Results

Oligonucleotide Microarray-Based Mapping of Fast Neutron Bombarded Mutants. The Affymetrix ATH1 genome microarray contains 22,746 probe sets representing ≈23,750 Arabidopsis genes (14). Each probe set has an average of 11 perfect match probes (www.affymetrix.com/products/arrays/specific/arab.affx). These probe sets were used to identify genomic deletions by hybridization of both mutant and WT genomic DNA to oligonucleotide-based microarrays. Note that genomic DNA, instead of mRNA, was subjected to biotin labeling and hybridization to microarray genechips in this microarray-based cloning strategy, because the mRNA abundance of many genes is too low to be efficiently and reliably detected upon hybridization to oligonucleotide microarrays. Furthermore, mRNA level changes could be the result of secondary effects of a mutant. If one or more probes show decreased hybridization intensity in the mutant array compared to the WT array, a deletion may exist within the genomic DNA corresponding to that probe set. To determine the efficiency of ATH1 genome arrays to identify a genomic deletion by this microarray-based cloning method, we first analyzed the locations of probe sets covering the ORFs of annotated genes (Fig. 1A ). The 5′ most-located probes (5′-MP) were analyzed (Fig. 1 A and B ) for each gene. The average 5′-MP is located 42% downstream of the ORF start codons and shows a broad distribution for the genes represented on the ATH1 genechip (Fig. 1B ). The average 3′ most-located probe (3′-MP) of a probe set (Fig. 1 A and C ) is located 95.4% downstream of the ORF start codons and is mainly close to the stop codon of genes (Fig. 1C ). Assuming an average fast neutron deletion size of 1–4 kb (15, 16), we estimated that a substantial number of deletion mutant genes could be cloned, considering that the average Arabidopsis gene size is ≈2 kb (17). In some cases, deletion sizes can be >5 kb (18–20), which may cover more than one gene, thus increasing the deletion identification efficiency using ATH1 microarray-based mapping. These bioinformatic analyses indicated that the Affymetrix ATH1 microarrays could be used for developing rapid microarray-based cloning of fast neutron bombarded mutants.

Recessive Mutant FN1148 Has Strong Shoot Na⁺ Overaccumulation and Salt-Sensitive Phenotypes. A Na⁺ overaccumulation mutant FN1148 was isolated in our laboratory within the ICP elemental profiling screen (2) (http://www-biology.ucsd.edu/labs/schroeder/supplemental.html). As shown in Fig. 2A , ICP profiling of individual M₃ mutant progeny derived from the original FN1148 mutant showed significant Na⁺ overaccumulation in shoots (P = 4.56e ^-10, n = 10 M₃ plants). ICP profiling of M₃ FN1148 mutant plants was also performed in potting soil without added heavy metals, but with one time flooding of trays with 10 mM NaCl at 3 weeks of age (see Materials and Methods). Similar Na⁺ overaccumulation in shoots of the M₃ progeny of the FN1148 mutant was observed (n = 10). Further analyses showed that, when grown on minimal medium containing agar plates supplemented with 40 mM NaCl, the FN1148 mutants showed curled, smaller size rosette leaves (Fig. 2B ) compared to WT plants (Fig. 2C ). The salt sensitive phenotype of the FN1148 mutant (Fig. 2B ) correlates with Na⁺ overaccumulation in rosette leaves (Fig. 2 A ). No dramatic visible morphological phenotype of FN1148 was observed without exposure to excess Na⁺ under the imposed conditions (data not shown).

After a single backcross of FN1148 to its WT parent (Col-0), the Na⁺ overaccumulation phenotype was scored in the back-crossed F₁ and F₂ generations. No ion profile phenotype was observed in 22 F₁ plants (data not shown). In the F₂ progeny, six plants showed the ICP phenotype, whereas 28 showed a WT Na⁺ accumulation level, suggesting that a single recessive locus is responsible for the ion profile phenotype (P = 0.322102, χ² = 0.980392).

Identification of a 523-bp Genomic Deletion in AtHKT1 Gene by Microarray-Based Mapping. To test the hypothesis that oligonucleotide-based microarrays can be used to identify deletions, we performed two different types of chip hybridization experiments. Genomic DNA extracted directly from the original M₃ FN1148 mutant without backcrossing was used in one experiment. In the other experiment, pooled genomic DNA of five individual backcross-derived Na⁺ overaccumulation F₂ mutant plants was used for MABS analyses. Genome-wide mutant DNA hybridization intensities were compared to two control chips in which WT genomic DNA was hybridized to ATH1 chips. As expected, the experiment using the M₃ FN1148 mutant DNA pool showed many putative deletions as judged by lack of or poor hybridization intensity compared to the WT control. Significant numbers of putative deletions were also identified in the experiment using MABS analysis of the Na⁺ overaccumulation F₂ plant DNA pool.

Reduced hybridization to individual oligonucleotide probes is likely due to experiment-to-experiment variability and variation in the hybridization efficiency between individual probes. To identify potential deletions in the presence of significant noise resulting from individual oligonucleotide hybridizations, the two experiments were compared to identify common deletions of ≥2 adjacent probes in both experiments. Among the >250,000 individual probes analyzed, only four potential common deletions were identified in the following ATH1 probe sets: 245156_at, 255812_at, 266145_at, and 266146_at (Fig. 3A ). These four potential deletions showed reduced hybridizations to two neighboring oligonucleotide probes in the two mutant arrays compared to the two WT arrays (Fig. 3A , grouped red signals of two adjacent probes in two experiments).

Further analyses of the absolute hybridization reduction (WT - mutant) of the four potential deletions showed that only the hybridization reduction in probe set 255812_at showed large absolute reductions in two adjacent probes in both experiments (Fig. 3B ). The other three putative loci were found to include exon probes that are separated by an intron, which may have decreased the hybridization signal, thus decreasing the hybridization signal-to-noise ratio. PCR analyses showed that, of the four putative mutations, the deletion within probe set 255812_at is the only true deletion (Fig. 3C ). This probe set corresponds to the gene AtHKT1 (At4g10310). Two PCR primers located outside the potential deletion region were designed for PCR cloning of the genomic fragment carrying the deletion (Fig. 3D , black arrows). DNA sequencing identified a 523-bp genomic deletion within the AtHKT1 gene, which includes the last 72 bp of the second exon and the first 451 bp of the second intron (Fig. 3D , gray box). The genomic deletion includes a loss of the last 5 bp of probe 7 and the whole probe 8 (Fig. 3 A and B , probe set 255812_at, and D). Further RT-PCR analyses showed that the genomic deletion within the AtHKT1 gene in FN1148 resulted in an incorrectly spliced athkt1 mRNA (Fig. 3D , dotted box). This incorrectly spliced athkt1 mRNA lost the whole second exon, which caused a point mutation from methionine (ATG) to isoleucine (ATT) and the loss of 78 amino acid residues. No ORF shift was observed in the third exon (Fig. 3D ). AtHKT1 encodes a sodium transporter (21) with eight transmembrane domains (22). The deletion is located within transmembrane domains 6 and 7 and the following P-loop.

The findings that only one deletion was identified by comparing the experiment using FN1148 M₃ plants and the experiments using MABS analyses suggests that (i) comparative analyses of two independent experiments will aid considerably in identifying target deletions in the presence of hybridization noise of individual oligonucleotide probes. (ii) MABS analyses mask unrelated deletions, thus allowing rapid identification of ORF deletions responsible for a phenotype by using oligonucleotide microarrays.

Na⁺ Overaccumulation Phenotype Cosegregates with the AtHKT1 Deletion in FN1148. To further analyze whether the deletion within AtHKT1 is responsible for the Na⁺ overaccumulation phenotype of the FN1148 mutant, cosegregation analyses were performed by using newly selected plants from the F₂ segregating population derived from a backcross of FN1148 to WT Col-0. ICP profiling showed that 5 of 31 F₂ plants showed a dramatic Na⁺ overaccumulation phenotype in leaves (Fig. 4A Upper). PCR experiments were then performed to determine which of the individual plants showed a homozygous deletion in the AtHKT1 gene. The homozygous deletion was only identified in the same five individuals that showed the leaf Na⁺ overaccumulation phenotype (Fig. 4A Lower, PCR/A). These results support the hypothesis that the deletion within the AtHKT1 gene is responsible for the ICP phenotype of the FN1148 mutant.

Transgenic Expression of AtHKT1 Gene Complements Salt Sensitivity of FN1148 Mutant. An AtHKT1 cDNA construct driven by a native AtHKT1 promoter fragment was transformed into the FN1148 mutant. Three independent transgenic lines expressing the WT AtHKT1 mRNA were subjected to whole plant salt sensitivity analyses. In the presence of 40 mM Na⁺, the FN1148 mutant showed a significant salt sensitive phenotype (Fig. 4B ): wilted, chlorotic, or brown rosette leaves. In contrast, all of the three transgenic lines (e.g., Fig. 4C ) grew similar to WT plants (Fig. 4D ). In the absence of Na⁺ stress, no difference was observed between the FN1148 mutants (Fig. 4E ), the complementation lines (e.g., Fig. 4F ), and WT plants (Fig. 4G ). These results showed that AtHKT1 complements the salt sensitivity of the FN1148 mutant.

Comparative Analyses Between Different Salt-Sensitive Mutants. Comparative ICP analyses between FN1148, the salt overly sensitive mutant sos1 (23), and a T-DNA insertion allele athkt1–1 (24) (kindly provided by P. M. Hasegawa and A. Rus, Purdue University, West Lafayette, IN) were pursued. As shown in Fig. 5, Na⁺ overaccumulation in stems and rosette leaves was observed in both the FN1148 mutant and the athkt1–1 T-DNA insertion mutant. Moreover, in roots of FN1148 and athkt1–1, less Na⁺ was detected compared to the WT (Fig. 5; P < 0.0008 FN1148, P < 0.014 athkt1–1). In the salt overly sensitive sos1 mutant, which lies in a different transporter gene (25), overall Na⁺ accumulation in stems, leaves, and roots was increased compared to the WT (Fig. 5). These results showed that the FN1148 mutant shows similar Na⁺ overaccumulation in shoot and Na⁺ underaccumulation in root phenotypes to the athkt1–1 allele and other athkt1 alleles (26, 27).

Discussion

Microarray-Based Mapping Is an Efficient and Rapid Mutant Gene Isolation Method. Here, we describe the development of a microarray-based rapid cloning strategy and the successful application of this method to clone ion profile mutant genes (2). For traditional map-based cloning of recessive ion profile mutants, which generally have quantitative phenotypes (2), ≈2,000 cross-originated F₂ plants would have to be ICP profiled to obtain 500 homozygous F₂ mutants for fine mapping. Moreover, ≈2,500 F₃ plants might need to be ICP profiled to confirm the 500 F₂ mutants due to the quantitative nature of ICP phenotypes (see Introduction). In contrast, 34 F₂ backcrossed plants were ICP profiled here to obtain approximately six homozygous mutants, five of which were pooled for MABS analysis and the microarray-based mutant gene identification (see Materials and Methods). By hybridizing genomic DNA of both pooled mutant and WT plants and comparative analyses of two mutant and two control oligonucleotides microarrays, we have successfully identified a 523-bp genomic deletion in the AtHKT1 gene of the Na⁺ overaccumulation mutant FN1148. A mutant Ca²⁺/calmodulin-dependent protein kinase gene was identified by using a microarray approach based on mRNA abundance (28). The general applicability of this elegant mRNA hybridization-based study may be limited because some mutations may affect the transcript abundance of other genes and because of the low mRNA levels of many expressed genes.

Recent studies of other athkt1 alleles showed Na⁺ accumulation phenotypes similar to that of the FN1148 athkt1 allele (26, 27). The presented findings further confirm the predicted function of the AtHKT1 Na⁺ transporter in controlling Na⁺ distribution between shoots and roots (26, 27) under diverse growth conditions (Figs. 2 A , 4A , and 5). The microarray-based cloning strategy requires only backcrossing to the parental ecotype, thus circumventing natural variation related phenotypes resulting from crosses between two ecotypes, which is required for map-based cloning. Importantly, the rapid cloning method described here can be expanded to cloning of deletion mutant genes in other organisms with extensive microarray coverage, including human disease genes.

MABS Analysis Greatly Enhances Microarray-Based Cloning. The fast neutron mutagenized seeds used here have an average of 60–70 deletions per genome (www.arabidopsis.com/main/cat/seeds/M2/FN/!2fb.html). After one backcross, 50% of the mutagenized genome is replaced by WT genome, which suggests that an average of 30–35 deletions are still present in each F₂ genome. To identify a mutation by microarray analyses of individual mutant lines, additional backcrosses would be necessary to remove additional deletions. MABS analysis can enable mutant locus identification after a single backcross as demonstrated here, or after two backcrosses.

Conventional bulked segregant analysis usually is conducted by crossing a mutant to a different ecotype to generate two DNA pools from F₁ and F₂ mutant plants, between which only the target mutation region would show a genetic polymorphism, thus allowing identification of a molecular marker that is linked to a target mutation region (4). MABS analysis allows direct identification of deleted genes (Fig. 3). In a selected F₂ mutant pool, only the target mutation region and the neighboring linked regions are homozygous, whereas other mutations are arbitrary and would be heterozygous in a pool of sufficiently large size. When hybridized to microarrays, the heterozygous deletions will be masked by the corresponding WT genome. Thus, with MABS analysis, only the homozygous target deletion can be detected by comparing the hybridization intensity of both the WT pool and backcrossed F₂ mutant pool independent of how many unrelated heterozygous deletions are still present after one backcross.

Genomic DNA was pooled from five selected F₂ mutants for MABS analysis in the present study. With a pool of 10 F₂ mutants, the probability of linking an unlinked molecular marker to a target mutation is theoretically 2 × 10^-6 (2(1 - [1/4]ⁿ)[1/4]ⁿ), where n is number of pooled individuals (4), assuming that a single WT locus copy is sufficient to mask the corresponding unlinked deletions. However, our data show that, because of hybridization noise, independent replicate experiments are needed to identify the target mutation locus. Furthermore, there is a possibility that a deletion that is not responsible for the phenotype might be genetically closely linked to a target deletion. Such closely linked deletions will also be identified by microarray-based MABS analyses. Complementation analyses (Fig. 5 B–D ) will be necessary to rule out unrelated deletions.

Hybridization Noise in Oligonucleotide Microarray-Based Mutant Cloning. For microarray analyses of mRNA expression using the ATH1 chips, 11 probes contribute to the expression value of each gene, and therefore variation in individual probe hybridization intensities is “averaged out” by statistical analyses (www.affymetrix.com). However, in the present analysis, genomic DNA, which has more nucleotide sequence complexity compared to mRNA, was used for chip hybridization, and we searched >250,000 perfect match probes for intensity changes at the individual probe level. Therefore, noise of genomic DNA hybridization to individual probes is a more significant problem in MABS analysis. To identify deletions, we initially searched for putative deletions that spanned at least two adjacent probes. No potential deletions were identified when we searched both experiments for a ≥40% decreased signal. However, when using a less stringent criterion (at least two adjacent probes with ≥30% decreased signal), significant numbers of potential deletions were identified in each individual experiment, which are in part due to noise in the hybridization signal of individual probes. Then, the two experiments were compared to analyze for common probes with lower hybridization signals compared to WT controls. Only four common potential deletions were identified (Fig. 3A ). These findings suggest that replicates using independent DNA samples are necessary for rapidly identifying deletions in the presence of significant hybridization noise. Note that typically larger fast neutron deletions spanning more than two probes would be easier to identify than the FN1148 mutant, which deletes only 5 bp in one probe and an additional entire probe (Fig. 3 A, B, and D ). If a deletion covers only one probe, more than two independent microarray experiments and possibly increased hybridization stringency would greatly aid in overcoming hybridization noise problems. In addition, rough mapping of a mutant locus to reduce the number of probes analyzed, if natural variation does not occur for the analyzed elements, could be used in conjunction with chip mapping.

Improved Design of Oligonucleotide Microarrays for GeneChip-Based Cloning. The ATH1 probe coverage is sufficient for mapping a substantial number of fast neutron-generated mutants (Fig. 1) based on the large size of fast neutron deletions (15, 16, 18–20). However, there is a significant probability that some small and 5′ ORF deletions may not be detectable by ATH1 microarray-based hybridizations if these small deletions do not fall within the probe sets. Furthermore, even a large deletion may not be detectable if most of the deletion falls within an intron. Improved design of oligonucleotide arrays could address these limitations. For example, exons could be tiled contiguously with oligonucleotides to cover all of the possible deletions within exons. In addition, two intron probes could be included for each intron/exon junction to cover deletions that fall mainly within introns. In addition, longer oligonucleotide probes would greatly reduce the above discussed hybridization noise by increasing hybridization specificity, particularly when applying this method to isolate mutant genes in organisms with a large genome size. Oligonucleotide microarrays with both improved probe coverage and longer oligonucleotide probes are becoming publicly available (www.chem.agilent.com, www.nimblegen.com, and www.affymetrix.com).

In summary, a practical microarray-based masking bulk segregant cloning strategy was developed and successfully used to rapidly clone a shoot Na⁺ overaccumulation mutant gene. We also analyzed important sources of noise and demonstrated and discussed methods to circumvent these problems. The presented method is applicable to large-scale ion profile mutant cloning and also to isolation of genes that confer mutant phenotypes requiring intricate analyses of each individual, such as metabolic and cellular signaling mutants.