In a previous study, two regions in the soybean genome comprising approximately 0.5 Mb each surrounding the soybean cyst nematode resistance loci, rhg1 and Rhg4, were used to identify syntenic regions in the Medicago genome 32. Because there was a 2 cM gap in the first region, these sequences were referred to as synteny blocks 1a, 1b, and 2 32. The syntenic regions in barrel medic also totaled about 1 Mb, though they were scattered into smaller contigs. For example, synteny block 1b in barrel medic contained two additional gaps 32. In barrel medic, there were two segmental duplications (block 2i and 2ii) that were both syntenic to soybean synteny block 2 32.

Genes were predicted in the 1 Mb barrel medic and soybean sequence contigs using FGENESH 33. Both the dicot plants (Arabidopsis) and the Medicago (legume plant) matrixes were used and their outputs compared 33. Using the legume matrix, 229 and 217 genes were predicted for the barrel medic and soybean sequences, respectively (Additional data file 1). These represent significantly more but shorter genes (exons) compared with the Arabidopsis matrix outputs. However, the legume matrix prediction also resulted in more base-pairs in the exons (increases of 10.3% and 8.2% for barrel medic and soybean, respectively; Additional data file 1). These results clearly demonstrate that gene prediction output is sensitive to the training matrix and highlight the importance of experimental means in verifying and improving computational gene prediction. For simplicity, we selected the gene prediction from the legume matrix for further analysis.

We designed two independent sets of overlapping 36-mer oligonucleotide probes offset by five nucleotides to represent both DNA strands of the 1 Mb syntenic barrel medic and soybean sequences (see Materials and methods). Each set of probes was synthesized into a single array based on Maskless Array Synthesis technology 891034. The barrel medic and soybean arrays were hybridized in parallel with target cDNA prepared from six organ types of each plant, namely, root, nodule, stem, leaf, flower and developing seed. Fluorescence intensity of the probes was correlated with the genome position by alignment of the probes to the chromosomal coordinates (Figure 1). Transcriptional analysis of the syntenic regions was then achieved by examining expression of the predicted genes and systematically screening for TARs.

We used a method based on the binomial theorem to score the tiling array data obtained from the six organ types to detect transcription of the predicted genes 10. Analysis of the tiling array data detected 193 out of 229 (84%) and 176 out of 217 (81%) predicted genes in at least one of the six organ types in barrel medic and soybean, respectively (Figure 2a), indicating that most predicted gene loci are transcribed. Among the six organ types, detection rates of predicted genes ranged from 48% (flower) to 75% (nodule) in barrel medic, and from 60% (root) to 76% (flower) in soybean (Figure 2b). Interestingly, the gene detection rate in the nodule was the most similar between both species (74.7% and 73.3% in barrel medic and soybean, respectively; Figure 2b). These results suggest that transcription of the predicted genes from the 1 Mb syntenic sequences between barrel medic and soybean is, to a large extent, differentially regulated in the two species, which was further investigated (see below).

We next scored tiling microarray data blind to the annotated genes and identified 499 and 660 unique TARs in barrel medic and soybean, respectively (see Materials and methods). The barrel medic and soybean TARs exhibited distinct overall organ specificity. Compared with TARs in barrel medic, soybean TARs in general were detected in more tissue types (Figure 3a), implying a more constitutive expression pattern. Furthermore, roughly equal numbers of barrel medic (181) and soybean (187) TARs were detected in just one organ type. These TARs were detected in barrel medic mainly from stem and leaf while nodule and root were the most abundance source in soybean (Figure 3b). Thus, these TARs appear to represent organ-specific transcriptional activities that differ in the examined sequences between barrel medic and soybean.

Aligning against the predicted genes, 188 (38%) and 305 (46%) barrel medic and soybean TARs intersect with an exon. The remaining 311 (62%) barrel medic and 355 (54%) soybean TARs are located outside of or antisense to the predicted exons and are referred to as non-exonic TARs. The distributions of TARs detected in barrel medic and soybean in the different annotated genome components are illustrated in Figure 4a. Interestingly, the relative proportion of TARs in each annotated genome component is largely comparable to results from a whole-genome tiling array analysis in rice 10. This observation indicates that predicted exons account for less than half of the transcriptome detected by tiling arrays in rice and legume plants, despite their different genome sizes and distinct genome organization. Furthermore, a significant portion of TARs was found antisense to the predicted genes in both barrel medic (14%) and soybean (16%) (Figure 4a), which adds to previous tiling array analysis in Arabidopsis 5 and rice 8910 in showing that antisense transcription is an inherent property of the plant genomes.

The non-exonic TARs were further analyzed in terms of their physical location relative to the predicted genes. In this analysis, genome regions were divided into eight different configurations against the predicted exons (Figure 4b). Interestingly, in almost all antisense configurations, there were more TARs in soybean than in barrel medic (Figure 4b), suggesting that antisense transcription is more prevalent in soybean than in barrel medic. This analysis also revealed a surprisingly large number of intergenic TARs (36 in barrel medic and 45 in soybean) located in close proximity on the antisense strand 5' to the start of a predicted gene (Figure 4b). Because the predicted genes do not include untranslated regions, it is conceivable that transcripts derived from these TARs and the corresponding genes are arranged in a divergent antisense orientation and could potentially form duplex transcript pairs.

The binomial theorem-based method used to detect gene transcription does not assign a value to the expression level and is only useful for present calls 35. Therefore, we used a median polishing-based method that fits an additive linear model 36 to determine differential expression of the predicted genes in the six examined organ types and to assess the relative deviation of gene expression level in each organ type (see Materials and methods). In barrel medic, 67 (29%) of the 229 predicted genes were identified as differentially expressed (p < 0.001) among the six examined organ types (Figure 5). In soybean, 72 (33%) of the 217 predicted genes displayed differential expression (Figure 5).

Precise transcriptional and developmental controls are required for the establishment of the complex interaction between the nitrogen-fixing rhizobia and plant cells in the nodule. To begin to understand the transcriptional program in nodules, we identified and compared genes specifically expressed in the nodule. Within the syntenic regions in barrel medic, 11 (16%, including one duplicated gene) differentially expressed genes showed higher transcription levels in the nodule than in the other five organ types (Additional data file 2). In soybean, there were 10 (14%) differentially expressed genes showing higher transcription levels in the nodule (Additional data file 3). Nodule-enhanced expression levels of six randomly selected genes in soybean were all confirmed by RT-PCR analysis (Figure 6a), indicating that the median polishing-based method used to score the tiling data is accurate in detecting organ type-specific transcripts. A particular example is illustrated in Figure 6b. This gene (Gm_121) is homologous to the Ljsbp gene from Lotus japonicus that encodes a putative selenium binding protein 37. In situ hybridization analysis revealed that the Ljsbp transcripts were localized in the young nodules, the vascular tissues of young seedpods and embryos 37, which is consistent with the tiling array and RT-PCR data on the soybean ortholog (Figure 6b).

In soybean, all but one of the detected nodule-enhanced genes are known genes (Additional data file 3). In contrast, only three of the 11 nodule-enhanced genes detected in barrel medic match with a known gene while the other eight genes have no assigned functions (Additional data file 2). When the nodule-enhanced genes detected in barrel medic and soybean were compared for synteny, six of the ten soybean genes were found to have a collinear counterpart in barrel medic, although transcription of the collinear genes in barrel medic was not nodule-enhanced (Additional data file 3). Consequently, there was only one gene encoding a TGACG-binding transcription factor that is collinear as well as specifically expressed in the nodule in both species.

The barrel medic and soybean sequences interrogated by the tiling microarray are highly syntenic. In the previous report, a total of 68 pairs of genes were found to be collinear with both the gene order and orientation conserved between barrel medic and soybean homologs 32. In the current study, we were able to identify 78 collinear gene pairs based on the gene prediction output from the legume matrix.

To begin to obtain information on the variation in gene expression between barrel medic and soybean, which is important for defining transcriptional regulatory networks that contribute to their phenotypic variations 38, we examined the expression pattern of the collinear genes. To this end, we used the transcription level deviation in the six organ types for each collinear gene as a parameter to profile gene expression patterns. Consistent with the fact that most genes were not differentially expressed in different organ types, a majority of the collinear genes showed relatively small organ type deviation (Figure 7). However, a number of collinear genes exhibited drastic variation in transcription levels across the organ types. In barrel medic, the most conspicuous example is a group of genes that are down-regulated in the seed but up-regulated in the stem. In soybean, the root exhibited the greatest gene expression variation (Figure 7). Importantly, the transcription pattern of these collinear genes is not conserved in the reciprocal species, suggesting that the regulatory sequence of these genes is under positive selection.

Barrel medic (Medicago truncatula cv. Jemalong A17) seed was treated with concentrated H₂SO₄ for 10 minutess, rinsed with water and then allowed to germinate on moist filter paper at room temperature for a week. Seedlings 1-2 cm in length were planted in soil and maintained in the greenhouse with nitrogen-free plant nutrient solution as previously described 45. Soybean (Glycine max cv. William 82) seed was directly sown in soil and maintained in the greenhouse with nitrogen-free plant nutrient solution as described by Subramanian et al. 46.

The Rhizobium bacterium Sinorhizobium meliloti 1021 and Bradyrhizobium japonicum USDA110 was used to inoculate barrel medic and soybean plants, respectively. The bacteria were grown in a yeast extract-mannitol medium for three days at 28°C as previously described 47. The bacterial cells were then suspended in nitrogen-free nutrient solution to an OD₆₀₀ of 0.08 and used to water four-week-old plants. This flood-inoculation step was repeated after two weeks. The nodules were collected three weeks after the second treatment. Each nodule was separated from the roots with sharp tweezers and placed on dry ice immediately. The stem, root, and leaf organs were harvested from four-week-old plants that were maintained with nitrogen containing plant nutrient solution. The same plants were maintained until maturity for collection of the flower and seed organs.

Soybean sequences from four bacterial artificial chromosomes (BACs) were obtained from GenBank (accession numbers: AX196294.1, AX196295.1, AX196297.1, and AX197417.1). The BACs AX196294 and AX196295, and AX196297 and AX197417 form two contigs. There is a physical gap (represented by 100 Ns) between AX196294 and AX196295, and an approximately 50 Kb overlap between AX196297 and AX197417. Thus, the two contigs represent a total of 977 Kb of non-redundant sequences. Putative homologs to these soybean sequences in barrel medic were identified from sequenced barrel medic BACs as previously reported 32. A total of 12 BACs (accession numbers: AC141115.22, AC149303.10, CR378662.1, CR378661.1, AC142498.20, AC146585.18, AY224188.1, AC146706.8, AY224189.1, AC146705.11, AC144644.3, and AC146683.9) were identified. The sequences were aligned and merged in regions of sequence overlap on the basis of ≥99% identity 32. The merged barrel medic sequences form 7 contigs and total 1,060 Kb. Genes were predicted from soybean and barrel medic sequence contigs using FGENESH 33. Both the dicot plants (Arabidopsis) and the Medicago (legume plant) matrix was used and their output compared.

Both the barrel medic and soybean tiling arrays were produced on the Maskless Array Synthesizer platform as previously described 6834. Briefly, tiling paths consisting of 36-mer oligonucleotides offset by five nucleotides were designed to represent both DNA strands of the selected barrel medic and soybean genome sequence. Probes were synthesized at a feature-density of 390,000 probes per array in a 'chessboard' design 34. Microarray production and storage were carried out previously described 6834.

Total RNA and mRNA were sequentially isolated using the RNeasy Plant Mini kit (Qiagen, Valencia, CA, USA) and the Oligotex mRNA kit (Qiagen), respectively, according to the manufacturers' recommendations. mRNA from different organ types was reverse transcribed using a mixture of oligo(dT)₁₈ and random nonamer primers 634, during which amino-allyl-modified dUTP (aa-dUTP) was incorporated. The aa-dUTP decorated cDNA was fluorescent labeled by conjugating the monofunctional Cy3 dye (GE Healthcare, Piscataway, NJ, USA) to the amino-allyl functional groups in the cDNA. We used 2 μg dye-labeled targets for hybridization as previously described 6834. Tiling microarray design and experimental data are available in the NCBI Gene Expression Omnibus under the SuperSeries GSE10151, which is composed of subset series GSE10055 and GSE10056 for the barrel medic and soybean arrays, respectively.

Raw microarray data were first Log₂ transformed and then quantile normalized for the six organ types. A sign-test was used to determine transcription of the predicted gene 1035. First, probes lying within the exons of a predicted gene were checked to determine if their intensity was greater than the median of all probes in the array. Next, we determined whether or not the number of probes with intensity above the median was more than expected by chance alone. The probability, p, of obtaining h probes with intensity above median out of N probes is given by the equation:

p = 0 .5 N   ∑ i=h N ( N i ) MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfeBSjuyZL2yd9gzLbvyNv2Caerbhv2BYDwAHbqedmvETj2BSbqee0evGueE0jxyaibaiKI8=vI8GiVeY=Pipec8Eeeu0xXdbba9frFj0xb9Lqpepeea0xd9q8qiYRWxGi6xij=hbbc9s8aq0=yqpe0xbbG8A8frFve9Fve9Fj0dmeaabaqaciGacaGaaeqabaqabeGadaaakeaacaqGWbaccaGae8xpa0Jaaeimaiaab6cacaqG1aWaaWbaaSqabeaacaqGobaaaOGaaGPaVpaaqahabaWaaeWaaeaafaqabeGabaaabaGaaeOtaaqaaiaabMgaaaaacaGLOaGaayzkaaaaleaacaqGPbGaaeypaiaabIgaaeaacaqGobaaniabggHiLdaaaa@3FED@

Finally, genes with a p-value smaller than 0.05 were considered as detected.

Determination of differential gene expression was carried out based on the Tukey's median polish procedure, which fits an additive linear model 1036. The expression level of a given gene in the six organ types, M, is estimated by the equation:

O_ij = M + a_i + v_j + e_j

where O_ij is the observed intensity of the i^th probe in the j^th organ type, a_i is the probe affinity effect of the i^th probe, v_j is the organ type variation of the j^th organ type, and e_j is the experimental error. A series of iterations of residue subtraction was performed using Tukey's median polishing until the matrix reaches a stable status, and the organ type deviation v was determined for each organ type. To obtain the false negative error rate, a permutation test was applied. The intensities of every probe were randomly shuffled among the organ types and median polishing repeated 1,000 times. A significant level at p < 0.001 was selected for organ-specific expression calling.

To identify TARs in each organ type, the Log₂-transformed, quantile-normalized data that were anchored to the chromosome coordinates were scanned from the 5' end of each DNA strand one probe at a time. TARs were identified as regions with length >70 nucleotides (spanning roughly 14 probes) with no 2 consecutive probes having an intensity below the cutoff of 11.0.

To identify collinear genes, repetitive sequences were masked and tandemly duplicated genes were counted as one. The BLAT 48 program was then used to search the protein sequences of the predicted genes between barrel medic and soybean. The gene pairs with more than 10 amino acids mapped and more than 40% identity over the entire mapped sequences were identified as potential collinear genes, which were further manually inspected for gene order and orientation.