A Very Oil Yellow1 Modifier of the Oil Yellow1-N1989 Allele Uncovers a Cryptic Phenotypic Impact of Cis-regulatory Variation in Maize

Forward genetics determines the function of genes underlying trait variation by identifying the change in DNA responsible for changes in phenotype. Detecting phenotypically-relevant variation outside protein coding sequences and distinguishing this from neutral variants is not trivial; partly because the mechanisms by which DNA polymorphisms in the intergenic regions affect gene regulation are poorly understood. Here we utilized a dominant genetic reporter to investigate the effect of cis and trans-acting regulatory variation. We performed a forward genetic screen for natural variation that suppressed or enhanced the semi-dominant mutant allele Oy1-N1989, encoding the magnesium chelatase subunit I of maize. This mutant permits rapid phenotyping of leaf color as a reporter for chlorophyll accumulation, and mapping of natural variation in maize affecting chlorophyll metabolism. We identified a single modifier locus segregating between B73 and Mo17 that was linked to the reporter gene itself, which we call very oil yellow1 (vey1). Based on the variation in OY1 transcript abundance and genome-wide association data, vey1 is predicted to consist of multiple cis-acting regulatory sequence polymorphisms encoded at the wild-type oy1 alleles. The vey1 locus appears to be a common polymorphism in the maize germplasm that alters the expression level of a key gene in chlorophyll biosynthesis. These vey1 alleles have no discernable impact on leaf chlorophyll in the absence of the Oy1-N1989 reporter. Thus, the use of a mutant as a reporter for magnesium chelatase activity resulted in the detection of expression-level polymorphisms not readily visible in the laboratory.

ABSTRACT Forward genetics determines the function of genes underlying trait variation by identifying the change in DNA responsible for changes in phenotype. Detecting phenotypically-relevant variation outside protein coding sequences and distinguishing this from neutral variants is not trivial; partly because the mechanisms by which DNA polymorphisms in the intergenic regions affect gene regulation are poorly understood. Here we utilized a dominant genetic reporter to investigate the effect of cis and trans-acting regulatory variation. We performed a forward genetic screen for natural variation that suppressed or enhanced the semi-dominant mutant allele Oy1-N1989, encoding the magnesium chelatase subunit I of maize. This mutant permits rapid phenotyping of leaf color as a reporter for chlorophyll accumulation, and mapping of natural variation in maize affecting chlorophyll metabolism. We identified a single modifier locus segregating between B73 and Mo17 that was linked to the reporter gene itself, which we call very oil yellow1 (vey1). Based on the variation in OY1 transcript abundance and genome-wide association data, vey1 is predicted to consist of multiple cis-acting regulatory sequence polymorphisms encoded at the wildtype oy1 alleles. The vey1 locus appears to be a common polymorphism in the maize germplasm that alters the expression level of a key gene in chlorophyll biosynthesis. These vey1 alleles have no discernable impact on leaf chlorophyll in the absence of the Oy1-N1989 reporter. Thus, the use of a mutant as a reporter for magnesium chelatase activity resulted in the detection of expression-level polymorphisms not readily visible in the laboratory.

Chlorophyll biosynthesis cryptic variation cis-acting complex traits epistasis
Gene function discovery via mutant analyses focuses on linking phenotype alterations to gene variants. Forward genetics has been of great value to understand biological systems but is predominantly useful for determining functions for genes with alleles that cause large phenotypic impacts. Natural variants, including those that encode alleles of relevance to adaptation and fitness of the organism, are often of small effect (Fisher 1930;Orr 1998Orr , 2005. Mutant alleles with conditional impacts on phenotypes through genetic interactions or modifiers, as well as alleles of small individual effect are difficult to study. Further, when studies of natural variation identify loci that have not been previously associated with a biological process, it can be difficult to definitively associate such genetic variants with physiological and biochemical mechanisms. A forward genetics approach that uses a mutant phenotype as a reporter for genetic interactions can be used to detect modifiers in natural populations and expose cryptic variation affecting traits of interest (Johal et al. 2008). This Mutant-Assisted Gene Identification and Characterization (MAGIC) approach is particularly efficient in species where outcrossing is easy, such as maize (Zea mays). Although this approach can be applied to any mutant with a quantifiable phenotype, it is convenient when a dominant mutant allele is used as a reporter because natural variants that encode enhancers or suppressors of a given mutant phenotype can be detected in F1 crosses. Any germplasm collection, diversity panel, or line-cross population can serve as the variable parents in these crosses to provide segregating alleles for mapping loci that alter mutant phenotype expression. Natural variants discovered by this type of genetic screen have an experimental link (e.g., genetic modifiers) of the processes affected by the mutant reporter allele. Thus, this approach speeds the assignment of a mechanism to natural variation in the germplasm. Indeed, one can consider this as a screen for epistatic or contingent gene action. MAGIC screens have identified loci from maize involved in the hypersensitive response (Penning et al. 2004;Chintamanani et al. 2010;Olukolu et al. 2013Olukolu et al. , 2014Olukolu et al. , 2016 and plant development (Buescher et al. 2014), among others. In these cases, in the absence of a mutant allele, no phenotype was previously associated to the modifying alleles demonstrating the remarkable efficiency of this genetic screen to detect epistatic interactions Thus, this approach is a powerful way to both characterize cryptic variation in genomes and construct genetic pathways affecting phenotypes of interest.
The easy visual scoring and simplicity of quantifying chlorophyll make chlorophyll biosynthetic mutants an excellent reporter for MAGIC screens. Chlorophyll is a major component of central metabolism in plants, which can produce phototoxic intermediates during both synthesis (Hu et al. 1998;Huang et al. 2009) and breakdown (Gray et al. 1997;Mach et al. 2001;Gray et al. 2002;Yang et al. 2004;Sindhu et al. 2018), and its levels are carefully regulated in plants (Meskauskiene et al. 2001). Enzymatic conversion of protoporphyrin IX into magnesium protoporphyrin IX by Magnesium Chelatase (MgChl) is the first committed step in chlorophyll biosynthesis (Von Wettstein et al. 1995). MgChl is a hetero-oligomeric enzyme consisting of three subunits (I, D, and H) that are conserved from prokaryotes to plants. The MgChl subunit I is encoded by the oil yellow1 (oy1; GRMZM2G419806) gene in maize and encodes the AAA+-type ATPase subunit that energizes the protein complex (Fodje et al. 2001). Weak loss-of-function alleles of oy1 result in recessive yellowgreen plants while complete loss-of-function alleles result in a recessive yellow seedling-lethal phenotype (Sawers et al. 2006). The semi-dominant Oy1-N1989 allele carries a leucine (L) to phenylalanine (F) change at amino acid position 176 (L176F). Heterozygous plants carrying one Oy1-N1989 allele and one wild-type allele are oil-yellow, but Oy1-N1989 homozygotes lack any MgChl activity and die as yellow seedlings with no chlorophyll accumulation (Sawers et al. 2006). The orthologous L-.F mutation (encoded by Oy1-N1989) was identified in barley (L161F) and also results in a pale-green phenotype as a heterozygote and yellow seedling-lethal phenotype as a homozygote with no detectable chlorophyll (Hansson et al. 1999). The biochemical basis of this semi-dominant mutant allele was studied by creating a mutant MgChl subunit I in Rhodobacter (BCHI), with the orthologous amino acid change at position 111 of wild-type BCHI (Hansson et al. 2002). The L111F mutation converted BCHI into a competitive inhibitor of MgChl that reduced enzyme activity by fourfold when mixed 1:1 with wild-type BCHI (Hansson et al. 2002;Lundqvist et al. 2013). This conserved leucine residue is between the ATP-binding fold created by the Walker A and B motifs of MgChl subunit I (Hansson et al. 2002;Sawers et al. 2006;Lundqvist et al. 2013), and its substitution with phenylalanine was deleterious to dephosphorylation activity. The ATPase activity of wild-type MgChl complex is directly proportional to the magnesium chelation reaction. However, complexes assembled from MgChl subunit I with the L-.F change exhibit reduced ATPase activity and no ability to chelate Mg 2þ ions into protoporphyrin IX (Hansson et al. 1999(Hansson et al. , 2002Sawers et al. 2006). The absence of MgChl activity displayed by the mutant BCHI carrying L111F substitution (BCHI L111F ) demonstrates that this amino acid change results in a dominant-negative subunit which uncoupled ATP hydrolysis and magnesium chelation (Hansson et al. 2002;Lundqvist et al. 2013).
We screened maize germplasm for cryptic variation using the Oy1-N1989 mutant as a dominant reporter for chlorophyll biosynthesis. We hypothesized that alteration in the quantity of biochemically active MgChl complex should change chlorophyll content and plant color in heterozygous Oy1-N1989 mutants. For instance, in a population of heterozygous Oy1-N1989/oy1 F1 plants, an amino acid change in the wild-type OY1 protein that alters the dissociation constant (kD) of OY1 for the other protein subunits of MgChl should contribute to variance in chlorophyll biogenesis. Similarly, a cis-acting expression QTL (eQTL) that increases expression of the wild-type oy1 allele should result in assembly of more active MgChl complexes and increase chlorophyll content in F1 mutant plants. Thus, chlorophyll content of Oy1-N1989 mutants should be modulated by the stoichiometry of the wild-type and mutant OY1 proteins present in the MgChl complex in heterozygous Oy1-N1989/oy1 plants affected either by protein structure variation or abundance changes.
We introgressed the maize Oy1-N1989 mutant allele into the B73 inbred background and maintained it as a heterozygote (Oy1-N1989/ oy1:B73). While crossing this mutant to multiple backgrounds, we detected genetic variation in the Oy1-N1989/oy1 mutant phenotype expression between the maize inbred lines B73 and Mo17. The phenotype of the Oy1-N1989 mutant heterozygotes was suppressed in the B73 background. However, F1 hybrids with Mo17 dramatically enhanced it. We carried out genetic mapping experiments using five F1 populations. In each of these mapping experiments, we identified a quantitative trait locus (QTL) of large effect on chromosome 10 linked to the oy1 locus itself. The inheritance of the traits, proposed allele expression bias at oy1 due to a putative cis-acting regulatory element, implications of the discovered cryptic variation, and the utility of this study, in general, are discussed.

Plant Materials
The Oy1-N1989 mutant allele was acquired from the Maize Genetics Cooperation Stock Center (University of Illinois, Urbana-Champaign, IL). The original allele of the Oy1-N1989 mutation was isolated from a r1 c1 colorless synthetic stock of mixed parentage (Myron Gerald Neuffer, personal communication). This allele was introgressed into B73 for eight generations by repeated backcrossing of B73 ear-parents with Oy1-N1989/oy1 pollen-parents and is maintained as a heterozygote (Oy1-N1989/oy1:B73) by crossing to wild-type siblings.
Genome-wide association (GWA) mapping: For this experiment, Oy1-N1989/oy1:B73 plants were crossed to 343 inbred lines that included 249 inbreds from maize association panel (Flint-Garcia et al. 2005), and 94 inbred lines that included 82 Expired Plant Variety Protections (ExPVP) lines from the Ames panel (Romay et al. 2013). Pollen from Oy1-N1989/oy1:B73 plants were used for these crosses except for the popcorn lines in the maize association panel, where Oy1-N1989/oy1:B73 plants were used as an ear-parent to avoid the crossing barrier due to gametophytic factor GA1-S (first described by Correns in 1902) (Mangelsdorf and Jones 1926;Lauter et al. 2017). This panel of 343 inbred lines is referred to as maize diversity lines (MDL). The full list of IBM, Syn10, BM-NILs, and MDL used to develop F1 hybrid populations are provided in Tables S1-S4.

Field Trials
All field experiments were performed at the Purdue Agronomy Center for Research and Education (ACRE) in West Lafayette, Indiana. All F1 populations described below were planted as a single plot of 12-16 plants that segregated for both mutant and wild-type siblings. Plots were sown in a 3.84 m long row with the inter-row spacing of 0.79 meters and an alley space of 0.79 meters. No irrigation was applied during the entire crop season as rainfalls were uniformly distributed for satisfactory plant growth. Conventional fertilizer, pest and weed control practices for growing field maize in Indiana were followed. Progenies of Oy1-N1989/ oy1:B73 pollen-parents crossed with B73 and Mo17 were planted as parental checks in each block of every experiment. The testcross F1 populations with IBM were evaluated in a single replication in the summer of 2013 with each range treated as a block. In 2016, the testcross F1 populations with Syn10 lines were evaluated as two replications in a randomized complete block design (RCBD) with each range divided into two blocks. The testcross F1 progenies with BM-NILs and parents (B73 and Mo17) were planted in a RCBD with five replications in 2016. In the same year, F1 populations with MDL were also evaluated with three replications planted in a RCBD. Each replication of MDL F1 population was divided into ten blocks of the same size, and parental checks were randomized within each block.

Phenotyping and Data Collection
Maize seedlings used for destructive and non-destructive chlorophyll quantification were grown under greenhouse conditions using mogul base high-pressure sodium lamps (1000 Watts) as the supplemental light source for L:D cycle of 16:8 and temperature around 28°(day-time) and 20°(night-time). Destructive chlorophyll measurements were performed on the fresh weight basis in 80% acetone solution using a UV-VIS spectroscopic method (Lichtenthaler and Buschmann 2001). Non-destructive chlorophyll measurements were performed in the middle of the same leaf avoiding the midribs using the procedure described below.
For the field-grown experiments, mutant siblings in the suppressing genetic backgrounds were tagged at knee height stage with plastic tags so that they can be easily distinguished from the wild-type siblings at later developmental stages. All the F1 families segregated for the mutant (Oy1-N1989/oy1) and wild-type (oy1/oy1) siblings in approximately 1:1 fashion. For each F1 family, two to four plants of each phenotypic class were picked at random for trait measurements. Non-destructive chlorophyll content in the maize leaves was approximated using a chlorophyll content meter model Inc.,Hudson,NH). The measurements were expressed as chlorophyll content index (CCM). Measurements were taken on the leaf lamina of the top fully expanded leaf at two time points. First CCM measurements were taken at 25-30 days after sowing (expressed as CCMI) and the second at 45-50 days after sowing (expressed as CCMII). For each trait, measurements were performed on both mutant (reported with a prefix MT) and wild-type (reported with a prefix WT) siblings. Besides using primary trait measurements of CCMI and CCMII on mutant and wild-type siblings, indirect CCM measurements were also calculated and expressed as ratios (MT/WT) and differences (WT-MT) of CCMI and CCMII. Phenotypic data of all the CCM traits in the F1 populations with both bi-parental populations, BM-NILs and MDL are provided in Tables S1-S4.

Public Genotypic and Gene Expression Datasets
Public marker data for the IBM was obtained from www.maizegdb.org (Sen et al. 2010). A total of 2,178 retrieved markers were reduced to 2,156 after the removal duplicates. Approximately 13.3% of the marker data were missing. The reads per kilobase of transcript per million mapped reads (RPKM) for the oy1 locus were obtained from a public repository of the National Science Foundation grant (GEPR: Genomic Analyses of shoot meristem function in maize; NSF DBI-0820610; https://ftp.maizegdb.org/MaizeGDB/FTP/shoot_apical_meristem_data_ scanlon_lab/). These data consist of normalized read counts (expressed as RPKM) of the maize genes from the transcriptome of shoot apex of 14 days old IBM seedlings.
Marker data for Syn10 lines was obtained from Liu et al. (2015). The Syn10 lines were genotyped at 6611 positions (B73 RefGenv2) with SNP markers covering all ten chromosomes of maize. The entire set of markers were used for linkage analysis as there was no missing data. All B73-Mo17 NILs in both the B73 and Mo17 recurrent backgrounds that had introgression of the critical region from the opposite parent were selected for QTL validation. Genotyping data of the BM-NILs to choose informative lines and perform QTL validation was obtained from Eichten et al. (2011).
Genotypic data for the MDL used in this study to perform GWA were obtained from third generation maize haplotypes (HapMap3) described in Bukowski et al. (2018). Out of the 343 inbred lines that were used to develop F1 hybrids for phenotyping, only 305 lines were genotyped as part of HapMap3. The HapMap3 consists of over 83 million variant sites across ten chromosomes of maize that are anchored to B73 version 3 assembly. After obtaining these genotypic data, variant sites were filtered for # 10% missing data and minor allele frequency of $ 0.05 using VCFtools (Danecek et al. 2011). This filtered SNP dataset was used for GWA analyses. A summary of the variant sites before and after the filtering procedure is in Table S5. The genotypic data for the set of 305 accessions were obtained from Romay et al. (2013). These genotypes consist of 681 257 SNPs (physical positions from B73 RefGenv2) obtained using a GBS protocol (Elshire et al. 2011) covering all ten chromosomes of maize. This marker dataset was filtered for # 10% missing data and minor allele frequency of $ 0.05 using TAS-SEL (Bradbury et al. 2007), reducing the marker number to 150 920 SNPs. This genotypic dataset was solely used to compute principal components (PC) and a kinship matrix to control for population structure and familial relatedness in a unified mixed linear model, respectively (Yu et al. 2006). To test for cis-eQTL at the oy1 locus in the maize diversity lines used for GWA mapping, normalized count of OY1 expression derived from the germinating seedling shoots was obtained from http://www.cyverse.org (Kremling et al. 2018).

Statistical Analyses
QTL mapping: Line-cross phenotypes and markers were used to detect and localize QTL using the R/QTL package version 1.40-8 (Broman et al. 2003). Trait means were used for the QTL analyses. Single interval mapping (SIM) was used for all traits, although composite interval mapping (CIM) was carried out with remarkably similar results (data not shown). Statistical thresholds were computed by 1000 permutations for each trait (Churchill and Doerge 1994).
Genome-wide association study (GWAS): Preliminary data analysis was done using JMP 13.0. Statistical corrections on the raw phenotypic data were performed by determining the most significant terms in the model using analysis of variance (Fisher 1921). Genotype and replication were used as a random effect in a linear mixed-effects model built in the lme4 package (Bates et al. 2015) implemented in R (R Core 2014) to calculate the best linear unbiased predictor (BLUP) for each trait. Broad-sense heritability (line mean basis) were calculated using BLUP values, using the method described by Lin and Allaire (1977). BLUP estimates for each trait were used to perform GWAS. GWAS was done using a compressed mixed linear model implemented in the R package GAPIT (Zhang et al. 2010;Lipka et al. 2012). HapMap3 SNPs were used to calculate genotype to phenotype associations. As explained before, kinship and population structure estimates were obtained for the same population using the second subset of 150 920 SNPs to correct for spurious associations. The Bonferroni correction and false discovery rate (FDR) adjustments were used to compute a statistical threshold for the positive association to further control for false positive assessment of associations (Holm 1979;Benjamini and Hochberg 1995).

Molecular Analyses
Genotyping: The recombinants in selected Syn10 lines and the BC1F1 population were detected using three PCR-based markers. Two markers detecting insertion polymorphisms flanking the oy1 locus and one dCAPS marker at oy1 locus were designed for this purpose. Genotyping at insertion-deletion (indel) marker ftcl1 (flanking an indel polymorphism in intron 4 of ftcl1; GRMZM2G001904) was performed with forward primer 5'-GCAGAGCTGGAATATGGAATGC-3' and reverse primer 5'-GATGACCTGAGTAGGGGTGC-3'. Genotyping at indel marker gfa2 (flanking an indel polymorphism in the intron of gfa2; GRMZM2G118316) was performed with forward primer 5'-ACGGCTCCAAAAGTCGTGTA-3' and reverse primer 5'-ATG-GATGGGGTCAGGAAAGC-3'. A polymorphic SNP in the second intron at oy1 was used to design a dCAPS forward primer 5'-CGCCCCCGTTCTCCAATCCTGC-3' and a gene-specific reverse primer 59-GACCTCGGGGCCCATGACCT-39 using a web user interface at http://helix.wustl.edu/dcaps/dcaps.html (Neff et al. 2002). The PCR products from polymerization reactions with the dCAPS oligonucleotide at oy1 were digested by PstI restriction endonuclease (New England Biolabs, MA, USA) and resolved on 3.5% agarose gel.
Allele-specific expression analyses: Allelic bias at transcriptional level was quantified using the third leaf of maize seedlings at the V3 developmental stage. Total RNA was extracted using a modified Phenol/ Lithium chloride protocol (Eggermont et al. 1996). Total RNA was subjected to DNase I treatment using Invitrogen Turbo DNA-free kit (Catalog# AM1907, Life Technologies, Carlsbad, CA) and 1 mg of DNase treated RNA from each sample was converted to cDNA using oligo dT primers and a recombinant M-MLV reverse transcriptase provided in iScriptTM Select cDNA synthesis kit (catalog# 170-8896, Bio-Rad, Hercules, CA) according to the manufacturer's recommendations. Besides the cDNA samples, genomic DNA samples were also prepared as a control to test the sensitivity of the assay. Genomic DNA controls included a 1 1 (F1 hybrid), 1 2, and 2 1 mixture of B73 and Mo17. PCR was conducted using gene-specific forward primer 5'-CAACGTCATCGACCCCAAGA-3' and reverse primer 5'-GGTTAC-CAGAGCCGATGAGG-3' for 30 cycles (94°for 30s, 56°for 30s, 72°f or 30s and final extension for 2 min) to amplify the OY1 gene product. These primers flank SNP252 (C-.T), which is the causative mutation of Oy1-N1989, and SNP317 (C-.T) which is polymorphic between B73 and Mo17 but monomorphic between the B73 and Oy1-N1989 genetic backgrounds. Corresponding PCR products were used to generate sequencing libraries using transposon-mediated library construction with the Illumina Nextera DNA library preparation kit, and sequence data were generated on a MiSeq instrument (Illumina, San Diego, CA) at the Purdue Genomics Core Facility. The SNP variation and read counts were decoded from the sequenced PCR amplicons by alignment of the quality controlled reads to the oy1 reference allele from B73 using the BBMap (Bushnell 2014) and the GATK packages (DePristo et al. 2011). Additional analysis was performed using IGV (Robinson et al. 2011) to manually quality-check the alignments and SNP calls. Read counts at polymorphic sites obtained from GATK was used to calculate allele-specific expression. Genomic DNA control samples showed bias in the read counts in a dosage-dependent manner. DNA from F1 hybrids between B73 and Mo17 resulted in 1:1 reads at oy1 demonstrating no bias in the assay to quantify expression.

Data Availability
The public genotypic and expression datasets leveraged in this study are open access at the data repositories mentioned above. The input files containing genotypes and phenotypes of both line-cross populations that were used for QTL detection are provided in the supplemental section for re-analyses. All of the supplemental data (Figures S1-S9, Tables S1-S23, and File S1-S2) was deposited at figshare. The seed stocks described in this study are available upon request. Supplemental material available at Figshare: https://doi.org/10.25387/g3.7370948.

RESULTS
Mo17 encodes an enhancer of the semi-dominant mutant allele Oy1-N1989 of maize The Oy1-N1989 allele was recovered from a nitrosoguanidine mutant population in mixed genetic background. The molecular nature of the mutation is a single non-synonymous base pair change (Sawers et al. 2006). Heterozygous Oy1-N1989 plants have the eponymous oil-yellow color but are reasonably vigorous and produce both ears and tassels. During introgression of the semi-dominant Oy1-N1989 allele into B73 and Mo17 inbred backgrounds, we observed a dramatic suppression of the mutant phenotype in F1 crosses of the mutant stock (obtained from Maize COOP) to the B73 background. In contrast, crosses to Mo17 enhanced the mutant phenotype. The difference in phenotype expression was stable and persisted in both genetic backgrounds through all six backcross generations observed to date. To further explore and quantify this suppression, B73, Mo17, as well as Oy1-N1989/oy1:B73, crossed with each of these inbred lines were grown to the V3 stage in the greenhouse. To improve upon our visual assessment of leaf color and provide quantitation, optical absorbance was measured using a Chlorophyll Content Meter-200 plus (CCM; Opti-Sciences, Inc), a handheld LED-based instrument. CCM is predicted to strongly correlate with chlorophyll and carotenoid contents. To confirm that our rapid phenotyping with the CCM would accurately assess chlorophyll levels, we measured the absolute pigment levels using a UV-VIS spectrophotometer (destructive sampling) on the same leaves used for CCM measurements. The non-destructive CCM measurements and the absolute pigment contents displayed a strong positive correlation with a R 2 value of 0.94 for chla, chlb, and total chlorophyll ( Figure S1). Given this high correlation of maize leaf greenness between the rapid measurement using CCM-200 plus instrument and absolute pigment contents quantified using UV-VIS spectrophotometer, we performed all chlorophyll measurements of Oy1-N1989 enhancement discussed in later results using CCM values.
In the greenhouse grown seedlings, chlorophyll accumulation was below the level of detection in the Oy1-N1989 homozygotes using CCM and spectrophotometric method. In wild-type plants, CCM measurements were slightly higher in B73 than Mo17, but the spectrophotometric method did not identify any significant difference in the amount of chlorophyll a (chla), chlorophyll b (chlb), total chlorophyll, or carotenoids between these two genotypes (Table S6). We detected a mild parent-of-origin-effect for both CCM and absolute amounts of chla, chlb, total chlorophyll, and carotenoids in the wild-type siblings of our F1 crosses. These plants had slightly higher chlorophyll accumulation when B73 was used as the pollen parent (Table S6). However, no such effect of parent-of-origin was observed for the mutant heterozygotes (Oy1-N1989/oy1) and reciprocal hybrid combinations of crosses between Oy1-N1989/oy1:B73 and Mo17 were indistinguishable. Further, both CCM and absolute chlorophyll contents were higher in the Oy1-N1989/oy1:B73 plants compared to the mutants in B73 x Mo17 hybrid background. Thus, there was a strong effect of genetics on chlorophyll pigment variation in mutants, that went opposite to predictions for hybrid vigor.
We tested the progenies from the crosses of Oy1-N1989/oy1:B73 with B73 and Mo17 inbred lines in the field. Consistent with our previous observation, mutant heterozygotes in the B73 inbred background were substantially greener than heterozygotes in the Mo17 x B73 F1 hybrid background ( Figure 1, Table S7). The increased severity of Oy1-N1989 heterozygotes in the Mo17 genetic background was stable across generations and observed even after six backcrosses ( Figure S2). No suppressed mutant plants were observed during any generation of backcrossing into Mo17 (data not shown). Thus, the expected negative impact of the Oy1-N1989 allele on chlorophyll pigment accumulation was dramatically suppressed by the B73 background compared to Mo17 genotype. vey1 maps to a single locus that co-segregates With the oy1 allele of Mo17 in DH, RIL, BC1F1 and NIL families derived From B73 and Mo17 To identify the genetic basis of the suppression of Oy1-N1989 allele in B73, we performed a series of crosses to four mapping populations. In each case, we crossed pollen from heterozygous Oy1-N1989 plants to a population of recombinant lines as ear-parents ( Figure 2). We chose two public populations, IBM and Syn10, to map all modifiers altering the severity of the Oy1-N1989 phenotypes. IBM and Syn10 differ in the number of rounds of intermating, and therefore vary in the number of recombinants captured and genetic resolution of trait localization (Lee et al. 2002;Hussain et al. 2007). Each F1 progeny of the testcross segregated approximately 1:1 for wild-type (oy1/oy1) and mutant heterozygote (Oy1-N1989/oy1) in the hybrid genetic background with B73. Mutant heterozygote siblings in both (IBM and Syn10) F1 populations, chlorophyll approximation using CCM measured at an early (CCMI) and late (CCMII) (Tables S8 and S9; Figures S3  and S4). A similar trend was also observed for the CCMI and CCMII in the mutant F1 siblings. The CCM measurements in the wild-type and mutant siblings did not display any significant correlation. To control for variation in CCM observed due to the genetic potential of each line that was independent of the Oy1-N1989 modification, we divided the mutant CCM trait values by the congenic wild-type sibling CCM values to derive ratio for both time points. We also calculated differences between congenic wild-type and mutant CCM values. Each of the direct measurements, as well as the ratio-metric and difference values, were used as phenotypes to detect and localize QTL.
Summary of the peak positions of all QTL passing a permutation computed threshold are presented in Table S10 for the IBM F1 populations and Table S11 for the Syn10 F1 populations. All mutant CCM traits, all mutant to wild-type CCM ratios, and all differences between mutant and wild-type CCM measurements identified a single QTL of large effect on chromosome 10. We name this suppressor locus very oil yellow1 (vey1). A plot of the log10 of odds (LOD) score with permutation calculated threshold for CCMII from the mutant siblings in the Syn10 F1 population is plotted in Figures 3C and 3D. Other mutantrelated traits in Syn10 and IBM F1 populations produced similar plots (data not shown). One additional minor effect QTL was identified from these data, and it only influenced the early chlorophyll measurements in wild-type siblings of the IBM F1 population (Table S10). Interestingly, vey1 mapped to the same genetic position as oy1 locus itself ( Figures 3E and 3F) and there was no influence of this locus on CCM traits in wild-type siblings. These analyses indicate that the vey1 QTL encodes a single locus with an effect contingent upon the presence of the Oy1-N1989 allele.
The high penetrance of the MT_CCMII trait allowed us to interpret a discordance between the marker genotype and F1 mutant phenotype of high or low CCM categorization as recombination between vey1 and the given marker. We classified each F1 mutant hybrid in the Syn10 and IBM F1 populations as high or low CCMII using the bimodal distribution. We then compared the marker genotypes at each marker linked to vey1 with the phenotypic categories. Table S12 shows the mutant trait values, marker genotypes, and phenotypic categories for the nine F1 Syn10 lines with recombinants within the vey1 region (between the flanking markers 10.90.5 and 10.95.5). A summary of the recombinants between the markers and the phenotypic class at each marker position across the vey1 region is presented in Figures 3E and 3F. Genotypes at marker 10.93 perfectly predicted CCMII trait expression in Syn10 F1 mutant siblings and recombinants placed this QTL within the 227 kb interval between markers 10.94.5 and 10.90.5 ( Figure 3E). This vey1 region includes the oy1 locus itself, suggesting that the Mo17 allele of oy1 may enhance the impact of the Oy1-N1989 allele. Three Syn10 DH lines recombinant within this critical region were genotyped with additional markers developed at ftcl1 (GRMZM2G001904), and gfa2 (GRMZM2G118316). Three recombinants were detected between vey1 and ftcl1 but no recombinants were recovered between the gfa2 indel marker and vey1. A SNP marker used in the Syn10 map construction (10.94.5) at the distal end of the gfa2 locus identified one recombinant ( Figure 3E). Thus, vey1 is localized between ftcl1 and gfa2. This region includes oy1, ereb28, a small region of gfa2 (from marker 10.94.5), and ftcl1 proximate to oy1. A similar fine-mapping attempt with the IBM F1 population provided no additional resolution ( Figure 3F). The genotyping of 576 BC1F1 individuals from (B73 x Mo17) x Oy1-N1989/oy1:B73 crosses failed to recover any recombinants between vey1 and gfa2 ( Figure 3E).
We further explored background influence on the vey1 alleles from B73 and Mo17 using a series of BM-NILs that contained the vey1 region introgressed into a homogeneous background of either B73 or Mo17. Progeny from crosses of BM-NILs with Oy1-N1989/oy1:B73 displayed the expected mutant CCM phenotype based on the vey1 introgression regardless of the recurrent parent ( Figure S6 and Table S3). These data confirm the expectation of the QTL mapping but offered no additional recombinants within the identified vey1 QTL.
Thus, the formal list of candidate genes for the vey1 QTL is the oy1 gene itself and the three most closely linked loci: ftcl1; Figure 1 The chlorophyll pigment accumulation differs in the Oy1-N1989/oy1 heterozygotes in the B73 and B73 x Mo17 hybrid backgrounds. Representative wild-type and mutant siblings from (A) B73 x Oy1-N1989/oy1:B73, and (B) Mo17 x Oy1-N1989/oy1:B73 F1 crosses in the fieldgrown plants. Measuring stick in panels A and B is 243 cm. (C) Non-destructive chlorophyll approximation in mutant and wild-type siblings at an 3 weeks (CCMI) and 6 weeks (CCMII) after planting; data for each class of genotype is derived from 39 replications planted in RCBD. Asterisks indicate statistical significance between the means in each genotypic category at P , 0.01 using student's t-test.

Figure 2
The crossing scheme used to map Oy1-N1989 enhancer/suppressor loci in IBM and Syn10 populations. Red, blue and white colors indicate B73, Mo17, and missing genotypes, respectively. The heterozygous tester shows chromosome 10 with a black spot indicating Oy1-N1989 mutant allele. F1 progenies depicted here shows hypothetical genotypes of chromosome 10 for each F1 wild-type and mutant siblings.
gfa2; ereb28. The ortholog of Zmftcl1 (62% protein identity) from Arabidopsis thaliana is required for folate metabolism (Pribat et al. 2011). The maize gene gfa2 is uncharacterized, but mutation of the Arabidopsis ortholog caused defects in megagametogenesis including failures of polar nuclear fusion in the female gametophyte and synergid cell-death at fertilization (Christensen et al. 1998(Christensen et al. , 2002. The third linked gene, ereb28 (Apetela2-Ethylene Responsive Element Binding Protein-transcription factor 28) has a highly conserved AP2/EREB domain. This gene has a very low expression level and is localized only to the root tissue of maize (https://www.maizegdb.org/gene_center/ gene?id=GRMZM2G544539#rnaseq).

Controlling for the vey1 QTL did not reveal additional modifiers of Oy1-N1989 phenotype expression
The non-normality of some of the trait distributions and apparent thresholds prompted us to explore additional QTL models. No additional QTL were recovered by implementing two-part threshold models (Broman et al. 2003) for any of the traits (data not shown). Similarly, two-way genome scans also failed to detect any statistically significant genetic interactions with vey1 or between other loci. In both the IBM and the Syn10, the region encoding vey1 exhibited substantial segregation distortion with the B73:Mo17 alleles present at 120:72 in the IBM and 175:76 in the Syn10 population. This uneven sample size will reduce the power to detect epistasis with vey1 but would not limit the detection of additional unlinked epistatic modifiers of Oy1-N1989.
We used the top marker at vey1 as a covariate to control for the contribution of this allele to phenotypic variation and performed a onedimensional scan of the genome (Broman et al. 2003). In our previous naïve one-dimensional scans, the large effect of vey1 partitioned into the error term and might reduce our power to detect additional unlinked QTL(s). By adding a marker linked to vey1 as a covariate, this term will capture the variance explained by vey1 and could improve detection of additional QTL(s) of presumably smaller effect. In both populations, use of a vey1 linked marker as a covariate did not identify any additional QTL for any trait (data not shown). Thus, modification of the Oy1-N1989 phenotype by vey1 was inherited as a single QTL, acting alone.

GWAS for chlorophyll content in maize diversity lines (MDL) and Oy1-N1989/oy1 F1 genotypes identifies vey1
We undertook GWA mapping of Oy1-N1989 severity to search for additional loci and potentially identify recombinants at vey1. A population of 343 lines was crossed to Oy1-N1989/oy1:B73. This generated MDL x Oy1-N1989/oy1:B73 F1 populations segregating 1:1 for mutant and wild-type siblings in hybrid genetic background. There was total separation between mutant and wild-type siblings for the CCMI and CCMII traits ( Figure S5C). The mutant severity of some F1 families was similar to the F1 progeny of Mo17 x Oy1-N1989/oy1:B73 ( Figure  S7). Pairwise correlations of the CCM trait measurements at two time points (CCMI and CCMII) in the wild-type siblings displayed statistically significant positive relationship (Table S13). CCM traits were much more strongly correlated in the mutant F1 siblings, similar to the B73 x Mo17 F1 populations. However, weak positive correlations were also observed between mutant and wild-type CCM measurements in the MDL F1 populations. Broad-sense heritability estimates for CCMI and CCMII were high for the mutant and ratio traits, whereas wild-type siblings had much lower repeatability (Table S14).
GWA was performed using the 305 inbred lines from MDL crosses that were present in the HapMap3 SNP data set (Bukowski et al. 2018). The variation in MT_CCMI, MT_CCMII, and their ratios identified a region encoding multiple SNPs that passed a multiple test correction (see Methods) on chromosome 10 at the site of the oy1 gene. No other loci passed the corrected threshold in the GWA analysis. No statistically significant SNPs were detected for the CCM traits of wild-type siblings. The plot showing the negative log10 of the p-values from GWA tests for all the SNPs for MT_CCMII trait is graphed in Figure 4A. A closer view of the SNPs within the region encoding the vey1 locus on chromosome 10 are plotted in Figure 4C. A summary of the GWAS results for mutant CCM and ratio traits is presented in Table S15. The top association for the mutant CCM traits was a SNP at position 9161643 on chromosome 10 (S10_9161643) located just 3' of the oy1 protein-coding sequence. S10_9161643 displays high allelic frequency in our population (f = 0.49). Thus, it appears that variation at the oy1 locus may be responsible for the suppression of the Oy1-N1989 mutant allele in the diverse panel of maize inbred lines analyzed in this experiment. Analysis of the LD between S10_9161643 and the other variants in this region identified no other variants with r 2 greater than 0.5 despite the relatively strong associations between many SNPs and the CCM traits ( Figure 4E). LD was substantially higher for SNPs encoded toward the telomere from oy1 than toward the centromere, with a strong discontinuity of LD at the 3'-end of oy1. Given the relatively low p-values calculated for multiple SNPs in the area, this raises the possibility that multiple alleles contribute to the suppression of the Oy1-N1989 phenotype. To test this hypothesis, the genotypes at SNP S10_ 9161643 were used as a covariate, and the genetic associations were recalculated. If SNPs segregate independently of S10_9161643 and contribute to Oy1-N1989 suppression, the p-values of association test statistics should stay significant. On the contrary, those SNPs that have relatively low p-values due to linkage with S10_9161643 should become statistically insignificant in the covariate model. When these analyses were done, low-frequency variants at the 5' end of the oy1 locus were identified as the most significant SNPs and passed a chromosome-wide multiple test correction ( Figure 4B and 4D). This result suggests that there are multiple alleles capable of modifying the Oy1-N1989 mutant phenotype in the MDL panel. The top SNP on chromosome 10 in the covariate model of MLM was detected at position 9179932 (S10_9179932), and the allele associated with Oy1-N1989 suppression is a relatively low-frequency variant (f = 0.08). It remains formally possible that the SNP S10_9179932 is not causative and merely in LD with a causative polymorphism, and the second locus is fortuitously present in recombinant haplotypes. Analysis of LD of S10_9179932 with other SNPs in 500 kb window detected multiple SNPs of low allelic frequency that were in high LD (r 2 0.85) toward the 5' end of oy1 ( Figure 4F). Consistent with the strong discontinuity of LD at 3' end of oy1 with S10_9161643, we observed discontinuity of LD with S10_9179932 at 5' end of the oy1 locus. The analyses suggest that S10_9161643 and S10_9179932 are not in LD with each other and can act independently. The MLM model that considered only these two SNPs accounted for 29% of the variation in CCMII measurements of the mutant siblings (Table S15).
Given that the MLM model using S10_9161643 as a covariate detected S10_9179932 as the most significant association, we tested the phenotypic outcome of the four possible haplotypes at these two SNPs. We observed that the four haplotypes varied only for mutant CCM traits, with haplotypes AG and CA being the most favorable (highest CCM mean) and least favorable (lowest CCM mean), respectively (Table S16). Alleles at these two SNPs affected CCMI and CCMII in the mutant plants, consistent with additive inheritance for two polymorphisms. The additive effect of the SNPs is consistent with independent cis-regulatory alleles at oy1 modifying the Oy1-N1989 mutant phenotype. Consistent with the strong enhancement caused by crossing Mo17 to Oy1-N1989/oy1:B73, the Mo17 oy1 locus encodes the most severe, and relatively rare, CA allele combination, and B73 encodes the most suppressing AG allele combination (Table S16). Thus, the line-cross mapping was performed with inbred lines that carry phenotypically extreme allele combinations of these two SNPs in the vicinity of the oy1 locus.

Oy1-N1989 is a semi-dominant chlorophyll mutant and enhanced by reduced function at oy1
Maize seedlings that are double heterozygotes for Oy1-N1989 and hypomorphic oy1 allele (chlI-MTM1) are more severe than isogenic Oy1-N1989/oy1 siblings (Sawers et al. 2006). To confirm that the reduced oy1 function could determine the differential sensitivity to Oy1-N1989, we crossed dominant and recessive mutant alleles of oy1 to each other. The recessive weak hypomorphic allele oy1-yg was obtained from the Maize COOP in the unknown genetic background. The homozygous oy1-yg plants were crossed as a pollen-parent with both B73 and Mo17 to develop F1 material that would segregate the mutation. The F1 plants were then crossed to Oy1-N1989/oy1:B73 as well as backcrossed to the oy1-yg homozygotes in the original mixed background. These crosses allowed us to recover plants that had Oy1-N1989 in combination with the wild-type oy1-B73, wild-type oy1-Mo17, and mutant oy1-yg alleles. Chlorophyll contents were determined using CCM at 21 and 40 days after sowing in the field. The Oy1-N1989 allele was substantially enhanced when combined with the oy1-yg allele, demonstrating that reduced function of the oy1 allele in Mo17 could be the genetic basis of vey1 QTL (Figures 5A and 5B). A summary of these data are presented in Table S17. This result is similar to the one described by Sawers et al. (2006), where a reduction in chlorophyll content was observed when Oy1-N1989 allele was combined with a recessive allele of oy1. We also noticed a similar drop in chlorophyll accumulation in the oy1-yg homozygotes as oppose to oy1-yg heterozygotes with wild-type oy1 allele from both B73 and Mo17 in the BC1F1 progenies ( Figure 5C and Table S17). However, we did not observe any significant difference in the oy1-yg heterozygotes with B73 and Mo17 wild-type oy1 allele. Selfed progeny from Oy1-N1989 heterozygotes segregated for yellow-seedling lethal Oy1-N1989 homozygotes with no detectable chlorophyll by either CCM or spectrophotometer quantification (Table S6). Therefore, consistent with previous work (Hansson et al. 2002;Sawers et al. 2006), the Oy1-N1989 is a dominant-negative neomorphic mutant allele with no evident MgChl activity under the tested conditions. Based on these genetic data, any QTL resulting in decreased expression of oy1 or an increased proportion of mutant to wild-type gene product in the Oy1-N1989/oy1 heterozygotes can increase the severity of the mutant phenotype.
No coding sequence difference in OY1 accounts for vey1 inheritance Our reliance on SNP variation leaves us open to the problem that linked, but the unknown non-SNP variation can be responsible for vey1. Given that reduced oy1 activity enhanced the phenotype of Oy1-N1989, we sequenced the oy1 locus from Mo17 and B73 to determine if coding sequence differences could encode the vey1 modifier. The only nonsynonymous changes that distinguish these two alleles is at the site of the previously reported in-frame 6 bp insertion (Sawers et al. 2006), which adds alanine (A) and threonine (T) amino acid residues to the OY1 protein. PCR amplification of oy1 locus in 18 maize inbred lines, as well as the Oy1-N1989 allele, was performed. Sequencing of the PCR products confirmed the absence of the 6 bp insertion in Oy1-N1989 allele reported by Sawers et al. (2006). In addition, multiple inbred lines Figure 4 The Manhattan plots of SNP associations with MT_CCMII in MDL x Oy1-N1989/oy1:B73 F1 population. Genome-wide association of (A) MT_CCMII with no marker covariate, (B) MT_CCMII using S10_9161643 as a covariate. Close-up views of the SNP associations in the vey1 region on chromosome 10 for (C) Panel A, (D) Panel B. Arrows in panels A-D identify SNPs S10_9161643 and S10_9179932. Horizontal solid red and hashed red lines in panels A-D indicate the genome-wide Bonferroni cut-off at P , 0.05, and hashed golden line in panels C-D is a chromosomewide FDR-adjusted threshold at P , 0.05. Linkage disequilibrium of all SNPs in a 450 kb region flanking oy1 with SNPs (E) S10_9161643, and (F) S10_9179932. Vertical lines in panels C-F from left to right represent genomic position of ftcl1, ereb28, oy1 (green), and gfa2.
including B73, CML103, and CML322 also carried this 6 bp in-frame deletion. A polymorphism within the 6 bp insertion was also found that resulted in an alternative in-frame insertion encoding an alanine and serine (S) codon in Mo17 and five other inbred lines. Thus, three alleles at this site were found to be a common variant in OY1 gene product. These allelic states of oy1 did not explain the phenotypic severity of CCM trait value in the F1 mutant siblings (Figure 6). The allelic state of oy1 at this polymorphic site in 18 maize inbred lines and the average CCM trait values in the wild-type and mutant siblings of their respective F1 progenies with Oy1-N1989/oy1:B73 are summarized in Table  S18. Five inbred lines, including Mo17, resulted in dramatic enhancement of the mutant CCMI and ratio of CCMI phenotypes of F1 plants crossed to Oy1-N1989/oy1:B73. These enhanced genotypes encoded all three possible alleles at oy1. In addition, the suppressing inbred lines also encoded all three possible alleles. Besides this 6 bp indel, three inbred lines had few more variants in OY1 protein. An enhancing inbred line CML322 had two missense mutations that lead to amino acid change at position 321 (D-.E), and 374 (S-.I). A suppressing inbred line NC358 had one amino acid change at position 336 (D-.G) and the enhancing inbred Tzi8 had a 15 bp in-frame deletion leading to the removal of five amino acids (VMGPE) in the third exon of the coding sequence. Even considering the additional alleles at oy1 found in few maize inbred lines, these results suggest that the only coding sequence polymorphism at oy1 between B73 and Mo17 could not be the genetic basis of vey1. This result leaves the two additive top SNPs in cis with oy1 as the most likely cause of cis-acting regulatory variation.
Expression level polymorphism at oy1 co-segregates with suppression of Oy1-N1989 mutant phenotype Measurements of mRNA accumulation from oy1 in the IBM was publicly available in a previously published study (Li et al. 2013(Li et al. , 2018. Out of 105 IBM lines that were assessed for expression level in the past study, 74 were among those tested for chlorophyll accumulation in the Oy1-N1989 F1 hybrid populations. Using a genetic marker linked to the oy1 locus, we determined that a cis-acting eQTL controlled the accumulation of OY1 transcripts in IBM shoot apex (Figure 7). A summary of these data are presented in Table S19. This cis-acting eQTL conditioned greater expression of the B73 allele and explained 19% variation in OY1 transcript abundance in the IBM. Given the enhancement of Oy1-N1989 by the oy1-yg allele, a lower expression of the wild-type oy1 allele from Mo17 is expected to enhance the phenotype of Oy1-N1989 ( Figure 5). In addition, OY1 RPKM values obtained from the shoot apex of IBM were able to predict the CCM trait values in the mutant but not the wild-type siblings in IBM F1 population with Oy1-N1989/oy1: B73 (Figure 7). This result suggests that inbred lines with increased MgChl subunit I transcripts available for protein production and MgChl complex assembly could overcome chlorophyll accumulation defects caused by the Oy1-N1989/oy1 genotype in IBM F1 hybrids. Consistent with this, a full linear regression model that included both the isu085b marker (cis-eQTL) genotypes and the residual variation in RPKM at OY1 did a better job in predicting CCMI and CCMII in the IBM mutant F1 hybrids than the isu085b marker by itself. If the cis-eQTL at oy1, which results in differential accumulation of OY1 transcripts in the IBM inbred lines can affect allele-specific expression in the F1 hybrids, it could explain the better performance of the IBM mutant F1 hybrids with the B73 allele at vey1.
A previous study of allele-specific expression in the F1 hybrid maize seedlings identified expression bias at oy1 toward B73 in the hybrid combinations of B73 inbred line with PH207 and Mo17 but not Oh43 (Waters et al. 2017). We used two SNP positions, SNP_252 and SNP_317, to explore the allele-specific expression of OY1 in our materials. SNP_252 is the causative polymorphism for the Oy1-N1989 missense allele while SNP_317 is polymorphic between B73 and Mo17, but monomorphic between Oy1-N1989 and B73. As the original allele of the Oy1-N1989 mutation was isolated from a r1 c1 colorless synthetic stock of mixed parentage, this raises the possibility that the same cis-acting regulatory variation that lowered expression of OY1 from PH207 and Mo17 when combined with the B73 allele might also be present in the oy1 allele that was the progenitor of Oy1-N1989. We tested this possibility by using the SNPs that distinguish B73, Mo17, and the Oy1-N1989 alleles to measure allele-specific expression in each of the hybrids. Consistent with the previous data (Waters et al. 2017), we observed a biased expression at oy1 toward the B73 allele in the B73 x Mo17 F1 wild-type hybrids (Table 1). Extended data from this experiment is provided in Table S20. In the B73 isogenic crosses, transcripts from the Oy1-N1989 and B73 wild-type alleles accumulated to equal levels in the heterozygotes, indicating that the suppressed phenotype of the mutants in B73 background was not due to a lowered expression of Oy1-N1989 relative to the wild-type allele. Remarkably, mutant siblings from the reciprocal crosses between Oy1-N1989/oy1: B73 and Mo17 resulted in greater expression from the Oy1-N1989 allele than the wild-type oy1 allele of Mo17. Allele-specific bias at oy1 was Figure 5 The single locus test of oy1 showing the interaction between wild-type alleles of oy1 from B73 and Mo17 with the semi-dominant (Oy1-N1989), and recessive mutant allele (oy1-yg) at oy1. (A) Mutant (two severity groups) and wild-type individuals segregating in a cross (B73 x oy1yg/oy1-yg) x Oy1-N1989/+:B73. White-fill arrows indicate Oy1-N1989/+ plants (pale-green and suppressed), whereas yellow-fill arrows indicate Oy1-N1989/oy1-yg (yellow-green and severe) plants. CCM measurements of the test crosses at 21 and 40 days after planting in the (B) Mutant siblings (Oy1-N1989/oy1-yg and Oy1-N1989/+) of (Mo17 x oy1-yg/oy1-yg) x Oy1-N1989/+:B73, and (B73 x oy1-yg/oy1-yg) x Oy1-N1989/+:B73 crosses, (C) Mutant (oy1-yg/oy1-yg) and wild-type (oy1-yg/+) siblings of (Mo17 x oy1-yg/oy1-yg) x oy1-yg/ oy1-yg and (B73 x oy1-yg/oy1-yg) x oy1-yg/ oy1-yg crosses. Asterisk in panels B-C indicate statistically significant difference between the genotypes in a given cross at P , 0.01 using student's t-test.
significantly higher toward the Oy1-N1989 allele in the Oy1-N1989 mutant heterozygotes in the B73 x Mo17 hybrid background compared to B73 isogenic material. Thus, in mutant hybrids, overexpression of Oy1-N1989 relative to the wild-type oy1 allele in Mo17 could account for increased phenotypic severity.
If vey1 is encoded by an eQTL, then PH207 should encode an enhancing allele and Oh43 should encode a suppressing allele of vey1. We tested this genetically by producing F1 progenies in crosses of PH207 and Oh43 by Oy1-N1989/oy1:B73 pollen. Oh43 was evaluated in our initial screening and also as part of the MDL panel used for GWAS. In both experiments, the F1 hybrids between Oh43 and Oy1-N1989/oy1:B73 suppressed the mutant phenotype, suggesting that Oh43 is a suppressing inbred line (CCM values in Tables S4 and S18). B73 x PH207 F1 hybrids were missing from our previous datasets. We crossed PH207 ears with pollen from Oy1-N1989/ oy1:B73 plants. The F1 hybrids from this cross were analyzed in the greenhouse at seedlings stage along with F1 hybrids of B73 x Oy1-N1989/oy1:B73 and Mo17 x Oy1-N1989/oy1:B73 F1 progenies as controls. PH207 was an enhancing inbred genotype as mutant heterozygotes in a PH207 x B73 F1 genetic background accumulated less chlorophyll than mutants in the B73 isogenic background (Figure S8 and Table S21).  oy1 and vey1). Asterisk indicate significant difference in the mean between two groups using student's t-test at P , 0.01. Linear regression of OY1 expression in IBM RILs on CCMII in the (B) Wild-type and (C) Mutant siblings derived from IBM x Oy1-N1989/oy1:B73 crosses. Figure 6 The distributions of CCM trait measurements in the F1 progenies of a sub-set of maize inbred lines crossed with Oy1-N1989/oy1:B73 divided using three allelic variants in the oy1 coding sequence of the respective inbred line. Phenotypic distribution of (A) MT_CCMI, (B) WT_CCMI, (C) MT_CCMII, and (D) WT_CCMII. Symbols "-", "AS", and "AT" on the X-axis of each panel denote deletion of 6 base pairs (bp), insertion of amino acid residues Alanine-Serine (AS), and Alanine-Threonine (AT), respectively, in a given inbred line. Three inbred lines including B73 carried "-" allele, six inbred lines carried "AS" insertion, nine inbred lines carried "AT" insertion. No statistically significant differences were found among three categories using ANOVA.
We further leveraged the normalized expression data of OY1 in the emerging shoot tissue of the maize diversity lines (Kremling et al. 2018) and used the top two additive SNPs (S10_9161643 and S10_9179932) at vey1 from GWAS to test if these cis-variants of oy1 affect its expression. Alleles that suppress Oy1-N1989 linked to either SNP were associated with the greater abundance of OY1 transcripts. Plants carrying the B73-like allele combination (A and G at marker S10_9161643 and S10_9179932, respectively) showed the highest OY1 abundance in the emerging shoots of diverse maize inbred lines, whereas, plants with the Mo17-like allele combination (C and A at S10_9161643 and S10_ 9179932, respectively) showed lowest OY1 count (Table S22). Consistent with the additive suppression of leaf greenness in Oy1-N1989 mutants by the alleles at S10_9161643 and S10_9179932 as discussed previously (Table S16), these alleles were also additive for their impacts on OY1 transcript abundance (Table S22). This observation is consistent with the hypothesis that increased OY1 abundance can overcome the negative effect of the Oy1-N1989 allele. It is likely that multiple phenotypically-affective cis-acting regulatory polymorphisms at oy1 are responsible for the vey1 locus.
The effect sizes of the gene expression changes observed in the IBM, diverse maize inbred lines, and allele-specific expression in hybrids are quite modest, resulting in 10% of differences in oy1 accumulation. If these changes in wild-type OY1 transcript accumulation are responsible for suppression of Oy1-N1989, then the severity of the mutant phenotype (as indicated by CCM) in the MDL F1 population should correlate with OY1 expression level. As expected, we observed a statistically significant positive correlation between the mutant derived CCM traits and OY1 counts (Table S23). Wild-type CCM in the MDL F1 population did not show any significant correlation with OY1 abundance in the emerging shoot tissue of maize inbred lines. These correlations are in agreement with the lack of any QTL at this locus controlling wild-type chlorophyll levels, and the epistatic relationship between vey1 and Oy1-N1989.

DISCUSSION
The semi-dominant mutant allele Oy1-N1989 encodes a dominantnegative allele at the oy1 locus (Sawers et al. 2006). In a heterozygous condition, the strength of the negative effect of this allele on the MgChl enzyme complex depends on the wild-type oy1 allele. Thus, the Oy1-N1989 allele can sensitize maize plants to variation in MgChl and expose a phenotypic consequence for genetic variants that are otherwise invisible. Similar methodology has been adopted previously in maize to gain the genetic understanding of various traits (Chintamanani et al. 2010;Olukolu et al. 2013Olukolu et al. , 2014Buescher et al. 2014). Thus, genetic screens based upon semi-dominant mutant alleles as reporters offer a cost-effective and robust approach to map QTL(s) for metabolic pathways of interest by leveraging the publicly available germplasm.
Alleles with dramatic fitness consequences may be visible to researchers working with population sizes in the thousands, such as in GWAS. By contrast, evolutionarily-relevant genetic variation may have minimal phenotypic effects in laboratory and agronomic field conditions. The cryptic variation observed by these mutant-contingent QTL approaches may not be fitness-affecting, and interpreting mutantconditioned phenotypes as non-neutral variation would be a mistake. It is, of course, possible that neutral variants may result in increased severity of a mutant phenotype due to changes not physiologically relevant for all alleles of that reporter gene present in a species. Nevertheless, it can inform us about the allelic variation in the species and pathway topology via gene discovery.
In the current study, use of bi-parental mapping populations derived from the same inbred lines but developed using different breeding schemes provided the opportunity to compare the effect of additional rounds of random interbreeding in the development of mapping population on the genetic resolution. The comparative fine mapping of vey1 in the Syn10 population that employed ten rounds of random mating provided far better localization of the vey1 QTL than the IBM populations that was derived from four rounds of random mating. This observation demonstrates the benefits of increased recombination during random intermating of early generations in QTL localization. Based on these results, as future RILs are generated, we recommend increased intermating in the early generations followed by DH induction rather than relying on further recombination during the self-pollination cycles of RIL development.
This study illustrated a complementary use of line-cross QTL mapping and GWAS to explore trait genetics. As expected, GWAS provided a fine-scale genetic resolution and corroborated the mapping of vey1. GWAS identified two SNPs, one in the 5' and another in the 3' intergenic DNA, proximate to oy1 that represent candidate quantitative trait nucleotides (QTN). The signal detection by multiple unlinked SNPs in GWAS may indicate a complex set of phenotypically-affective alleles at the oy1 locus. Alternatively, it could be an artifact of strong associations with markers that are tightly linked to an unmeasured causative variant but unlinked or in repulsion to each other. Consistent with two QTN, rather than fortuitous linkage of tag SNPs with a single causative polymorphism, alleles at the two SNPs additively influenced chlorophyll contents in the mutant siblings. Future work will be required to identify the nucleotide changes sufficient to reproduce the vey1 phenotype.
We used a non-destructive, inexpensive, and rapid phenotyping method to approximate leaf chlorophyll with high accuracy. Previous studies have highlighted the importance of benchmarking indirect measurements or proxies for traits of interest. For instance, near infrared reflectance spectroscopy (NIRS) that estimates major and total carotenoids in maize kernels could replace sensitive, accurate, cumbersome, expensive, slow, and destructive measurements by HPLC (Berardo et al. 2004).
The connecting letter report indicates the statistical significance calculated using ANOVA with post-hoc analysis using Tukey's HSD with P , 0.01.
b Data obtained from Waters et al. (2017).
Comparisons of HPLC and NIRS measurements resulted in a correlation of 0.85 for total carotenoids (Berardo et al. 2004). A subjective visual score for yellow color in maize kernels was not adequate, yielding a correlation of 0.12 with HPLC measurements of total carotenoids (Harjes et al. 2008). Thus, using mutant alleles as reporters to develop methodologies that rely on non-invasive multispectral data as proxies for absolute quantification of biochemical compounds will accelerate studies on gene function and allele discovery. This can enable genetic studies that are currently deemed unfeasible due to the arduous task of phenotyping large populations for traits only visible in the laboratory.
How could a 10% change in wild-type OY1 expression affect chlorophyll biosynthesis in the Oy1-N1989/oy1 mutant heterozygotes? Magnesium chelatase (MgChl) is formed by a trimer of dimers of MgChl subunit I interacting with the other subunits of MgChl complex. A previous study found that addition of mutant or wild-type BCHI protein to pre-assembled MgChl complexes resulted in altered reaction rates due to differences in subunit turnover, which occurred on a minutes timescale (Lundqvist et al. 2013). This subunit turnover and reformation of the complex dynamically exchange mutant and wild-type BCHI subunits over time. Therefore, any net increase in the amount of wild-type OY1 in the reaction pool, for instance, due to higher transcription of wild-type oy1 allele will allow a higher rate of magnesium chelatase activity and result in more chlorophyll biosynthesis. The observation of stronger affinity and greater dissociation rate of BCHI L111F subunits (orthologous to the L176F change encoded by Oy1-N1989) for the wildtype subunits (Hansson et al. 2002) suggests that exchange of BCHI monomeric units in the magnesium chelatase complex might also differ based on the structure of BCHI protein (Lundqvist et al. 2013). In the AAA+ protein family, the ATP-binding site is located at the interface of two neighboring subunits in the oligomeric complex (Vale 2000). Since a dimer of functional MgChl subunit I proteins are required for the complex to carry out the MgChl activity (Lundqvist et al. 2013), approximately 1 in 3 dimers of assembled MgChl subunit I will be active in a 1:1 mixture of wild-type and BCHI L111F : Indeed, complexes made from reaction mixtures with equal proportions of wild-type and BCHI L111F subunits resulted in 26% of the enzyme activity of an equivalent all-wildtype mix (Lundqvist et al. 2013). Therefore, we expect that decreasing expression of the wild-type oy1 subunit by 10% and creating a 0.9:1.1 mixture of wild-type OY1 and mutant OY1-N1989 protein, respectively, would result in 21% activity compared to the activity of all-wildtype mixture. Likewise, increasing the wild-type oy1 expression by 10% would result in 30% activity of MgChl compared to the all-wildtype mixture. This dosage-sensitivity is a general feature of protein complexes (Birchler and Newton 1981;Grossniklaus et al. 1996;Veitia 2003;Birchler and Veitia 2012), and the semi-dominant nature of Oy1-N1989 is dosage sensitive. Taking these observations and proposed models on the dynamics of molecular interaction between the wild-type and mutant BCHI protein subunits (especially BCHI L111F ) into account, it is formally possible that a small change in the expression of wild-type OY1 can have a significant impact on the magnesium chelatase activity of heterozygous Oy1-N1989/oy1 plants. The increase in magnesium chelatase activity due to the even small relative increase in the proportion of wild-type OY1 transcripts over the mutant OY1-N1989 transcripts will read out as a proportional, presumably non-linear, increase in chlorophyll accumulation.
What is vey1? Variation at the vey1 locus appears to be the result of allelic diversity linked to the oy1 locus. The only remaining possibilities are regulatory polymorphisms within the cis-acting control regions of oy1. Previous studies have utilized reciprocal test-crosses to loss-of-function alleles in multiple genetic backgrounds to provide single-locus tests of additive QTL alleles in an otherwise identical hybrid background (Dilkes et al. 2008). Protein-null alleles of oy1 isolated directly from the B73 and Mo17 backgrounds could be used to carry out a similar test. The intergenic genomic region in maize is spanned by transposable elements and can be highly divergent between different inbred lines due to large insertion/deletion polymorphisms (SanMiguel and Bennetzen 1998). Consistent with this, inbreds B73 and Mo17 are polymorphic at the region between oy1 and gfa2. These two maize inbred lines share 12 kb of sequence interspersed with numerous large insertions and deletions that add 139 kb of DNA sequence to Mo17 as compared to B73 (data not shown). However, we did not find any conserved noncoding sequence (CNS) in this region (data not shown). It is conceivable that recombinants at oy1 between B73 and Mo17 themselves could be identified and used to test the effects of upstream or downstream regulatory sequences. The recombinant haplotypes encoding all four possible alleles at the top two SNPs identified in the GWAS indicate that Mo17 may be a strong enhancer due to more than one causative polymorphism. Allele-specific expression at oy1 locus in the Oy1-N1989 heterozygous mutants demonstrated the existence of functional cis-acting regulatory polymorphism between Oy1-N1989 and both wild-type oy1 alleles in B73 and Mo17. In addition, oy1 is affected by a cis-eQTL in the IBM and MDL. Together with our other data, the allele-specific expression at OY1 that was visible when we re-analyzed the data from Waters et al. (2017) led us to propose that vey1 encodes a cis-acting regulatory DNA sequence variation ( Figure  S9). Sequence comparisons outside the protein coding sequence of the gene can be quite challenging, especially in maize, as it exhibits limited to poor sequence conservation between different inbred lines (SanMiguel et al. 1996). Thus, distinguishing phenotypically-affective polymorphisms from the neutral variants, e.g., via transgenic testing, is not trivial. As a result, biochemical and in vitro studies are the best tool for functional validation of these polymorphisms (Wray et al. 2003). Similar experiments have been done to characterize the role of DNA sequence polymorphisms in cis on the expression of downstream genes in the case of flowering locus T in Arabidopsis and teosinte branched1 in maize (Adrian et al. 2010;Studer et al. 2011).

Relevance to research on transcriptional regulation
Since their discovery, the role of regulatory elements in gene function has been recognized as vital to our understanding of biological systems (McClintock 1950(McClintock , 1956a(McClintock ,b, 1961Peterson 1953;Jacob and Monod 1961). Gene regulation and gene product dosage are at the forefront of evolutionary theories about sources of novelty and diversification (Ohno 1972;King and Wilson 1975). Transcriptional regulation of a gene can be as important as the protein coding sequence (Wray et al. 2003). For instance, complete knock-down of expression of a gene by a regulatory polymorphism will have the same phenotypic consequence as the non-sense mediated decay of a transcript harboring an early stop codon (Willing et al. 1996). We do not have a set of rules, analogous to codon tables, for functional polymorphisms outside the coding sequence of a gene. Thus, detecting expression variation and tying it to phenotypic consequence, especially in the absence of CNS, remains a challenge to this day. Several eQTL studies in eukaryotes have found abundant cis and trans-acting genomic regions that affect gene expression (Brem et al. 2002;West et al. 2007;Li et al. 2013Li et al. , 2018. Expression polymorphisms are a potential source of phenotypic variation (Gibson and Weir 2005), and multiple studies detected expression polymorphisms co-segregating with phenotypic changes, including contributions to species domestication (Clark et al. 2006;Salvi et al. 2007;Schwartz et al. 2009;Lemmon et al. 2014). However, the majority of the cis-eQTLs only exhibit a moderate difference in gene expression. Detecting such variants and linking them to visible phenotypes may require detailed study using approaches focused on the specific biological process affected by the gene product. We do not yet have a standardized experimental tool for these purposes. As such, we cannot simultaneously identify and characterize the phenotypic impact of most cis-acting eQTLs. The vey1 polymorphism detected in the current study co-segregates with a cis-eQTL at oy1 in IBM (Li et al. 2013(Li et al. , 2018 and diverse maize inbred lines (Kremling et al. 2018). The high heritability of the alternative and direct CCM phenotype allowed us to scan large populations at a rapid rate to identify the genomic regions underlying the cis-acting regulatory elements and study allelic diversity in the natural population of maize. We propose that the approach we have taken is not likely to be unique to oy1. Therefore, we propose that all semi-dominant mutant alleles can be used as reporters to not only detect novel cis-acting gene regulatory elements but also functionally validate previously-detected cis-eQTL(s) from genome-wide eQTL studies.

ACKNOWLEDGMENTS
Authors acknowledge the help of the staff at Purdue ACRE farm for assistance with planting and crop management of all the field experiments described in this study. R.S.K was supported by USAID Feed the Future CGIAR-US Initiative grant "Heat stress resilient maize for South Asia through a public-private partnership" to G.S.J.. This work was supported by NSF grant 1444503 awarded to G.S.J. and B.P.D. We especially thank the Muehlbauer lab for making their data open to the public, and the USDA for the MaizeGDB website, which allowed us to reanalyze the eQTL data.