The S. tuberosum clone DRH195 and the C. sativa cultivar Suneson were used in this study. DRH195 is a diploid S. tuberosum Phureja F1 derived from a cross between a homozygous doubled monoploid DM 1-3 516 R44 (DM) and a heterozygous diploid RH89-039-16 (RH) (Pham et al., 2020; Zhou et al., 2020). Potato plants were propagated in vitro using nodal cuttings in tissue culture on Murashige and Skoog (MS) medium (MS basal salts plus vitamins, 3% sucrose, 0.7% plant agar, pH 5.8) (Murashige and Skoog, 1962) and grown in culture tubes in growth chambers at 22°C with an average light intensity of 200 μmoles m^-2 s^-1 under a 16h photoperiod. Nine Suneson FAD2 Cas9 edited plants were obtained from a previous study (Jiang et al., 2017); seven empty-vector control camelina lines were generated in this study ( Supplementary Table S1 ). Camelina sativa cv. Suneson was grown in a growth chamber from seed at temperatures of 22/18°C (day/night), 40% relative humidity with a light intensity of 300 µmol m^-2 s^-1 under a 16h photoperiod.

Genome assemblies and annotations for DM 1-3 516 R44 (DM) and RH89-039-16 (RHv3) were retrieved from Spud DB (Pham et al., 2020; Zhou et al., 2020; http://spuddb.uga.edu/). Whole genome sequencing data for DRH195 was retrieved from the Sequence Read Archive (SRA) of the National Center for Biotechnology Information (NCBI) under SRR4018191.

MUMmer v4.0.0’s nucmer function was used for global nucleotide alignments with the following configuration: -c 100 (Marçais et al., 2018). Global alignments were filtered using MUMmer’s delta filter to remove alignments less than 20,000 bases. SNPs between alignments were collected using MUMmer’s show-snps filtering to remove SNPs from ambiguous alignments with -C. SNPs were filtered further in R version 4.3.0 using dplyr version 1.1.2 (Wickham et al., 2023) to remove missing values and identical variants between RH haplotypes producing the final set of RH haplotype specific variants.

Whole genome shotgun reads were cleaned using Cutadapt v2.1 (Martin, 2011) to trim low-quality regions using a minimum base quality of 20 and a minimum read length of 100 bp. Picard v2.18.27 was used to convert cleaned fastq reads into an unmapped BAM using FastqtoSam and adapter sequences were marked using Mark Illumina Adapter and SamToFastq, with CLIPPING_ATTRIBUTE=XT and CLIPPING_ACTION=2 (https://github.com/broadinstitute/picard). Genomic reads were mapped to the DM reference assembly in paired-end mode, flagging secondary hits (-M), using BWA-MEM v0.7.17 (Li, 2013). Alignments were filtered to only retain properly paired reads and alignments to chromosomes 1-12 using SAMtools’ v1.7 view command (Li et al., 2009). Picard’s MergeBamAlignment was used to set metadata as well as allow for any number of insertion or deletion mutations by setting MAX_INSERTIONS_OR_DELETIONS = -1. Duplicate reads were marked using Picard’s MARKDuplicates. Reads surrounding insertion/deletions were identified and realigned using GATK’s v3.8.1 RealignerTargetCreator and IndelRealigner, respectively (McKenna et al., 2010). GATK’s Haplotypecaller v4.1.4.1 was used to call variants using default configuration. Variants were flagged using GATK v4.1.4.1 Variant Filtration using with the following expression: QD < 2.00 & MQ < 50.00 and flagged variants were removed using SelectVariants -exclude. Variants were filtered further in R version 4.3.0 using dplyr version 1.1.2 to retain variants overlapping the RH haplotype specific variant set and removing variants that did not match either of the two RH haplotype variants. A sliding window of 20 variants with 80% congruence was used to assign RH haplotype bins.

A custom script using Biopython version 1.79 in Python v3.10.4 was used to construct the DRH195 assembly (Cock et al., 2009). A gene annotation set was created for DRH195 using LiftOff version 1.6.3 (Shumate and Salzberg, 2021). Ideograms were created in R version 4.2.0 using the package chromPlot (Oróstica and Verdugo, 2016).

Members of the polyphenol oxidase gene were identified by aligning previously annotated StPPO1-9 protein sequences in DMv3.4 from Chi et al. (2014) using BLASTP version 2.10.0+ with at least 90% sequence homology (The Potato Genome Sequencing Consortium, 2011; Altschul et al., 1997). Gene family members were assigned using phylogenetic inference with the Maximum-Likelihood method in MEGA X (Tamura et al., 2021).

For potato, a double sgRNA construct was assembled into the pTRANS_220d binary vector using modular assembly as described by Čermák et al. (2017). We designed two sgRNAs in the ORF in conserved regions of the StPPO gene family in potato. Based on the potato expression data, four StPPOs are expressed in tubers including StPPO1-4, of which, StPPO2 had the highest expression. Two single guide RNAs sgRNA1: CGCTTTGCCATATTGGAATTGGG and sgRNA2: AACACTAATGTACCGTCAAATGG were designed to target StPPO1, StPPO2-1 and StPPO3 using the CRISPR RGEN tools (Park et al., 2015). The two sgRNAs were cloned into pTRANS_220d using modular assembly (Čermák et al., 2017). A protoplast transient assay (Nadakuduti et al., 2019) was used to test the in vivo sgRNA editing activity which indicated that only sgRNA2 was active. For camelina, pTRANS_220d was modified to include DsRed2 which was used as the empty vector control. Vectors used for modular assembly and empty vector controls (https://www.addgene.org/browse/article/28189956/) were gifts from Dr. Daniel Voytas (University of Minnesota).

For potato, binary vectors were electroporated into Agrobacterium tumefaciens strain GV3101 pMP90 (Koncz et al., 1994). Agrobacterium-mediated transformation was performed using leaf and internode explants from four-week-old tissue culture plants as described previously (Jayakody et al., 2023). Transformation events (T0 lines) were selected and transferred to MS medium supplemented with 250 mg/l cefotaxime, 300 mg/l timentin and 50 mg/l kanamycin for rooting and selection. For camelina, floral dip transformation using vacuum infiltration of floral buds was performed according to Lu and Kang (2008).

For potato and camelina, DNA from transformation events was isolated from young leaves using the DNeasy Plant Mini Kit (Qiagen, Hilden, Germany). PCR for screening T-DNA insertion was carried out using the GoTaq DNA polymerase (Promega, Fitchburg, WI, United States) using primers designed to amplify an 853 bp region of Cas9 ( Supplementary Table S2 ) with the following thermocycler conditions: one cycle of initial denaturation for 3 min at 95°C, followed by 34 cycles for 30 s at 95°C, 45 s at 60°C and 1 min 30 sec at 72°C and a final extension of 5 min at 72°C.

PCR amplification of StPPO1, StPPO2-1 and StPPO3 for sequencing was carried out using the NEB Q5 DNA polymerase (New England Biolabs, Ipswich, MA, United States) using primers described in Supplementary Table S2 . PCR products were purified using QIAquick PCR Purification kit (Qiagen, Hilden, Germany) and sequenced at the Michigan State University Genomics Core. Chromatograms were analyzed for presence of indels near the target site using Synthego’s ICE CRISPR Analysis tool (Synthego Performance Analysis, 2019).

Deep sequencing of PCR amplicons was conducted on an Illumina MiSeq v2 Nano flow cell in a 2x250 nt paired-end format using amplicon sequencing primers described in Supplementary Table S2 . Paired end reads were trimmed using Cutadapt v2.1 to remove adapters and bases with a quality score less than 20. Paired reads were joined using BBMaps’s BBMerge program (Bushnell et al., 2017). Joined reads were aligned to both haplotypes of chromosome 8 from DRH195 using BWA-MEM, marking secondary alignments. Alignments were filtered to retain only primary alignments to the amplicon’s respective target using SAMtools v1.7. Retrieved reads were analyzed using CasAnalyzer (Park et al., 2017) with a comparison range of 100, minimum frequency of 25 and a 10 base WT marker. Sanger and deep sequencing were conducted by the Michigan State University Genomics Core. Multiple sequence alignments were visualized in MEGAX using the CLUSTAL algorithm for alignment (Tamura et al., 2021).

Camelina FAD2 CRISPR/Cas9 knockout events are described previously (Jiang et al., 2017). Briefly, F1 (sgRNA: GTCCAGTTTGTCCTCGGGTGG), R1 (sgRNA: CCACCGCAGTGTTTCAAACGCTC) and R2 (sgRNA: CCTCCCTCAGCCTCTCTCTTAC) events are either T5 or T6 generation events derived from T0 plants that had been independently transformed via agrobacterium-mediated floral dip transformation ( Supplementary Table S1 ). In each lineage, a different site homologous to all CsFAD2 homeologs was targeted by CRISPR/Cas9. The edits in the T5 and T6 events were confirmed by amplifying the F1, R1 and R2 target sites, respectively using NEB Phusion polymerase with primers listed in Supplementary Table S2 and then digested with AvaI, BtsI and BbvCI restriction enzymes, respectively, in addition to Sanger sequencing for some lines.

For potato, tissue was collected from leaves at 12-14 weeks from T0 events. Genomic DNA from potato and camelina events ( Supplementary Table S3 ) was isolated using the DNeasy Plant Mini Kit (Qiagen, Hilden, Germany). Whole genome shotgun sequencing (WGS) libraries were prepared and multiplexed using PerkinElmer NEXTFLEX Rapid XP DNA-Seq kit, then sequenced on an Illumina NovaSeq 6000 in paired-end mode by the Texas A&M AgriLife Research Genomics and Bioinformatics Service generating 150 nt reads ( Supplementary Table S3 ).

Whole genome sequencing reads were processed and aligned as described previously. GATK Haplotypecaller v4.1.4.1 was used to call variants using default parameters. Variants with the following characteristics were removed using GATK v4.1.4.1 Variant Filtration and SelectVariants: QD < 2.00, MQ < 50.00, DP<4, DP>50, AD <4. Variants overlapping controls were removed using BEDTools v2.3.0 subtract (Quinlan and Hall, 2010). Upset plots were created in R version 4.2.0 using the package UpSetR (Conway et al., 2017). Off-target sites were identified using Cas-offinder v2.4.0 allowing up to five mismatches in the spacer sequence for canonical and non-canonical PAM sites (Bae et al., 2014). SPAdes v3.15.5 was used for de novo assembly (Prjibelski et al., 2020).

To facilitate WGS analysis for detecting off-target gene editing in potato, a synthetic genome assembly was constructed for DRH195. First, a set of variants that could discern the two haplotypes in the heterozygous RH clone were identified by aligning chromosomes from both haplotypes independently to the respective chromosome in DM. SNPs between alignments were identified and filtered to retain unambiguous variants at shared locations that were unique to each RH haplotype. Then, to assign the RH haplotype inherited in DRH195, alternate alleles called from DRH195 WGS aligned to DM were compared to the RH haplotype specific variant set. Haplotype bins were assigned using a sliding window of 20 variants with 80% congruency.

A synthetic chromosome scale assembly for DRH195 was then constructed by assuming all chromosomes from the homozygous DM were inherited. For the RH haplotype, the assigned haplotype sequences were extracted from the RH genome assembly. Phasing of the haplotype inherited from RH uncovered recombination events on chromosomes 1,2,5,7,8 and 9, while the remaining chromosomes retained the entirety of one of either RH haplotypes ( Figure 1 ).

A previous study annotated all members of the polyphenol oxidase (PPO) gene family present in potato through a genome wide survey using the DMv3.4 reference assembly (Chi et al., 2014). The protein models designating StPPO1-9 from that study were used for phylogenetic inference to assign the homologous sequences in DMv6.1 and RHv3 ( Supplementary Table S4 ). This uncovered an additional PPO-like sequence in DMv3.4 that had not previously been described that was most like StPPO5 in sequence (StPPO5-2). The more contiguous DMv6.1 assembly revealed one additional StPPO3 and three additional StPPO7 copies that were not present in the DMv3.4 reference genome.

A single copy of StPPO1-9 was identified for each gene family member in haplotype 1 of the RHv3 assembly. The second haplotype of RHv3 contains two full length copies of StPPO2 like sequence in addition to two truncated StPPO2 like sequences, however, StPPO1, StPPO5, StPPO6 or StPPO7 like sequences were absent ( Supplementary Table S4 ; Figure 1 ). The region on chromosome 8 where the StPPO1-8 are present on RH haplotype 2 was inherited in DRH195 ( Figure 1 ). Although only one StPPO1 like sequence was identified in DMv6.1 and none in the RHv3 haplotype 2, chromatograms from sequencing StPPO1 in WT DRH195 indicated the presence of a second allele ( Supplementary Figure S1 ). To recreate the entire open reading frame of the second StPPO1 like sequence, a consensus sequence was created using the alternate alleles called from WGS of WT DRH195 aligned to StPPO1 in the DM assembly. This sequence retains the identical sgRNA2 target site and PAM sequence.

DRH195 was genetically transformed using CRISPR constructs with sgRNAs targeting StPPOs and the empty vector control. Kanamycin resistance was used for selecting transgenic events and PCR amplification of Cas9 was used to confirm T-DNA integration. Only events with clear kanamycin resistance and PCR confirmation were selected for further analysis. Five T0 empty vector control and 27 T0 events from CRISPR construct were generated for potato in this study. However, only 7 of the 27 T0 events had confirmed insertion/deletion mutations in at least one StPPO target site ( Supplementary Table S5 ).

Both alleles of StPPO1, StPPO2-1, and StPPO3 are targeted by sgRNA2 in DRH195 and were screened for mutations in T0 plants ( Figures 2A–C ). Given this abundance and the sequence homology between StPPO gene family members in potato, a semi-nested PCR was used to differentiate StPPO paralogs and alleles from each other. The first round of PCR used primers specific to both alleles of StPPO1, StPPO2-1 or StPPO3 in DRH195 ( Supplementary Table S2 ). To facilitate Sanger sequencing of amplicons, the second round of PCR tagged the M13 forward sequence to the 5’ end of the forward primer which was located ca. 300 bp upstream from the predicted sgRNA2 edit site. Indels were identified at target sites using Synthego’s ICE algorithm which identifies the positions where Sanger traces are discordant between wildtype and mutated amplicons (Conant et al., 2022; Synthego Performance Analysis, 2019).

To verify allelic representation, the same purified PCR products for Sanger sequencing were deep sequenced using Illumina sequencing. Chimeric PCR products indicative of mixed template amplifications was observed in events DRH195_1c StPPO1 and DRH195_6b StPPO3 from amplicon sequencing ( Figures 2A, C ). Chimeric PCR products were also seen in wild-type amplicon controls supporting that these are PCR artifacts and not recombination events. No T0 event had an edit in all alleles of the targeted StPPOs. In all cases except for event DRH195_7b, only one allele of one paralog was edited. Mosaic edits were observed in at least one allele of StPPO1 and StPPO3 for DRH195_7b ( Table 1 ). Although ICE could not differentiate between alleles, the predicted indel sizes were congruent with the results from deep sequencing ( Table 1 ).

A single copy of CsFAD2 is present on each of the homoeologous chromosomes 1, 15 and 19 designated by the following Suneson gene model IDs Camsa.SUN.01G012720.1, Camsa.SUN.15G013420.1 and Camsa.SUN.19G013580.1, respectively. Jiang et al. (2017) designed three independent sgRNAs for F1, R1 or R2 events, each of which targeted the three homeologs of FAD2. Suneson was transformed with the same empty vector control construct as potato but modified to include DsRed2 marker for seed selection ( Supplementary Table S1 ). PCR amplification of DsRed2 was used to confirm T-DNA integration and PCR positive events were phenotyped for DsRed-positive seed and selected for further analyses. Seven empty vector controls for camelina were generated in this study ( Supplementary Table S1 ). On-target editing was confirmed for Suneson CsFAD2 KO events Jiang et al. (2017) using the following restriction enzyme digestions: AvaI for F1 sites, BtsI for R1 sites and BbvCI for R2 sites. Events with resistant bands indicated disrupted restriction sites due to Cas9-editing and were selected for WGS analysis ( Supplementary Table S1 ).

For potato, the DRH195 synthetic genome assembly was used as the reference for WGS analysis. Across the samples, an average of 55x coverage for the haploid genome was obtained, except for event DRH195_2a which had 91x coverage ( Supplementary Table S3 ). Coverage was normalized for event DRH195_2a by taking a random subsample of reads using the median coverage from potato events, 58x. An average of 99.58% of reads mapped to the reference genome sequence for all samples. WGS analysis supported the assignment of variants identified through deep sequencing at the target site for the T0 events DRH195_13c, DRH195_2a, DRH195_10b and DRH195_5, with no additional variants at other on-target sites ( Table 1 , Figure 2 ). Deep sequencing of Cas9 event DRH195_7b indicated a mosaic edit in the RH and DM alleles of StPPO1 and StPPO3, but no variants were called at these locations from WGS ( Table 1 ). Although the expected StPPO1 and StPPO3 variants were not called, there were multiallelic variants, indicative of mosaic edits in both alleles of StPPO2-1 from WGS ( Figure 2B ). These results are consistent with T0 regenerated plants of the vegetatively propagated potato being a chimera of edited and wild-type alleles. On-target variations in potato ranged from a 17-base deletion to 1 base insertion, all within the seed sequence ( Table 1 , Figure 2 ). Larger deletions and more variable in sizes were more commonly detected in potato, the largest being a 17bp deletion ( Table 1 , Figure 2A ). This characteristic combined with the presence of microhomology at target sites indicates that MMEJ was the repair mechanism employed.

For camelina, the Suneson genome assembly was used as the reference (Fang et al., 2023) and~60x coverage was obtained with an average of 99.76% of reads mapping to the reference. Only variants that were unique to each CRISPR/Cas9 edited event from the respective wild-type and empty vector controls were considered for NHEJ outcomes.

No T5 or T6 camelina events had homozygous edits in all CsFAD2 homeologs suggesting that a complete loss of function of this trait may be lethal. All R1 events at their target sites had evidence of either fixed or a mosaic editing at all sites ( Figure 3 ). Transgenerational editing (TGE), which is continued editing throughout multiple generations, was observed in camelina events due to T-DNA integration and constitutive expression of CRISPR/Cas9 reagents as reported earlier (Impens et al., 2022; Jiang et al., 2017). In R1 events, a mosaic of wild-type and mutant reads were observed at most target sites. Jiang et al. (2017) also detected TGE at target sites into the T3 generation for R1 events. In this study, we observed that TGE continued into the T5 and even T6 generations in R1 events. Several homozygous sites had 1bp on-target insertions within the seed sequence, first 10 nts upstream of the 3’ end of the sgRNA ( Figure 3 ). These sites remained unaltered by additional TGE, supporting previous reports of CRISPR/Cas9 specificity and preference toward editing sites without variation in seed sequences (Liu et al., 2016). The rate of TGE varied between the F1, R1 and R2 events, with R1 events having the highest mutation rate. This is consistent with results reported by Jiang et al. (2017). Only a few reads suggest TGE in events at the F1 and R2 sites although a higher read depth would be needed to distinguish from sequencing error ( Figure 3 ).

Evidence of a complex variant was observed at the target site in Camsa.SUN.15G013420 in the T5 event SUN_R1E, as seen by clipped reads with partial homology to the reference ( Figure 3B ). To identify the full sequence composition of this variant, the WGS reads were used to create a de novo contig assembly that indicates 195 bases were replaced by a 166 bp insertion. This variant was then confirmed through cloning and sequencing. This insertion has no significant homology to any other region in the genome or to the vector. This variant is suggestive of a SDSA like repair mechanism in combination with cNHEJ.

Most on-target variants detected in camelina were 1bp insertions 3 bases upstream of PAM site at the predicted cut site for Cas9. Furthermore, no insertions larger than 1bp were detected in either species in this study. The preference toward one base pair insertion in camelina was also described in Jiang et al. (2017) where there was a noticeable enrichment in insertions over deletions at target sites with 99% of insertions being single nucleotide. Insertions of 1-3 bp are characteristic of cNHEJ mediated repair indicating a preference toward cNHEJ in camelina at all three target sites.

Sequence variation was observed between wild-type controls and the DRH195 and Suneson reference assemblies ( Table 2 ). For potato, there was a larger proportion of SNP variants compared to indels, with an average of 57,344 SNPs to 18,109 indels per transgenic event. This contrasts variation seen in camelina which had a larger proportion of indel variants to SNPs with an average of 4,041 SNPs to 36,379 indels per transgenic event. This trend is also seen in wild-type controls compared to the reference, with 5x more SNP to indel variants in potato versus 5x more indel to SNP variants in camelina. Most indel variants in camelina events are single nucleotide insertions or deletions ( Supplementary Figure S2 ). The consistency of indel mutations across all events and controls indicates that this variation is more likely related to common sequencing errors observed in Oxford nanopore derived genome assemblies (Oxford Nanopore Technologies, 2024) relative to the PacBio-derived RH assembly which is 50% of the overall DRH195 genome assembly.

Off-target sites containing up to 5 mismatches in the target sequence for canonical (NGG) and non-canonical (NGA, NAG) PAMs were identified in the DRH195 and Suneson assemblies using Cas-Offinder (Bae et al., 2014). Only variants that were unique to CRISPR/Cas9 edited events compared to the empty vector and wild type controls were considered for off-target analysis. In potato, two canonical NGG off-target sites identical to sgRNA2 were identified, both of which were on chromosome 8 of the RH haplotype. The first off-target site overlapped with the CDS of StPPO2-2 and the second off-target site overlapped with the 5’ UTR of the following gene model RHC08H2G1680.2. RHC08H2G1680.2 shares partial sequence homology to the 3’ end of the adjacent gene model StPPO4-3 suggesting that this may also belong to StPPO gene family. Analysis of variants across all 7 events indicated no edits in either of these off-target sites ( Table 3 ).

In the remaining canonical and non-canonical PAM off-target sites in potato, less than 0.1% of putative off-targets in any event contained a variant ( Table 3 ; Supplementary Table S6 ). Two canonical NGG off target sites with five mismatches in the target sequence contained a variant in events DRH195_1c and DRH195_13c, but manual inspection of alignments showed the same SNP variant was shared between the two transgenic events suggesting this as a tissue culture induced somatic variant. Seven non-canonical NGA sites with five mismatches in the target sequence contained a variant across all potato events, but only four of the variants were unique between samples. Manual inspection of alignments showed that all variants were SNPs which are not a common outcome of Cas9 dependent editing. Furthermore, the SNPs were supported by reads present in other samples or controls, but with an allele fraction below the threshold to be called a variant from WGS analysis. Therefore, these variants are also classified as background and not Cas9 dependent off target edits, resulting in no substantial evidence for off-target editing in potato.

In camelina, no additional canonical PAM off target sites with exact matches to the F1, R1 or R2 target sites were identified ( Table 2 ; Supplementary Table S6 ). Of the remaining canonical and non-canonical PAM off-target sites with up to five mismatches, only event SUN_R2A had putative off-targets variants, with less than 0.8% of off target sites containing a variant representing less than 0.7% of the total genetic variation in this event. The majority of these off-targets were in non-canonical NGA PAM sites with five mismatches ( Supplementary Table S6 ). Contrary to the results in potato, most of the off-target variants were short deletions. Out of the 77 off-target variants detected in SUN_R2A, 59 were 2bp deletions. Indels were the most common spontaneous variant type identified across all events and controls for camelina in this study ( Table 2 ). A total of 11% of all indels in SUN_R2A were 2bp deletions which was the third most frequent variant type in this event ( Supplementary Figure S2 ). As mentioned previously, Cas9 editing most often resulted in 1bp insertions for these camelina events (Jiang et al., 2017). Together, this suggests that the indels present at these putative off-target sites may likely be attributed to sequencing errors in the reference assembly, although further investigation into these sites is necessary to determine which variants are bona fide Cas9 dependent off-target mutations.

Most variants identified in the potato and camelina Cas9 events were outside of the putative on- or off-target sites and were unique to each event ( Figure 4 , Table 2 ). In both species most variants were intergenic ( Figure 5 ). Generally, variants are called in the euchromatic chromosome arms, but on chromosome 8 in DRH195, variants were called across the entire chromosomes, including the heterochromatic region across all CRISPR/Cas9-edited events ( Figure 6A ). The positions of the centromeres are not available for Suneson, but on nearly every chromosome there is a region where the number of indel variants dips, which may be a suggestion to the position of the centromeres ( Figure 6B ). In Suneson, the SNPs were called across the entire chromosome, whereas the indel mutations are localized in the presumed euchromatic regions. Overall, potato accumulated more Cas9 independent variants mostly likely due to tissue culture induced mutations than the seed propagated camelina ( Figure 4 , Table 2 ).