- American Society of Plant Biologists
Abstract
Maize starchy endosperm mutants have kernel phenotypes that include a brittle texture, susceptibility to insect pests, and inferior functional characteristics of products made from their flour. At least 18 such mutants have been identified, but only in the cases of opaque2 (o2) and floury2 (fl2), which affect different aspects of storage protein synthesis, is the molecular basis of the mutation known. To better understand the relationship between the phenotypes of these mutants and their biochemical bases, we characterized the protein and amino acid composition, as well as the mRNA transcript profiles, of nearly isogenic inbred lines of W64A o1, o2, o5, o9, o11, Mucuronate (Mc), Defective endosperm B30 (DeB30), and fl2. The largest reductions in zein protein synthesis occur in the W64A o2, DeB30, and fl2 mutants, which have ∼35 to 55% of the wild-type level of storage proteins. Zeins in W64A o5, o9, o11, and Mc are within 80 to 90% of the amount found in the wild type. Only in the cases of o5 and Mc were significant qualitative changes in zein synthesis observed. The pattern of gene expression in normal and mutant genotypes was assayed by profiling endosperm mRNA transcripts at 18 days after pollination with an Affymetrix GeneChip containing >1400 selected maize gene sequences. Compared with W64A sugary1, a mutant defective in starch synthesis, alterations in the gene expression patterns of the opaque mutants are very pleiotropic. Increased expression of genes associated with physiological stress, and the unfolded protein response, are common features of the opaque mutants. Based on global patterns of gene expression, these mutants were categorized in four phenotypic groups as follows: W64A+ and o1; o2; o5/o9/o11; and Mc and fl2.
INTRODUCTION
Endosperm texture is an important quality trait in maize, because it influences the shipping and handling characteristics of the grain, the susceptibility to insect pests, the yield of grits from dry milling, energy costs during wet milling, and the baking properties of the flour. The genetic, biochemical, and environmental factors that contribute to texture (i.e., hardness and vitreousness) are poorly understood (Chandrashekar and Mazhar, 1999). Kernel texture is related at least partially to the formation of a vitreous, or glassy, endosperm, and this is influenced by the protein content of the seed and the conditions of the kernel during maturation and desiccation (Kirleis and Stroshine, 1990; Mestres and Matencio, 1996). There appears to be a relationship between kernel hardness and storage protein synthesis in the endosperm, because factors that influence storage protein accumulation, such as nitrogen fertilization (Tsai et al., 1978) and deposition in protein bodies during endosperm development, are related to kernel phenotype (Coleman and Larkins, 1999).
The major storage proteins in maize endosperm are a group of prolamins known as zeins. Zeins are synthesized on rough endoplasmic reticulum (ER) membranes and accumulate in the ER as insoluble accretions called protein bodies (Larkins and Hurkman, 1978). There are three structurally distinct types of zein proteins, and they are distributed asymmetrically in protein bodies (Lending and Larkins, 1989; Woo et al., 2001). Mutations that alter zein synthesis can lead to protein bodies with abnormal morphology, size, or number, and they result in kernels with a soft, starchy texture. Mutations that reduce α-zein synthesis, such as opaque2 (o2), result in small, unexpanded protein bodies (Geetha et al., 1991), whereas those that reduce γ-zein synthesis, such as o15 (Dannenhoffer et al., 1995), lead to a smaller number of protein bodies. Conversely, the overproduction of γ-zein appears to enhance protein body number and result in the formation of more vitreous endosperm (Lopes et al., 1995; Moro et al., 1995). Other opaque mutants, such as floury2 (fl2), Mucuronate (Mc), and Defective endosperm B30 (DeB30), are associated with irregularly shaped protein bodies (Fontes et al., 1991; Coleman et al., 1997b). However, it is not clear whether the apparent link between a starchy endosperm phenotype and the synthesis and deposition of zeins is the result of a causal relationship or whether both are secondary, perhaps pleiotropic, effects of the mutations. For example, the o1 mutation appears to have little effect on zein synthesis (Nelson et al., 1965), yet it results in a soft, starchy endosperm. Furthermore, an opaque phenotype also has been observed as a result of nutritional, mechanical, and biotic stresses, and it is common among dek mutants (Lyznik and Tsai, 1989; Neuffer et al., 1997; O.A. Olsen, personal communication).
At least 18 mutations have been described that cause a soft, starchy endosperm phenotype (Thompson and Larkins, 1994). Only in the case of o2 and fl2 is the molecular basis for the mutation understood. o2 encodes a transcription factor that regulates genes encoding the 22-kD α-zeins as well as several other proteins (Damerval and De Vienne, 1993; Habben et al., 1993). fl2 corresponds to a point mutation in the signal peptide of a 22-kD α-zein protein (Coleman et al., 1997b). Failure to correctly process this signal peptide appears to cause the protein to become anchored to the ER membrane, leading to the creation of foci of hydrophobic α-zeins at the surfaces of protein bodies (Gillikin et al., 1997). This results in a dramatic increase in the synthesis of binding protein and other ER-resident chaperones, indicating an unfolded protein response in endosperm cells (Fontes et al., 1991; Coleman et al., 1997b; Shank et al., 2001). Biochemical responses in some opaque mutants appear similar to those in o2 and fl2. For instance, in endosperm of Mc and DeB30, ER-resident chaperones are induced dramatically (Zhang and Boston, 1992; Shank et al., 2001). However, a direct comparison of the various opaque mutants has been difficult because of their occurrence in different genetic backgrounds and the variation in penetrance of the mutant phenotype in these backgrounds (Balconi et al., 1998).
To better understand the nature of the mutations associated with a starchy endosperm phenotype, we developed nearly isogenic lines of a number of opaque mutants and compared their effects on protein synthesis and amino acid composition. We also compared the pattern of gene expression in these mutants by transcript profiling with Affymetrix GeneChip arrays. Our results reveal distinct, as well as shared, gene expression patterns in these mutants, and they provide a framework for investigating a common mechanism that underlies the opaque kernel phenotype.
RESULTS
To systematically compare the protein composition and patterns of gene expression between opaque mutants, we backcrossed the o1, o5, o9, o11, DeB30, and Mc mutations into the W64A inbred line. After six generations of backcrossing and two generations of self-pollination, each of the inbred lines appeared to be homogeneous, resembling the recurrent parent with respect to plant height, color, and tassel morphology and resembling the mutant with respect to an opaque endosperm phenotype. The independently developed nearly isogenic lines of W64A sugary1 (su1), fl2, and o2 (see Methods) were phenotypically consistent with these lines. The o5 and o9 plants tended to have yellow seedlings at the two- to four-leaf stage, although they greened as the plants developed. The yellow phenotype of o5 agrees with a previous description of this mutant (Neuffer et al., 1997). In addition, o5 and o9 developed more slowly than the wild-type and mutant inbred lines. o5 typically flowered 1 week later, whereas flowering in o9 usually was delayed by 10 to 14 days.
We measured the protein and amino acid contents of the endosperm from these mutants to determine if they manifested large qualitative or quantitative differences in zein and nonzein protein composition compared with the wild type (Table 1). This analysis demonstrated a significant degree of similarity among the inbred lines: the wild type and o1, o9, o11, and DeB30 mutants each contained ∼12% protein, whereas the o5, Mc, and fl2 mutants were very close to this value (i.e., 11.5 to 11.8% protein). The o2 mutant contained 10.1% protein; this amounts to a reduction in total protein of ∼15%. There was variability among the mutants with regard to the partitioning of proteins between the zein and nonzein fractions. The amount of nonzein protein in o1 and o5 was nearly identical to that in the wild type (2.1 mg/100 mg dry weight). The largest increases in nonzein proteins were in o2, o11, DeB30, and fl2, which contained ∼70% more protein than the wild type. Nonprotein nitrogen was most increased in o2, which is consistent with the high level of free amino acids in this mutant (Wang and Larkins, 2001). The amount of nonprotein nitrogen in o1 was similar to that in the wild type, whereas that in most of the other mutants was intermediate between the wild type and o2.
Protein and Amino Acid Composition of Mature Endosperm of W64A+ and Selected Opaque Mutants
With the exception of o1, each of the mutants had a higher Lys content than the wild type, but there were three distinct levels of increase: o2 had more than twice the amount in the wild type; o11, DeB30, and fl2 had 1.8 times the amount in the wild type; and o5, o9, and Mc had 1.4 times the amount in the wild type. A similar pattern was observed with respect to increases in Arg and Asn/Asp content. Among the amino acids reduced in the opaque mutants, except o1, were Gln/Glu, Leu, and Pro, the most abundant amino acids in zein proteins. The reduction of these amino acids generally was reciprocal to the increase in Lys.
SDS-PAGE was used to compare qualitative and, to some degree, quantitative differences in specific proteins among the opaque mutants. To standardize these comparisons, proteins from equal weights of endosperm flour of each genotype were extracted and partitioned into zein and nonzein fractions (Wallace et al., 1990). Figure 1A shows a Coomassie blue–stained gel showing the total zein fraction obtained from the wild-type and opaque mutant inbred lines. With the exception of o2, which showed the characteristic reduction in 22-kD α-zeins (Lee et al., 1976), there were no obvious differences in the major zein components detected among these mutants. However, in Coomassie blue–stained gels, the less abundant zein proteins are not always identified easily (Woo et al., 2001), and the very abundant 22- and 19-kD α-zeins can interfere with the detection of similarly sized zein proteins (i.e., the 18-kD δ-, 16-kD γ-, and 15-kD β-zeins). Therefore, the proteins also were identified by immunodetection.
SDS-PAGE Analysis of Zein Proteins in the Wild Type and Opaque Endosperm Mutants.
(A) Coomassie blue–stained 12.5% polyacrylamide gel showing total zeins extracted from 750 μg of endosperm flour. Lane 1, W64A+; lane 2, o1; lane 3, o2; lane 4, o5; lane 5, o9; lane 6, o11; lane 7, DeB30; lane 8, Mc; lane 9, fl2.
(B) to (K) Immunoblots developed with antisera against the 50-kD γ-zein (B), 27-kD γ-zein (C), 16-kD γ-zein (D), 15-kD β-zein (E), total α-zein (F), 22-kD α-zein (G), 19-kD Bα-zein (H), 19-kD Dα-zein (I), 18-kD δ-zein (J), and 10-kD δ-zein (K). Protein samples in (B) and (C) were extracted from 15 μg of flour, whereas the others were extracted from 75 μg of flour. The antisera concentrations are described in Methods.
Immunoblotting with 50-kD (Figure 1B) and 27-kD (Figure 1C) γ-zein–specific antisera showed no differences in the sizes and no reproducible differences in the amounts of these proteins in wild-type and opaque mutant endosperms. With the exception of Mc, reaction with the 16-kD γ-zein antiserum also showed no differences in protein patterns (Figure 1D). Two weakly stained polypeptides were detected in Mc. One migrated at 16 kD, and the second migrated with an apparent molecular mass of 13 kD. A reduction of the signal detected with the 15-kD β-zein–specific antiserum was seen in o2 (Figure 1E).
Antiserum recognizing total α-zeins from W64A (Lending et al., 1988) detected differences in the proteins in o2 and fl2 but not in the other opaque mutants. There was a marked reduction in 22-kD α-zeins in o2; fl2 also showed a reduction in 22-kD α-zeins, as well as a mutant α-zein, the 24-kD fl2 gene product (Figure 1F, arrow) (Coleman et al., 1997b). Similar differences were observed with a 22-kD α-zein–specific antibody (Figure 1G); however, the 24-kD fl2 α-zein was not detected readily on this immunoblot. In addition to reaction with a major band at 19 kD, the 19-kD Bα-zein antiserum reacted with a slightly larger polypeptide in the normal and mutant genotypes (Figure 1H). This larger protein is most likely the product of the 19-kD B-2 α-zein gene, which has a 33–amino acid insertion (Woo et al., 2001). The pattern of immunostaining among the different opaque mutants was fundamentally similar, except in fl2, which had an additional, even slower migrating polypeptide. This is most likely a modified α-zein, as noted previously for this mutant (Lopes et al., 1994). Immunoreaction with the 19-kD Dα-zein antiserum revealed no obvious differences in proteins among the normal and mutant backgrounds (Figure 1I).
To specifically detect the 18-kD δ-zein, blots were probed first with an antibody that was raised against a 25-residue peptide derived from the B73 allele of 18-kD δ-zein (Woo et al., 2001). However, no signal was detected in any of the zein extracts (data not shown). Because the 18-kD δ-zein was found to show considerable allelic variation, the blots were probed with a different antibody (see Methods) that recognizes a wide spectrum of 18-kD δ-zein alleles (R. Jung, unpublished data). With the notable exception of o5, we did not detect 18-kD δ-zein–specific bands in W64A+ and in the nearly isogenic mutant lines with this antibody (Figure 1J). In o5, the blot showed a number of strong, but poorly resolved, bands. Some of these are larger than 18 kD, and it was not clear whether this is the result of a distortion (masking) because of coincident electrophoretic mobility of the very abundant α-zein polypeptides. In extracts from each of the inbred lines, the 18-kD δ-zein antibody cross-reacted with a 10-kD peptide, probably the 10-kD δ-zein, and a reduction of the 10-kD signal was seen in o2 (Figure 1J). The monospecific 10-kD δ-zein antiserum (Figure 1K) reacted similarly with protein from the wild type and the opaque mutants. In o2, there was again a notable reduction in the 10-kD δ-zein signal, as was the case with the 18-kD δ-zein antibody.
Figure 2A shows a Coomassie blue–stained polyacrylamide gel from SDS-PAGE separation of the nonzein protein extracts of W64A+ and the different opaque mutants. The protein in these samples was based on equal weights of endosperm flour, and the greater staining intensity in lanes 3, 7, and 9 reflects the increased amount of nonzein proteins in o2, DeB30, and fl2, respectively (Table 1). Recently, several novel globulin proteins—18-kD α-globulin, legumin1, and legumin2—were detected in maize endosperm (Woo et al., 2001; R. Jung, unpublished data). During extraction, these globulin proteins partition into the nonzein fraction (Wallace et al., 1990). We did not detect any qualitative or quantitative differences in the level of legumin1 or legumin2 in the wild-type and mutant inbred lines (Figures 2B and 2C, respectively). The anti-globulin antiserum revealed slight differences in the amount of this protein among some of the opaque mutants, and it routinely had a slightly slower migration in o5, suggesting altered processing or an allele that encodes a protein with a slightly larger molecular mass (Figure 2D).
SDS-PAGE Analysis of Nonzein Seed Proteins in the Wild Type and Opaque Endosperm Mutants.
(A) Coomassie blue–stained 12.5% polyacrylamide gel showing total nonzein proteins extracted from 1.5 mg of flour. Lane 1, W64A+; lane 2, o1; lane 3, o2; lane 4, o5; lane 5, o9; lane 6, o11; lane 7, DeB30; lane 8, Mc; lane 9, fl2.
(B) to (D) Immunoblots developed with antisera against legumin1 (B), legumin2 (C), and α-globulin (D). Protein was from 750 μg of flour. The antisera concentrations are described in Methods.
GeneChip Microarray Analysis of Maize Endosperm mRNA Transcripts
Endosperm mRNAs in W64A+ and the nearly isogenic opaque mutants were analyzed using a GeneChip microarray produced by Affymetrix. The GeneChip contained arrays corresponding to >1400 maize genes selected from the Pioneer Hi-Bred EST database (see supplemental data online), and sequences corresponding to the majority of these genes (or their alleles/gene family members) also were identified in endosperm cDNA libraries. The transcript level in developing B73 endosperm for several genes in this array was described previously (Woo et al., 2001).
Duplicate GeneChip microarrays were hybridized with copy RNA (cRNA) targets from two replicated kernel samples (see Methods). In typical experiments, each sample was prepared from endosperm tissue pooled from three ears to minimize the effect of biological variation. Because of the expense of the Affymetrix GeneChip, hybridization experiments with cRNA from each replica sample were conducted only once. The experimental conditions, including demonstration of deoxyribonucleotide probe excess, were optimized independently and demonstrated to provide highly reproducible results (M.K. Beatty, unpublished data).
The results obtained after hybridization of the GeneChip with cRNAs revealed a strong pleiotropic effect of the opaque mutations on gene expression. This result made the transcript data refractory to the normalization procedure recommended by Affymetrix (Wodicka et al., 1997). Therefore, to allow a direct comparison of the transcript data set from each genotype, successive regression analyses (see Methods) were used to normalize these data and to globally scale them to that of W64A+. This procedure generated scaled values for each gene and permitted a direct comparison of all data points, regardless of the experiment from which they originated. This resulted in normalized GeneChip intensities that varied in their cumulative values of expression intensity for the different mutants (Table 2). For example, the normalized cumulative expression intensity of W64A o2 arrays (121,365) was approximately one-third lower than that of W64A+ (176,300), perhaps as a direct result of the downregulation of highly expressed zein genes (Lee et al., 1976). The cumulative expression intensity of W64A o11 (268,500) was approximately one-third larger than that of W64A+, indicating increased expression of many genes (see below).
Maize GeneChip Ranges of Normalized Expression Intensities
In each analysis, ∼20% of the genes had negative expression intensities (Table 2), indicating that the matched oligodeoxyribonucleotide probe set had a lower hybridization signal than the mismatched set of probes. These negative values were reproducible within genotypes and did not result from a failure in hybridization or scanning. This result could be explained by the presence of nearly identical transcripts (alleles) with higher expression levels than the probe gene that hybridized to the mismatched set of probes. Reasons for the strong hybridization signals in the mismatch set are difficult to evaluate, because Affymetrix does not disclose the oligodeoxyribonucleotide sequences of matched and mismatched probes. However, in a few cases, oligodeoxyribonucleotide sequences were provided, and comparison of these sequences with those in public and proprietary databases by BLAST analysis supported the hypothesis that allelism is a major cause of the negative expression values (data not shown). Because of these ambiguities, expression values of <20, including negative values, were considered “not detected” and, according to Affymetrix conventions, were systematically set to 20 arbitrary expression units (baseline). Consequently, the expression of between 500 and 700 genes on the GeneChip was regarded to be at baseline and therefore not interpretable in any of the endosperm transcript profiling experiments. This reflects the presence of many genes on the microarray for which expression is below the detection limit or that are not expressed in endosperm tissue. Interestingly, the number of expressed genes detected in the mutants was 10 to 15% greater than that in the wild type. Approximately 10% of the genes on the microarray were expressed at >0.1% of the cumulative expression intensity, and ∼1.5% of the genes were expressed at >1% of the cumulative expression intensity. In W64A+, the latter consisted of mostly seed storage protein transcripts, whereas in the mutants, they also included other types of mRNAs (see below). This result reflects the abundance of a given mRNA relative to its concentration in each of the total mRNA populations. The results with o2 stand out among these measurements, because the expression of 266 genes was >0.1% of the cumulative expression intensity, which is twice the number measured in the wild type (135 genes).
Table 3 shows a subset of the expression data (the entire data set is presented in the supplemental data online). It is easily recognized that endosperm storage protein genes had the highest levels of transcripts. The relative expression intensities in W64A+ for the 19-kD Bα-zein genes (Table 3, rows 153 to 155) and the 22-kD α-zein genes (Table 3, rows 149, 150, 156, and 157) were in the range of 5000 and 10,000 arbitrary expression units. The cumulative intensity of the array analysis for W64A+ was ∼175,000 arbitrary expression units; therefore, provided that expression units and transcript abundance are related linearly, each of the α-zein genes apparently accounted for ∼2.5 to 5% of the total transcript population. These data are consistent with our previous estimates of transcripts for the corresponding genes in W64A+ (Marks et al., 1985) and B73 endosperm (Woo et al., 2001). Similarly, in accordance with the literature (Kodrzycki et al., 1989; Lohmer et al., 1991), in W64A o2, the GeneChip expression intensities of the 22-kD α-zein genes and the gene encoding the ribosome-inactivating protein (Table 3, row 222) were diminished between 200- and 3-fold, respectively, compared with that of W64A+.
Affymetrix GeneChip for Maize (Subact)
The α-zeins, as well as other zein genes on the GeneChip, are members of closely related multigene families, but the results demonstrate that the expression of these genes can be evaluated independently. For example, the expression intensities of the 22-kD α-zein Z1 gene were 4823, 20, and 2650 in W64A+, W64A o2, and W64A o5, respectively (Table 3, row 149). Expression intensities of the 22-kD α-zein Z5 gene, which is 93% identical to the Z1 gene, were 5756, 20, and 20 in W64A+, W64A o2, and W64A o5, respectively (Table 3, row 150). The fact that the expression of these two nearly identical genes was detected differentially (i.e., different intensity levels are detected in o2 compared with o5) validates the utility of the GeneChip assay for endosperm mRNA transcript profiling.
Comparison of Endosperm mRNA Transcripts in W64A+, W64A su1, and the Nearly Isogenic Opaque Mutants
Eighteen to 22 days after pollination (DAP) endosperm was chosen for comparison of mRNA transcripts between genotypes, because by this stage the majority of cells have ceased division and take on a mature differentiation state, reflecting a high level of starch and storage protein synthesis. Furthermore, high-quality RNA can be prepared from endosperm at this period of development. Several control experiments were performed to assess biological variation in gene expression as a result of ear sampling and the potential differences in gene expression measurements that result simply from allelic variation. Figure 3A shows a scatterplot analysis comparing transcripts in two different ears of 22 DAP W64A+. The regression analysis of these data yielded a very high coefficient of determination (R2 = 0.996), indicating nearly genetic identity between the samples and a high degree of reproducibility in the Affymetrix analysis of the transcripts. These data were replotted in Figure 3B, which shows an expanded view of the genes with expression intensities between 20 and 1000. The small degree of scatter among these data further illustrates the similarity in the level of gene expression between the two samples. Because the latter presentation provided a significant degree of resolution for most of the expressed genes on the microarray, it was used routinely to globally compare transcript data sets.
Assessment of Endosperm Gene Expression Using Maize Affymetrix GeneChip Microarrays.
Each graph shows a scatterplot and a linear regression plot comparing expression intensities of endosperm-specific transcripts. R2 is the coefficient of determination.
(A) Comparison of endosperm-specific transcripts obtained from two independent ears of W64A+.
(B) Expanded view of the plot shown in (A), with expression intensities between 20 and 1000 arbitrary units to emphasize expression intensity differences between the two ears.
(C) Comparison of endosperm-specific transcripts obtained from the inbred lines W64A+ and Oh43+.
(D) Expanded view of the plot shown in (C), with a maximum of 1000 arbitrary units.
(E) Comparison of endosperm-specific transcripts obtained from W64A+ and from the starch biosynthetic mutant W64A su1. The plot is an expanded view with a maximum of 1000 arbitrary units.
(F) Comparison of endosperm-specific transcripts obtained from W64A+ and the W64A o2 mutant. The plot is an expanded view with a maximum of 1000 arbitrary units.
Figure 3C shows a similar type of comparison with RNA transcripts from 18-DAP endosperm of W64A+ and Oh43+. In this case, the coefficient of determination is 0.862, indicating a much greater degree of variability in signal intensity than was observed in the W64A+ comparison described above. The variability in these samples is much more apparent in the presentation of the data in Figure 3D, which shows the expanded view of genes with expression intensities between 20 and 1000. Regression analysis of the data from the Oh43 replicates yielded a higher coefficient of determination (R2 = 0.99), demonstrating that the differences between W64A and Oh43 cannot be accounted for by a greater variance between the Oh43 replicates. This variability in transcript profiling most likely is a consequence of allelic variation between inbred lines and indicates the importance of using nearly isogenic lines to compare variability in gene expression among the mutants.
Figure 3E presents a comparison of mRNA transcripts in 18-DAP W64A+ and W64A su1. su1 encodes a starch-debranching enzyme (isoamylase), and the su1-1 mutant leads to increased Suc concentration, decreased amylopectin content, and a collapsed endosperm phenotype at maturity as a result of the synthesis of water-soluble phytoglycogen (James et al., 1995). Comparison of mRNA transcripts in 18-DAP endosperm of the wild-type and su1 endosperms revealed very few differences in gene expression, in spite of the severity of the kernel phenotype at maturity. Although su1 is considered a pleiotropic mutant (Giroux et al., 1994), the linear regression analysis of these data produced a coefficient of determination (R2 = 0.998) that is as high as that produced by the comparison between the two wild-type ears (cf. Figures 3A and 3E). Only 21 genes were found to have an expression level that is at least twofold greater than that of the wild type, and the expression of no gene was <50% that of the wild type (Table 3). Most of the genes that showed an increased level of expression, such as molecular chaperones and wound-induced and Hyp-rich proteins, are related to stress responses. The transcript level of su1, like that of other genes encoding starch biosynthetic enzymes, was not altered in this mutant, in agreement with the finding of James et al. (1995) for the reference allele (su1-ref) used in this study.
The o2 gene has a much greater impact than su1 on gene expression in 18-DAP endosperm. As shown in Figure 3F, the expression of a large number of genes was different from that of the wild type, as reflected by the coefficient of determination (R2 = 0.71) in the scatterplot. Sixty genes were upregulated more than threefold relative to the wild type, and these accounted for ∼15% of the total gene expression intensity measured by the microarray (see supplemental data online). These genes appear to function in a number of pathways, although it was difficult to extrapolate a unifying explanation. However, 19 upregulated genes belong to the complex of genes related to stress responses, molecular chaperones, and protein turnover. Transcripts for genes encoding ubiquinol–cytochrome c reductase, glutathione S-transferase, β-glucosidase, and a zeamatin-like protein each accounted for >1% of the cumulative gene expression intensity on the chip. The expression of 66 genes was reduced more than threefold compared with that of the wild type. Among these are genes encoding zein storage proteins (all 22-kD α-zeins, one 19-kD Dα-zein, 15-kD β-zein, and 10-kD δ-zein), ribosome-inactivating protein (RIP), aldolase, enolase, orthophosphate dikinase, actin, and α-tubulin. In contrast to zein proteins, transcripts of two nonzein seed proteins that typically are expressed preferentially in the embryo (germin-like protein and globulin1) were increased. The transcript of the o2 gene was reduced fivefold. Notably, many of the downregulated genes are involved in carbon metabolism and carbohydrate metabolism. Some genes putatively involved in amino acid metabolism were upregulated (e.g., Ser o-acetyltransferase), but genes involved in the synthesis of branched amino acids (e.g., ketol-acid reductoisomerase and acetohydroxyacid synthase) were downregulated. Of the 66 downregulated genes, transcripts of 36 were reduced below the detection limit of 20 expression intensity units. The downregulated genes in o2 accounted for ∼30% of the expressed gene intensity units in W64A+.
Figure 4 shows a data set illustrating the effect of the six other opaque mutants on gene expression in 18-DAP endosperm (DeB30 was not included in this analysis because it was unavailable during this growing season). A cursory examination of the scatterplots demonstrates that these mutants, like o2, are much more pleiotropic than su1. Of the six opaque mutants, o5, o9, o11, and fl2 showed the greatest variation in gene expression relative to the wild type, with regression analysis values of approximately R2 = 0.85; o1 and Mc were more similar to the wild type, with regression analysis values of R2 = 0.93. In o1, 35 genes were upregulated more than threefold (see supplemental data online), and most of these encode putative chaperones (HSP90, HSP82, HSP70 [two different accessions], calreticulin, and petide disulfide isomerase) and stress-related proteins (peroxidase, proteinase inhibitor, and Gly- and Hyp-rich proteins [three different accessions]). Transcripts of the latter three genes were increased dramatically in o1 relative to the wild type. The expression of one gene encoding eEF1A was increased fourfold, and expression of a gene encoding Asp aminotransferase was increased eightfold. Twenty-two genes, including Ala aminotransferase, two ribosomal proteins, zeathionin genes (two different accessions), orthophosphate dikinase, and RIP, were downregulated at least twofold in o1. There were no large changes in seed storage protein gene expression in o1.
Comparisons of Endosperm Gene Expression between W64A+ and the Nearly Isogenic Opaque Mutants Using Maize Affymetrix GeneChip Microarrays.
Each graph is a scatterplot and a linear regression plot comparing expression intensities of endosperm-specific transcripts. The plots show a maximum of 1000 arbitrary expression units to emphasize expression intensity differences. R2 is the coefficient of determination.
(A) Comparison between W64A+ and W64A o1.
(B) Comparison between W64A+ and W64A o5.
(C) Comparison between W64A+ and W64A o9.
(D) Comparison between W64A+ and W64A o11.
(E) Comparison between W64A+ and W64A Mc.
(F) Comparison between W64A+ and W64A fl2.
In o5, the expression of 67 genes was increased more than threefold relative to the wild type. o5 showed upregulated expression of many molecular chaperones and the stress-related genes upregulated in o1, but in addition to these, expression of HSP22, TCP1, cis-trans prolyl-isomerase (cyclophilin), BET1, and phenylalanine ammonia lyase (PAL) also were increased (see supplemental data online). Other genes upregulated in o5 include a 14-3-3 homolog, Asp aminotransferase, cytoskeleton-related proteins (α-tubulin, actin, and eEF1A), and legumin1. Among the 26 genes downregulated more than twofold in o5 are Trp synthase, Ala aminotransferase, metallothionein, ubiquitin-conjugating enzyme, orthophosphate dikinase, RIP, and zeathionin (two different accessions). Several genes encoding zein storage proteins are among those downregulated in o5: the 50-kD γ-zein, 19-kD D1 and D2 α-zeins, and four 22-kD α-zeins (fl2 transcript disappeared, and three others were reduced by half).
In o9, the expression of 58 genes was increased more than threefold relative to the wild type (see supplemental data online). Most of these (51) are identical to those upregulated in o5 (Table 3). Nineteen genes were downregulated more than twofold compared with the wild type, and 17 of these were identical to those affected similarly in o5. These included several zein protein genes, the 50-kD γ-zein, and the 19-kD D1 and D2 α-zeins but only one of the 22-kD α-zeins (fl2).
Alterations in the expression pattern of genes in o11 were similar to those in o5 (62) and o9 (51), the most upregulated and downregulated genes (77 and 5, respectively) (Table 3; see also supplemental data online). The only downregulated storage protein gene in o11 encodes the 19-kD D1 α-zein. The legumin1 gene and the 18-kD δ-zein were upregulated.
In fl2, the expression of 37 genes was increased threefold relative to the wild type, whereas the expression of 22 genes was reduced at least twofold (see supplemental data online). Among the overexpressed genes are several that encode stress-related transcripts, including PAL, the chaperones affected in o5, o9, and o11, as well as HSP60, a second binding protein accession, and calreticulin. Notably, as a group, the expression of ER-related chaperones was remarkably higher than that in the other opaque mutants (Table 3). Genes encoding cytoskeletal proteins (actin, eEF1A, and α-tubulin) and genes related to protein turnover (polyubiquitin, MUBG9 ubiquitin, and subtilisin-like protease) also were upregulated. Downregulated genes included starch-branching enzyme1, ribosomal proteins, and glutathione S-transferase. Water channel protein genes also were downregulated, in contrast to o2, in which they were upregulated. The expression of three zein genes was reduced by ∼50%: 19-kD Bα-zein, a 22-kD α-zein that is not the fl2 gene, and the 16-kD γ-zein gene.
Relative to the wild type, there were 57 upregulated genes but only 7 downregulated genes in the Mc mutant (see supplemental data online). Genes showing greater than normal levels of expression included stress-related sequences and the chaperones affected in the other opaque mutants, and in particular the putatively ER-localized chaperones affected in fl2. Two 14-3-3 gene homologs, as well as a duplicated domain structure protein gene, were expressed at increased levels. Also increased in expression were genes associated with protein turnover, including a 26S protease subunit, subtilisin protease, and polyubiquitin, and genes encoding cytoskeletal proteins, such as eEF1A. Conspicuously, there were no zein genes among the five sequences downregulated in Mc.
A cluster analysis (Eisen et al., 1998) of the gene expression phenotypes of these mutants revealed a close relationship between o5, o9, and o11; the fl2 and Mc mutants also were related. The o1 and o2 mutants showed no meaningful grouping with the other opaque mutants (data not shown). A pairwise comparison of selected mutants is shown in Figure 5 . The comparison of gene expression in o5 and o2 (R2 = 0.66) showed that these mutants are as dissimilar to one another as they are to the wild type (Figure 5A). By contrast, comparison of o5 and o9 (R2 = 0.93; Figure 5B) and o5 and o11 (R2 = 0.98; data not shown) showed that these mutants share fundamentally similar patterns of gene expression. Figure 5C shows a comparison of the gene expression patterns in o2 and fl2. In this case, the coefficient of determination (R2 = 0.56) for the scatterplot shows that these mutants induce quite different patterns of gene expression in 18-DAP endosperm. The opaque mutant with the closest relation to fl2 is Mc (Figure 5D). The coefficient of determination between fl2 and the wild type is R2 = 0.84, and that between fl2 and Mc is R2 = 0.89. This finding indicates that fl2 and Mc share a subset of genes with expression patterns that are not similar to those of the wild type. This conclusion is corroborated by comparison of the gene expression data in Table 3.
Gene Expression Differences and Similarities between Selected Opaque Mutants.
Each graph is a scatterplot and a linear regression plot comparing expression intensities of endosperm-specific transcripts. R2 is the coefficient of determination.
(A) Comparison between W64A o5 and W64A o2.
(B) Comparison between W64A o5 and W64A o9.
(C) Comparison between W64A fl2 and W64A o2.
(D) Comparison between W64A fl2 and W64A Mu.
DISCUSSION
The objective of this research is to understand the cause of the starchy kernel phenotype in maize opaque endosperm mutants. In the case of o2 and fl2, the basis of the genetic defect is known (Schmidt et al., 1990; Coleman et al., 1997b), but the mechanisms by which these two very different types of mutations result in a similar mature endosperm phenotype is unclear. To compare the patterns of gene expression and protein synthesis in o2 and fl2 with those of other opaque mutants, it was essential that the mutations be introgressed uniformly into a common genetic background. The nearly isogenic W64A+ and opaque mutant inbred lines were created by recurrent backcrossing, and they are phenotypically uniform and appear to be genetically homogeneous. Besides displaying similar height, leaf shape, and color, as well as a number of other whole-plant phenotypic characteristics, their seeds are similar in size and protein content. These results are expected, because after six backcross generations the mutant inbred lines should share, on average, ∼99% of the recurrent parent genome. Most of these mutants show a strict endosperm phenotype, but in the cases of o5 and o9, it is possible that the mutations are not seed specific. These mutant plants show retarded development; consequently, the opaque endosperm phenotype could result from deleterious effects of the o5 and o9 mutations on seed development.
Although an opaque phenotype often is tacitly associated with an increased endosperm Lys concentration, a comparison of these mutations in the W64A genetic background revealed marked differences in the contents of essential amino acids, and Lys in particular. The o1 mutant has an amino acid composition nearly identical to that of the wild type, a result consistent with that of Nelson et al. (1965). Each of the other opaque mutants has a higher endosperm Lys concentration than the wild type, although the increases are not as great as in o2. The o2 mutant has the highest Lys content, followed by o11, DeB30, and fl2. Except for o1, the amount of the nonzein protein fraction in these mutants is increased, as in o2, and the amount of the zein protein fraction is decreased. Balconi et al. (1998) reported that o1 increased the endosperm Lys content of the A69Y inbred line as much as o2. They also found the Lys contents of A69Y o9 and o11 to be significantly higher that those we observed in W64A. The differences between these two sets of data could be explained by the response of the inbred backgrounds to these mutations and, potentially, allelic variation among the mutants. However, like the samples used by Balconi et al. (1998), our o1, o9, and o11 mutants were from the Maize Genetic Cooperation (Urbana, IL), and they are likely to result from the same alleles (Neuffer et al., 1997).
Significant differences in total protein and zein proteins were found in most of the opaque mutants. The largest reductions in zein synthesis occurred in the W64A o2, DeB30, and fl2 mutants, which had ∼35 to 55% of the wild-type level of storage proteins. Zeins in W64A o5, o9, o11, and Mc were within 80 to 90% of the amount found in the wild type. Besides o2 and fl2, significant qualitative changes in zein synthesis also were observed in o5 and Mc (see below). We previously identified several wild-type inbred lines that contain less total protein (7 to 9%) and similar amounts of zein proteins (5.5 mg/100 mg of endosperm) as these opaque mutants. Furthermore, the o1 mutant contains nearly identical amounts of zein and nonzein proteins as the wild type. Consequently, neither the amount of zein protein synthesis nor the reduction in total protein content adequately explains the opaque endosperm phenotype of these mutants.
In only a few of these mutants did we detect considerable qualitative and quantitative differences in individual seed proteins. In o2 and fl2, we observed the well-documented changes in the amount and/or size of the 22-kD α-zeins. We also found a reduction in the amount of 10-kD δ-zein and of 15-kD β-zein in o2, which was not reported previously. Likewise, the reduced level of 16-kD γ-zein in Mc was not described previously, nor was the detection of polymorphic forms of this protein. Based on our recent experiments, it appears that the Mc mutation likely results from a defective allele encoding a 16-kD γ-zein protein (C.-S. Kim, unpublished data). With the exception of Mc, we were unable to identify qualitative changes in seed storage proteins that could explain the starchy endosperm phenotype of the other opaque mutants.
An interesting change in one seed protein is the presence of the 18-kD δ-zein in o5 endosperm (Figure 1). W64A+ and the nearly isogenic opaque mutants did not accumulate detectable levels of the 18-kD δ-zein polypeptide, and these inbred lines appear to express this gene at a low level. However, the 18-kD δ-zein gene in o5 is an unlikely candidate for a mutated allele of the O5 gene: the o5 mutation is associated with a whole-plant phenotype (see above), and the 18-kD δ-zein is expressed exclusively in the endosperm. There are two possible explanations for the higher expression of the 18-kD δ-zein in o5. First, the normally suppressed 18-kD δ-zein gene (Swarup et al., 1995) could be derepressed in o5, perhaps as a result of the mutation. Second, a strong δ-zein allele could have been carried along during the backcrossing process. A genetic linkage between o5 and the 18-kD δ-zein gene would not be expected, because the 18-kD δ-zein maps to chromosome 6L and o5 maps to chromosome 7L (Coleman et al., 1997a; Woo et al., 2001). However, another polymorphism detected in o5 is an electrophoretically slower migrating α-globulin peptide (Figure 1). The α-globulin and the 18-kD δ-zein loci are linked closely on the long arm of maize chromosome 6 (Woo et al., 2001). This observation implies that these alleles were transmitted from the o5 donor parent, despite the six backcross generations.
The development of an Affymetrix GeneChip containing maize genes made it possible for us to analyze the patterns of gene expression in these mutants in a much more global manner than was possible by comparing qualitative and quantitative changes in protein and amino acid composition. The use of high-density oligodeoxyribonucleotide microarrays has emerged as an important tool for the simultaneous examination of large numbers of expressed genes (Knight, 2001). However, a major limitation of this technology is its “closed” nature (i.e., only preselected genes are probed, and it is not possible to discover new genes). The maize genome contains an estimated 80,000 expressed genes, and approximately one-fifth of these are transcribed in the endosperm (Woo et al., 2001). Despite a growing collection of maize ESTs in GenBank, only a fraction of the corresponding genes have been sequenced completely. Consequently, it is currently not feasible, technically or economically, to produce microarrays that include the majority of maize genes.
The profile of endosperm transcripts was obtained with an Affymetrix GeneChip microarray based on the sequence information of >1400 maize genes; generally, these are alleles of the B73 inbred line (see Methods). More than 80% of these genes have annotations (i.e., have been characterized functionally in maize or other species and closely match functionally known genes). These genes were selected to cover a wide range of metabolic pathways and cellular and physiological processes (see supplemental data online). The GeneChip also contains probes for members of multigene families, notably, members of the different seed protein gene families (Table 3). The hybridization results show a high degree of resolution between these probes. This specificity illustrates an important advantage of oligodeoxyribonucleotide-derived microarrays compared with arrays that use larger DNA fragments immobilized on glass slides (Knight, 2001). The latter usually do not distinguish between closely related members of multigene families. A disadvantage of the selectivity of GeneChips, at least for the types of experiments conducted here, is the allele sensitivity of the probes. Our experiments were performed with W64A cRNA samples, but the majority of the probes on the microarray were derived from B73 alleles. When scanned after hybridization, ∼20% of these probes produced negative values (Table 2). This occurred because of the stronger hybridization signals obtained with the control probes for each gene (i.e., from the sets of mismatched oligodeoxyribonucleotides). Consequently, the W64A samples contained a number of cRNA species (alleles or homologs) that matched more closely the control than the target probes. Because of this ambiguity, we did not consider such results informative for this analysis. This suggests a general difficulty with array studies that rely on exact sequences and knowledge of homologs.
ESTs corresponding to the majority of genes (or their alleles) on the maize GeneChip are represented in maize endosperm cDNA libraries, and the level of transcripts corresponding to several of these genes was studied previously in detail (Woo et al., 2001). Therefore, the use of the maize GeneChip to examine endosperm gene expression appeared to be feasible, and the presence of well-characterized genes provided important experimental controls.
The analysis of the GeneChip hybridization results demonstrated that these experiments provide a reasonably thorough assessment of endosperm mRNA transcript abundance. The cRNA was synthesized with biotin-labeled CTP and UTP nucleotides; consequently, the labeling differences between cRNA species of varying GC content were minimal. The majority of the oligonucleotide targets (75%) accounted for <5% of the total cRNA expression intensity, and 1% (mostly zeins) accounted for 60% of the endosperm transcript abundance. These data closely reflect the mRNA concentrations expected for developing endosperm tissue (Woo et al., 2001). Furthermore, the expression intensities (given in normalized expression units) of each gene are a reflection of their relative RNA concentrations, although the caveats stated above regarding allelic variation and multigene families must be considered.
The data obtained from the GeneChip hybridizations showed a high degree of reproducibility, as illustrated by the comparison of transcripts from kernels of two 22-DAP ears of W64A+. Even with ears harvested from different locations and during different seasons, the results were highly reproducible, provided the endosperm samples were of a similar developmental stage and of the same genetic background. By comparison, samples from different genetic backgrounds were different to a surprising degree, in some instances surpassing the variation in transcript levels in samples from the wild type and the nearly isogenic mutant lines. This finding illustrates the importance of using nearly isogenic lines to compare patterns of gene expression between mutants.
Surprisingly, the different opaque mutations had a greater impact on gene expression in 18-DAP W64A+ endosperm than su1, even though this starch mutant has a striking effect on carbohydrate accumulation by 18 DAP (Doehlert and Kuo, 1994; Singletary et al., 1997). It is possible that pronounced changes in gene expression in su1 occur at later stages of kernel development. However, an independent GeneChip profiling study of five different starch biosynthetic mutants, including su1, in the Oh43 background at 26 DAP also revealed a smaller degree of gene expression changes than was observed in our study of the opaque mutants (G.W. Singletary, unpublished data). Thus, it appears that, as a group, the opaque mutations have a much more pleiotropic effect on gene expression than do mutations that affect starch synthesis.
Although the opaque mutations used in this study influence the expression of a large number of genes in 18-DAP endosperm, the degree of the pleiotropic effect varied among the mutants. o1 has the smallest effect on global patterns of gene expression (Figure 4), consistent with the relatively small differences in protein and amino acid composition in this mutant compared with the wild type (Table 1). By contrast, the large changes of protein and amino acid synthesis in o2 (Table 1) are associated with large changes in the patterns of gene expression (Figure 3). These observations are consistent with the role of O2 as a transcriptional activator (Schmidt et al., 1990). The O2 protein is known to regulate the expression of genes that encode the 22-kD α-zein gene family (Schmidt et al., 1992), RIP (Lohmer et al., 1991), and orthophosphate dikinase (Maddaloni et al., 1996), and our microarray analysis corroborates these observations. It also suggests that O2 regulates a number of other genes, including 10-kD δ-zein, 15-kD β-zein, apos-pory-associated protein homolog (two different accessions), ARSK1 protein kinase homolog, and many others (see supplemental data online). We identified 30 genes in W64A+ that have expression levels below baseline in o2, and it is possible that these genes are regulated by the O2 transcription factor. In o2, there is a large cumulative reduction in gene expression (49,000 expression intensity units) as a consequence of the downregulation of zein genes. However, the expression of 131 genes that were below the level of detection in W64A+ was significantly increased in o2. Among these are a large number of genes associated functionally with stress, cellular perturbation, and pathogen responses (Table 3). This major shift in expression from zein to nonzein genes (Table 1) is consistent with changes in patterns of protein synthesis in the endosperm that were described previously in this mutant (Damerval and De Vienne, 1993; Habben et al., 1993; Damerval and Le Guilloux, 1998).
We reported previously that the synthesis of eEF1A is increased at least twofold in W64A o2, and the increased accumulation of this protein is highly correlated with the Lys content of the endosperm (Habben et al., 1995; Sun et al., 1997). The GeneChip hybridization data showed that transcripts encoding eEF1A are below detectable levels in this mutant. To exclude the possibility of error that might occur as a result of hybridization with the mismatched set of probes (see above), we visually inspected the array and observed no significant hybridization to the mismatched eEF1A sequences. Therefore, the observed downregulation of this gene in W64A o2 most likely reflects selection of an allele for the GeneChip that is not highly expressed in this mutant. eEF1A is encoded by ∼15 genes in maize, and these are expressed differentially in various tissues (Carneiro et al., 1999). Understanding the basis for the differences in eEF1A gene expression reported here and in other experiments (Habben et al., 1995; Carneiro et al., 1999) will require an Affymetrix array containing all of the maize genes that encode eEF1A.
The fl2 mutation is the consequence of a defective signal peptide on a 22-kD α-zein protein (Coleman et al., 1997b). The failure to cleave this sequence leads to the unfolded protein response, which is manifested by a dramatic increase in ER-resident molecular chaperones (Boston et al., 1991). The effect of fl2 on chaperone gene expression, as well as the expression of a number of stress response genes, is documented clearly by the results from the GeneChip hybridization analysis. Among the opaque mutants examined here, this pattern of gene expression is matched most closely by Mc, which also shows evidence of the unfolded protein response phenotype by having significant increases in the expression of ER-resident chaperones and a similar set of stress response genes. The changes we observed in the amount and size of the 16-kD γ-zein proteins in Mc, in conjunction with the observation of abnormally formed protein bodies in this mutant (Zhang and Boston, 1992), provide support for the hypothesis that Mc, like fl2, is related to a perturbation of zein protein deposition in protein bodies.
One group of genes consistently affected in all of the opaque mutants has been implicated routinely in stress responses, and it includes genes that encode molecular chaperones, cell wall proteins, and wound- and pathogen-activated proteins (Hammond-Kosack and Jones, 1996). The upregulation of these genes is a strong indicator of the deleterious nature of the opaque mutations and their perturbation of endosperm cell function. The penetrance of this phenotype in o2 mutants can vary, depending on the genetic background. This is well documented in the case of the so-called “modified” o2 maize genotypes, which develop a vitreous kernel phenotype (Vasal, 1994). The vitreous phenotype of modified o2 mutants is controlled by several independent loci (o2 modifiers), and their mechanism of action is not understood (Lopes et al., 1995). However, our preliminary data from experiments comparing o2 and modified o2 endosperm show a general reduction in the stress response of gene expression (data not shown). It is possible that an intervention in the stress response pathway constitutes the basis of o2 modifier gene activity. It remains to be determined whether or not the phenotype of the other opaque mutants can be ameliorated by these genes.
The opaque appearance of maize kernels results from their failure to transmit light. It is assumed generally that opacity is a consequence of storage protein distribution, structure, or accumulation. However, opacity could result from any change of mature endosperm organization that increases light scattering. Conceivably, opacity could be a consequence of phenomena as distinct as a change in a developmental “switch” that influences physiological maturity and disruption in the ability of starch grains to form regular arrays as a consequence of alterations in their size, shape, or chemistry. Our data call into question the assumption involving the role of storage proteins and raise a fundamental question regarding the origin of the opaque phenotype.
Although seed protein profiles appeared unchanged in some opaque mutants (Figure 1), alteration in the abundance of at least one zein transcript in each mutant (Table 3) prevents us from excluding the hypothesis that opacity is a consequence at least in part of an alteration in protein body constitution. However, our data are consistent with at least two likely and testable alternative models. The su1 mutant, which also has a distinct phenotype at maturity but maintains a vitreous endosperm, shows neither significant changes in gene expression nor alterations in zein transcript abundance (Table 3). However, all of the opaque mutants show a dramatic departure from the wild-type gene expression phenotype, making it reasonable to propose that this shared trait is responsible for opacity. Indeed, as in the zein RNAs, at least some stress response transcripts are altered in all of the opaque mutants. Thus, the hypotheses that an increased stress response or altered zein levels cause the opaque phenotype are equally parsimonious explanations. By contrast, despite the fact that both the o2 and fl2 mutations have direct effects on zein accumulation or structure, they have dissimilar effects on gene expression. Meanwhile, o5, o9, and o11 have very similar effects on gene expression, but as a group they are equally dissimilar to o2, the wild type, and fl2. Together, these data suggest that multiple mechanisms lead to opacity, with differing effects on the pattern of gene expression in the endosperm. If this is so, the shared aspects of the gene expression pattern could be explained as the consequences of producing a “misorganized” endosperm structure.
Along with the gene expression analysis presented in this work, classic genetic experiments could be used to test these hypotheses. If the mechanisms by which the o5, o9, and o11 mutations disrupt light transmission are related, we predict that these mutant alleles would exhibit epistatic interactions. Furthermore, suppressors of other opaque mutants, such as the o2 modifiers (Lopes et al., 1995), should not affect o5, o9, and o11. Double mutants of o5, o9, and o11 with opaque mutants outside of this phenotypically correlated cluster should be additive. However, if the basis of the mutant phenotype is zein misexpression or some other shared alteration in gene expression, such as an increased stress response, o2 modifiers also should modify the o5, o9, o11, and fl2 phenotypes and epistatic interactions would occur both within and between the phenotypically clustered mutant alleles.
The mRNA transcript profiling experiments of opaque mutants with the Affymetrix GeneChip suggest the strengths as well as the weaknesses of using gene arrays to compare patterns of gene expression. Although these experiments may not identify the genetic basis of a mutant, they provide a list of genes (some with deduced functions) that are clearly affected (or not) by the mutation, and they allow us to recognize and, to a certain degree, quantify complex (pleiotropic) changes in gene expression. This is important and potentially valuable information. Unfortunately, in many cases, our knowledge of the role played by these genes is limited. Besides our ignorance of gene functions, understanding the implications of changes in the patterns of gene expression is confounded by the fact that transcript concentration is only one aspect of gene expression. Furthermore, transcript levels of a mutated allele are not necessarily affected in the mutant. This is demonstrated in our study, for instance, by the levels of the fl2 and su1 transcripts in their respective mutant endosperms. Nevertheless, the concentration of the o2 transcript in W64A o2 endosperm is diminished greatly. Another limitation of gene microarray data is the closed nature of the experiments, which are limited to the predetermined set of genes. “Open” technologies for mRNA transcript profiling, such as Serial Analysis of Gene Expression-related techniques (Shimkets et al., 1999), address these limitations by providing the opportunity to analyze all of the expressed genes, regardless of whether they have been isolated. We are using these approaches at present to further dissect the pleiotropic effects of the maize opaque endosperm mutations.
METHODS
Maize Mutants
o2 and fl2 inbred lines of W64A maize (Zea mays) were obtained from C.Y. Tsai (Purdue University, West Lafayette, IN) and were maintained at the University of Arizona (Tucson) since 1988. The su1 mutant in the W64A background, and the o1, o5, o9, and o11 mutants in different germplasm backgrounds, were obtained from the Maize Genetics Stock Center at the University of Illinois (Champaign-Urbana). The DeB30 and Mc mutants were provided by Francesco Salamini (Instituo Sperimentale per la Cerealicoltura, Bergamo, Italy). The o1, o5, o9, o11, Mc, and DeB30 mutants were converted to the W64A background by backcrossing six times, followed by two self-pollinations. Plants used for the protein analysis were grown at the University of Arizona Campus Agricultural Center in 1998.
SDS-PAGE and Immunoblot Analysis
Endosperm flour was prepared by the methods of Wang and Larkins (2001). After pooling flour from 15 kernels, the protein in 50-mg aliquots was separated into zein and nonzein proteins and analyzed by SDS-PAGE, as described by Habben et al. (1993). Polyacrylamide gels were either stained with Coomassie Brilliant Blue R250 or blotted onto nitrocellulose using either a Trans-Blot Semi-Dry Transfer Cell or the Mini-Trans-Blot Cell (Bio-Rad, Hercules, CA). Zeins and other seed proteins were immunodetected (Harlow and Lane, 1988) with monospecific rabbit antisera that react with total α-zeins (1:5000 dilution), the 19-kD B α-zein and D α-zein proteins (1:10,000), the 22-kD α-zeins (1:10,000), the 50-kD (1:10,000), 27-kD (1:10,000), and 16-kD γ-zein proteins (1:5000), the 15-kD β-zein (1:5000), the 18- and 10-kD δ-zeins (1:5000), 18-kD α-globulin (1:3000), legumin1 (1:3000), and legumin2 (1:5000). Antiserum reacting with maize legumin2 was developed with a bacterially expressed His tag fusion (Harlow and Lane, 1988) of the entire legumin2 coding sequence. The antibody against the 18-kD δ-zein was developed with a bacterially expressed His tag fusion of a partial 18-kD δ-zein polypeptide (nucleotides 71 to 125). This 18-kD δ-zein antibody also cross-reacts weakly with 10-kD δ-zein polypeptides. All other antisera were as described previously (Woo et al., 2001). The total α-zein antiserum was described by Lending et al. (1988).
Protein and Amino Acid Analyses
Protein analyses were performed with endosperm flour from mature kernels as described previously (Moro et al., 1995). Amino acid analysis (after performic acid oxidation) was performed at the protein analytical facility of the University of Iowa (Iowa City). Measurements were made with pooled samples of 15 kernels of each genotype, and the data presented are the means of two independent assays.
RNA Preparation and Hybridization
The maize inbred lines Oh43+, W64A+, and the W64A nearly isogenic mutant lines of su1, o1, o2, o5, o9, o11, Mc, and fl2 were grown in the summer of 1998 in adjacent field plots at the Pioneer Hi-Bred International genetic nursery in Johnston, IA. Plants were hand pollinated, and a minimum of six well-filled ears of each line were harvested 18 days after pollination (22 days after pollination in one case) and frozen immediately in liquid nitrogen. Kernels were taken from the center of each ear, and the endosperm was dissected from the embryo and pericarp. To minimize the effect of biological variation between ears on gene expression, equal numbers of dissected endosperms from three ears were pooled and treated as one sample; thus, a minimum of two replicated samples was used for each experiment. To compare ear-to-ear variation, an analysis was done with two W64A+ ears harvested at 22 days after pollination.
Total RNA was prepared from frozen, ground endosperm tissue using the Trizol reagent (Gibco BRL, Gaithersburg, MD) according to the manufacturer's suggested protocol. Poly(A)+ RNA was isolated using the PolyA Tract mRNA Isolation kit IV (Promega, Madison, WI), and double-stranded cDNA was synthesized using the SuperScript Choice System (Gibco BRL). Copy RNA probes were synthesized subsequently with biotin-conjugated CTP and UTP ribonucleotides (MEGAscript T7 kit; Ambion, Austin, TX) and purified with RNeasy (Qiagen, Valencia, CA). Amplified copy RNA (12 μg) was fragmented, combined with nonplant controls, and hybridized to the GeneChip arrays according to Affymetrix (Santa Clara, CA) protocols. The hybridized microarrays were washed and stained with a streptavidin R-phycoerythrin conjugate (Affymetrix Fluidics Station) and then scanned with a Hewlett-Packard Gene Array Scanner (Palo Alto, CA). Each image was inspected visually for hybridization artifacts and manufacturing defects.
Maize GeneChip Design and Data Analysis
The maize GeneChip microarrays manufactured by Affymetrix contained 64,324 oligodeoxynucleotides derived from maize cDNA sequences. A table of 1442 genes is presented in the supplemental data online. The cDNA sequences represented a selection of various maize genes that are expressed differentially or ubiquitously. Most of these genes have assigned or hypothetical functions based on homology with known genes. Approximately 200 genes had no recognized function as of January 2002. Twenty oligodeoxyribonucleotide pairs, consisting of overlapping match and mismatch probes (20 nucleotides long), were synthesized for each gene. A small number of genes were represented by <20 oligodeoxyribonucleotide pairs. However, these sequence sets generated inconsistent results, and the corresponding data are not reported. The specific design of the oligodeoxyribonucleotide sequences used is a trade secret of Affymetrix and was not available for our evaluation.
Affymetrix GeneChip Analysis Suite 3.0 was used to evaluate the hybridization signals and determine the average difference values (expression intensities) between perfectly matched oligodeoxyribonucleotide probes and the 1-bp mismatches for each probe set. To compare gene expression in W64A+ and the mutant genotypes, it was necessary to standardize the expression data. Because of the pleiotropic effects of the opaque mutants on gene expression, the data sets obtained in this study were refractory to the data-scaling procedures recommended by Affymetrix (http://www.affymetrix.com/products/is.html) (Wodicka et al., 1997). Instead, the fluorescence intensity data, presented as arbitrary “expression” units, were scaled globally using W64A+ as the reference, as described below. Arbitrary expression data from each of the experiments were organized into spreadsheets on Excel (Microsoft, Redmond, WA). All calculations and manipulations then were performed within Excel.
First, data from the two independent replications of each genotype were compared with each other by standard linear regression, and a regression coefficient was calculated. This regression coefficient (slope) was used as a scaling factor to normalize replicate data sets to each other, and the two normalized data sets were averaged. Second, each of the averaged mutant (treatment) data sets was scaled to the averaged W64A+ (control) data set using a two-step regression analysis. The first step identified 2% of the genes among the mutant and reference sets with the most outlying data points. These genes were omitted before the second step of the analysis, and the regression coefficient obtained from the second comparison was used as a scaling factor to normalize the average expression units for each measured gene to that in W64A+. Finally, after these scaling steps were performed, relative intensity data with a numeric value <20, including negative values, were considered not detected and, according to Affymetrix conventions, systematically set to a baseline of 20 arbitrary expression units. A “trimmed” mean is more appropriate to this situation, in which there are outliers, such as zeins, that unduly influence the mean. This scaling procedure more appropriately considers the distribution of these data sets (Wonnacott and Wonnacott, 1990). The globally scaled, normalized values are presented in the supplemental data online and reflect the abundance of a particular gene relative to the total mRNA populations. These values permitted a more conservative, direct comparison of all data points, regardless of the experiment from which they originated. By contrast, most notably with the W64A o2 data set, the normalization method recommended by Wodicka et al. (1997) resulted in scaling the data such that the majority of the “treatment” expression units appeared different from the expression units of the “control.”
Materials Distribution Statement
Novel materials described in this article will be made available for noncommercial research purposes upon acceptance and signing of a material transfer agreement. In some cases, such materials may contain or be derived from materials obtained from a third party. In such cases, distribution of material will be subject to the requisite permission from any third-party owners, licensors, or controllers of all or parts of the material. Obtaining any permission will be the sole responsibility of the requester.
Acknowledgments
We are grateful to our colleagues who made critical suggestions to improve the manuscript, in particular David Galbraith at the University of Arizona. We are indebted to Paul Anderson and Larry Beach at Pioneer Hi-Bred International, who encouraged this work. We thank Virginia Dress and David Hu for technical support and Sean Yang and Carol Hendrick for bioinformatics support. We further thank colleagues in the Pioneer Hi-Bred Analytical Biochemistry Department and the Dupont-Pioneer genomics groups for their expert support. A special thanks to Jeff Kraft for assistance with the immunoblots and figure preparation. This research was supported by grants from the Department of Energy (DE-FG03-96ER20242) and Pioneer Hi-Bred International to B.A.L.
Footnotes
Article, publication date, and citation information can be found at www.plantcell.org/cgi/doi/10.1105/tpc.003905.
↵ Online version contains Web-only data.
- Received April 16, 2002.
- Accepted July 18, 2002.
- Published September 26, 2002.
References
