During the generation of the Affymetrix Soybean Genome Array GeneChip the previously identified 66 H. glycines parasitism genes 1012162122 were grouped by the manufacturer into 46 contigs, which are represented by 62 probesets on the GeneChip (some contigs have multiple probesets) (Additional file 1). Using our previously described Affymetrix microarray data set, which encompasses all life stages of H. glycines except adult males 35, we determined the developmental expression patterns of all 62 probesets representing these 46 contigs. All of the parasitism gene probesets displayed significant expression changes during the H. glycines life cycle at a false discovery rate (FDR) of 5%. We performed statistical analyses as described in Methods for all parasitism gene probesets, which identified two clusters based on gene expression patterns (Figure 1). H. glycines parasitism genes either reached a maximum mRNA abundance in infective J2, remained steady until parasitic J2 and dropped off steeply thereafter (Cluster 1) or they reached a maximum in their mRNA expression levels in parasitic J2 and then fell steadily after the J3 stage (Cluster 2). The gland-specific expression patterns of these genes have been determined in previous studies 162122. Apart from very few exceptions, parasitism genes expressed in the subventral glands of the nematode fell into Cluster 1, whereas parasitism genes from the dorsal gland could be grouped into Cluster 1 or Cluster 2 (Additional file 1). The majority of H. glycines parasitism genes, particularly those expressed in the dorsal esophageal gland, were novel and did not share similarities with any known sequences. Therefore, the expression profiles obtained here are important tools to infer putative protein functions judging from the time points at which these genes are upregulated. Genes that are important during host invasion and early sedentary phases, which includes the early events of syncytium induction (i.e., infective J2/parasitic J2), should be represented in Cluster 1, because this cluster showed an expression maximum in these two stages (Figure 1). Genes that are relevant during later stages of parasitism like the later stages of syncytial development, syncytium maintenance and feeding, are expected in Cluster 2, whose members were expressed at a relatively high level also in late stages (J3, J4, adult females). However, it was remarkable that such diverse proteins as secreted cellulases and pectate lyases with an obvious role in plant cell wall degradation (i.e., during early phases of parasitism) had a strikingly similar expression pattern to genes encoding secretory proteins that are believed to influence the host cell's physiology (e.g., chorismate mutases, ubiquitin extension proteins) or for which no function during parasitism can be ascribed yet (e.g., chitinase, venom allergen proteins). The expression profiles of these genes showed a clear downregulation after the parasitic J2 stage, which is the beginning of the sedentary phase of the nematode and of feeding site development. Furthermore, those parasitism genes that were upregulated in later life stages did not match any other known sequences, apart from two exceptions: annexin and a cellulose-binding protein (Additional file 1). Annexins bind to calcium-dependent phospholipid membranes, and while their role during parasitism remains unknown, the cellulose-binding protein binds to a plant pectin methylesterase and functions in cell wall modifications during syncytium development 39.

In addition to analyses of gene expression changes throughout the whole life cycle, in which each life stage was compared to all others, we also performed statistical analyses to identify gene expression changes in consecutive life stages to isolate specific shifts in gene expression. The results of these studies showed that most parasitism genes were differentially expressed when eggs were compared with infective J2 as well as when infective J2 were compared with parasitic J2. All of these differentially expressed genes were upregulated (Table 1). This pattern drastically changed from parasitic J2 to J3, where only a smaller subset of genes was differentially expressed and, interestingly, these were all down-rather than upregulated.

To identify new H. glycines genes that are potentially involved in host-parasite interactions we identified all genes that encode secretory proteins. Specifically, we searched for signal peptide-coding regions in the consensus sequences of all 7,530 H. glycines probesets on the Affymetrix Soybean Genome Array GeneChip as described in Methods. This analysis identified 633 unique H. glycines genes that encode proteins with a putative signal peptide but lack a transmembrane helix and have an open reading frame (ORF) of at least thirty amino acids after the predicted signal peptide cleavage site (Additional file 2). However, it can be assumed that H. glycines possesses more gene products with signal peptides than we could detect here because a significant portion of ESTs lacks a complete 5'-end so that no signal peptide-coding region could possibly be identified. All known parasitism genes are included in these 7,530 probesets, and our detection protocol for secretory protein coding genes re-identified all but seven of the known sixty-six parasitism genes analyzed here, giving a re-discovery rate of 89%. The probesets of genes that were missing did not meet all of our stringent selection criteria.

Because most parasitism proteins have no database hits to known proteins, we determined whether the signal peptide-encoding genes identified here differ in their likelihood of having database matches compared to other genes. For this purpose, we conducted BLASTX searches of all 6,860 unique H. glycines genes underlying the 7,530 probesets of the Affymetrix GeneChip against the non-redundant GenBank database. We found that 52.8% (334/633) of the gene products predicted to be secreted had matches when using a cutoff value of 1e-05 (Additional file 2). Similarly, 58.5% (3,644/6,227) of the gene products not thought to be secreted had matches. Even though this difference of 5.7% is relatively small, we found a significant difference when the relative BLASTX scores were compared. Out of all 6,860 unique genes, 3,978 had matches, and resulted in a median BLASTX score of 310 (min 41, max 5,249). Using only the 334 gene products of secreted candidates, the median BLASTX score was reduced to 129 (min 27, max 3,299), which suggests that as was the case for previously reported parasitism genes, our newly identified cohort of secretory proteins containing potential parasitism proteins evolved more rapidly than non-secreted sequences.

To analyze with which organisms H. glycines signal peptide-bearing proteins share conservation, we sorted the top BLASTX hits meeting a cutoff of 1e-05 by organism (Figure 2). Approximately 26% (166 genes) of the 633 H. glycines genes with putative signal peptide-coding regions matched Caenorhabditis elegans or Caenorhabditis briggsae sequences, followed by 12% that matched other H. glycines genes and 7% that aligned best with sequences from animals other than nematodes. However, 147 out of the total 166 Caenorhabditis matches represented unknown or hypothetical proteins and, similarly, 47 out of 78 H. glycines sequences were novel, i.e., they did not share similarity with other known sequences.

Using statistical analyses, all Affymetrix probeset expression profiles obtained for the 633 signal peptide-encoding genes separated into nine expression clusters (Figure 3) as described in Methods. When all experimental data, i.e., expression data throughout all life cycle stages, were taken into consideration, 94% of these probesets were differentially expressed (FDR 5%) during the life cycle. When conducting statistical analyses of expression data for consecutive life stages in pairwise comparisons, we determined that 49% of the Affymetrix probesets representing the 633 secretory protein genes were differentially expressed (FDR 5%) during the transition from eggs to infective J2. Similarly, 48% of these Affymetrix probesets were differentially expressed from infective J2 to parasitic J2 and 33% from parasitic J2 to J3. The fewest significant changes (10%) occurred between J3 and J4, followed by the comparison between J4 and adult females (20%) (Table 1). Taken together, the clusters for overall differentially expressed probesets as well as the stage by stage comparisons showed that the strongest expression changes in genes that code for secretory proteins took place at the transitions into and out of the infective J2 stage, which marks the preparations for host invasion and a parasitic lifestyle.

Previous studies showed that a substantial proportion of the known cyst nematode parasitism proteins share a high degree of similarity with bacterial 101113 and fungal 111213 proteins and not to the non-pathogenic Caenorhabditis spp. These observations led to the suggestion that certain nematode genes that are important for a parasitic relationship with the host might have been acquired by horizontal gene transfer from microbes 71038. Further studies demonstrated that other cyst nematode secretory proteins with potential involvement in parasitism have a striking similarity to plant proteins 15164041, which could hint at mimicry of plant regulatory proteins by cyst nematodes to alter the physiology of the host. Therefore, we identified H. glycines proteins that showed significant sequence similarity with microbe or plant proteins by manually sorting the results of the BLASTX searches of all 6,860 H. glycines genes against the non-redundant GenBank database mentioned above. We isolated all matches to proteins from plants, microbial phytopathogens/phytosymbionts (termed phytomicrobes here), soil-living microbes and other microbes. A counter-selection protocol ensured that highly conserved sequences that are present in diverse organisms (including Caenorhabditis spp.) were removed in order to discard sequences that are unlikely to be involved in parasitic relationships (see Methods). These analyses revealed that, using our criteria, 29 H. glycines protein sequences were conserved in plants, 41 in microbial phytomicrobes, 33 in soil-living microbes, and 53 in other microbes, resulting in a total of 156 such proteins (Additional files 3, 4, 5, 6). For example, we identified a plant-like H. glycines gene whose translated product had similarity to beta-amylase from Arabidopsis (HgAffx.12554.1), which is interesting because genes encoding beta-amylases have so far only been found in microbe and plant genomes, but not in animals. Furthermore, we found H. glycines genes whose products were similar to a potato protein induced in the feeding site (giant-cells) of the root-knot nematode (Meloidogyne) (HgAffx.13422.1), or were involved in RNA interference (RNAi) like a Zwille/Pinhead-like protein from rice (HgAffx.13330.1). Phytomicrobe-like H. glycines sequences matched sugar metabolizing enzymes from Agrobacterium tumefaciens, e.g., mannitol-2-dehydrogenase (HgAffx.18502.1) and sucrose hydrolase (HgAffx.9663.1, HgAffx.12954.1), as well as enzymes like glutamine synthetase (HgAffx.18955.1) and phosphoribosyltransferase (HgAffx.23512.1) or aldose-1-epimerase (HgAffx.18360.1) from Mesorhizobium to name just a few.

It is particularly interesting to discern if these H. glycines genes potentially are secreted from the nematode. Therefore, we examined how many of these H. glycines genes were part of the secretome identified in this study. Of the 29 plant-conserved H. glycines genes, three encoded a putative signal peptide. Similarly, four of those conserved in phytomicrobes, four of those conserved in soilmicrobes and four in other microbes (Additional file 7) contained signal peptide sequences. However, these predicted proteins did not necessarily meet all our other, more stringent criteria. E.g., some proteins did have a putative transmembrane helix or did have less than 30 amino acids after the predicted signal peptide cleavage site (see Methods). Also in these analyses, one needs to remember that a significant portion of the ESTs lacks a complete 5'-end so that no signal peptide-coding region could possibly be detected. Additional file 7 shows the overlap between H. glycines contigs identified as parasitism genes, secretory protein-encoding or plant/microbe-like. Interestingly, some of those H. glycines genes that did fulfill all our criteria for secreted plant-like proteins matched for example plant genes encoding histone deacetylase 2 from Arabidopsis (HgAffx.19783.1) or the Meloidogyne-induced giant-cell protein-like protein from potato (HgAffx.13422.1). H. glycines proteins with similarity to microbial gene products that passed our signal peptide selection process matched among others a flavin adenine dinucleotide (FAD)-linked oxidase from Arthrobacter (HgAffx.18477.1), histidine triad nucleotide-binding family protein 1 (HIT1) from Chaetomium globosum (HgAffx.11878.1) or a hypothetical protein from Gibberella zeae (HgAffx.17226.1). To our knowledge, none of these genes encode proteins with signal peptides in the respective plant or microbe species to which the H. glycines sequences matched.

We further grouped all identified plant- and microbe-like H. glycines genes represented by their respective probesets into distinct expression clusters (Additional files 8, 9, 10, 11). In order to identify specific shifts in gene expression, we performed pairwise comparisons of consecutive life stages of H. glycines for genes that were conserved in plants, phytopathogens/phytosymbionts, soilmicrobes and other microbes regardless of whether we could find a signal peptide-encoding sequence or not. The results of these analyses are summarized in Table 1. It is evident that the vast majority of these probesets was differentially expressed when eggs were compared with infective J2 and infective J2 with parasitic J2 and that up- and downregulated probesets were represented in about equal proportions in these two stage-wise comparisons. This means that these plant and microbe-like H. glycines genes are strongly regulated at the onset of parasitism.

Histone modifications are an important regulatory element in gene expression 42. We found here a H. glycines histone deacetylase-2 (HDA2) probeset (HgAffx.19783.1.S1_at) with an Arabidopsis HDA2 (AAM34784.1) as best BLASTX match for its consensus sequence, rather than a homologous gene in nematode species, including the fully sequenced C. elegans or C. briggsae genomes. Interestingly, this H. glycines HDA2 has a putative signal peptide and lacks a transmembrane helix, which makes it a putatively secreted protein. To our knowledge, there are no prior reports of signal-peptide-containing histone deacetylases. Cyst nematodes like H. glycines are known to possess plant-like proteins with signal peptides that are normally not secreted in plants or other organisms, e.g., SKP1 or chorismate mutases 7. Hence, a secreted Arabidopsis-like histone deacetylase would be an exciting new example of how cyst nematodes could modify gene expression of plants by altering the epigenome of their host cells.

To analyze this H. glycines gene product further, we constructed a multiple alignment and a phylogenetic tree for the translated Affymetrix consensus sequence of this probeset (HgAffx.19783.1.S1_at) and HDA2 homologs of selected plant and nematode species using CLUSTAL W 43. It can be clearly seen that the H. glycines HDA2 is not only more similar to Arabidopsis HDA2, but to other plant HDA2s as well, and that the respective homologs of other nematode species are very different from plant HDA2s (Figure 4, Additional file 12).

In addition to the 7,530 probesets corresponding to H. glycines mRNAs, the Affymetrix Soybean Genome Array GeneChip contains 37,500 probesets for soybean (Glycine max) mRNAs and 15,800 probesets for mRNAs of Phytophthora sojae, an oomycete pathogen of soybean plants. We found 576 G. max and 134 P. sojae probesets that hybridized strongly and repeatedly to H. glycines probes in three independent experiments (Additional files 13, 14) and that showed differential expression (Figures 5, 6). A variety of BLAST searches of cross-hybridizing probesets (BLASTX, BLASTN against non-redundant GenBank; BLASTN against dbEST_other database) detected only a negligible number of Caenorhabditis hits and did not lead to matches for cyst nematode sequences, but primarily soybean and oomycete or fungus sequences as best hits (data not shown). These results indicate that the cross-hybridizing probesets in all likelihood originated from soybean plants and P. sojae and were not H. glycines contaminations in cDNA libraries of these two organisms because we did not find a single hit with a bona fide H. glycines database entry, which otherwise would have to have occurred. The possibility of falsely annotated H. glycines ESTs can be ruled out based on the same results. To determine whether in fact there are highly conserved cyst nematode sequences that could cross-hybridize with gene sequences of soybean or P. sojae, we conducted BLASTN searches of the cross-hybridizing soybean and P. sojae probesets against an in-house database of all cyst nematode (Heterodera spp. and Globodera spp.) sequences. Of 576 soybean probesets, 119 (20.7%) indeed matched known cyst nematode genes and of 134 P. sojae probesets 12 (9.0%) in fact had matches with known Heterodera spp. or Globodera spp. genes (Additional files 13, 14). In other words, in all likelihood, sequence conservation was responsible for the observed cross-hybridization results.

To complement the BLAST searches of the cross-hybridizing soybean and P. sojae probesets, we used InterProScan 44 to identify known protein domains. We found 336 soybean probesets that aligned to 207 unique InterPro domains and 83 P. sojae probesets that aligned to 70 unique InterPro domains, respectively. A summary of the 25 most abundant InterPro domains for both organisms can be viewed in Table 2.

Taken together, the soybean and P. sojae probesets that surprisingly cross-hybridized with H. glycines probes potentially provide a lead to an additional set of nematode genes from which novel parasitism-associated genes can be isolated and confirmed. Ultimately, this aspect can be advanced further only with a future release of a complete H. glycines genome sequence, because the recent deposition of a large number of genome fragments did not allow an exhaustive and conclusive analysis (data not shown).

The microarray data analyzed here is based on our previously published experiments 35. Briefly, for that study, we planted forty pots of Kenwood 94 soybean seed under greenhouse conditions in three replications and all nematodes used within a given replication were from the same experimental setup and the same batch of eggs. Two weeks after planting, each pot was inoculated with 15,000–20,000 H. glycines strain OP-50 51 infective J2. The inoculum was collected from 4 day old hatch chambers, each containing about two million H. glycines OP-50 eggs. From the same batch of eggs used in the hatch chamber, 50,000 eggs were collected and flash frozen in liquid nitrogen, for use as the egg stage hybridization probe. After four days, an aliquot of 50,000 hatched infective J2 was flash frozen in liquid nitrogen, for use as the infective J2 stage probe. The remainder of hatched infective J2 was divided among the forty pots for seedling inoculation. Four days after infection, parasitic J2 were collected from twelve pots. Eight days after infection, another twelve pots were harvested for collection of J3 juveniles and fourteen days after infection, a further ten pots were used to collect J4 juveniles. Finally, twenty-one days after infection, the final six pots were harvested for collection of adult females.

All data analyzed here is based on our previous results 35. For that study, frozen nematode tissue was disrupted with frozen zirconia beads (BioSpec Products, Bartlesville, OK) in a beadbeater (BioSpec Products, Bartlesville, OK). RNA was isolated using the Versagene kit (Gentra Systems, Minneapolis, MN) as described 35. RNA quality and concentration of each sample were determined by RNA Nanochip on a 2100 Bioanalyzer (Agilent Technologies Inc, Palo Alto, CA) and by a NanoDrop spectrophotometer (NanoDrop Technologies, Wilmington, DE). Standard procedures for reverse transcription and labeling of the probes and for hybridization and scanning of the GeneChips were followed by the Iowa State University GeneChip Facility.

For our previous study 35, from which the microarray data analyzed here has been drawn, we measured expression using 18 Affymetrix Soybean Genome Array GeneChips (3 replications × 6 life stages) using a randomized complete block design with replications as blocks. In that study, the Affymetrix signal data were transformed with the natural log (ln) and normalized by median centering prior to the analysis. The normalized data for each gene were analyzed separately using a standard linear model with fixed effects for replications and stages. To test for a difference in expression between life stages for each probeset, we performed F tests, which resulted in p-values. A q-value was calculated for each p-value following 52 and was used to maintain approximate control of the false discovery rate (FDR) at 5% by declaring q-values at or below 0.05 significant. For a detailed description of biological sample preparation, RNA extraction and experimental design see our previous study 35 and 53.

Clustering was implemented to group the observed expression patterns of differentially expressed genes. We estimated the mean normalized expression level for each probeset in all six life stages. The resulting six estimated values were standardized to have mean 0 and standard deviation 1 within each probeset. The Euclidian distance between any pair of standardized expression profiles was used as a measure of dissimilarity in all clustering algorithms. This approach considers genes with similar expression patterns to be close in six-dimensional space and is equivalent to using (1-r)^0.5 as the measure of dissimilarity, where r is the Pearson correlation coefficient between non-standardized expression profiles. All clusters and related figures were generated using the free open-source statistical software package R. Hierarchical agglomerative clustering using average linkage to measure the dissimilarity between clusters was carried out using the R function hclust from the R cluster library.

The microarray data used here has been validated in our previous study 35 by quantitative real-time PCR (qRT-PCR). For that study, we examined the expression patterns of six genes representing different expression patterns for each of the five consecutive pairs of life stages (egg/infective J2, infective J2/parasitic J2, parasitic J2/J3, J3/J4, J4/female), totaling thirty different genes. The template used for the qRT-PCR was the same biological material used for the microarray hybridizations, and the reactions were performed in technical triplicates. As detailed in 35, we found qualitative agreement for 28 out of 30 tested genes between our GeneChip and qRT-PCR results.

The nucleotide consensus sequences of all 7,530 H. glycines probesets (freely available at Affymetrix 54) were translated into the three forward reading frames (for sense or S1 probesets) or into the three reverse reading frames (for antisense or A1 probesets). Additionally, all nucleotide probeset consensus sequences were translated beginning with the first and second start codons, so that for each probeset up to five translations were generated. All translations were then analyzed for the presence of a signal peptide using SignalP 3.0 55. Only those translations were kept for which the C-score, Y-max, S-max, S-mean and D-score of the SignalP neural network output were positive and for which the SignalP hidden Markov model predicted a signal peptide. We then filtered these translations and kept only those that had at least thirty amino acids after the predicted signal peptide cleavage site. The signal peptides of these translations were cleaved off and the remaining amino acid sequences were checked for the presence of transmembrane helices with the TMHMM software 56. All translations for which a transmembrane helix was predicted were removed. Probeset nucleotide consensus translations that passed this identification protocol were sorted into probesets, from which a final number of genes was determined by taking probeset variants as designed by Affymetrix into account.

Soybean and P. sojae probesets were considered as cross-hybridizing if their respective signal intensities were called 'present' by Affymetrix' GCOS software (Affymetrix, Santa Clara, CA) in all three replications.

All 6,860 unique Affymetrix H. glycines gene sequences (contigs) were aligned against the non-redundant GenBank database (downloaded November 2005) using the following parameters in BLASTX with a post-processing cutoff value of 1e-05: filter = seg, lcfilter, W = 4, T = 20, E = 100, B = 25, V = 25, topcomboN = 1, golmax = 10. For Additional data file 1, all H. glycines parasitism gene nucleotide sequences (as found in GenBank) were aligned against the non-redundant GenBank database using standard NCBI BLASTX parameters with a post-processing cutoff value of 1e-15 (dated March 2008). To analyze cross-hybridizing G. max and P. sojae probesets, the respective Affymetrix probeset nucleotide sequences were aligned against the non-redundant GenBank database (downloaded December 2005) using the following parameters in BLASTX with a post-processing cutoff value of 1e-10: filter = seg, lcfilter, W = 4, T = 20, E = 100, B = 10, V = 10, topcomboN = 1, golmax = 10. Additionally, the same G. max and P. sojae probeset sequences were aligned against the non-redundant GenBank database (downloaded February 2006) using the following parameters in BLASTN with a post-processing cutoff value of 1e-05: M = 1, N = -1, Q = 3, R = 3, lcmask, golmax = 10, topcomboN = 1, filter = seg, B = 100, V = 100, as well as against the 'est_other' database of dbEST (all ESTs other than human and mouse, dated May 12, 2006) using the following parameters in BLASTN with a post-processing cutoff value of 1e-05: M = 1, N = -1, Q = 3, R = 3, lcmask, golmax = 10, topcomboN = 1, filter = seg, and against an in-house cyst nematode nucleotide database (dated March 2006) 35 using the following parameters in BLASTN with a post-processing cutoff value of 1e-05: M = 1, N = -1, Q = 3, R = 3, lcmask, golmax = 10, topcomboN = 1, filter = seg, B = 20, V = 20.

BLASTX search results (cutoff value 1e-05) of 6,860 Affymetrix contigs against the non-redundant GenBank database were manually sorted into best matches to sequences from plants, phytopathogens/phytosymbionts, soil-living microbes and 'other' microbes. As a counter selection, all matches were removed from further analyses for which a Caenorhabditis spp. hit was within 15% of the BLAST score of the best plant or microbe alignment.

InterProScan was run using InterPro data files dated November 2005 (iprscan_PTHR_DATA_12.0.tar). InterProScan translated all 576 cross-hybridizing soybean and all 134 cross-hybridizing P. sojae probesets in six frames and then ran its suite of domain finding tools. We required a minimum translation length of 20 amino acids to be considered by InterProScan, and we used the EGC.0 translation table. Due to the six frames translation, each probeset typically had several alignments amongst the significant open reading frames (ORFs) found in the translation. We kept, as representative of each probeset, the single longest aligning ORF that contained an InterPro domain, even though InterProScan may have found several ORFs for each probeset with alignments to some domain or motif. Results were parsed into files representing expression clusters.

CLUSTAL W 43 was used to align selected HDA2 sequences and to construct a phylogenetic tree. For the phylogenetic tree, the 'neighbor joining' output format from CLUSTAL W was chosen and the 'correct distances' and 'ignore gaps' options were turned off.

All Affymetrix Soybean Genome Array GeneChip raw and normalized data files analyzed here were deposited previously in the MIAME-compliant ArrayExpress database 57 by these authors 35 and are freely available under accession number E-MEXP-1110.