The pDGB3α2 Tnos:nptII:Pnos—U6-26:gRNA:scaffold—P35S:hCas9:Tnos (GB4587) vector was built following the standard GoldenBraid assembly strategy as described by Vázquez-Vilar et al. (2021). The gRNA targeting the GF tomato locus (Solyc08g080090) was designed using the Benchling CRISPR tool² and it was selected to target the third exon. The level 1 GB construct pDGB3α1 U6-26:gRNA:scaffold was confirmed by restriction enzyme (RE) analysis and sequencing; level >1 constructs, which incorporated the hCas9 and nptII transcriptional units, were obtained through bipartite BsmBI- or BsaI-mediated reactions and were checked by RE-analysis. All GB parts are available at http://gbcloning.upv.es under the correspondent GB database ID, and their names and sequences are reported in Supplementary Table 1. The vector was finally transferred to A. tumefaciens LBA4404 electrocompetent cells.

Stable transformation of ‘MoneyMaker’ and ‘San Marzano’ was performed according to the protocol reported by Ellul et al. (2003), with minor modifications. In vitro culture was performed in a long day growth chamber (16:8 light:dark cycle) at 24°C, 60–70% humidity and 250 μmol photons m^–2 s^–1. For initial assessment of editing events, 150 mg of leaf tissue were collected from three different leaves of in vitro fully grown, rooted plantlets, frozen in liquid nitrogen and stored at −80°C until DNA extraction. For ‘MoneyMaker,’ a second editing assessment was performed on a selected pool of genotypes growing in the greenhouse, using 150 mg of plant material from two leaves and two fruits from each plant.

Genomic DNA was extracted according to a CTAB protocol (Murray and Thompson, 1980) and treated with RNAse A to remove RNA contamination. DNA quality was checked by 0.8% (w/v) agarose gel electrophoresis and quantification was carried out with Qubit 2.0 (Life Technologies, Carlsbad, CA, United States) based on the Qubit dsDNA HS Assay (Life Science).

Mutation frequencies at the target and off-target sites were evaluated according to an adapted version of the 16S Metagenomic Sequencing Library preparation protocol provided by Illumina (16S Sample Preparation Guide: https://support.illumina.com/documents/documentation/chemistry_documentation/16s/16s-metagenomic-library-prep-guide-15044223-b.pdf). Putative off-target sites were selected using CasOFFinder³ as representatives of the different possible configuration and position of mismatches relative to the PAM. Their characteristics are listed in Supplementary Table 2. DNA amplifications were carried out using the KAPA HiFi HotStart ReadyMix PCR Kit (Kapa Biosystems, Boston, MA, United States). Dual indexing was done using the Nextera XT system (Illumina, San Diego, CA, United States) using 16 i5 indexes (S502-S522) and 24 i7 indexes (N701-N729), enabling the multiplexing of 384 individual libraries. DNA sequencing was performed with an Illumina MiSeq sequencer (Illumina Inc., San Diego, CA, United States) and 150 bp paired-end reads were generated. After sequencing, adapter sequences were removed and reads less than 50 nucleotides long were discarded using Trimmomatic v0.39 (Bolger et al., 2014). Processed reads were analyzed for possible CRISPR-Cas9 editing events with CRISPResso2 (Clement et al., 2019), with -min_paired_end_reads_overlap to 1, leaving the other options by default.

A selected pool of T₀ ‘MoneyMaker’ plants underwent a second genetic screening 3 months after having been transferred to the greenhouse, sampling two leaves and two fruits per plant: their genotypes were evaluated through Sanger sequencing and chromatogram decomposition analysis using TIDE (TIDE | Web tool: https://tide.nki.nl/). The T₁ and T₂ selected progenies were screened throughS anger sequencing for gf genotyping and through Illumina Amplicon Sequencing at putative off-target loci. All oligonucleotide and primer sequences used in this work are listed in Supplementary Table 3.

In the T₀ generation, five edited events of the ‘MoneyMaker’ variety were chosen, grown in the greenhouse and their seeds collected to identify, through amplification of the Cas9 and nptII genes, T₁ individuals in which the T-DNA had been segregated from the mutated gf locus. Five transgene-free edited individuals were selected; among these, two lines (2B19 and 12A41) were selected as élite for being homozygous for gf alleles of interest, and their T₂ progeny was phenotypically evaluated. Their fruits were sampled at six ripening stages, from mature green (MG) to breaker (Br), and every 2 days until 8 days after breaker (Br+2 to Br+8). For each ripening stage and for each line (2B19, 12A41, and WT), three samples were evaluated; each sample consisted of a pool of four fruits. Fruits were collected, snap-frozen in liquid nitrogen and ground using the IKA^® A11 basic analytical mill (IKA, Staufen, Germany).

Extractions of isoprenoids in fruits at six stages of ripening were performed as described previously by Diretto et al. (2020). Briefly, for each sample 5 mg of freeze-dried powder were extracted with chloroform (spiked with 50 mg/l DL-α-tocopherol acetate as internal standard) and methanol (2:1 by volume); 1 volume of 50 mM Tris buffer (pH 7.5, containing 1 M NaCl) was then added and the samples were kept for 20 min on ice before a centrifugation step at 15,000 g for 10 min at 4°C. The hypophase was collected and the aqueous phase was re-extracted with the same amount of spiked chloroform; the two organic phases were merged and dried by speedvac and the resulting pellets were resuspended in 50 μl of ethyl acetate. For each sample, at least two independent extractions were performed. LC-DAD analyses were carried out using an Accela U-HPLC system (Thermo Fisher Scientific, Waltham, MA, United States). LC separations were performed using a C30 reverse-phase column (100 x 3.0 mm) from YMC (YMC Europe GmbH, Schermbeck, Germany) with mobile phases composed by methanol (A), water–methanol (20:80 by volume) containing 0.2% ammonium acetate (B) and tert–methyl butyl ether (C). The gradient was 95% A : 5% B for 1.3 min, followed by 80% A : 5% B : 15% C for 2.0 min and by a linear gradient to 30% A : 5% B : 65% C over 9.2 min. UV–visible detection was performed continuously from 220 to 700 nm with an online Accela Surveyor photodiode array detector (PDA; Thermo Fisher Scientific). Mass ionization was performed with an atmospheric-pressure chemical ionization (APCI) probe, operating in both + and – voltage conditions. Nitrogen was utilized at 20 and 10 units as sheath and auxiliary gas, respectively. The vaporizer and capillary temperature were set at 300 and 250°C, respectively. The discharge current was 5.5 μA, while S-lens RF level was set at 50. A mass range of 110/1,600 m/z was used both in positive and in negative voltage with the following parameters: resolution set at 70,000; microscan, AGC target and maximum injection time equal to, respectively, 1, 1x106 and 50. All solvents used were LC-MS grade quality (CHROMASOLV^® from Sigma-Aldrich, Saint Louis, MO, United States). Different isoprenoid classes (carotenoids, chlorophylls, tocochromanols, and quinones) were identified based on the accurate masses and by comparison with authentic standards, when available, and quantified based on the internal standard (IS) amounts (thus named Fold IS); finally, gf data were normalized and expressed relatively to WT data (thus reported as Fold WT).

The analysis of volatile compounds was performed by means of headspace solid phase microextraction (HS-SPME) coupled to gas chromatography and mass spectrometry (GC/MS) following the protocol reported by Rambla et al. (2017). Roughly, 500 mg of frozen tomato pericarp powder were introduced in a 15 ml glass vial and incubated with a closed cap at 37°C for 10 min in a water bath. Then, 500 ml of an EDTA 100 mM, pH 7.5 solution and 1.1 g of CaCl₂.2H₂O were added, mixed gently and sonicated for 5 min. One mL of the resulting paste was transferred to a 10 ml screw cap headspace vial with silicon/PTFE septum and analyzed within 10 h. Volatile compounds in the headspace were extracted by means of a 65 μm PDMS/DVB solid phase microextraction fiber (SUPELCO, Bellefonte, PA, United States). Volatile extraction was performed automatically by means of a CombiPAL autosampler (CTC Analytics, Zwingen, Switzerland). Vials were first incubated at 50°C for 10 min with agitation at 500 rpm. Then, the fiber was exposed to the headspace of the vial for 20 min with the same conditions of agitation and temperature. Volatiles were desorbed at 250°C during 1 min in splitless mode in the injection port of a 6890N gas chromatograph (Agilent, Santa Clara, CA, United States). After desorption, the fiber was cleaned in an SPME fiber conditioning station (CTC Analytics) at 250°C for 5 min with helium flow. Chromatography was performed on a DB-5ms (60 m, 0.25 mm, 1.00 μm) capillary column (J&W, Agilent) with helium as carrier gas at a constant flow of 1.2 ml/min. Temperatures of GC interface and MS source were 260 and 230°C, respectively. Oven programming conditions were 40°C for 2 min, 5°C/min ramp until 250°C and a final hold at 250°C for 5 min. Data was recorded in a 5975B mass spectrometer (Agilent) in the 35–250 m/z range at 6.2 scans/s, with 70 eV electronic impact ionization. Chromatograms were processed by means of the Enhanced ChemStation E.02.02 software (Agilent). Compound identification was performed by comparison of both retention time and mass spectrum with those of pure standards. All the standards were purchased from Sigma-Aldrich except 1-nitro-2-phenylethane, which was obtained from Apin Chemicals (Compton, United Kingdom). For quantitation, one specific ion was selected for each compound, and the corresponding peak from the extracted ion chromatogram was integrated. The criteria for ion selection were the highest signal-to-noise ratio and being specific enough to provide good peak integration in that region of the chromatogram. An admixture reference sample was prepared for each season by mixing thoroughly equal amounts of each sample. A 500 mg aliquot of the admixture was analyzed regularly (one admixture for every six to seven samples) and processed as a regular sample as part of the injection series. This admixture contained all the compounds identified in the samples, and it was used as a reference to normalize for temporal variation and fiber aging. Finally, the normalized results (corrected for temporal variation and fiber aging) for a particular sample were expressed as the ratio of the abundance of each compound in that sample to those present in the reference admixture.

For photosynthetic activity assays, seeds of the ‘MoneyMaker’ genotypes 12A26 and 12A50 were sown in 10 l pots containing a substrate mixture of horticultural substrate and perlite (3:1 v/v). For each genotype, four plants were used. All plants were grown in a glass greenhouse in Mallorca during late winter and early spring (mean temperature 25°C, with maximum temperature of 33°C). Plants were irrigated every 2 days to full capacity and with Hoagland’s solution 50% once a week, to avoid growth restrictions related with water or nutrient deficit. Leaf gas-exchange and chlorophyll a fluorescence were measured 2 months after seed germination simultaneously with an open infrared gas-exchange analyzer system equipped with a leaf chamber fluorometer (Li-6400–40, Li-Cor Inc., United States). Measurements were performed in two types of leaves: (1) young fully expanded leaves, located in the upper part of the plant, and (2) basal leaves, being among the oldest but not presenting chlorotic or necrotic symptoms. Along the measurements, the vapor pressure deficit (VPD) ranged between 1.1 and 2.8 kPa, with a mean of 2.1 kPa. Measurements were performed from 09:00 to 12:00. Environmental conditions in the leaf chamber consisted of a photosynthetic photon flux density of 1,500 μmol m^–2 s^–1 (with 10% blue light), and a leaf temperature of 25°C. Measurements were performed after inducing steady-state photosynthesis for at least 5 min at an ambient CO₂ concentration (C_a) of 400 μmol CO₂ m^–2 s^–1. The quantum efficiency of the photosystem II (PSII)-driven electron transport was determined using Equation (1):

where F_s is the steady-state fluorescence in the light (PPFD 1,500 μmol photon m^–2 s^–1) and F′_M the maximum fluorescence obtained with a light-saturating pulse (8,500 μmol photon m^–2 s^–1) (Genty et al., 1989). Leaf greenness was measured using a chlorophyll meter (SPAD 502, Konica Minolta, Osaka, Japan) simultaneously with leaf gas exchange measurements in the same leaves.

The susceptibility of tomato fruits to B. cinerea was assayed by inoculating ripe fruits (Br+8) of the 2B19 and 12A41 T₂ lines, together with WT tomatoes, with 10 μl of a suspension of 8 × 10⁵ conidia ml^–1, through a pipette tip. Two inoculations were performed for each fruit at two diametrically opposed spots around the equator. Six fruits were inoculated for the WT, six for line 2B19 and five for line 12A41. Infection area was measured 7 days post infection (dpi) with graph paper. Leaf susceptibility assays were performed by infecting the detached 5th and 6th compound leaves of 5-week-old plants of the 2B19 and 12A41 T₂ lines, together with WT leaves, with 5 μl of a suspension of 10⁶ B. cinerea conidia ml^–1, in four distant spots per leaf. For each genotype, six leaflets were infected. Prior to infection, leaves were rinsed in sterile distilled water and, once infected, they were kept in Petri dishes in a high-humidity chamber. The progression of the infection was checked daily, and 7 dpi the necrotic area around the infection spot was measured using ImageJ (Schneider et al., 2012). Comparable amounts of leaf tissue in the area included between infection spots was collected 7 dpi. Genomic DNA was extracted using the protocol reported by Khang et al. (2007). The abundance of pathogen DNA was estimated in three samples per genotype through qPCR by amplifying the B. cinerea intergenic spacer (IGS) as reported by Suarez et al. (2005).

IBM SPSS Statistics for Windows, version 25 (IBM Corp., Armonk, NY, United States) was used to perform univariate analyses. Multivariate statistical analyses were performed with MetaboAnalyst 5.0⁴ and the SIMCA-P version 13.0 software (Umetrics, Malmö, Sweden).

Despite not having annotated functional domains, GF proteins possess a highly conserved N-terminal region of about 150 amino acids, which targets the protein to the chloroplast, a highly conserved central core and a conserved cysteine-rich motif in the otherwise variable C-terminal region. A gRNA was selected to target the third exon, with the Cas9 cut site at position 1749, in correspondence with the conserved central core and in proximity to the known gf ² (pos. 1768) and gf (pos. 1789) alleles (Figure 1A). The gene editing module carrying the human codon optimized SpCas9, the gRNA and the selection marker (nptII) transcriptional units (Figure 1B) was assembled according to the GB 4.0 syntax. Our strategy was directed at producing loss of function mutations similar to the spontaneous mutations characterized for the GF locus (Barry and Pandey, 2009), which are all predicted to result in truncated forms of the GF protein, with similar effects on mutant visual phenotypes.

For each cultivar (‘MoneyMaker’ and ‘San Marzano’), 15 independent T₀ plants regenerated on selective media were analyzed. Genotyping was performed through high-throughput Illumina amplicon sequencing of the target locus. Transgene integration was confirmed for all analyzed plants by PCR amplification of the Cas9 gene on tomato genomic DNA. The average gene editing efficiency was 71.38% for ‘MoneyMaker’ and 66.66% for ‘San Marzano’ (Figure 2A). Editing levels for individual mutants are reported in Table 1. It is worth noting that in most cases the observed differences were between plants in which gene editing had occurred with very high efficiency and those in which it had occurred at very low rates or not at all, pointing to a divergence in transgene expression rather than in efficiency of the editing machinery. Excluding plants with editing levels below 25% (from which it would be less likely to retrieve an edited progeny), i.e., four plants in the ‘MoneyMaker’ dataset and five for ‘San Marzano,’ efficiencies were consistently similar (‘MoneyMaker’ 89%, ‘San Marzano’ 90.66%) and standard deviations were greatly reduced (Figure 2A). Differences in editing efficiency between the two cultivars were not significant either in the general population, or in the subset of plants with over 25% editing. The visual phenotype of ripe fruits also showed similar traits between the two cultivars (Figure 2B). Chlorophyll retention was visible in all fruit tissues and in the green coloration of the placenta.

The most frequent allelic configuration in edited plants was biallelic (with two different edited alleles) both for ‘MoneyMaker’ (60%) and ‘San Marzano’ (40%) (Figure 2C). 20% of ‘San Marzano’ mutants were chimeric, contrary to the ‘MoneyMaker’ population, where no chimeras were present. The percentage of WT, unedited individuals was 20% for both cultivars. Edited alleles ranged in size from an insertion or deletion of 1 nucleotide to a 123 bp deletion (Figure 2D). Ten different alleles were found in ‘MoneyMaker’ plants and six in ‘San Marzano.’ The most common mutation was a single nucleotide insertion (44% for ‘MoneyMaker,’ 60% for ‘San Marzano’) and, specifically, the insertion of a thymine represented 18% of total edited alleles in ‘MoneyMaker’ and 35% in ‘San Marzano,’ making it the most common mutation of our gene editing at the targeted GF locus (29% of the total edited alleles found in the two cultivars, combined). Small deletions (<10 bp) represented the second most frequent editing outcome (41% in ‘MoneyMaker,’ 40% in ‘San Marzano’). Larger deletions were found at lower frequencies: in ‘MoneyMaker,’ a 20 bp and a 123 bp deletions in, respectively, 2 and 9% of the total edited alleles. In ‘San Marzano,’ no deletions larger than 5 bp were found. Also, a single base substitution (T → A) was found in ‘MoneyMaker’ (4%). For the ‘MoneyMaker’ cultivar, three edited lines (2B, 3C, and 10B, corresponding to MM3, MM6, and MM11 in Table 1, respectively) were subjected to a second molecular screening 3 months after having been transferred to the greenhouse, with the aim to monitor Cas9 activity across plant growth and development, sampling two leaves and two fruits per plant (Supplementary Figure 1). For lines 3C and 10B, which already carried only mutated alleles, no further changes were detected at the target locus. In line 2B, which initially showed a 57.42% editing efficiency, Cas9 activity increased during plant growth, reaching almost complete editing in some of the sampled tissues (96.3% in one of the analyzed fruits). This indicates that, where a WT allele is still present, Cas9 can keep on introducing changes. While no new alleles were introduced, this additional activity span resulted in an enrichment of the +G and +T alleles already detected in plant 2B.

Five putative off-target sequences were selected among those identified by CasOFFinder, as representatives of different kinds and numbers of mismatches. Their characteristics are listed in Supplementary Table 2. In general, no sequences can be found in the tomato genome with less than three mismatches with respect to the gf gRNA used in this study. Among the selected putative off-targets, three correspond to non-coding regions of the tomato genome, while two correspond to annotated genes. Targeted deep sequencing at these loci proved that indeed no unspecific editing was detectable, confirming existing data on the high specificity of Cas9-mediated gene editing in plants (Figure 3; Maioli et al., 2020). To understand the general impact of off-target effects, for each cultivar the total variation at putative off-target sites was analyzed. Mutation rates for all five off-targets were consistently below 1% in both cultivars, and the rare detectable mutations were not indels, but substitutions, likely due to SNPs or base misattributions. Notably, no significant differences were found between edited and WT plants for these loci.

Two ‘MoneyMaker’ plants (2B and 12A, MM3 and MM13 in Table 1, respectively) were selected for further characterization. Both T₀ plants were biallelic: MM3 carried two different single-base insertions (+T and +G), while MM13 carried a 4 bp deletion and the larger deletion found in the dataset (123 bp). Segregant, transgene-free T₁ progenies were selected for both lines through amplification of the Cas9 and nptII genes and stable inheritance of the edited gf alleles was confirmed by PCR and Sanger sequencing. T₁ plants 2B19 and 12A41 were selected as transgene-free, homozygous lines to yield a T₂ progeny used for in-depth metabolic characterization and for pathogen susceptibility assays. Line 2B19 carries a single base insertion (+T), while line 12A41 carries a 123 bp deletion. In addition to genotyping the GF locus, T₁plants and their progenies were also screened at the five putative off-target loci identified before: lack of mutations was confirmed, which led us to confidently treat these progenies as isogenic lines suitable for faithful phenotypic characterization. Their genotypes are summarized in Supplementary Table 4.

The development of tomato fruits of the 2B19 and 12A41 lines was monitored during ripening, from the mature green (MG) stage to eight days after breaker (Br+8). A distinctive phenotype associated with the gf mutation, characterized by a green placenta and darker pericarp and mesocarp determined by chlorophyll retention alongside carotenoid accumulation, was clearly visible in fruits starting from breaker (Br) (Figure 4A). The levels of a total of 29 isoprenoid compounds (chlorophylls, carotenoids, tocochromanols, and quinones) were monitored and significant differences were found between gf mutants and WT fruits throughout the ripening process for all classes of compounds (Figure 4B and Supplementary Data Files 1, 2).

Based on the evidence of an association between gf/sgr genotypes and pathogen resistance in Arabidopsis, M. truncatula, soybean and cucumber (Mecey et al., 2011; Ishiga et al., 2015; Chang et al., 2019; Wang et al., 2019), we tested the susceptibility of tomato leaves and fruits to B. cinerea, an important tomato pathogen causing the grey mold disease. A significant difference was observed for both organs. The extension of necrotic spots in tomato leaves was measured at 7 dpi (Figure 5A), finding a significant difference between WT and gf mutants of both lines, with infection area reduced on average by 83% in line 2B19 and by 66% in line 12A41 (Figure 5B). The proliferation of B. cinerea on infected tissues was estimated by qPCR amplification of the pathogen DNA (Figure 5C), finding a 50% reduction in gf leaves of both mutant lines. In ripe tomato fruits sampled at 7 dpi, the average infection area was reduced by 56% in line 2B19 and by 68% in line 12A41 (Figure 5D).

Parallel to the metabolite profiling of the 12A41 and 2B19 genotypes, analyses were conducted on two different T₂ populations of the 12A line (12A26 and 12A50, both homozygous for a 4 bp deletion at the GF locus) to assess whether the CRISPR-induced gf mutations affected photosynthetic activity, net CO₂ assimilation rate, PSII efficiency and chlorophyll content. Young and basal leaves were analyzed in WT and gf plants (Supplementary Figure 4). While chlorophyll content in basal leaves was approximately 15 times higher in gf than in WT leaves, net CO₂ assimilation in young expanded leaves did not vary significantly between genotypes, and in basal leaves CO₂ assimilation dropped in a similar manner in both gf and WT plants. The same trend was observed for PSII, where a slightly increased efficiency in young expanded leaves was not maintained with senescence. This is consistent with the “cosmetic” nature of gf mutations in terms of photosynthetic activity (Myers et al., 2018). The recapitulation of the gf phenotype also includes the unchanged photosynthetic capacity of mutant leaves, despite substantially increased chlorophyll accumulation.

The potential of CRISPR-based systems to introduce target mutations at selected loci in a specific and efficient manner is well established in tomato (Yu et al., 2017; Hashimoto et al., 2018; Li T. et al., 2018; Lu et al., 2021). In this work, we were able to assess the specificity, reproducibility and robustness for the editing of the GF locus. The consistency of the editing pattern is a valuable feature for breeding, since it makes this tool reliable for the introduction of this trait in different genetic backgrounds. In both analyzed cultivars, editing occurred with comparable efficiencies and resulted in the introduction of similar mutations, with a prevalence of small indels. In both ‘San Marzano’ and ‘MoneyMaker,’ the most common mutation was the introduction of a single nucleotide. We confirmed that no mutations were detectable at putative off-target loci. In plant research, off-target activity has been extensively investigated at different depths (from standard PCR to whole genome resequencing), highlighting how it is most often negligible, if not entirely undetectable (Hahn and Nekrasov, 2019). The very techniques used to obtain edited crops, such as in vitro regeneration, can in themselves induce a small number of random mutations often referred to as somaclonal variation. WGS of edited progenies detects a degree of variation consistent with such mutation rates (Nekrasov et al., 2017; Modrzejewski et al., 2020), and can be a valuable tool for the validation of an edited line with respect to its wild-type equivalent. An advantage of sexually propagated species, such as tomato, is that the extremely rare unwanted mutations can be easily segregated in selected progenies. Thanks to its precision and the possibility of segregating the targeted loci from the transgene, CRISPR represents the “cleanest” tool in our hands at this time to introduce tailored mutations in selected crops.

The robustness of a phenotype linked to a mutation in a gene depends on the consistency with which it can be introduced and maintained, and this is very dependent on the target locus. Gene editing techniques have helped establish the exact function of some genes whose mutations are linked to known phenotypes. An interesting example is that of important tomato ripening mutants such as RIN, NOR, and CNR (Gao et al., 2019, 2020): gene editing of these loci failed to fully recapitulate the phenotypes from which they were first described, and this in turn allowed a more accurate elucidation of their role and of the complex network which regulates fruit ripening. Edited lines can be considered truly isogenic lines and allow to reveal the phenotypes associated to a gene of interest by direct comparison with the line they derive from, and this contrasts with spontaneous mutants, for which original lines are often difficult to track, or with introgression lines that may contain other mutations linked to the one of interest. A targeted SDN-1 mutation allows to identify, in a more polished manner, the effects of a particular allele in a defined genetic background. We compared the predicted effects on the GF protein encoded by the edited gf alleles and those deriving from the five spontaneous mutations described by Barry and Pandey (2009; Figure 6). The mutation carried by line 2B19 (the insertion of a T at position 1750 in the GF locus) results in the introduction of a premature stop codon at positions 1781–1783; the gf ² allele (the insertion of an A at position 1768) causes translation to stop at the same position. The 123 bp deletion of line 12A41 causes the elimination of a substantial region of exon 3, along with 23 nucleotides from the second intron. A similar rearrangement happens for the gf⁵ allele, where a 1,163 bp deletion removes the third and most of the fourth exons. In both cases, this results in the early termination of the protein. In gf⁵ mutants, GF is truncated at Q97; in 12A41 mutants the protein sequence diverges from the WT at the same Q97 position and ends at V113. 2B19 thus faithfully reproduces a mutation already present in the tomato gene pool, and 12A41 closely mimics another such mutation.

Our two lines, despite the structural differences of their mutations, show remarkable consistency in their phenotypes, something which is not true, for example, of RIN mutants obtained with CRISPR/Cas9, which showed different degrees of ripening delays. The reproducibility of a phenotype depends on various factors, including the role of the mutated protein product in cellular metabolism and the complex web of interactions in which it takes part, and locus architecture (for example, in RIN spontaneous mutants, RIN is truncated and fused in frame to a portion of the associated MACROCALYX locus, which causes it to switch from an activation to a repression function). For regulation, this means that the phenotypic consequences of a mutation should be carefully evaluated on a case-by-case basis, irrespective of the method used to obtain them. The ability to replicate a known mutation and, with it, a known phenotype, reinforces the argument of robustness for gene editing of the GF locus in tomato.

The correlation between gf and increased carotenoid levels in tomato was described by Luo et al. (2013), Li X. et al. (2018), and Ma et al. (2022). In the model proposed by Luo et al. (2013), the phytoene synthase responsible for fruit carotenoid biosynthesis (PSY1) is negatively regulated by native GF, either post-transcriptionally by direct interaction, or indirectly through modulation of PIF1. Mutations in the GF locus lift this negative regulation, increasing phytoene synthase activity to levels comparable to those of a constitutively overexpressed PSY1: gf mutants accumulate higher levels of carotenoids, especially phytoene, lycopene and β-carotene. The metabolic profile of our edited fruits fits this model well, showing an increase of metabolites of the early steps of the carotenoid pathway. Comparable levels of lycopene and α- and β-carotene were also observed by Li X. et al. (2018) in gf tomato fruits at Br+7. Xanthophyll levels are less affected by the mutation: this could be explained by the fact that in gf plants the retention of chlorophyll does not correspond to increased or continued photosynthetic activity, making it unnecessary for cells to accumulate greater quantities of antenna pigments.

Chlorophyll retention is the most apparent metabolic change in our edited fruits. Chlorophyll a and b concentrations are greatly increased in mutants at all stages of ripening compared to the WT, but still decrease by the end of fruit ripening. Alternative mechanisms for chlorophyll degradation might be in place and, especially, two other loci in tomato might be implicated in starting chlorophyll breakdown. In addition to the described GF locus (Solyc08g080090), tomato encodes a staygreen-like (SGRL) protein (Solyc04g063240). SGRL proteins are known to intervene in chlorophyll degradation in leaves prior to senescence and under stress conditions (Rong et al., 2013; Sakuraba et al., 2014; Wu et al., 2016). Its specific role in tomato has not been elucidated yet, but SlSGRL is known to take part in chlorophyll breakdown in an ABA-dependent manner (Yang et al., 2020). Alignment of GF against the tomato proteome also identifies another locus (Solyc12g056480) which encodes a putative staygreen protein with a 72.69% similarity with GF (Matsuda et al., 2016). Unfortunately, it is not possible to make a direct comparison between our gf mutants and those of Luo et al. (2013) and Li X. et al. (2018) for chlorophyll retention. The first did not detect differences in chlorophyll accumulation between mutants and WT fruits, something which might be due to an incomplete disruption of GF activity by RNAi, while the second did not investigate this feature. However, the visual phenotype reported by Li X. et al. (2018) is indicative of chlorophyll retention during ripening.

Unexpectedly, we found that gf fruits accumulate considerably higher levels of vitamin E, especially at Br+4 and Br+8. High chlorophyll levels should in principle reduce vitamin E biosynthesis during fruit ripening, inhibiting the recycling of the chlorophyll-derived phytol toward tocochromanol biosynthesis. Previous studies have reported simultaneous increases in carotenoids and tocochromanols, as in the case of tomato fruits overexpressing PSY1 (Fraser et al., 2007), a biochemical phenotype occurring together with chlorophyll degradation. However, it is possible that the activation of the carotenoid pathway provides an abundance of precursors which are additionally channeled toward tocochromanol biosynthesis. An overview of the general isoprenoid pathways can be seen in Figure 7. Overall, these data suggest the existence of a complex and not yet fully understood equilibrium between chlorophyll synthesis/degradation, phytyl salvage pathway and synthesis and accumulation of other isoprenoid classes. However, it is worth noticing that these changes mainly affect plastidial (carotenoids and tocochromanols) rather than cytosolic (quinones) isoprenoids, thus indicating a reduced capacity of gf to affect carbon exchanges between MVA and MEP pathways. Notably, the acceptable daily intake (ADI) for vitamin E is 12 mg/die, and its concentration in tomato is on average 1 mg/100 g fw.⁵ The increase of vitamin E (especially α-tocopherol) in ripe fruits to up to around 10-fold represents a significant enrichment of this class of compounds with respect to WT fruits, meaning that just 120 g of gf fruits would be needed to meet the ADI.

The implication of gf in pathogen susceptibility in tomato had not been previously reported. Our findings in tomato fruits and leaves are consistent with the characterization of staygreen/greenflesh mutants in other crop species, namely soybean, Medicago and cucumber. A clear elucidation of the molecular mechanisms underlying this phenotype is still lacking. However, Wang et al. (2019) do propose a model in which gf is regarded as a R gene with a loss-of-susceptibility mechanism. In this model, the reduced chlorophyll breakdown of gf mutants limits the amount of ROS and phototoxic Chl catabolites produced upon pathogen infection, thus limiting symptom development. The fact that, despite not being involved in increased photosynthetic activity, light harvesting complexes are still present in plastids, together with an increase in carotenoids and vitamin E, might contribute to limiting oxidative damage and to the creation of a cellular environment capable of containing pathogen growth.