Heritability of On- and Off-Target LRR-XII Mutations

Seed was obtained from 22 of the T0 lines, and the alleles identified in T0 were observed segregating in T1. The results are summarized in Table I. New alleles were identified in T1 progeny derived from three lines (LRR-1, -11, and -23), indicating continued activity of the Cas9 nuclease. All other lines contained the same mutant alleles identified in T0, indicating that all wild-type alleles were mutated in the T0 generation and did not undergo additional mutations. Null segregants for the Cas9 T-DNA also were identified in the T1 generation. These transgene-free progeny contained the same alleles as the T0 parents and T1 siblings, further confirming the presence of heritable mutations.

Lines without mutations in T0 were checked for new mutations in T1, but no new on-target mutants were identified. However, two lines (LRR-11 and -23) contained off-target mutations in T1 that were absent in the T0 generation (Table I). LRR-11 SL2.50ch04 6829670 and LRR-23 Solyc02g070910 (Fls2.2; Rosli et al., 2013) had overlapping chromatograms only in T1 plants containing the T-DNA, indicating that continued activity of Cas9 resulted in the additional mutations. The LRR-23 Solyc02g070890 (Fls2.1; Rosli et al., 2013) gRNA has a single G-to-A mismatch at the corresponding Fls2.2 locus, 3 bp 5′ from the PAM site. No Fls2.2 mutations were recovered in Cas9-free T2 plants, even though their Cas9-positive T1 parent contained mutations, suggesting that the observed off-target mutations at Fls2.2 are somatic events.

A single mutant allele was detected in the T0 generation at five of the target loci, which suggested instant homozygosity at these targets. However, three of these putative homozygous loci could not be amplified by PCR from a subset of T1 individuals. By designing primers to amplify an ∼3-kb portion of the genomic region surrounding Solyc03g118330, a 2,112-bp deletion was detected in the LRR-4 T0 and T1 plants. In LRR-2 at the Solyc06g006040 target, a 609-bp insertion was observed only in T1 plants homozygous for the insertion, as the initial PCR conditions favored the smaller (4-bp deletion) product in heterozygous individuals. Several LRR-4 T1 individuals also failed to amplify Solyc04g014900, suggesting another allele that could not be observed. Despite numerous attempts to amplify around the targeted region, this allele could not be resolved. Therefore, true instant homozygosity appears to have occurred only at two of the targeted loci.

Unintended Targets

During mutation analysis, it became clear that the scoring parameters used to select the LRR-XII gRNAs incorrectly scored two of the incorporated gRNAs, leading to perfectly matched, unintended targets (LRR-11 Solyc04g014650 and LRR-6 SL2.50ch00:17387212). Both unintended targets had heritable mutations, one of which is in the LRR-XII gene family (Table I).

Errors in the reference genome also contributed to unintended targets. There are multiple contig breaks in the tomato reference genome near and within several of the LRR-XII gene family members. While the LRR-2 and LRR-6 gRNAs were specific to the reference sequence, perfect target sites are in fact present in the tomato genome. For the LRR-2 gRNA, sequence data from a BAC contig ({"type":"entrez-nucleotide","attrs":{"text":"AC254430","term_id":"491550638","term_text":"AC254430"}}AC254430) revealed three perfect target sites, one in Solyc06g006020 and two that are ∼300 bp upstream of the predicted start codons of Solyc06g006050 and Solyc06g006070 (Supplemental Fig. S1). Sequence analysis of each of these sites in the LRR-2 line indicated that heritable mutations were present at each of these unintended targets. Interestingly, the 609-bp insertion at the Solyc06g006040 target site is an inversion of part of the first intron of Solyc06g006020. The exact alleles produced at all the target sites have not been completely resolved. As these four genes occur on a single cluster on chromosome 6, it is likely the simultaneous DSBs resulted in larger chromosomal alterations that prevent PCR amplification. LRR-6 also contains a mutation in a locus adjacent to the target gene Solyc06g006020 (Table I). Given the similarity between the two sequences, it is possible that this locus was produced by a tandem duplication. However, it is unclear if this locus represents an unidentified LRR-XII family member or is a pseudogene.

Fls2.1 Is Likely Required for Age-Dependent PTI in Tomato

There are two tandemly duplicated homologous genes in tomato of the Arabidopsis flg22 receptor gene FLS2. The line LRR-23 contains a 2-bp deletion in one of these genes, Fls2.1, resulting in a premature stop codon in the third LRR domain. A single 2-bp deletion was observed in T0 as well as all 19 T1 plants sequenced, indicating a true homozygous event. As described above, somatic mutations were observed in the off-target site on Fls2.2 in T1 plants containing Cas9. All 29 T1 plants tested failed to respond to flg22 in a reactive oxygen species (ROS) assay, while the response to flgII-28, which is conferred by Fls3, was maintained (Fig. 2A; Supplemental Fig. S2A). When vacuum inoculated with Pseudomonas syringae DC3000∆avrPto∆avrPtoB, there were greater disease symptoms (Fig. 2B) and an approximately 1-log increase in bacterial growth in T1 LRR-23 plants compared with the M82sp+ wild-type control (Fig. 2C; Supplemental Fig. S2, B and C). The difference in bacterial growth was greatest, but more variable, in younger leaves, as has also been observed in Arabidopsis fls2 mutants (Zipfel et al., 2004). Together, these data indicate that Fls2.1 plays the primary role in the response to flg22 in tomato leaves.

Aberrant Phenotypes in the LRR-XII Lines

Six T1 lines had obvious segregating aberrant phenotypes. A subset of T1 progeny from LRR-1 had thin, spindly plants with long internodes and failed to produce flowers (Supplemental Fig. S3, A and B), plants from LRR-4 terminated after the fourth true leaf with wiry leaflets (Supplemental Fig. S3, C and D), LRR-8 had serrated cotyledons (Supplemental Fig. S3E), LRR-11 had pale seedlings that usually died a few weeks after germination or remained dwarfed (Supplemental Fig. S3F), and LRR-14 had anthocyanin-free plants (Supplemental Fig. S3G). Segregation analysis suggested that these characteristics were simple recessive traits (data not shown). However, none of the observed mutations cosegregated with these phenotypes (Supplemental Fig. S3H). Off-target mutations were checked in lines LRR-1, -4, -8, -11, and -14. Line LRR-14 did not contain any mutations at the target locus in the T0 or T1 generation. While line LRR-11 did contain off-target mutations (Table I), none of the mutations segregated with the pale/dwarf phenotype. Because of the stunted, bleached phenotype in this line, mutations were checked at each of the three PDS target sites included in the gRNA library, and no mutations were found. The presence/absence of the T-DNA also was evaluated, but again, there was no association of the T-DNA insertion with the aberrant phenotypes (Supplemental Fig. S3H).

We considered the possibility that some of the aberrant phenotypes in the LRR lines could be explained by transient T-DNA mutations (i.e. transient expression of the T-DNA, without genomic integration), as we observed in the transporter lines (see below). As transient mutations could occur at least at any of the 165 target loci, spread across 55 genes that are approximately 3 kb in length, we opted to use the long read length of single molecule, real time (SMRT) sequencing to screen for these mutations. Fifty-eight amplicons covering the 54 LRR-XII genes and four potential off-target genes were sequenced from T0 and T1 plants in LRR-1, -4, -11, and -14. The T1 plants selected for sequencing showed the respective aberrant phenotypes and, therefore, should be homozygous for the mutated genes. Across all samples, at least five aligning circular consensus sequences (CCS) were obtained for 42 to 55 of the 58 amplicons (Supplemental Table S3). All but one of the previously identified mutations were detected, indicating that this custom amplicon-sequencing approach was suitable for identifying mutations (Supplemental Table S4). While we did observe some candidate mutations at gRNA target sites, none were homozygous in the T1 individuals and are likely false positives. Based on these observations, it seems most likely that the aberrant phenotypes are due to somaclonal variation resulting from the transformation and tissue culture processes (Larkin and Scowcroft, 1981; van den Bulk et al., 1990).

Higher Order Library

Having demonstrated that transformation with a pooled CRISPR library could generate a targeted collection of mutant lines, a second CRISPR library was designed. To increase throughput, plasmids containing three gRNAs were constructed. The library also was limited to 18 genes, to increase the probability of obtaining multiple independent mutations in each of the target genes, and with two gRNAs per gene to reduce the cost of oligonucleotide synthesis. The 18 targeted genes encode putative transporters, and we reasoned that some of these may be susceptibility factors, as their expression is suppressed by the plant during PTI (perception of flgII-28; Rosli et al., 2013) and induced by the pathogen effectors AvrPto and AvrPtoB, potentially to promote access to water and/or nutrients.

For library construction, the 36 gRNAs were divided into three pools with six genes targeted per pool. Three separate DNA assembly reactions were used to generate the pools, and the reactions were used as templates for PCR with gRNA-position-specific primers (Fig. 3A). During initial construction attempts, it became apparent that the PCR amplification step was biasing the relative abundance of the gRNAs. Therefore, a lower number of PCR cycles (14) was used to minimize this bias. The final assembly was cloned into E. coli at 118× coverage (∼1,400 clones), introduced into A. tumefaciens, and transformed into the tomato genotype Rio Grande-PtoR.

Numerous Combinations of Mutated Genes Were Obtained

Fifty-nine unique events (transporters 1–59) were obtained with the transporter library (Fig. 3B). Three events did not contain mutations or the T-DNA and are likely nontransformed escapes. Thirty-five of the 36 gRNAs were observed in the T0 plants (Supplemental Table S5). The distribution of the gRNAs was much greater than in the LRR-XII library, likely due to the PCR amplification step during library construction. For every event, all of the 36 targeted loci were amplified and Sanger sequenced. Forty-two events (71%) contained mutations, with each event containing a unique set of mutant-gene combinations. Fifteen out of the 18 transporter genes were mutated at least once, and 14 of the genes were mutated at least three times (Fig. 3B; Supplemental Table S5). Seven of the events contained T-DNAs with single gRNAs, with each of these coming from the third gRNA position and none resulting in mutations. Ten (4%) of the incorporated gRNAs had errors in the protospacer that likely prevented mutations at their target sites.

There was a wide range of editing efficiencies for the different gRNAs. Sixteen of the incorporated gRNAs had no mutations at all, although seven of these were only counted two or fewer times, and their failure to induce mutations may be attributed to a small sample size. Approximately half of the total mutations were derived from gRNAs in the first position, 31% from the second, and 17% from the third (Supplemental Table S5), suggesting that there may be a position effect within the gRNA array. Two genes that were not mutated, Solyc06g060620 and Solyc08g066940, had 11 and 16 gRNAs, respectively, suggesting that these are ineffective gRNAs. Alternatively, there might be selection against mutations in these genes. Thus far, 21 T1 lines have been checked, and heritable mutations have been confirmed in 19 of them (Supplemental Table S6).

Transiently Expressed T-DNA-Induced Mutations

Ten of the 59 events contained mutations at gRNA target sites in which the respective gRNAs were not observed after sequencing the incorporated gRNAs (Fig. 3B). All mutations were located between the third and fourth nucleotides upstream of the PAM, a hallmark of Cas9-mediated gene editing. These results suggest that transient expression of the T-DNA, without genomic integration, led to these mutations. Such possible transient T-DNA mutations accounted for 23 of the 93 mutations (25%; Supplemental Table S5). For example, no gRNAs could be amplified from transporter-5, -10, and -11, but mutations were detected in each of them and in the T1 progeny of transporter-5 and -11. Since rearrangements of the incorporated T-DNA might prevent amplification of the gRNA amplicons used for sequencing, and therefore, detection of gRNAs, PCR with primers designed to amplify a minimal portion of the gRNA (227 bp) was performed. Again, these three events tested negative (data not shown). These events also lack the two Cas9 amplicons, and transporter-10 and -11 failed to amplify the nptII selectable marker, suggesting that these events are escapes. The seven other transient T-DNA events contained other gRNAs, with and without mutations at their target genes.

Transient Expression

The use of pooled A. tumefaciens strains permitted the identification of genes that were mutated during apparent transient expression of the T-DNAs. Other groups have reported successful targeted mutations without the incorporation of DNA or RNA with biolistic delivery methods or with biolistic or protoplast transfection of ribonucleoprotein complexes (Woo et al., 2015; Svitashev et al., 2016; Zhang et al., 2016; Liang et al., 2017). The observations of transiently expressed T-DNA leading to CRE recombinase activity (Ghedira et al., 2013) and Cas9-induced DNA mutations reported here and by others (Iaffaldano et al., 2016) indicate that A. tumefaciens-mediated delivery of T-DNA is another viable means of obtaining transgene-free gene-edited plants. This may be useful for plant species that are better suited to A. tumefaciens-mediated transformation and regeneration over biolistic or protoplast approaches. It will be interesting to determine the rate of mutagenesis in the absence of selection or if selection against events with incorporated T-DNAs will be needed. Alternatively, lines with a knockout in polymerase Θ could be utilized to produce plants resistant to stable integration but still support transient T-DNA expression of nucleases (van Kregten et al., 2016).

However, the presence of transient T-DNA modifications does present a limitation to the pooled library approach. A library with a large number of gene targets may be desirable under certain conditions (e.g. species, phenotype, transformation efficiency, etc.), but any candidate lines would need to be checked at every target gene because sequencing the incorporated gRNAs would not identify all possible mutation sites. In the transporter library, it would not have been possible to so rapidly associate Solyc04g005070 with the aberrant phenotype observed in transporter-10 and -38 if only the incorporated gRNAs were considered. Such transient T-DNA mutations may not be a problem in all A. tumefaciens-mediated plant transformation systems where transient effects are limited (Ghedira et al., 2013).

Practical Considerations

Genome-wide CRISPR libraries are routinely used in animal cell culture systems (Shalem et al., 2015). While it is tempting to extend the pooled A. tumefaciens approach presented here to a genome-wide screen in plants, there are a number of technical and practical limitations that would need to be overcome: namely, the occurrence of transient T-DNA mutations. If every potential gRNA target site needs to be checked in every plant, a significant amount of time and money would be required for each plant. However, if transient T-DNA modifications could be eliminated, or reduced, only targets identified by the incorporated gRNA(s) would need to be checked. This issue also may be overcome by using a sufficiently large population or if sequencing costs are minimal.

Identifying incorporated gRNAs becomes a laborious task when multiple T-DNAs are incorporated into an event or the same Pol III promoter is repeated, as was done for the transporter library. Several rounds of cloning and Sanger sequencing were required, and we were still unable to confidently identify all gRNAs in two of the T0 plants (Supplemental Table S7). A gRNA amplicon-sequencing approach would be more practical for large numbers of events.

While CRISPR systems are generally quite efficient in plants, the results from the transporter library clearly show that there is room for improvement in gRNA and/or vector design. Optimization of library preparation by further reducing the number of PCR amplification cycles for a more even gRNA distribution would likely have improved the outcome. Certainly, the use of more efficient gRNAs also would be an advantage: 13 of the gRNAs resulted in no mutations at all, and many of the most abundant gRNAs had mutation rates lower than 20%. However, despite the variety of gRNA scoring algorithms available for animal systems (Haeussler et al., 2016), there are currently no plant-validated gRNA activity predictions available.

There also is an apparent negative correlation between gRNA position and gRNA efficiency. In position 1, the gRNAs had an average efficiency of 40%, in position 2, 32%, and in position 3, 14%. This trend also is consistent for transient T-DNA mutations: 45% of all transient mutations came from position 1, 36% from position 2, and 18% from position 3. In all, 51% of all mutations came from the first position (Supplemental Table S5). Multiplexing with CRISPR arrays using identical Pol III promoters is a common strategy (Belhaj et al., 2013; Endo et al., 2016; Vazquez-Vilar et al., 2016), but none have reported a decrease in efficiency associated with gRNA position in the array. It is not immediately clear why the repeated use of a Pol III promoter would lead to a decrease in efficiency across the array. The gRNA array appears to be stably integrated, as all but two of the events (transporter-13 and-27) had the same number of gRNAs in each pool. If gene silencing were the cause, then the effects should be spread across all gRNA cassettes. The use of different Pol III promoters in a gRNA array might improve the mutation efficiency and would be a benefit when amplifying the gRNAs from an array, as a PCR across the entire array results in a ladder of products and was part of the reason why identifying the gRNAs was challenging. It will be worthwhile to test a number of hypotheses from this observation, as the use of multiplexing CRISPR reagents will surely expand in the future.

The quality of the reference genome is an important consideration when setting up a CRISPR screen. In plants, LRR genes are well known to occur in clusters of tandemly duplicated genes, which can make their assembly and annotation difficult (Andolfo et al., 2014). We experienced this issue as the guide in LRR-2 has (at least) four targets in a single cluster on chromosome 6. These unintentional targets would have been overlooked if not for the insertion of the Solyc06g006020 intron into Solyc06g006040.

Tomato Genotypes and Transformation

Tomato transformation was performed at the Boyce Thompson Institute transformation facility as described previously (Frary and Van Eck, 2005). The seed used for transformation was a mixture of M82 (determinate) and M82sp+ (indeterminate; Krieger et al., 2010). The primers sp175F and sp1088R (Supplemental Table S8) were used to amplify Solyc06g074350, and lines with sequences identical to the reference were scored as M82 and those differing were scored as M82sp+ (Table I). The genotypes were confirmed by growth habit in the greenhouse. The transporter library was transformed into the tomato genotype Rio Grande-PtoR by the plant transformation facility at the University of Nebraska. The E. coli library was used in a triparental mating to obtain a library of A. tumefaciens. Rescued plasmids from 10 clones were sequenced to ensure that no plasmid was overrepresented.

Genotyping and Mutation Analysis

DNA was extracted from tomato leaves using a modified CTAB method (Murray and Thompson, 1980; 20 g L⁻¹ CTAB, 100 mm Tris, 81.82 g L⁻¹ NaCl, and 2 mm EDTA, pH 6). Primer sequences can be found in Supplemental Table S8. PCR was performed with either GoTaq master mix (Promega) or Apex taq red Master Mix (Genesee Scientific) in 10-µL reactions using the recommended PCR conditions. Sanger sequencing was performed at the Cornell Biotechnology Resource Center. Chromatograms were aligned to target sequences in Geneious. Overlapping chromatograms at the cleavage site indicated a potential mutation, and the online program TIDE was used to estimate mutations and allele frequencies (Brinkman et al., 2014).

Oxidative Burst Assays

ROS assays were performed as described previously (Clarke et al., 2013) with slight modifications. Briefly, 100 nm of each peptide was used to elicit the ROS response. Four leaf discs per plant were used for flg22 treatment, and two leaf discs per plant were used for water and flgII-28 controls.

Plant Inoculations

Bacterial growth and disease assays were performed as described previously (Mathieu et al., 2014) with slight modifications. Pseudomonas syringae pv tomato DC3000∆avrPto∆avrPtoB was grown for ∼24 h on Kings B solid medium supplemented with rifampicin. Bacteria were scraped from the plate and resuspended in sterile 10 mm MgCl₂ and diluted to OD₆₀₀ = 0.35. Three milliliters of this suspension was used to inoculate 3 L of 10 mm MgCl₂ containing 0.002% Silwet L-77, resulting in a solution of bacteria at 3.5 × 10⁵ CFU mL⁻¹. Plants were submerged in the bacterial solution, a vacuum of −80 kPa was applied for 2 min, and plants were allowed to dry standing up before being transferred to a growth chamber. Visual inspection of disease and photographs were taken 3 d after infiltration. For bacterial growth assays, three independent experiments were performed over 1 month. For each experiment, six plants per genotype of approximately equal size were used. A lower inoculum of 3.5 × 10⁴ CFU mL⁻¹ was used, and 2 d after infiltration, three leaf discs were taken from each leaf with a 6-mm-diameter cork borer for the youngest leaves and an 8-mm-diameter cork borer for the other leaves. Leaf samples were ground in 1 mL of sterile 10 mm MgCl₂, and serial dilutions were plated to count bacterial colonies.

Boric Acid Treatments

A solution of 100 µm boric acid (Sigma-Aldrich) was sprayed daily on T0 events transporter-3 and -29 such that the leaves were completely wet with the solution.

SMRT Sequencing

Primers for the amplicons used for SMRT sequencing can be found in Supplemental Table S8. Amplicons were designed to be 3 to 5 kb in length and validated by Sanger sequencing and alignment to the reference sequence. Each gene-specific forward and reverse primer was appended with a 5′ block and the PacBio Universal Sequences (forward primer, 5′-GCAGTCGAACATGTAGCTGACTCAGGTCAC-3′; and reverse primer, 5′-TGGATCACTTGTGCAAGCATCACATCGTAG-3′). Approximately 10 ng of genomic DNA was used as a template for PCR and amplified in 10-µL reactions with 2x HotStart Ready Mix HiFi polymerase (KAPA Biosystems) and 0.2 µm of each primer under the following conditions: 95°C for 3 min; 32 cycles of 98°C for 20 s, 62°C for 30 s, and 72°C for 4 min; followed by 72°C for 5 min. Each amplicon was amplified separately and confirmed by electrophoresis. Within each line, approximately equal amounts of amplicons were pooled, and products were column purified. A second PCR was used to add indexing primers (PBbarcode; Supplemental Table S8) with universal sequences (Full-Length 16S Amplification, SMRTbell Library Preparation and Sequencing; Pacific Biosciences) with 200 pg of purified amplicons under the following conditions: 20-μL reactions at 95°C for 3 min; 15 cycles of 98°C for 20 s, 62°C for 30 s, and 72°C for 4 min; followed by 72°C for 5 min. Amplification was verified by agarose gel and column purified. All amplicons were then pooled in equal concentrations and sent to the Icahn School of Medicine at Mount Sinai for sequencing. After sequencing, reads were demultiplexed, and approximately 1,000 CCS reads were generated per line. The reads were aligned to the 58 target amplicons using the Geneious function Map to Reference using the default conditions under the sensitivity setting High Sensitivity/Medium. Alignments were inspected manually for mutations at all target sites.

We thank Brian Bell, Jay Miller, Nick Vail, Jacob Wszolek, and Lesley Middleton for plant care; Laurens Pauwels for helpful comments on the article; Hernan Rosli for RNA-seq data analysis; Noe Fernandez-Pozo and Surya Saha for bioinformatics support; Joyce Van Eck and Patricia Keen for the LRR-XII library transformation; Thomas Clemente and Shirley Sato for the transporter library transformation; Robert Sebra, Diane Castillo, and Leah Newman for assistance with the design, sequencing, and analysis of the SMRT amplicons; Konrad Thorner and Samantha Mainiero for technical help; and Simon Schwizer, Christine Kraus, and Sarah Hind for technical advice.