A combination of directed evolution and in silico predictions were used by Precision BioSciences to re-engineer I-CreI (including a peptide linker to join the two dimers as a monomer) to bind and cleave 22bp sequences specific to atp1 as previously described (Smith et al., 2012; Smith and Jantz, 2013). Initial confirmation of mitoARCUS specificity was conducted using a split-GFP assay in CHO cells as described in Zekontye et al. (2021).

The repurposed atp1 construct (nATP1), codon optimized for expression as a tobacco nuclear gene and including the sequences encoding the transit peptide and 3’-UTR of ATP2 ( Figure S1 ), was custom synthesized by GenScript (Piscataway, NJ). nATP1 was cloned downstream of an enhanced CaMV 35S promoter and introduced into plant expression vector pCAMBIA1300 (Abcam) that contains an hptII selectable marker gene. The atp1-targeting vector SP3379 contains two mitoARCUS cassettes: ATP-5/6, which was cloned downstream of a chimeric Figwort Mosaic Virus – Peanut Chlorotic Streak Caulimovirus promoter (Acharya et al., 2014) and upstream of the termination sequences of the soybean Ubiquitin3 gene; and ATP-7/8, under the transcriptional control of an enhanced CaMV 35S promoter and termination sequences mediated by the soybean MYB2 gene ( Figure S2 ). Transport of the ARCUS enzymes into the mitochondria was facilitated by a transit peptide designated M20, derived from a putative Arabidopsis HNH endonuclease (GenBank accession KAG6466785). SP3379 contains an nptII selectable marker gene.

Tobacco haploids were generated by crossing variety K326 with pollen from N. africana (Burk et al., 1979). Vector pCAMBIA1300-35S:nATP1 was introduced into Agrobacterium tumefaciens strain LBA4404 and transformed into haploid K326 leaf discs using standard protocols (Horsch et al., 1985) and hygromycin selection. Several independent T₀ haploids were screened using semi-quantitative PCR to identify individuals accumulating the highest levels of nATP1 transcripts. Selected haploids were chromosome doubled using the midvein culturing technique (Kasperbauer and Collins, 1972) to fix the transgene to homozygosity. Several doubled haploid individuals were again screened by semi-quantitative PCR to confirm high transgene expression, and plant 35S:nATP1#25 was chosen for further experiments. MitoARCUS vector SP3379 was transformed into 35S:nATP1#25 using the same Agrobacterium-based transformation protocol, but with kanamycin selection. As an EV control, a K326 doubled haploid plant transformed with the pCAMBIA1300 vector alone was also used as a host for transformation with SP3379.

For PCR analyses, total cellular DNA was isolated from tobacco leaf tissue using the MP Biomedicals FastDNA Kit. Primer sequences and reaction conditions for all PCR experiments are listed in Table S1 . DNA sequence analysis of PCR amplification products was conducted by the NCSU Genomic Sciences Laboratory (https://research.ncsu.edu/gsl).

Tobacco pollen was stained within intact anthers as previously described (Peterson et al., 2010) and imaged using a Labomed CxL microscope with an AmScope MU1000-HS digital camera. Toluidine Blue-O (TBO; Sigma-Aldrich) staining was performed on developmentally staged anthers or ovaries as previously described (Strable et al., 2020). Briefly, TBO was dissolved in 1% sodium borate (w/v) to make a 0.5% TBO staining solution (w/v). Slide-adhered microtome sections (10 µm) were deparaffinized in Histo-Clear (National Diagnostics). Slides were hydrated through a graded ethanol series, 1 min each (100%, 100, 95, 95, 70, 50, distilled water) and stained in 0.5% TBO staining solution for 1 min. Slides were then dehydrated through a graded ethanol series, 30 sec each (distilled water, 50%, 70, 95, 95, 100, 100) and Histo-Clear (three times, 1 min each). Slides were dried briefly and mounted with Permount (Fisher) and a cover slip. A Leica DM4B microscope equipped with a DMC6200 digital camera was used to capture images. Images of intact anthers and stigmas, and tobacco seeds germinated on solid MS media were taken using a dissecting microscope (Nikon SMZ-U) with an AmScope MU1000-HS digital camera.

A scaled-up version of the initial steps of the protocol described by Ahmed and Fu (2015) was used to enrich the preparations for mitochondria prior to DNA exaction. Briefly, thirty grams of leaf tissue from lines 35S:nATP1/Δatp1#8, #16, and #22 were ground on ice using a pre-chilled grinding medium (0.5M sucrose, 1 mM EDTA, 70 mM KH₂PO₄, pH 7.5) supplemented with 0.8% BSA and 0.1% β-mercaptoethanol. After filtering through two layers of Miracloth (Sigma-Aldrich), the homogenate was centrifuged at 1800g for 7 min to pellet intact nuclei, chloroplasts, and cellular debris. The supernatant was centrifuged at 17,200g for 20 min to pellet a fraction enriched in mitochondria. The pellet was resuspended in grinding media plus 0.8% BSA and the cycle of discarding the pellet of a low-speed spin (1800g), resuspension, and retention of the pellet after a high-speed spin (17,200g) was repeated two more times. DNA was extracted from the final pellets using the NucleoBond High Molecular Weight DNA Kit according to the manufacturer’s instruction (Macherey-Nagel).

Library preparation and PacBio sequencing was conducted by Eremid Genomic Services (Kannapolis, NC) using the SMRTbell prep kit 3.0 (PacBio), following the manufacturer’s instruction. In brief, for each sample, 10 µg of high molecular weight genomic DNA was fragmented using g-TUBE (Covaris), ligated to a SMRTbell barcoded adapter 3.0 and size selected using AMPure PB beads (PacBio) to remove fragments shorter than 5 kb. The integrity of each library was analyzed on TapeStation (Agilent), pooled in equimolar concentration, and loaded by diffusion at 350 pM into one SMRTcell-8M. Sequencing was performed on the Sequel-IIe system HiFi-CCS mode (PacBio).

HiFi reads were mapped to N. tabacum mitochondrial reference genome BA000042 using Minimap2 (version 2.24-r1122) with the flags -ax map-hifi –sam-hit-only (Li, 2018; Li, 2021) to estimate the number of reads of potential mitochondrial origin. The alignment results were summarized with samtools (v 1.12) to calculate the percentage of reads that aligned to BA000042 (Danecek et al., 2021). Read coverage was analyzed using the Integrative Genomics Viewer (IGV, v 2.12.0) (Robinson et al., 2011) and the R package karyoploteR (v 1.20.3) (Gel and Serra, 2017). To help reduce the number of NUMTs and chloroplast sequences included in analyses of the mapped reads, reads that were more than 33% clipped were removed. These reads were identified using the SamJdk program from JVarkit (Lindenbaum, 2015).

For each of the three Δatp1 genotypes, the HiFi reads that mapped to the reference genome were assembled into contigs using hifiasm with the flag -l 0 (version 0.16.1-r375)(Cheng et al., 2021; Cheng et al., 2022). Each assembly was then aligned to BA000042 using nucmer (v mummer-4.0.0rc1) with the -c 250 flag (Delcher et al., 2002; Kurtz et al., 2004) to identify potential mitochondrial contigs. From this set, contigs that shared greater than 99.99% identity with the reference (based on BLASTN) were classified as representing the mitochondrial genome; contigs whose homology to BA000042 was less than 98% were classified as NUMTs. In six cases, a contig shared greater than 99.99% sequence identity to the reference genome over a region spanning more than 100 kb, but was connected to sequences at the 3’ or 5’ end that were clearly derived from NUMTs. These contigs were hand curated to trim the non-mitochondrial sequences. BLASTN alignments of genuine mitochondrial contigs against BA000042 were conducted to identify the portions of the genome that were missing in the Δatp1 mutants, as well as define the nature of the specific junctions 5’ and 3’ of atp1 where the mutant mitochondrial genomes differed from the reference. To visualize the alignments between mitochondrial contigs and BA000042, circular plots were generated with Circos v 0.69-6 (Krzywinski et al., 2009) using nucmer with the -c 250 flag, followed by delta-filter -q (v mummer-4.0.0rc1) to obtain the best hit alignments.

We selected the atp1 gene encoding the α-subunit of the mitochondrial ATPase complex for allotopic expression in tobacco for our proof-of-concept experiments. Our rationale for atp1 was: (1) unlike most proteins encoded by essential genes within the mitochondria, the Atp1 protein is soluble, and thus less likely to encounter folding issues that may arise as the protein is being imported from the cytosol; and (2) there is precedence for atp1 functioning properly as a nuclear gene, as it is located in the nuclear genome in animals. Another potential advantage of atp1 is that the ATP2 gene that encodes the related β-subunit of the mitochondrial ATPase is a nuclear gene, and thus a favored source for acquiring important elements needed to convert atp1 into a functional nuclear gene. Specifically, the transit peptide of the Nicotiana plumbaginafolia ATP2 protein has been shown to be effective in transporting foreign proteins into the mitochondria (Chaumont et al., 1994; Duby et al., 2001) and the 3’- untranslated region (UTR) of yeast ATP2 plays a key role in enhancing ATP2 transport into the mitochondria through mediating the association of ATP2 transcripts with the organelle (Fox, 2012).

Despite the extensive efforts that have been devoted to the study of plant promoters, we found no promoter that was characterized as being expressed highly in all tissues except those associated with male reproduction. In the absence of such information, we proposed the best candidate to be an enhanced 35S Cauliflower Mosaic Virus (CaMV) promoter, given that expression of this promoter was shown to be lacking in the tapetum and sporogenous cells of developing anthers in dicots (Plegt and Bino, 1989). Furthermore, in maize transformed with a glyphosate tolerance gene driven by an enhanced 35S promoter, male reproductive tissues were the only tissue type reported to be unprotected from the herbicide in mature plants sprayed with glyphosate (Feng et al., 2014).

RNA editing is the phenomenon whereby select nucleotides within mitochondrial transcripts are post-transcriptionally modified (predominantly C to U base transitions), giving rise to mature transcripts whose encoded proteins differ from that predicted by the corresponding DNA sequence (Small et al., 2020). The tobacco atp1 gene has six locations where RNA editing has modified codons in the mature transcript (GenBank accession BA000042). Because of this, when repurposing atp1 to function within the nucleus, we based the redesign of the gene on the cDNA sequence as opposed to the mitochondrial DNA sequence. Codon usage is another consideration in allotopic expression of mitochondrial genes, as the translational machinery within the organelle is procaryotic-like (Artika, 2020). To avoid the potential of unfavorable codons limiting expression of the repurposed atp1 gene, a variant was custom synthesized using codons favored in the tobacco nucleus.

The repurposed atp1 construct, designated 35S:nATP1 ( Figure 1A ), includes an enhanced 35S promoter upstream of sequences encoding the transit peptide of the tobacco ATP2 protein (including the first 12 aa of the mature protein as these sequences are required for optimal transport; Duby et al., 2001) followed by the codon-optimized atp1 cDNA. The presumed 3’-UTR and termination signal of the tobacco ATP2 gene were also used. 35S:nATP1 (nucleotide sequence, Figure S1 ) was cloned into expression vector pCAMBIA1300 which contains the hptII selectable marker gene.

Transgenic plants were generated by transforming haploid plants of tobacco variety K326 with the 35S:nATP1 construct. Eighteen hygromycin resistant T₀ individuals were screened for transgene expression using semi-quantitative RT-PCR ( Figure S3 ). Individuals judged to be expressing nATP1 to the greatest extent were chromosome doubled to produce a fixed, high expressing nATP1 line (35S:nATP1#25) used for atp1 editing.

Custom-designed meganuclease enzymes developed via a genome editing platform termed ARCUS (Precision BioSciences) were used to target and cleave atp1. ARCUS enzymes are created by protein engineering of the homing endonuclease I-CreI and have been successfully designed to target and cleave the mitochondrial genome of mice (termed mitoARCUS; Zekontye et al., 2021). I-CreI exists as a dimer, but mitoARCUS enzymes contain an engineered linker that fuses the two engineered monomers into a single polypeptide. One unique property that distinguishes ARCUS enzymes from other genome editing platforms is the production of 4 bp 3’-overhangs upon cleavage. This presented a unique opportunity to test whether precise deletions in the mitochondrial genome could be created via the cleavage and ligation of compatible sticky ends by two mitoARCUS enzymes targeting sequences in close proximity, as opposed to the somewhat random deletions observed by researchers using mitoTALENs. A single plant transformation vector, designated SP3379, was developed with two distinct ARCUS cassettes (ATP-5/6 and ATP-7/8) designed to target the tobacco atp1 gene ( Figures 1B, C ). SP3379 possesses an nptII gene to enable kanamycin selection ( Figure S2 ). Both ARCUS enzymes were predicted to leave the same 4 bp 3’-overhang following cleavage of their respective target sites ( Figure S1A ), the re-ligation of which would result in a 674 bp deletion. To facilitate entry of the ARCUS enzymes into the mitochondria, sequences encoding a transit peptide designated M20 were introduced immediately upstream of those encoding the ARCUS proteins, as this transit peptide was shown to be effective in transporting ARCUS-GFP fusions into the mitochondria of tobacco protoplasts ( Figure S4 ).

Doubled haploid K326 line 35S:nATP1#25 was transformed with vector SP3379. As a control, leaf discs from an empty vector (EV) K326 line containing only the pCAMBIA1300 vector were also transformed. Fifteen T₀ plants were recovered from the 35S:nATP1/SP3379 transformation experiment and nine plants were recovered from K326 EV transformed with SP3379. To test whether atp1 had been altered in any of the plants transformed with the mitoARCUS constructs, PCR primers were designed corresponding to atp1 sequences flanking both the ATP-5/6 and ATP-7/8 recognition sites. As a control, primers specific for the mitochondrial gene atp6 were also included in each PCR reaction. Remarkably, in all fifteen T₀ plants in the 35S:nATP1 background the intensity of the atp1 amplification product was greatly reduced in comparison to atp6; in contrast, all nine plants recovered in the K326 EV line showed atp1 and atp6 amplification products of similar intensity (results summarized in Table S2 ). In no instance was an atp1 amplification product observed that possessed a 674 bp deletion indicative of the perfect removal and re-ligation of the sequences between the ATP-5/6 and ATP-7/8 target sites.

There are two plausible explanations for the observation of atp1 amplification products being greatly diminished, but not fully eliminated in 35S:nATP1/SP3379 plants: (1) the cleavage of atp1 yielded heterogeneous populations of mutated and WT atp1 genes; or (2) the minor amplification products in these plants are false positives, corresponding to nuclear-encoded mitochondrial DNA sequences (NUMTs). Over the course of evolution, large portions of the mitochondrial genome have become incorporated into the nuclear genome (Lough et al., 2015; Yoshida et al., 2017). Although these presumably nonfunctional sequences have accumulated polymorphisms over time, it can be difficult to differentiate a true mitochondrial gene from an NUMT using PCR-based assays (Arimura et al., 2020).

As an initial attempt to discriminate between true atp1 sequences and atp1-like NUMTs, we sought to design more specific primers based on NUMTs found in the tobacco genome. Utilizing BLASTN searches on draft genome sequences of N. tabacum with atp1 as the query sequence, over 30 scaffolds or contigs were found that shared greater than 93% sequence identity to the atp1 sequence. Three forward and three reverse primers were designed based on the NUMT information and tested for their ability to amplify atp1 in 35S:nATP1/SP3379 events. Although considerable variability was observed in atp1 band intensity among the different primer combinations, in no case was the complete absence of a band observed ( Figure S5 ). For subsequent experiments, we typically used the primer pair that regularly yielded the faintest PCR product (atp1_F3 and atp1_R2; Table S1 ). PCR analysis of seven 35S:nATP1/SP3379 events and five EV control plants using this primer pair is shown in Figure 1D .

As a direct test of the nature of the faint amplification product observed in 35S:nATP1/SP3379 plants, the PCR products generated by one of the atp1 primer pairs (atp1_F1 and atp1_R1) from two 35S:nATP1/SP3379 events (#8 and #16) and a WT K326 control were cloned into a plasmid vector. Sequencing runs were conducted on 22 – 24 independent plasmids from each of the three cloning experiments ( Data S1 - S3 ). None of the amplification products from plants 35S:nATP1/SP3379#8 or 35S:nATP1/SP3379#16 (hereafter referred to as 35S:nATP1/Δatp1#8 and 35S:nATP1/Δatp1#16, respectively) were 100% identical to the WT atp1 sequence. Instead, each contained polymorphisms that could be found in one or more atp1-like NUMT ( Data S4 ). Interestingly, there were no NUMTs on a tobacco genomic contig that perfectly coincided across the entire read of a given PCR amplification product, suggesting that additional atp1-like NUMTs exist in the tobacco genome that are not present in current drafts and/or the existing tobacco contigs were not assembled accurately in these regions. In contrast, 21 of 24 cloned PCR products amplified from the WT control were identical to atp1. These results suggest that an intact atp1 gene is no longer present in the mitochondrial genomes of plants expressing mitoARCUS enzymes. To gain insights into the nature of the nine plants in the K326 EV background that were recovered, we assayed these for the presence of the nptII selectable marker gene. Seven of the nine plants were negative for nptII, suggesting they were escapes ( Table S2 ). It is likely that the mitoARCUS cassettes were not expressed well in the two nptII-positive plants.

T₀ plants were transferred to soil and grown indoors until 10 – 15 cm in height, then transferred to large pots and grown to maturity in a greenhouse. Nine of the fifteen 35S:nATP1/Δatp1 plants appeared phenotypically normal during vegetative growth. The other six, however, displayed a reduced growth and/or a misshapen leaf phenotype with mosaic patterning ( Figure 1E ). By ten weeks after transplant, the phenotypically normal 35S:nATP1/Δatp1 individuals had grown to a typical height and had flowered, while the abnormal plants continued to show a slow growth phenotype ( Figure S6 ). The first distinguishing characteristic between otherwise normal looking 35S:nATP1/Δatp1 plants and WT (or EV control) plants became apparent upon flower formation. At full maturity (Stage 12 as defined by Koltunow et al., 1990) the petals of 35S:nATP1/Δatp1 individuals were less expanded at the top ( Figures 2A , S7A ). Examination of the anthers revealed even more dramatic differences. The anthers of 35S:nATP1/Δatp1 plants appeared malformed and failed to dehisce ( Figures 2B , S7B ), while the pistils and stigmas appeared normal. Pollen staining within intact Stage 10 anthers, followed by their release by pressing the coverslip firmly on the slide, revealed plentiful pollen in WT tobacco (both free and still associated with the anther). In contrast, no viable pollen grains were detected in the 35S:nATP1/Δatp1 anthers ( Figure 2C ).

Toluidine Blue-O staining of paraffin-embedded histological sections provided a more detailed analysis of the differences between normal anthers and those observed in 35S:nATP1/Δatp1 plants. WT tobacco anthers contain two distinct halves, or theca, separated by the vascular bundle. 35S:nATP1/Δatp1 anthers contained a single kidney-shaped theca ( Figure 2D ). At Stage 1, microspores had already formed in pollen sacs of the two locules within each theca of WT anthers; in contrast, pollen sacs were collapsed with no evidence of microspore formation in 35S:nATP1/Δatp1 anthers. At Stage 5, pollen sacs had become apparent in the mutant anthers, but were lacking microspores. At Stage 10, the two locules within each theca had fused and were filled with pollen in the WT anthers. The two locules had similarly fused within the single theca of the mutant anthers by Stage 10, with still no evidence of pollen grains. A morphological distinction that was apparent throughout anther development was the appearance of an exaggerated stomium region in 35S:nATP1/Δatp1 anthers. The enlarged stomium is apparent both on the histological sections ( Figure 2D ) and close-up pictures of the mutant anthers (the whitish triangular sections of the 35S:nATP1/Δatp1 anthers in Figure 2B ). The stomium is the specialized cell layer of the anther that ruptures upon dehiscence, facilitating the release of mature pollen grains (Keijzer, 1987). Despite its exaggerated structure, the stomium in Δatp1 mutant anthers remained intact throughout anther maturation. Collectively, these results suggest that 35S:nATP1/Δatp1 plants are male-sterile, and because the perturbation is attributable to a mutation in the mitochondrial genome, they therefore manifest a CMS phenotype.

To propagate the 35S:nATP1/Δatp1 CMS materials, 35S:nATP1 and WT K326 plants were used as pollen parents in crosses to the T₀ individuals. Minimal evidence of ovary development was observed in unfertilized flowers of 35S:nATP1/Δatp1 plants ( Figures 3A , S6C , and S8A ). In contrast, ovary development and capsule formation in fertilized 35S:nATP1/Δatp1 individuals appeared similar to that observed in self-fertilized WT and EV controls (tobacco is typically a self-fertilizing species). Subtle differences were observed, however, between the seed collected from 35S:nATP1/Δatp1 X 35S:nATP (or WT) crosses compared to the controls, as seed from the crosses appeared to be both smaller and lighter in color ( Figures 3B , S8B ). One-hundred seed weight counts revealed that seeds from WT and 35S:nATP plants weighed 7.3 to 14.4 times more than seeds from the crosses ( Figure 3C ). In total, crosses were made between 35S:nATP1 and/or WT K326 pollen parents and eight of the 35S:nATP1/Δatp1 T₀ individuals (35S:nATP1/Δatp1#1, #3, #8, #11, #12, #21, #22, and #23). When these seeds were placed on soil no germination was observed.

To better observe the differences between WT seeds and those obtained from 35S:nATP1/Δatp1 X 35S:nATP crosses, seeds were germinated on solid MS media. At one day-after-plating (DAP), WT seeds were opaque. Seeds from 35S:nATP1/Δatp1 X 35S:nATP crosses, however, were smaller and translucent, with most of the seeds containing a small, dense interior region ( Figure 3D ). By four DAP, small plantlets had emerged from the WT seeds, but even at seven DAP the seeds from the crosses appeared no different than those at one DAP. Although several hundred seeds from 35S:nATP1/Δatp1 X 35S:nATP crosses were plated on MS media, no germination was observed. Histological staining was also conducted during seed development. At 14 days-after-fertilization (DAF), torpedo-shaped embryos were visible, surrounded by layers of developing endosperm cells in self-fertilized WT plants ( Figure 3E ). By 21 DAF, when tobacco seeds are fully mature, there were very few non-stainable sectors within the WT tobacco seeds. In contrast, at both 14 and 21 DAF, the majority of the volume within the developing 35S:nATP1/Δatp1 X 35S:nATP seeds remained unstained, and there was little evidence of endosperm development. At both stages there were clusters of darkly staining cells that we speculate to be embryo-like structures.

Due to the inability to perpetuate Δatp1 T₀ events through seed, tissue culturing of leaf discs followed by plant regeneration was conducted to propagate these materials. Three tissue culture-maintained lines were selected for PacBio-based sequence analysis: 35S:nATP1/Δatp1#8, 35S:nATP1/Δatp1#16, and 35S:nATP1/Δatp1#22. Lines 35S:nATP1Δ/atp1#8 and #22 were selected as representatives of T₀ events that displayed normal growth habits while 35S:nATP1Δatp1#16 was chosen from among the events noted for slow growth and short stature, a phenotype observed in both the original T₀ plant as well as subsequent plants derived from tissue culture. A picture of tissue culture-propagated plants of similar age representative of the three selected lines is shown in Figure S9 .

Due to the large size of the tobacco genome (4.5 Gb), ground leaf preparations were subjected to several rounds of differential centrifugation to enrich the crude fractions for mitochondria prior to DNA extraction for sequencing. PacBio sequencing of the DNAs isolated from 35S:nATP1/Δatp1#8, #16, and #22 yielded read counts of 878,706, 543,041, and 911,783, respectively ( Table S3 ). Of these, 3.4 to 4.9% displayed sequence homology to the tobacco mitochondrial reference genome BA000042, which included an assortment of legitimate mitochondrial DNA reads, NUMTs, and even chloroplast DNA, as fragments of the chloroplast genome are scattered throughout plant mitochondrial genomes (Kubo and Newton, 2008). To enhance for genuine mitochondrial DNAs, reads that were more than 33% clipped were removed prior to visualizing read coverage to the reference genome. A gap was observed in the region of atp1 in all three 35S:nATP1/Δatp1 lines when the reads were mapped to BA000042 ( Figures 4A–C ). Despite the reads being greatly diminished in the region of atp1, a small number did span across the entire reading frame in 35S:nATP1/Δatp1#8 (n=5), #16 (n=7), and #22 (n=4) mitochondria. Alignment of the ORFs corresponding to atp1 from each of these reads ( Data S5 ) to the WT atp1 sequence revealed polymorphisms that were shared with NUMTs predicted to be located in the nuclear genome ( Data S4 ) as well as the PCR products of the low representation atp1-like sequences observed when events 35S:nATP1/Δatp1#8 and #16 were amplified using primers intended to be specific for atp1 ( Data S2 , S3 ). In general, the PacBio atp1-like NUMTs matched more closely to the PCR products than the NUMTs in GenBank, again suggesting that these regions of the tobacco draft genome are incomplete or inaccurately assembled. The PacBio-based NUMT that shared the highest sequence homology to WT atp1 differed only by three polymorphisms. Four of the PCR products amplified from event 35S:nATP1/Δatp1#8 ( Data S2 ) and two PCR products from event 35S:nATP1/Δatp1#16 ( Data S3 ) corresponded perfectly with this PacBio NUMT.

The genuine mitochondrial-specific PacBio reads assembled into five predicted contigs in lines 35S:nATP1/Δatp1#8 and #22, and four predicted contigs in line 35S:nATP1/Δatp1#16 ( Table S3 ). Circos plots aligning the predicted contigs to BA000042 showed that collectively, every region of the WT reference genome was represented in the predicted contigs, except the region surrounding atp1 ( Figure S10 ). Examination of the contigs in these areas revealed predicted 5’ recombination junctions in the region between the cox3 gene and atp1, and 3’ recombination junctions in the region between atp1 and a trnM gene ( Figure 4D ). Specifically, the mitochondrial genomes of lines 35S:nATP1/Δatp1#8, #16, and #22 were predicted to harbor deletions of 8,101 bp, 6,773 bp, and 4,593 bp, respectively. Although a small number of uncharacterized ORFs are found in this region, no gene of known function other than atp1 is found in the sequences predicted to be deleted. Because the extent of the deletion observed in line 35S:nATP1/Δatp1#16 was smaller than that found in 35S:nATP1/Δatp1#8, the slow growth phenotype of the former cannot be simply explained by a larger proportion of the mitochondrial genome having been deleted.

Close investigation of the predicted recombination junctions in the Δatp1 mutant genomes revealed regions of homology ranging from 6 bp to 188 bp ( Figure 5 ). For line 35S:nATP1/Δatp1#8, a 188 bp sequence between cox3 and atp1 appeared to have recombined with a homologous 188 bp sequence (differing by only 3 polymorphisms) found over 57 kb away in reverse orientation ( Figure 5B ). Downstream of atp1, a recombination event between an inverted 9 bp repeat sequence separated by 55 bp was predicted to result in a large 12,719 bp inverted repeat. Events 35S:nATP1/Δatp1#16 and #22 shared the same predicted recombination event in the region 5’ of atp1 ( Figures 5C, D ). A 148 bp sequence between cox3 and atp1 is 100% identical to a 148 bp sequence, in reverse orientation, found in the middle of the large 4,713 bp Repeat 2 (Rep2) that is located three times on the master reference genome ( Figure 5A ). From the sequence configurations predicted in the contigs of lines 35S:nATP1/Δatp1#16 and #22, as well as alignments of individual PacBio reads, it appeared that recombination could occur between the 148 bp sequence upstream of atp1 and any of the three Rep2 copies in the reference genome. The recombination event predicted in the region 3’ of atp1 in line 35S:nATP1/Δatp1#16 appeared to have occurred across a 6 bp sequence of microhomology (ACCACC) between sequences located over 32 kb away ( Figure 5C ). Finally, the 3’ recombination event in line 35S:nATP1/Δatp1#22 was predicted to have occurred across an 8 bp inverted repeat separated by 421 bp, giving rise to a large 13,947 bp inverted repeat ( Figure 5D ).

PCR and DNA sequence analysis was used to validate the predicted recombination events in the three Δatp1 lines. For line 35S:nATP1/Δatp1#8, PCR primers were designed that specifically amplified both the 5’ and 3’ predicted recombination events, and DNA sequence analysis of the amplification products validated the configuration of the sequences predicted by the contigs ( Figure S11A ). Similarly, the predicted 3’ recombination junctions for events 35S:nATP1/Δatp1#16 and #22 were confirmed by PCR and DNA sequence analysis ( Figure S11B ). As mentioned above, the predicted 5’ recombination event(s) shared by lines 35S:nATP1/Δatp1#16 and #22 could either extend from the copy of Rep2 that lies ~15 kb downstream of atp1, or from either of the two distal locations where Rep2 abuts directly with Rep1 ( Figure 5A ). Interestingly, when primers were designed to test each of those options, amplification and sequence validation were not restricted to lines 35S:nATP1/Δatp1#16 and #22, but were also observed in DNAs isolated from 35S:nATP1/Δatp1#8 and WT K326 ( Figure S11C ).

In their characterization of the tobacco mitochondrial genome, Sugiyama et al. (2005) predicted the organization of the genome to be multipartite, comprised of the master circle, as well as subgenomic species generated via recombination across each of the three large repeats (Rep1, Rep2, and Rep3). In addition, they speculated that it was also possible that additional subgenomic species may exist as a result of recombination across any of the numerous smaller repeats scattered throughout the genome. Our results suggest that there is indeed a subgenomic species in tobacco mitochondria that is generated through recombination between the 148 bp sequence between cox3 and atp1, and the homologous sequences in reverse orientation found in Rep 2. Our observations suggest that the preservation of essential mitochondrial genes (such as cox3) in the region immediately upstream of atp1 in lines 35S:nATP1/Δatp1#16 and #22 may have not occurred through recombination across DNAs broken by the mitoARCUS enzymes, but rather through the phenomenon known as substoichiometric shifting (SSS), where subgenomic mitochondrial DNA species in low representation can become amplified (Arrieta-Montiel and Mackenzie, 2010).

We speculated that the presence of naturally occurring subgenomic species with the recombination junctions predicted 5’ of atp1 as shown in Figures 5C, 5D could be further tested through examination of the 35S:nATP1/Δatp1#8 PacBio sequence database. Because the 148 bp direct repeat involved in recombination within Rep2 repeats lies immediately upstream of the 188 bp repeat used to create the unique 5’ break junction characterized in line 35S:nATP1/Δatp1#8, there should be PacBio reads in this background that could confirm the presence of these subgenomic species. BLASTN analysis of the PacBio database revealed 11 reads consistent with this hypothesis ( Data S6 ). In nine cases, the sequences at the end of Rep2 were followed by those beginning at position 276,635 (i.e., recombination occurred through the central Rep2); for the other two reads, the sequences at the end of Rep2 were followed by Rep1 (positions 75,135 or 425,885).

Although our original attempts to germinate seeds obtained from T₀ 35S:nATP1/Δatp1 individuals fertilized with WT or 35S:nATP1 pollen failed to produce a single plant, as tissue-culture-derived clones of the T₀ plants became available, even more crosses were made using 35S:nATP1 pollen. Seed quantities ranging from 200 – 500 mg were obtained from crosses to the following seven individual 35S:nATP1/Δatp1 lines: 35S:nATP1/Δatp1#1, #3, #8, #11, #21, #22, and #23. Consistent with our previous observations, no germination was observed from seeds obtained from six of the seven crossing combinations. However, a total of six plants emerged from an estimated 50,000 seeds planted from the cross 35S:nATP1/Δatp1#8 X 35S:nATP1 ( Figure 6 ). To confirm that the plants were not contaminants, each individual was genotyped using atp1 primers. All six plants showed the diagnostic pattern of diminished amplification of atp1-like sequences in comparison to the co-amplified atp6 control ( Figure 6B ). Using primers directed against sequences encoding the ARCUS meganucleases and the nptII selectable marker gene, it appeared that the genome editing construct was lost through segregation in two of the six F₁ plants, and retained in the other four ( Figures 6C, D ). All six F₁ generation individuals were also assayed using primers specific for the 5’ and 3’ recombination junctions defined in the original 35S:nATP1/Δatp1#8 parent. Interestingly, only three of the six plants showed amplification of the unique 5’ recombination junction, and none of the plants showed the diagnostic amplification product originally observed at the 3’ recombination junction ( Figures 6E, F ). Because subgenomic species created by recombination across the 148 bp region between cox3 and atp1 and Rep2 were shown to exist in 35S:nATP1/Δatp1#8 mitochondria, as described above, we speculated that SSS may have resulted in the replacement of species created through recombination across the adjacent 188 bp region with the pre-existing species generated via recombination across the 148 bp repeat shared with Rep2. Semi-quantitative PCR analysis supported this conjecture, as the three F₁ plants that were PCR positive for the 35S:nATP1/Δatp1#8-specific 5’ recombination product amplified the two subgenomic variants shown in Figure S11C weakly, while those that failed to amplify the former, strongly amplified the latter (compare Figure 6E with Figure S12 ).

When grown to maturity, five of the six 35S:nATP1/Δatp1#8 X 35S:nATP1 F₁ progeny looked phenotypically normal through the onset of flowering, and one individual displayed a reduced growth habit ( Figure S13 ). Upon flowering, all six F₁ individuals produced flowers that did not fully expand, and developed mishappen anthers that failed to shed pollen ( Figures 6G, H ).