2
At HMS2, sgRNA AS+243NT/dCas9 was able to reduce SUT650 transcript levels significantly ( Figure 2F,G). Since SUT650 is a substrate for the major cytoplasmic 5'À3' exonuclease Xrn1 ( Figure 2F, lanes 1 and 2), we investigated whether the CRISPRi-induced SUT650 reduction was due to direct repression of SUT650 transcription and/or a reduction in SUT650 stability. Whilst SUT650 was not stabilised upon XRN1 deletion in the strain expressing AS+243NT/dCas9, supporting previous studies that CRISPRi operates via a transcriptional block  ( Figure 2F, lanes 3 and 6, transcript B), we detected a shorter antisense transcript ( Figure 2F, lane 6, transcript B t ). To map transcript B t , we used a series of Northern blotting probes across the locus ( Figure 2H). With antisense-specific probes H1 and H2 in the HMS2 gene promoter and 5' coding region, respectively, we could detect antisense transcript B t , indicating that it extends into the HMS2 sense promoter. However, probe H4 at the 3' end of the HMS2 gene was unable to hybridise to B t , indicating that B t initiates upstream of this site. So whilst CRISPRi successfully blocked the original SUT650, a new Xrn1-sensitive unstable antisense transcript (B t ) was created. This is potentially due to changes in the chromatin structure caused by AS+243NT/dCas9 binding to this region (Horlbeck et al., 2016b;Isaac et al., 2016) or by blocking SUT650, creating an environment that is permissive for transcription initiation (Murray et al., 2012). We note that using probe H4 we were also unable to detect SUT650 initiating from its WT start site but terminating at the AS+243 NT binding site, due to its small size (~243 nt).

CRISPRi-induced GAL1 AS repression does not affect the GAL1 sense transcript
We tested whether GAL1 AS repression by AS+112NT/dCas9 affected the GAL1 sense transcript.
Previously we mutated a TATA-like sequence to ablate antisense transcription but observed no difference in GAL1 sense transcript levels in the population under repressive or activating conditions (Murray et al., 2015). Using CRISPRi, we also observed no significant change in GAL1 sense transcript levels or size ( Figure 3A,B). Additionally, we observed no leaky expression of the GAL1 sense transcript under repressive conditions (glucose) when the GAL1 AS transcript was blocked To support a strand-specific transcriptional block of antisense transcription, GAL1 sense transcript polyA site usage was unaffected by sgRNA AS+112NT/dCas9 ( Figure 3C). Furthermore, XRN1 deletion in this strain also did not affect polyA site usage, ruling out a partial doublestranded transcriptional block and subsequent destabilisation of the resulting truncated sense transcript ( Figure 3C).

CRISPRi-induced SUT650 repression truncates the HMS2 sense transcript
Previous work shows that reducing SUT650 transcription increases HMS2 sense transcript levels (Nguyen et al., 2014). However, blocking SUT650 by AS+243NT/dCas9 did not similarly increase HMS2 sense levels ( Figure 3D, lanes 1 and 3). To our surprise, in addition to the full-length HMS2 sense transcript (A), we detected a considerably shortened HMS2 sense transcript (A t ). Since the combined levels of the truncated and full-length transcripts are similar to those in the strain not expressing an sgRNA, independent of the presence of XRN1 ( Figure 3D, lanes 1 and 3 or 2 and 6), we hypothesised that the AS+243NT/dCas9 complex bound within the HMS2 coding region may be causing partial premature sense transcription termination, leading to transcript A t . Thus, we mapped the 3' end of HMS2 S by Northern blotting using a series of strand-specific probes across the locus ( Figure 3E). Both transcripts A and A t could clearly be detected using the probes upstream of the AS+243NT/dCas9 binding site (probes H1, H2, H3), but the truncated transcript was undetectable using probe H4 downstream of this site. Thus, the transcriptional block caused by AS+243NT/dCas9 is not strand-specific and re-defines the transcriptional landscape at the HMS2 locus by creating a chromatin environment that is suitable for both transcription initiation and termination. To see if we could more directly target and block HMS2 sense transcripts, we designed three sgRNAs to bind the non-template strand adjacent to PAM sites near the HMS2 sense TSS (S+77NT, S+106NT, S +179NT) and monitored their effects on both HMS2 sense and antisense transcripts (Figure 3- CRISPRi-induced antisense repression at HMS2 and GAL1 is not as efficient as previous methods We compared the repression efficiency of CRISPRi with our previously used genetic interventions. At GAL1, mutation of a TATA-like sequence within the ADH1 terminator greatly reduced the level of the GAL1 antisense transcript ( Figure 4A,B) (Murray et al., 2015). Lowered GAL1 AS transcript levels could result from either reduced transcription or transcript destabilisation (or a combination of the two). To differentiate between these possibilities, we deleted the major cytoplasmic 5'À3' exonuclease XRN1. Since XRN1 deletion in the TATA mutant strain only slightly increased GAL1 AS levels, this suggests that most repression was at the level of transcription rather than altering transcript stability ( Figure 4B,C). By contrast, XRN1 deletion in the strain expressing AS+112NT/dCas9 did result in some upregulation of GAL1 AS ( Figure 4B,C), indicating that antisense transcript repression was not as great in this strain. Unlike at HMS2 ( Figure 2F,G), this stabilised GAL1 AS transcript was the same size as in the control and so likely represents stabilisation of the transcripts escaping repression rather than novel transcripts. Next we compared the efficacy of AS+243NT/dCas9 with previous experiments to ablate SUT650, where we replaced the entire HMS2 coding region, including the antisense TSS with the URA3 coding region (HMS2:URA3) (Nguyen et al., 2014). This successful block of SUT650 transcription increased levels of the HMS2 sense transcript, the di-cistronic HMS2-BAT2 transcript and consequently decreased levels of the downstream gene, BAT2, due to transcriptional interference ( Figure 4D-F). Whilst CRISPRi resulted in a similar decrease in SUT650 as HMS2:URA3, neither the HMS2, BAT2 nor HMS2-BAT2 di-cistronic transcripts were altered ( Figure 4F). The discovery that whilst AS+243NT/dCas9 represses SUT650, a new unstable HMS2 antisense transcript is induced and the sense transcript is prematurely terminated could explain why blocking SUT650 using CRISPRi and the URA3 gene body replacement strategies did not give the same results ( Figure 4F). Thus, CRISPRi is not as effective as a genetic mutation in reducing levels of either the GAL1 or HMS2 AS transcripts.

Concluding remarks
Although CRISPRi has been used to strand-specifically repress antisense transcription at GAL10 (Lenstra et al., 2015) and GAL1 (this work), here we demonstrate that this is not true at HMS2. This may reflect the vastly different transcription levels for the galactose-inducible genes (high) compared to HMS2 (low). Furthermore, a study in human K562 cells found that CRISPRi-induced repression did not correlate with which strand was targeted at 49 genes, although the mechanism behind this observation was not investigated (Gilbert et al., 2014).
This work extends our previous hypothesis (Nguyen et al., 2014), allowing us to propose that, rather than acting as a roadblock, the binding of the sgRNA/dCas9 complex at HMS2 creates a new chromatin environment, either directly or indirectly via the blocking of antisense transcription, which  (purple) showing the level of HMS2 sense (black arrowhead, transcript A) in the no sgRNA control (snR52::URA3 with dCas9) strain and strains expressing the indicated antisense-targeting sgRNAs. A truncated sense transcript (A t ) was also detected in the strain expressing AS+243NT. C and C' are read-through HMS2-BAT2 transcripts (see [Nguyen et al., 2014]). Positions of the rRNA are indicated by the short horizontal lines. Ethidiumbromide-stained rRNA is used as loading control. (E) Northern blots with a series of HMS2 sense-specific probes detecting the regions indicated in the schematic. The position of the sgRNA AS+243NT/dCas9 binding site is shown (green vertical line). Positions of the rRNA are indicated by the short horizontal lines. Ethidium-bromide-stained rRNA is used as loading control. DOI: https://doi.org/10.7554/eLife.29878.008 The following figure supplement is available for figure 3:  (Murray et al., 2015). In the TATA mutant construct, a TATA box within the engineered ADH1 terminator (T) is scrambled and the level of GAL1 antisense transcript is reduced. (B) A Northern blot probed for the GAL1 antisense transcript (black arrowhead) comparing the two methods of reducing antisense at this locus in the presence (top blot) or absence (bottom blot) of XRN1. All samples Figure 4 continued on next page is permissible for transcription initiation and termination. Thus, new transcription units are generated that result in novel sense and antisense transcripts of varying stabilities that are therefore not always detectable. This redefinition of the transcriptional landscape highlights the levels of controls and locus understanding needed before results from using CRISPRi can be interpreted.

Materials and methods
Key resource Continued on next page Figure 4 continued were run on the same gel and the blots were exposed to film for the same time. Samples were taken from cells grown in glucose (t = 0) and at the indicated times after transfer to galactose-containing media (min). Positions of the rRNA are indicated by the short horizontal lines. Ethidium-bromidestained rRNA is used as loading control. (C) Quantification of the Northern blot in (B) at t = 0. N = 2, errors represent the SEM, *p=0.026, n.s. nonsignificant. See Source data 1, tab 12. (D) A schematic of the HMS2-BAT2 locus, showing the HMS2 sense and antisense transcripts (A and B), the HMS2-BAT2 di-cistronic transcripts (C and C') and the BAT2 sense transcript (D). Cells switch between sense-and antisense-dominant states in which the indicated transcripts are produced (Nguyen et al., 2014). (E) The HMS2-BAT2 locus and associated transcripts in the wild-type (WT) strain and strains that repress the SUT650 transcript. In the HMS2:URA3 strain, the entire coding region of HMS2 is replaced by that of URA3. (F) Northern blots comparing the efficiency of SUT650 reduction by replacement of the HMS2 coding region with that of URA3 versus CRISPRi with sgRNA AS+243NT and the relative effects of these two strategies on the HMS2 sense and BAT2 transcripts shown in (E sgRNA design PAM sequences within 300 bp of the TSS in question were identified and the 20 bp immediately adjacent to these were used to design the variable regions of the sgRNAs. These sequences were run through an off-spotter algorithm (https://cm.jefferson.edu/Off-Spotter//) to minimise any off-target effects (Pliatsika and Rigoutsos, 2015). The sequence for the engineered constant region and SUP4 terminator were taken from Mali et al., 2013). The HMS2 sgRNA templates were synthesised as two DNA oligonucleotides to be joined by click chemistry, with the majority of the constant region on one oligonucleotide that could be used for all reactions. To increase the efficiency of the click reaction, the click linkage was designed between a dCpT dinucleotide within the Cas9 handle region. The sequences for the HMS2 and GAL1 sgRNA templates are shown in Supplementary file 1A. Full-length sgRNA templates were also synthesised for the GAL1 sgRNAs and as a control for the HMS2 click chemistry constructs (Eurofins Genomics, Germany).

Click chemistry
Single-stranded DNA oligonucleotides complementary to the constant and variable sgRNA regions were synthesised by standard phosphoramidite oligonucleotide synthesis with 3'-alkyne or 5'-azido modifications respectively (El-Sagheer et al., 2011) (Supplementary file 1A). The click chemistry reaction was performed with a Cu 1 catalyst and 24-30 nt splint DNAs . PCR was performed for second strand synthesis, amplification and incorporation of sequences homologous to the site of insertion in S. cerevisiae. PCR efficiencies were obtained using qPCR performed three times on a Corbett 6000 Rotorgene with four serial 10-fold dilutions in triplicate. Analysis was performed using Rotor-Gene Q Series Software (Qiagen, Hilden, Germany).

Strain construction
dCas9, without fusion to a transcriptional repressor domain, was expressed from a plasmid (# 46920 Addgene ) under the control of the TDH3 promoter. The DNA templates for sgRNA production were integrated in place of an exogenous URA3 cassette immediately downstream of the splice site in the endogenous snR52 locus using homologous recombination followed by 5-FOA selection. Correct insertion was confirmed by genomic DNA Sanger sequencing. SNR52 is a Pol IIItranscribed C/D box small nucleolar RNA (snoRNA) gene and so does not undergo extensive posttranscriptional processing such as capping and polyadenylation, and transcripts from this locus are retained in the nucleus. Additionally, the locus contains a self-splicing site that produces a mature transcript without additional machinery, allowing precise production of a mature sgRNA without any unwanted extensions. Primer sequences for strain construction are listed in Supplementary file 1B.

Yeast growth
The strains used in this study are listed in Supplementary file 1C. Strains were grown to mid-log at 30˚C in complete synthetic media lacking leucine (for dCas9 plasmid selection). For experiments studying the GAL1 locus, yeast were grown to mid-log in rich media (YP 2% D/YP 2% Gal) so that the GAL1 induction kinetics were similar to what we had observed previously (Murray et al., 2012;. dCas9 expression was unaffected by the temporary absence of plasmid selection (Figure 1figure supplement 1).

Yeast growth controls
Assessment of growth in liquid culture (CSM-leucine) was achieved using a Bioscreen spectrophotometer that automatically measures the optical density of cultures at 30˚C every 20 min for 24 hr. Doubling times were extracted from the gradient of the curves during logarithmic growth.

Northern blotting
Northern blotting was performed as before (Murray et al., 2012;Nguyen et al., 2014) using asymmetric PCR or in vitro transcription with T7 RNA polymerase to generate the radioactive strand-specific probes for GAL1 and HMS2, respectively. Primers for these reactions are listed in Supplementary file 1D. Northern blots were quantified using Fiji/ImageJ (Schindelin et al., 2012), and images were acquired using a FLA 7000 phosphorimager (GE Healthcare). Images in the figures are scans of exposures to X-ray film.

3' RACE mapping
Mapping of the 3' end of the GAL1 sense transcript was performed as (Nguyen et al., 2014). Primers are listed in Supplementary file 1F.
replicate three supporting Figure 2-figure supplement 2. All raw data can be found at