An F₂ mapping population consisted of 269 individuals from a cross between “snake melon” (PI 435288) and C940-fe. The population was genotyped and scored, and the chlorosis trait associated with the fefe mutation was mapped to an 8 cM region of linkage group 8 (LG8) (Ramamurthy and Waters, 2015). To fine map the fefe mutation, a second, 288 individual F₂ mapping population was developed from a cross between “pocket melon” (PI 536481) and C940-fe. This F₂ mapping population was grown in hydroponics as described below prior to scoring leaf chlorosis in F₂ plants as “0” (chlorotic) or “1” (normal). The normal and chlorotic fefe mutant F₂ plants were distinguishable 1–2 weeks after planting in nutrient solution. DNA was isolated from a single young leaf from each plant (Kang et al., 1998), and quantity and purity of DNA was assessed using spectrophotometry at 260 and 280 nm. DNA was diluted to 25 ng/μl and stored at -20°C until genotyping.

A total of 112 SSR markers for LG8 from the consensus genetic map (Diaz et al., 2011) and markers for LG8 provided by Syngenta on the ICUGI website¹ were tested, and in total, 27 markers were polymorphic and were used for genotyping the F₂ mapping population (Table 1). PCR reactions were performed in a final volume of 10 μl with 1×Taq buffer [(16 mM (NH₄)₂SO₄, 67 mM Tris–HCL (pH 8.8 at 25°C), 0.1% stabilizer], 2 mM MgCl₂, 0.15 mM dNTP, 1 μM each primer, 0.2 U Taq DNA polymerase (Bioline USA Inc., Taunton, MA, United States), and 20 ng DNA. The cycling conditions were: an initial cycle at 94°C for 3 min, followed by 40 cycles at 94°C, 30 s, 55–58°C, 30 s and 72°C, 30 s, and a final extension step at 72°C, 5 min. PCR products were visualized with UV light after electrophoresis in 3% superfine resolution agarose (Amresco LLC, Solon, OH, United States) gels with 1× TBE (0.9 M Tris-borate, 0.002 M EDTA, pH 8.0), stained with ethidium bromide.

Phenotype scores for 288 pocket melon × C940-fe F₂ individuals and their corresponding genotypes across 27 loci in population were entered into a spreadsheet as an input file for QTL analysis in R/qtl software (Broman et al., 2003). Genetic positions were deduced for markers that were not present in the consensus map (from Syngenta and Li et al., 2011). Data checking steps for genotyping were performed using standard R/qtl functions (Broman et al., 2003). To obtain a better resolution of the fefe locus, a joint linkage map consisting of 557 individuals in both snake melon X fefe and pocket melon X fefe populations was constructed using polymorphic markers (N = 35) on LG8. For joint map construction, genotypic (LG8) and phenotypic information from the two mapping populations was input for QTL analysis in R/qtl software (Broman et al., 2003). The genetic maps for individual and joint analysis were constructed using est.rf and est.map functions of R/qtl. Interval mapping was performed using the “scanone” function which detects a single QTL by performing genome wide scan with possible allowance for covariates, with a binary model to analyze the binary phenotype (i.e., fefe or FeFe).

Seeds were imbibed in germination paper soaked with 0.2 mM CaSO₄ for 4 days, then were transferred to hydroponic containers. Seedlings were placed in sponge holders in lids of black plastic containers, four plants per 750 ml solution, with continuous aeration. The nutrient solution had the following composition: 0.8 mM KNO₃, 0.4 mM Ca(NO₃)₂, 0.3 mM NH₄H₂PO₄, 0.2 mM MgSO₄, 25 μM CaCl₂, 25 μM H₃BO₃, 2 μM MnCl₂, 2 μM ZnSO₄, 0.5 μM CuSO₄, 0.5 μM Na₂MoO₄, 100 mM MES buffer (pH 5.5) and 10 μM Sprint 138 (Becker-Underwood, Ames, IA, United States). Plants were grown in a growth chamber with lighting provided by a mixture of incandescent and fluorescent sources at 250 μmol m^-2 s^-1 for a photoperiod of 16 h (on at 06:00 and off at 22:00). For the +/- Fe RNA-seq experiment, Edisto and fefe mutants were pretreated for 9 days on -Cu solution, and fefe mutants that had green leaves were used for treatments of 3 days duration in -Fe nutrient solution or 20 μM Sprint 138. The purpose for the -Cu pretreatment was to use only healthy fefe plants [since the fefe chlorotic phenotype can be rescued using -Cu treatment (Waters et al., 2014)].

Total RNA was extracted from roots using the Plant RNeasy kit (Qiagen, Hilden, Germany). RNA quality and concentration was determined by UV spectrophotometry. Sources of RNA samples were as described in the previous section. RNA-seq was performed at the University of Nebraska Medical Center Next Generation Sequencing Core Facility using an Illumina HiSeq 2000 instrument. Barcoded libraries were constructed from 3 μg of root total RNA, with three biological replicate libraries per treatment. Replicates were run in separate lanes, with a total of six samples from different treatments in each lane. The short reads are available as NCBI BioProject: PRJNA371826². Because melon and cucumber genomes are orthologous and the cucumber genome sequence and annotation is complete (Huang et al., 2009; González et al., 2010), the cucumber transcriptome was used as a reference for read mapping. Trimming of primers and adapters was performed using Trimmomatic (V0.32), read mapping was performed using BOWTIE2 (Langmead and Salzberg, 2012) with –local-N1 option, conversion of mapped reads into sam format was performed using SAMtools (Li et al., 2009) and extraction of read counts from sam files was performed using perl scripts, as previously reported (Waters et al., 2014). For gene expression analysis, the data matrix was imported into R and analyzed using the Bioconductor package DESeq (Anders and Huber, 2010). The count data was normalized for library size and then transformed using variance stabilization. Poisson distributions of normalized counts for each transcript were compared for different conditions using a negative binomial test. Differential expression was called for genes with a false discovery rate moderated q-value < 0.05 (Benjamini and Hochberg, 1995), and also showed a 1.0 log fold-change in expression and >10 reads in at least one treatment. De novo reconstruction of the bHLH38 transcript was performed using Trinity software (V.r20131110) (Haas et al., 2013). IGVviewer 2.3³ was used to view the reads that are mapped onto the reference sequence as applicable.

One microgram of DNase treated RNA (RNase-free DNase I, New England Biolabs, Ipswich, MA, United States) from roots of -Fe and -Fe/-Cu treated Edisto and fefe and from eight normal and eight mutant snake melon X fefe F2 roots was used for cDNA synthesis, using the High Capacity cDNA Reverse Transcription kit (ABI, Foster City, CA, United States) with random hexamers at 2.5 μM final concentration. The cDNA templates were PCR amplified using primers spanning the insertion: fefe_mrkr_F-5′-AGAAACTGAGTAATCCGGCGA-3′ and fefe_mrkr_R-5′ TCGACTTGCAGAAATTATCGA-3′.

Edisto bHLH38 (MELO3C019065) full length genomic sequence (2.334 kb promoter +2.32 kb gene) was PCR amplified using the primers bHLH38_Promoter_F 5′-TCCCTTTGAACCAATGATGG-3′ and bHLH38_XbaI_R 5′-GCATGATCTAGAACACATTGATATATATGGTTAATAA-3′. Phusion High-fidelity DNA polymerase (Thermo Scientific) was used for PCR amplification of bHLH38 following manufacturer’s instructions. An Xba1 restriction site was present in the promoter region at position 379 bp of 2.334 kb, and an XbaI site was in the reverse primer, underlined above. After XbaI restriction (NEB Biolabs), the resulting bHLH38 Xba1 fragment was cloned into the pHairyRed (Lin et al., 2011) destination vector. The Edisto-bHLH38 genomic construct was transformed into the K599 strain of Agrobacterium rhizogenes (generously provided by Dr. Christopher Taylor, The Ohio State University) using freeze-thaw transformation (Wise et al., 2006). K599 containing pHairyRed-Edisto bHLH38 genomic fragment or K599 containing pHairyRed (empty vector) was grown in YEP plates with streptomycin and kanamycin selection. Agrobacteria suspension was prepared as previously described (Kereszt et al., 2007), and plug preparation and inoculation of Agrobacteria into rock wool plugs was performed as described (Chabaud et al., 2006), except we used hydroponic liquid media (described above) instead of half-strength MS. Stem sections with one or two axillary nodes from 1-month-old fefe plants were cut and inserted into the hole in the rock wool plug, and the plants were covered in a humid chamber for 4–5 days under ambient light. The humid chamber was opened for dehydration treatment for several hours until the leaves were not turgid, and the humid chamber was closed. Hairy roots developed 2–3 weeks after transformation. Transgenic roots were distinguished from non-transgenic roots, based on the presence of DsRed fluorescence. Z-series images were acquired on a Nikon A1+ CLSM mounted on a Nikon 90i compound microscope. Excitation of DsRed was at 561 nm and emission was detected at 575–625 nm. Image series were projected to form a single image. The transmitted light images were simultaneously acquired, but only a single image plane is presented. At least two biological experiments were performed to obtain the transgenic plants.

Root ferric reductase assays were performed for 50 min on transgenic roots (positive for DsRed fluorescence) and non-transgenic roots of fefe plants that were grown on -Fe solution for 2 days, using 20 ml of an assay solution. The assay solution was composed of 0.1 mM ferrozine (3-(2-pyridyl)-5,6-diphenyl-1,2,4-triazine-4′,4′-disulfonic acid sodium salt; Sigma–Aldrich), 0.1 mM Fe(III)-EDTA and 1 mM MES buffer (pH 5.5) (Fisher Scientific, Fair Lawn, NJ, United States). Change of assay solution from colorless to purple indicates ferric-chelate reductase activity.

The Matchmaker^TM GAL4 Two-Hybrid System 3 (Clontech) was used for the yeast 2-hybrid experiment. Coding sequences for Edisto FIT, Edisto bHLH38 and the fefe bHLH38 with its 14 bp insertion were amplified from cDNA and were cloned into pGEM-T vector and were later cloned as EcoRI fragment into pGADT7 and pGBKT7 yeast two-hybrid vectors. Drop assays were performed by growing AH109 transformants on synthetic dropout (SD) liquid media without leucine (L) and tryptophan (T) at 29°C until an OD₆₀₀ of >0.5 is reached. Cultures were then diluted to an OD₆₀₀ of 0.1 and diluted in a 10× dilution series. For each dilution, 5 μl of cell suspension was spotted on SD media without leucine, tryptophan, histidine (H) and adenine (A).

The chlorotic phenotype (Figure 1) mapped to a single locus on LG8 using the snake melon X fefe F₂ mapping population (Ramamurthy and Waters, 2015) (Figure 2A). Using Syngenta marker positions, the fefe locus was mapped to a 6 cM (2-LOD support) (Figure 2A) region in the snake melon X fefe population. The percentage of variation explained by LG8 locus was approximately 59% (Figure 2A and Table 1). Due to the low polymorphism rate of 12% in the snake melon X fefe population, we could not further narrow down the fefe genetic interval with the markers available. Therefore, we also mapped the chlorosis phenotype in a pocket melon X fefe F₂ mapping population. Of 288 F₂ plants, 196 had normal green leaves and 92 had chlorotic leaves. 27 markers on LG8 were polymorphic in the parents. The distribution of the parental genotypes in the F₂ population was almost equal based on 27 polymorphic loci on LG8, with 20.7% pocket melon genotype, 30.1% C940-fe genotype and 42.2% heterozygous. The fefe locus was mapped to a 4 cM interval, with a LOD score of 62 (Figure 2B and Table 1), with 63% of the variation explained by the peak (Table 1). Since both populations in this study had a common parent, C940-fe (fefe), we used joint linkage analysis to refine the fefe genetic interval to a 1 cM peak on LG8 with a LOD score of 115 (Figure 2C and Supplementary Data S1A). The closest marker at the peak explained ∼61% of the variation (Figure 2C and Table 1).

Since the fefe gene was predicted to be a transcription factor (Waters et al., 2014), we explored possible candidates within the 1 cM confidence interval. The genetic interval spanning the fefe gene, between the markers DM0766 and DM0640, corresponded to two scaffolds (scaffold0068 and scaffold0036) of the melon genome draft (Garcia-Mas et al., 2012). There were 186 predicted genes (Supplementary Data S1B) in this interval. There were six genes annotated as transcription factors in the mapped interval, based on homology to Arabidopsis thaliana: AT3G56970.1 BHLH038; AT5G04150.1 BHLH101; AT3G57390.2 AGL18; AT5G62470.1 MYB96; AT3G57040.1 ARR9 (RESPONSE REGULATOR 9); and AT3G10760.1 myb family transcription factor. We ruled out AGL18, MYB96 and ARR9 as unlikely candidates for fefe based on their function (To et al., 2004; Verelst et al., 2007; Seo et al., 2011). The AT3G10760.1 myb family transcription factor is putatively involved in the fruit ripening process (Pillet et al., 2015). Two transcription factors in the interval, bHLH38 and bHLH101, are associated with regulation of Fe-uptake genes (Wang et al., 2007). To determine if these genes were polymorphic between fefe and WT Edisto, or were differentially regulated, we sequenced the cDNAs and observed the normalized RNA-seq read counts from two independent experiments. Sequencing the bHLH101 gene showed that it was not polymorphic between the mutant and WT plants. The expression of bHLH101 was extremely low in both WT and mutant plants, ranging from 0 to 3 total raw read counts (compared to average read counts of approximately 450) under Fe replete or Fe-deficient conditions, suggesting that bHLH101 can be considered not expressed. The bHLH38 transcript levels were much higher than bHLH101 both in WT and fefe, ranging from 965 to 3979 read counts under Fe replete or deficient conditions, respectively. Transcript abundance did not change significantly under Fe deficiency in Edisto and snake melon WT roots. Although the trend of Fe regulation appeared similar in WT and fefe, the fefe bHLH38 expression differed in the two RNA-seq experiments (Figure 3). In the first experiment, bHLH38 was not significantly up-regulated by Fe deficiency, but in the second experiment, bHLH38 abundance was about fourfold higher under Fe deficiency due mainly to an unusually low read count in the +Fe fefe sample. We also checked abundance of bHLH38 transcript by RT-PCR (data not shown) and confirmed that the transcript is not regulated by Fe deficiency.

To better understand the expression of the bHLH38 gene in fefe and WT plants, we visualized read mapping of fefe and Edisto bHLH38 against the reference cucumber CsbHLH38 transcript (Csa4M434480.1). The reads mapped uniformly to the reference CsbHLH38 transcript in Edisto (Figure 4A), but the fefe read counts were much higher at the beginning of the transcript and decreased abruptly at position 370 bp of the reference transcript (Figure 4B), for both Fe deficient and Fe sufficient roots, suggesting the presence of a transcript variant. We performed de novo assembly of the Edisto and fefe transcriptomes to reconstruct the bHLH38 transcripts. The assembled WT bHLH038 transcript was 943 bp, which includes predicted start and stop codons to include 309 deduced amino acids. The de novo assembly of the fefe mutant resulted in eight unique transcripts, which were longer than the WT transcript by 277–4789 bp. The extra length occurred beginning at 549 bp in the WT bHLH038 transcript, where we saw read mapping anomalies relative to the cucumber transcript. Both before 549 bp and after the insertion, from 550 bp onward, the fefe transcript sequence matched the WT transcript sequence. Using RT-PCR with primers spanning the insertion site to visualize cDNA, a single band was present in Edisto, whereas the fefe parent contained multiple bands, in agreement with the de novo assembly results, in -Fe, and -Fe/-Cu treatments (Figure 5A). Similar to Fe deficiency, Cu deficiency did not change the bHLH38 banding pattern. These results suggest that the fefe mutant was producing multiple insertion-containing bHLH38 transcripts, or one large transcript that had been differently or partially spliced. A cDNA laddering pattern was also observed in the snake melon X fefe F₂ mapping population (Figure 5B), but only in individual plants with the chlorotic fefe phenotype. The fefe cDNA band that was closest in size to the Edisto bHLH38 cDNA was sequenced, to reveal the presence of a 14 bp insertion. Relative to the start codon, the fefe transcript had a reading frame shift followed by a premature in-frame stop codon, and would produce a different deduced amino acid sequence (Supplementary Data S2) that could negatively affect protein structure and function.

To further investigate the source of the transcript insertions in fefe bHLH38, we amplified genomic bHLH38 in normal and fefe plants by PCR and sequenced the products. All WT lines produced an amplicon of 1.019 kb, while fefe produced a 6.076 kb fragment (Figure 6). The fefe bHLH38 contained a 5.057 kb insertion relative to WT Edisto (Supplementary Data S3). The sequence of the insertion had identical 278 bp sequences at both extremes of the insertion (Supplementary Data S3 and Figure S1). A BLAST search of the fefe genomic bHLH38 sequence against the melon reference genome had hits to six melon genomic scaffolds, unassembled sequences and scaffold 36. Length of the hits for fefe bHLH38 using BLAST search against the melon genome ranged between 272 and 1368 bp, and the total scores from BLAST search ranged between 608 and 2510. An NCBI conserved domain search identified the insertion as a long terminal repeat (LTR) Ty1-copia type retrotransposon. The fefe bHLH38 contained helix-loop-helix, polypurine tract, RNAseH1-RT-Ty1, Reverse transcriptase, Integrase, gag-polypeptide and primer binding site of LTR-copia type domains (Supplementary Figure S1). There were also large genomic bands in WTs Mainstream and pocket melon, of a similar but not identical size as the band in fefe (Figure 6). From the banding pattern and BLAST search, it appears that the retrotransposon in fefe-bHLH38 could be present in other loci in the melon genome, potentially as an intact sequence. A global BLAST search for sequences similar to fefe bHLH38 in the NCBI nucleotide database indicated that the fefe bHLH38 retrotransposon was specific to melon.

Since the FeFe gene is necessary for normal plant growth and Fe uptake only in roots (Waters et al., 2014), we tested whether the chlorotic fefe phenotype could be complemented with a normal copy of bHLH38. We transformed fefe plants with Agrobacterium rhizogenes to generate hairy roots containing the WT bHLH38 gene (Figure 7). The chlorotic phenotype of the fefe plants was rescued in the pHairyRed-Edisto-bHLH38 treated plants, but not in mock or empty vector treated plants (Figure 7A). The rescued fefe plant roots were positive for the dsRed reporter gene (Figures 7B–G), and were able to initiate ferric reductase activity under Fe deficiency (Figure 7H), suggesting that normal root Fe uptake responses were recovered in the transgenic fefe plant roots expressing Edisto-bHLH38.

To determine whether the bHLH38 protein interacts with the Fit protein, as it does in Arabidopsis (Yuan et al., 2008), we performed a yeast two-hybrid experiment. Yeast cells transformed with different combinations of bait and prey plasmids were tested for auxotrophic growth. The melon Fit protein tested positive for interacting with Edisto bHLH38, but did not interact with fefe bHLH38 (Figure 8). The Edisto bHLH38 protein was capable of forming a homodimer, although yeast growth was less robust than in other combinations (Supplementary Figure S2A). The fefe bHLH38 did not form homodimers (Supplementary Figure S2B). Yeast growth suggested that there was some degree of interaction between Edisto bHLH38 and fefe bHLH38. This test could not be used to determine whether melon Fit forms a homodimer, since Fit was capable of auto-activation, as indicated by growth of yeast transformed with pAD+pBD-FIT (Figure 8).