Data source and semantic data mining and integration for identification of stress associated genes

Five plant related ontologies, namely GO, TO, PO, GRO, EO were identified using the Gramene [29] and Gene Ontology [30] databases and were used to retrieve and identify sorghum genes that are functionally linked to plant phenotypes and directly or indirectly associated with drought tolerance. The data was uploaded to a local MYSQL database. Semantic queries pertaining to data expressed on the basis of a common vocabulary that leverage semantic information stored in ontologies were used to filter and retrieve the data from relational tables. To determine direct association, drought related ontology terms were first identified for each specified ontology including the number of genes that they represent for sorghum (Fig 1). Where direct association of sorghum gene-trait was not available from the respective ontologies, potential drought tolerant sorghum genes were captured using Ensembl BioMart [31] by transitive association, based on the putative functions of the sorghum gene orthologs in other three related species, namely, maize, rice and Arabidopsis. Ontology mapping was used to represent direct or transitive association of sorghum genes to multiple drought related ontology terms based on orthology functional relationships in maize and rice.

Once sorghum drought associated genes for all ontologies were identified and retrieved, those that were supported by all ontology terms in each ontology group were retained and merged to capture only unique entries. Further, genes supported by all ontology groups were used as an input for functional GO-enrichment (p-value < 0.01) using agriGO [32]. Investigation of gene-phenotype association was based on the correlation of genes to the enriched GO-terms.

Multiple responses of genes across stresses: Cross-talk and specificity

Using the same initial input as described above, functional correlation of drought responsive genes were compared with genes responsive to other stresses that include salt, cold, heat and oxidative stress. Sorghum drought specific and multiple stress responses were identified using the same procedure described in Fig 1. Genes were selected based on the extent of their association to each environmental stress under particular ontology terms and then filtered based on their enrichment significance level (P-value < 0.05). Where data was lacking for sorghum, closely related orthologs were used to retrieve gene association. Sorghum-rice orthologs were most employed because Gramene data source is comprehensive for rice gene association [20].

Cross-species comparative analysis: Correlating gene-trait association across species

Comparative analysis were determined based on GO associated drought responsive genes for all GO-domains across species. Ensembl BioMart [31] was used to trace sorghum orthologs in maize, rice and Arabidopsis based on the non-redundant genes identified for GO, TO, PO, GRO and EO with direct or transitive association to sorghum drought tolerance. Sorghum specific genes and those sharing attributes with other species were identified by determining cross-species gene functional association. Sorghum orthologs were compared against each other for specificity and for shared groups of orthology in relation to drought and other stress tolerances across species. Functionally conserved gene groups which are associated with drought tolerance in sorghum were detected by investigating attributes of orthologs in the respective species. Venny [33], an interactive tool for comparing list of genes with Venn Diagrams was used to display and visualize unique and common gene groups.

Integration of gene trait association with gene differential expression

Sorghum expression data related to drought stress were obtained from the National Center for Biotechnology Information (NCBI), Expression Omnibus (GEO) database [34], accession number GSE30249 [35] and GSE80699 [36]. To compare genes responding to multiple stresses across species and to detect the patterns of gene trait association with drought phenotypes that relied on tissue-specific differential gene expression, we also used maize drought expression dataset from GSE40070 [37]. To consolidate our analysis of multi-stress responses across species, we again used rice drought expression dataset from GSE57950 [38] and salt expression dataset from GSE73181 [39]. Drought and salt co-expressed genes were also identified in rice. Gene expression profiles for significantly expressed genes for all species based on drought and salt stresses are shown using a heat map and up-down regulated genes were visualized using a volcano plot for which a description of the P-value and fold-changes is shown in S1 Fig. Statistical significance was determined using parametric t-test (P-value < 0.01) to estimate the variance between subjects.

Functional-annotation and GO enrichment

Analysis of GO functional annotation was conducted using agriGO [32] where gene ID were used as input and Blast2GO V4.0 [40], a standalone software that locally incorporates repository using MYSQL DB, for sequence based analysis. Gene Ontology assignment was used to classify the functions of the selected sequences. Basic Local Alignment Search Tool (Blast, Blastp [41]) was employed to detect the sequences that were mapped against the non-redundant NCBI protein database for the best Blast hits. The functional classification and distribution of genes into main and sub-GO categories was determined, while the GO-terms were demonstrated and summarized based on the GO functional annotation.

Gene association and functional enrichment network were determined based on the enrichment level of GO terms (p-value, False Discovery Rate (FDR) < 0.05). The GO-terms with p-value < 0.05 were considered significantly enriched for all the 3 domains, namely BP, CC and MF. Similarly, enriched genes (FDR < 0.05) which exhibited strong association with their respective plant attribute from TO, PO, GRO and EO were also determined. To visualize stress related GO-term associated genes, scatter plots for multidimensional scaling of semantic similarities and gene enrichment map for functional network of the genes were generated using default values in ReviGO [42] and Cytoscape V3.3.0 [43] respectively.

Phylogenetic analysis

Sorghum protein sequences that represent 710 full length gene orthologs related with multiple stress responses in the 3 crop species (maize, rice and Arabidopsis) were retrieved from the Ensembl plant Compara multi-species database [44] for the purpose of multiple sequence alignment and phylogenetic analysis. These were subjected to a screening procedure to retain 450 genes (493 peptides) which were aligned using a multiple sequence alignment standalone tool ClustalW V2.1 [45]. In order to improve the overall sequence alignment, the iteration parameter was set to TREE to allow a retention of the resulting alignment if it was improved over the previous alignment at every iteration step. This was repeated to undertake progressive alignment for a number of rounds until the highest alignment score was met. The Maximum likelihood analysis alignment file obtained from ClustalW was used by ClustalW2 phylogeny (http://www.ebi.ac.uk/Tools/phylogeny/clustalw2_phylogeny/) to calculate a percentage sequence divergence based distance matrix and to generate a phylogenetic tree using the Nexus tree format and the neighbour-joining clustering method with the distance correction and the gaps exclusion parameters. The resulting tree file in the plain text format was uploaded, visualized and annotated in the iTOL [46].

Identification of genes associated with Quantitative Trait Loci (QTLs)

In order to evaluate if some of the genes identified in this study for sorghum multi-stress responses are co-localized within any of the QTLs previously identified for stress tolerance, we examined and compared the genomic coordinates of the target genes with the QTLs known for drought tolerance [47–49]. The target genes that fall within the QTLs genomic regions were considered as potential candidates for association with these QTLs. In addition, maize and rice QTLs fasta sequences release were obtained from the Gramene database [50] and compared with the nucleotide sequences of our target genes identified for maize and rice stress tolerance that were retrieved from the Phytozome database [51] using BioMart [31]. The latter were used as the query sequences to align with a local Blast database created from the maize and rice QTL sequences using Blastn [41]. Best Blast hits were selected by extracting a unique target hit per query sequence based on bit score, e-value (1e-10), % identity and length of alignment.

Semantic integration of data based on functional ontology

Semantic integration of information associated to sorghum and related model species stress perturbation resulted in a list of potential genes with direct and transitive relation to sorghum gene-trait association. This, however, also resulted in maize and rice specific gene-trait relationship. Where our query for relevant terms in the different ontologies yielded no existing information for sorghum, we opted to use transitive gene association to multiple traits through rice and maize orthologs. Therefore, transitive association of sorghum orthologs with drought related ontology terms was used for complementing sorghum related gene data to make sufficient association with multiple drought-related terms in several ontologies. We took advantage of the 19.6% direct and transitively associated sorghum orthologs of the identified 1709 candidate genes to construct the ontology mapping (Fig 2). This was employed to functionally validate the relevance of a total of 335 putative uncharacterized genes for drought response in sorghum (S1 Table).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Ontology map for sorghum gene association to multiple drought related terms.

The figure shows sorghum genes directly and transitively associated to multiple drought related terms based on functional ontologies. The information from EO, TO, GRO, PO and GO was used to investigate sorghum genes and orthologs in rice and maize associated with stress response. The map represents sorghum specific features for displaying class hierarchy against the ontologies under consideration and the orthologous genes from maize and rice. The hierarchical structure was designed to show multiparental relationships of sorghum genes with different ontology categories without including direct class hierarchy between maize or rice genes to the ontologies. This reveals the occurrence of multi-stress responsive sorghum specific genes and orthologous groups which are associated with GO cellular components for their localization. While the molecular functions and the biological processes of the sorghum specific genes and the orthologs are conserved, the ontology supports all these biological realities.

https://doi.org/10.1371/journal.pone.0192678.g002

Gene expression profiling based on stress conditions

Expression data was integrated with information from functional ontologies that demonstrated a successful association of drought and related stress responsive genes with phenotypes. A total of 46 significantly up-regulated sorghum genes from GSE30249 [35] were shown to have strong correlation with drought tolerance based on the evaluation of tissue type contribution to the gene expression. The evaluation of treatment effect revealed 42 significantly up-regulated genes under drought condition for which association from all plant attributes was determined (Fig 3A and S2 Table). These results show that there was a higher percentage of gene representation in tissue-specific expression under stress condition than with drought stimulation irrespective of tissue type, in agreement with the previous report [35]. Analysis of differential expression that shows significantly up-regulated genes was also demonstrated using volcano plot both for the evaluation of tissue type gene expression profiles and the treatment effect on experimental samples (S1 Fig). To consolidate our result, we further evaluated sorghum gene expression profiling, using additional experimental dataset (GSE80699) generated under drought condition for 2 leaf genotypes [36]. A total of 347 highly expressed genes were identified, of which 201 were assigned to enriched drought related GO-terms, p-value, FDR < 0.01 (S2 Table). Among up-regulated sorghum genes that were identified under drought stress, Auxin-responsive protein IAA30, Heat shock protein 81–2 and 90, Late Embryogenesis Abundant proteins-like (LEA), putative senescence-associated protein and Zinc finger family putatively expressed protein are just a few to name (S2 Table). A combination of sorghum drought specific non-redundant set of genes identified from the two dataset is presented in S2 Table.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Heat map showing differential gene expression.

This figure shows gene expression profiles based on sorghum drought stressed root and shoot tissues (a), the 22 most abundant GO terms enriched in maize leaf and ovary tissues under drought stress with the corresponding up-regulated maize genes and their respective sorghum orthologs expression patterns (b). Similarly, the figure shows the 22 common GO terms enriched in rice leaf and shoot under salt stress showing the corresponding up-regulated rice genes and their corresponding sorghum orthologs expression patterns (c). Sorghum orthologs expression patterns were added in (b) and (c) to show visual comparison of expression profiles for transitively associated genes between sorghum and maize and sorghum and rice separately. Parametric analysis of gene set enrichment was determined by the T-statistics based clustering frequency using MeV 4.48 [52], an R based software. The rows represent the genes (a), GO-terms and corresponding genes and orthologs and GO annotation (b) and (c), whereas the columns represent the biological samples. While the red color denotes the up-regulation, the green shows down-regulation of the genes in all the clustering panels. Hierarchical clustering, for instance show the patterns of expression in (a), by grouping the most up-regulated sorghum genes in the upper right corner, middle and lower left corner.

https://doi.org/10.1371/journal.pone.0192678.g003

Based on maize expression dataset, GSE40070 [37], the pattern of expression profiling that was determined using parametric t-Test, p-value, FDR < 0.01 resulted in a total of 300 genes expressed under drought stress of which 200 were tissue specific. Evaluation of tissue distribution of significantly expressed genes showed more up-regulated genes in the reproductive stage than in the vegetative, more likely concomitant with the trend of expression pattern in [37]. Conversely, 125 up-regulated genes were obtained from the treatment based grouping, out of which 100 genes with best fold-changes were selected for functional gene enrichment and GO annotation analysis in combination with the result from tissue based grouping. This resulted in 156 genes annotated for enriched drought associated GO-terms (S3 Table). This result shows an additional finding of drought expressed, tissue specific genes compared to what Kakumanu and colleagues had previously identified and presented in their final list [37]. The heat map for hierarchical clustering of 22 most abundant enriched drought related GO-terms including GO IDs is shown depicting the expression patterns of the maize genes and the corresponding sorghum orthologs (Fig 3B).

Analysis of the rice drought expression dataset, GSE57950 [38] showed 284 significantly expressed genes, p-value < 0.05 (S4 Table). Similarly, rice salt expression dataset, GSE73181 [39], revealed 164 tissue specific genes and 161 genes regulated irrespective of the influence of tissue on their expression pattern. Among tissue specific genes, 97 were annotated for enriched salt specific GO-terms, while among genes expressed regardless of tissue type, 36 were assigned to salt related GO-terms, p-value, FDR < 0.01 (S4 Table). However, when evaluated for multi-stress responses in rice, 84 genes were shown to be co-expressed both under drought and salt stresses. We show the heat map for hierarchical clustering of 22 most abundant enriched GO-terms with the corresponding GO IDs for drought and salt co-expression demonstrating expression patterns of the rice genes and the corresponding sorghum orthologs (Fig 3C).

Gene association across-environmental stresses: Cross-talk and specificity

Sorghum genes association for functional cross-talk and specificity was investigated for drought tolerance and other stresses (Fig 4A, 4B and 4C). Among the 169 genes initially identified as drought responsive in sorghum based on the GO biological process, about 56% were shown to be responsive in multiple environmental stresses (Fig 4A and S5 Table). Among these multi-stress responsive genes, about 69% were salt responsive, whereas 51 and 15% were responsive to cold and heat respectively. Again, 22, 11 and 4% of the genes were shown to have dual function towards salt and cold; salt and heat and cold and heat responses respectively. Interestingly, 2% of the genes were shown to have universal responses to all the stresses under investigation (Fig 4C and S5 Table).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Venn diagram to show functional cross-talk and specificity of genes for drought tolerance and other stresses.

Sorghum genes are shown in association with drought, salt, cold and heat stress related ontology terms of the biological process based on the datasets originated from Gramene database (a) and those in association with drought, ABA, cold and salt stresses based on sequence similarity search using expression dataset (b). Similarly, sorghum genes associated with ontology-terms of the biological process related to stress other than drought (salt, cold and heat and reactive oxygen species) based on data from Gramene database (c) are presented. The numbers displayed in the Venn diagram correspond to the number of genes. The superimposed regions of all circles show the number of genes shared in all the four species. The peripheral parts that don't overlap between circles show unique genes responsive to the respective stresses. Key to legend: RC—Response to cold; RH—Response to heat; ROS—Response to osmotic stress; RROS—Response to reactive oxygen species.

https://doi.org/10.1371/journal.pone.0192678.g004

Based on the biological processes of the GO category, 1116 sorghum candidate genes were identified to respond to 5 different abiotic stresses, viz., drought (169), salt (352), cold (221), heat (92) and oxidative stress (282) (Fig 4, Table 1 and S5 Table). A diverse functional array of sorghum gene association is characterized by the over-expression of some specific genes for multiple traits. For instance, 2 peculiar genes (Sb03g026070 and Sb09g030600) were identified to be expressed in all the 4 stresses, namely drought, salt, cold and heat (Fig 4A). Furthermore, many other genes were shown to have common expression in 2 or more abiotic stresses. For example, 2 genes (Sb01g037090 and Sb02g043450) for drought, cold and heat, 2 other genes (Sb03g039820 and Sb09g022290) for drought, salt and oxidative stress, again 2 other genes (Sb01g003880 and Sb10g023010) for drought and heat and 3 more genes (Sb0010s007790, Sb01g031520 and Sb10g022780) for drought and oxidative stress were found to be commonly expressed (S5 Table).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Description of the GO enrichment analysis with enrichment level of the GO-terms in decreasing order, the corresponding number of drought responsive genes involved and the associated traits in each GO-category.

https://doi.org/10.1371/journal.pone.0192678.t001

Similar results were also observed for the large number of genes interacting across environmental stresses. For example, 8 genes were shown to act commonly in 3 stresses: i) drought, cold and oxidative stress and ii) drought, heat and salt each (Fig 4A and 4C). Seventeen genes in drought and cold, 19 genes in drought, salt and cold and 35 other genes in drought and salt were commonly responsive (Fig 4A and S5 Table). The distribution and functional correlation of genes associated with abscisic acid (ABA), drought, salt and cold stresses as indicated in Fig 4B were basically dependent on the extent of sequence similarity. The pattern of functional association for genes that are purely drought responsive (Fig 4C) was depicted based on the enriched drought related GO-terms of the biological process. Stress specific expression of genes in all association was also shown (Fig 4A, 4B and 4C); for instance, 71, 232, 208, 120 and 53 genes were found to be uniquely specific to drought, oxidative stress, salt, cold and heat, respectively, as indicated in Fig 4A and S3 Fig.

Comparative gene association across-species

Among a total of 168 sorghum drought responsive genes identified based on sequence alignment, 90% were found to exhibit drought tolerance across species without expressing any sorghum specific genes (Fig 5). Sorghum genes' functional correlation with orthologs in other species showed that 11% were shared with maize only, nearly 5% with rice only and 5% with Arabidopsis only. Again, 12% of sorghum genes were shared with maize and rice in common and 15% with rice and Arabidopsis. Moreover, 34% of the total sorghum genes were commonly shared by all 3 species. This shows the presence of species specific and shared gene loci and probably functional conservation in closely and distantly related species of grass families. The total number of drought responsive genes represented in sorghum, maize, rice, and Arabidopsis were 335, 138, 214 and 613, respectively (Fig 5A and 5B; S1 Table). Such a representation of drought responsive genes in each species in this data was based on the relevant drought related terms in the EO, TO, PO, GRO and the GO (Table 2, S6 Table and Fig 5A). However, it is important to note that potential genes for drought tolerance in sorghum having shared functionality with closely related species were identified based on the putative functions of their orthologs in all related species using Blast algorithm and the Ensembl BioMart as described in the method. The identification of a relatively larger number of both shared drought responsive genes among all species and that of sorghum specific based on sequence similarity search using expression data (Fig 5B) than based on querying known genes in Gramene database (Fig 5A) suggests the presence of new biological information content in the expression dataset.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. Venn diagram showing distribution of shared drought responsive genes among species and specific genes to sorghum.

The figure shows a distribution of sorghum orthologous genes in the other 3 related species in association with drought related ontology terms based on existing data for known genes in Gramene database (a) and based on sequence similarity search using expression dataset (b). The numbers displayed in the Venn diagram correspond to the number of genes. Superimposed regions of all circles show the number of genes shared in all species under investigation. Overlapping regions between any 3 species indicate shared gene loci and functional conservation between the 3 of the 4 species while the shared regions between any 2 species involved show the shared gene loci and functional conservation in the 2 species. Parts that don't overlap between circles show unique drought responsive genes for each species.

https://doi.org/10.1371/journal.pone.0192678.g005

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Summarized description of functional association of genes with various drought related ontology terms identified at different stages based on step-wise screening procedure.

https://doi.org/10.1371/journal.pone.0192678.t002

Phylogenetic relationship

Phylogenetic tree of the sorghum specific and orthologous genes identified for drought response in the other 3 evolutionarily related species to sorghum is displayed in Fig 6. The tree represents evolutionarily related ortholog clades, with branch lengths showing the amount of genetic changes between the clades. Distinct classes of evolutionarily related genes were found to be conserved across species, while relatively few sorghum specific genes were shown to exist. This comparative sequence evolutionary pattern across species for drought response was depicted based on the protein sequences identified in sorghum and its close relative species (S7 Table). The number of protein sequences evaluated for functional conservation across species was dependent on the availability of the protein sequences queried per species in the existing database. For instance, a ClustalW phylogenetic tree of these proteins revealed 297 genes functionally conserved between sorghum and Arabidopsis (SOA) among initially identified (613) and combined (782) orthologs (S8 and S9 Tables). Of the initially identified sorghum orthologs in maize (SOM, 138 genes), 4.5% account for 3.3% of the total conserved genes, whereas 8% of the 214 initially identified and 1% of the combined sorghum orthologs in rice (SOR) altogether accounted for 4.3% of the total conserved genes.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 6. Circular representation of the polar formatted phylogenetic tree of the sorghum specific and orthologous genes.

Group of genes were color-coded by orthology group identified for drought response in the other species evolutionarily related to sorghum. The tree represents labels that were aligned with default leaf sorting. Branches represent evolutionarily related ortholog clades. Branch lengths for which 'ignored' setting was adjusted were represented each by the numbers in decimal and the bootstrap values in absolute numbers (S10 Table). The tree was reconstructed after removing the gaps using a bootstrap support of the 1,000 replicates to show the frequency of each internal node, clades in the tree. The red circular bootstrap symbol was used to indicate the bootstrap supported clades based on the values within the range of 100 (small dot)– 1000 (large dot) iterative replicates, where more than 75% of the clades showed the bootstrap above the commonly known threshold value (70%). The clades with the bootstrap values less than 5% were removed from the tree. The values for the robust bootstrap support were given in S10 Table. Key to legend for the colored ranges: SOA, Sorghum orthologs in Arabidopsis; SOM, sorghum orthologs in maize; SOMA, shared sorghum orthologs in maize and Arabidopsis; SOMR, shared sorghum orthologs in maize and rice; SOMRA, shared sorghum orthologs in maize, rice and Arabidopsis; SOR, sorghum orthologs in rice; SORA, shared sorghum orthologs in rice and Aabidopsis; Sorghum, sorghum specific genes; Sorghum_SOA, shared sorghum specific and sorghum orthologs in Arabidopsis; Sorghum_SOMA, shared sorghum specific and sorghum orthologs in maize and Arabidopsis; Sorghum_SOMR, shared sorghum specific and sorghum orthologs in maize and rice; Sorghum_SOMRA, shared sorghum specific and sorghum orthologs in maize, rice and Arabidopsis; Sorghum_SOR, shared sorghum specific and sorghum orthologs in rice; Sorghum_SORA, shared sorghum specific and sorghum orthologs in rice and Arabidopsis.

https://doi.org/10.1371/journal.pone.0192678.g006

When viewed in terms of the number of sorghum orthologs conserved among more than 2 species, we found 44 (9%) drought responsive genes in all the species, 63 genes (13%) among sorghum, rice and Arabidopsis, and 22 genes (4.5%) among sorghum, maize and rice. Furthermore, 19 genes (4%) were shown to be shared among sorghum, maize and Arabidopsis. On the other hand, 5.3% of the 169 sorghum genes that were initially identified as drought responsive remained sorghum-specific, while the rest were evolutionarily conserved in 1 or more other species (Figs 5 and 6). These cross species conserved orthologs were classified into 14 subclasses based on the number and the type of species in which the orthologs were commonly occurring. We show shared orthologous genes and the type of species in which these genes are conserved (Fig 6 and S9 Table).

Functional-annotation and enrichment of plant ontology terms

Based on the association of genes with drought related GO terms, 167, 148, 133 significantly enriched genes (Table 1; p-value, FDR < 0.05) were identified for all the 3 domain namely BP, CC and MF respectively. This was further filtered to 126 non-redundant genes supported by all GO-domains. Similarly, using the same method, 296, 1681, 98 and 712 enriched genes (p-value, FDR < 0.05) were filtered from TO, EO, PO and GRO respectively, which were shown to have strong association to the plant attributes (Fig 7; Table 2; S6 and S11 Tables). The combination of these makes a total of 2224 filtered non-redundant genes which were further screened down to 2118 enriched transcripts or 1820 genes (S2 Fig and S6 Table). The distribution of these sorghum orthologs based on ontology categories is depicted using a Venn diagram, where the EO was shown to contribute the highest proportion (72%), followed by GRO (30.6%), TO (12.7%), GO (5.4%) and PO (4.2%) with 7 genes shared by all ontologies (Fig 7).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 7. Venn diagram showing the sorghum orthologous genes identified in different plant ontology categories.

The diagram represents the distribution of gene association related to the five drought-associated plant ontology terms. The numbers in the overlapping portions represent the number of gene contributions shared by 1 or more ontology categories with the genes positioned in the center represented by all ontologies, while those shown on the peripheral portion represent the number of genes specific to each respective ontology category.

https://doi.org/10.1371/journal.pone.0192678.g007

Gene ontology assignments were employed to functionally group the genes. Based on the Blast2GO analysis of sequence homology, 2357 annotated sequences that had received Blast hits from the non-redundant NCBI protein database were classified into 28 functional groups under the main categories of the GO classification (S3 Fig and S11 Table). In the BP, the metabolic process, single-organism process, response to stimulus, biological regulation and regulation of biological process were noted to be dominant to which 227 genes (9.6%), 197 genes (8.4%), 118 genes (5.0%), 106 genes (4.5%), 100 genes (4.2%) were classified respectively. The rest functional groups of the BP contributed to the classification of a total of 783 genes (33%). In the CC of the GO category, the predominant categories were cell and cell part each accounted for the functional classification of 230 genes (9.8%). While the organelle, membrane and membrane part contributed for the classification of 172 genes (7.3%), 110 (4.7%) and 83 genes (3.5%) respectively, the rest functional groups of the CC accounted for the total of 101 genes (4.3%). On the other hand, in the functional category of MF, the binding and catalytic activity each predominantly accounted for the classification of 184 genes (7.8%) and 171 genes (7.3%) respectively. While the highest percentage of genes from binding, cell and cell part and metabolic process was noted, only few genes were detected from the category of molecular transducer activity, symplast and positive regulation of biological process in the main GO categories of the MF, CC and BP respectively. A graphical representation of significantly enriched GO-terms assigned to the identified genes that demonstrated strong association with drought-responses was demonstrated using scatter plots and GO annotation and classification (S3 Fig).

Gene functional enrichment network

Biological networks of gene association for which enriched GO-terms exist can be shown by using interactive biological networks [43] based on all deterministic factors attributed to the 3 GO-categories [53]. The gene enrichment network maps for selected 50 functionally enriched drought related GO-terms and their corresponding genes (P-value, FDR < 0.05) are shown in Fig 8A and 8B respectively. While mostly difficult to precisely interpret the functional networks and interactions of the genes, we opted to summarize the results as indicated in this article. The responses to stress, stimulus, chemical and abiotic stimulus and response to organic substances were shown to take the leading position with high significance of enrichment. The network denoted by the first 3 enriched GO-terms indicated above corresponds with the gene sets represented by 'Sb03g042500', the gene that also partly regulate the functional network of the primary metabolic process and the response to oxidative stress. Similarly the network represented by the last 2 enriched GO-terms corresponds with the gene sets represented by 'Sb04g030950' and 'Sb06g017490' respectively where the latter also coordinates the functional network and genetic interaction for leaf senescence. Moreover, the functional regulation of cellular and biological processes were shown to be controlled by the common set of genes represented by 'Sb03g030950', a gene that is also responsible for the functional network of response to abiotic stimulus. The biological network of the signal transduction was noted to be regulated by the set of genes represented by the gene, 'Sb01g007120' (Fig 8A and 8B; S12, S13 and S14 Tables).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 8. Enrichment network map for selected sorghum drought responsive genes.

Gene enrichment maps for selected 50 sorghum drought associated enriched GO-terms (a) and the corresponding genes (b) depict the biological networks of the genes that are involved in the regulation of cross talk in response to multiple stresses. Nodes denote a group of genes (gene-sets) or group of GO-terms and edges represent GO defined relations. The threshold level of the enrichment significance determines the appearance of the group of genes on the enrichment network map. The intensity of the node represents the level of significance of the enrichment and the size of the node correlates with the size of significantly enriched gene set that overlaps or makes the group of up-regulated enrichment. The p-value is included in the label of the nodes to indicate the level of enrichment significance. The network explains the corresponding gene function defined by the enriched GO-terms in the particular GO-category. The position of the nodes for the enriched gene set is not necessarily correlated with the position of the corresponding enriched set of GO-terms. The color usage for the node and edge is an arbitrary selection for proper contrast.

https://doi.org/10.1371/journal.pone.0192678.g008

It was also noted that the functional network of response to cold and temperature stimulus; response to heat and sesquiterpenoid biosynthetic process; response to water and water deprivation and response to osmotic and salt stresses were shown to be regulated by the interaction of set of genes each represented by Sb03g029570, Sb06g018220, Sb10g007610 and Sb09g022060 respectively. Likewise, the functional network of the hyperosmotic response, hyperosmotic salinity response and response to desiccation were noted to be controlled by the group of genes represented by Sb03g039820. The network for the negative regulation of signal transduction and regulation of abscisic acid mediated signaling pathway was however controlled by a common gene set denoted by Sb07g029150. Again, functional network for stomatal movement and regulation and molecular and signal transducer activity was detected to be controlled by gene sets each represented by Sb10g025470 and Sb10g029400, respectively.

Resistance from whole-plant to individual level components

This study identified multiple individual level components that represent various drought response mechanisms specifically related to tissue type and developmental stage. A total of 669 genes which were manifested via osmotic adjustment (OA, 52.6%), antioxidant capacity with response to oxidative stress (42%) and desiccation tolerance (5.4%) were identified for DT in general. In addition, 19 genes with cellular responses to water deprivation and 126 genes with physical response to water deprivation were identified as DA category. Furthermore, a remarkably high number of genes (2442) responsible for DE were also identified for which relevant genes associated with early booting and a gene “Sb03g003110”, a rice ortholog (BGIOSGA002217) with late booting character were filtered (S15 Table).

Identification of target genes associated with QTLs

This study has identified a total of 272 target genes which were associated with QTLs related to different traits including 62% of gene association from sorghum, 7.8% from maize and 30.2% from rice genes. One hundred and sixty-nine sorghum target genes initially identified for multi-stress tolerance were found to be associated with QTLs responsible for various traits. This includes 37 (21.9%) genes associated with drought adaption [47], 84 (49.7%) genes responsible for grain yield, flowering time, and stay-green traits [48] and 28.4% of the genes associated with seed dormancy [49] (Table 3; S16 Table). This study also identified 21 target genes in maize that are associated with drought QTLs of different agronomic purposes using sequence alignments based on Gramene QTL release [50]. The best hits were selected based on the % identity, e-value, bit-score. QTLs for abiotic stress tolerance (33.3%), biochemical (33.3%), developmental (4.8%), quality (4.8%) and yield (23.8%) traits were included. The abiotic stress tolerance QTLs were noted to play the main role in turgor pressure, stomatal conductance and abscisic acid concentration in plants, whereas the QTLs for the biochemical traits control the total soluble sugar content, ADP glucose pyrophosphorylase activity, peroxidase-71 and sucrose contents (S17 Table). In addition, other QTLs regulating female floral development, tenderness quality, ear number and seed weight were also identified in maize. Molecular markers linked to these QTLs and corresponding QTLs for the target maize genes are listed in S17 Table.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Description of sorghum target genes associated with QTLs known for drought tolerance.

https://doi.org/10.1371/journal.pone.0192678.t003

Furthermore, 82 target salt responsive genes that are associated with QTLs controlling different traits in rice were identified. The QTLs and the associated genes were subdivided into different categories based on the type of traits they control including abiotic stress tolerance (12%), anatomical (9.8%) and biochemical traits (12%), sterility or fertility (12.6%), yield (50%) and quality characters (1.2%). Two types of markers (RFLP, 56% and SSR, 44%) were identified to be genetically linked to the QTLs. The detailed description including QTLs and the corresponding marker IDs and species from which the markers originate is provided in S18 Table.

Conclusion

The methods used in this study could serve as a promising approach for data integration in multiple stress tolerance investigations across species. The information provided adds to the body of knowledge by providing researchers with a unique vanguard integrative data analysis system towards genetic dissection of complex polygenic traits. While the results have shown that genes with functionally relevant across species for multiple stresses have been successfully identified, the study may have more implications in comparative study of major cereal crops, thus providing insight into functional and evolutionary information. These data could, therefore, be used in comparative genomics and in breeding programs towards improving stress tolerance in sorghum and related species.