Background

Trypanosoma cruzi, the etiologic agent of Chagas disease in humans, is a protozoan parasite which assumes four morphological stages during its cycle in insect and mammalian hosts. In the reduviid insect vector, T. cruzi epimastigotes replicate extracellularly in the lumen of the gut. As the parasite stages reach the posterior end of the gut they attach to the wall of the rectum and convert to non-replicative, infective metacyclic trypomastigotes that are released in the feces when the bug takes a blood meal. The metacyclic trypomastigotes enter the mammalian host when they are rubbed into the bite wound or other open skin or mucosal surfaces and invade host cells. The intracytoplasmic parasites convert to the replicative, amastigote stage and undergo many rounds of division before transforming into elongated, motile trypomastigote stages which are released when the host cell ruptures. The trypomastigotes are disseminated in the blood and lymph where they may infect virtually any nucleated cell or be taken up by the insect vector to complete the life cycle. A detailed description of the T. cruzi life-cycle with accompanying diagram can be accessed at the web site for the Centers for Disease Control and Prevention http://www.dpd.cdc.gov/dpdx/html/trypanosomiasisamerican.htm.

There are no vaccines for Chagas disease, and specific chemotherapy for T. cruzi infection is hindered by adverse side-effects and questionable efficacy. In order to broadly screen for genes with relevant expression patterns, and to learn more about the biology of T. cruzi in general, we and others have conducted transcriptomic and proteomic analyses of T. cruzi life-cycle stages [1–11]. Although these analyses have highlighted potential drug and vaccine targets, they have been somewhat limited in scope, largely non-quantitative, and rarely correlated transcript and protein abundances. The latter issue is of particular importance for T. cruzi and other kinetoplastids, due to the generally accepted view that regulation of gene expression in the kinetoplastids is almost entirely post-transcriptional (reviewed in [12, 13]). In fact, only one kinetoplastid RNA pol II promoter has been described despite repeated attempts in many laboratories (reviewed i [14]). Other studies have implicated mRNA processing [15], translational repression [16–18], polysome recruitment [19], and codon adaptation [20] in the regulation of gene expression in the kinetoplastids, all processes that would be predicted to mitigate the role of mRNA abundance regulation in determining protein expression levels. Indeed, previous microarray studies in the kinetoplastids have revealed relatively modest numbers of genes whose transcript abundances are significantly regulated (reviewed in [21]). There is, however, plentiful evidence that transcript levels in these parasites are controlled by mRNA decay, involving 3'-UTRs and RNA-binding proteins [19, 22–28]. To determine the extent of mRNA abundance regulation in T. cruzi globally, we performed whole-genome, DNA microarray analysis of the four life-cycle stages of T. cruzi. The results of these analyses were compared with existing protein expression data for T. cruzi to determine the correlation, if any, between transcript and protein relative abundances in this human pathogen.

Results

T. cruzi displays significant stage-regulation of relative transcript abundances for thousands of its genes

Co-hybridization of cDNAs from each T. cruzi life-cycle stage with a reference cDNA sample comprised of all four life-cycle stages on oligonucleotide, whole genome microarrays, revealed that over 83% of the oligonucleotides detected transcript above background levels and were consistent between dye-swap replicates in at least three life-cycle stages (10,256/12,288). Significance Analysis of Microarrays (SAM) [29] determined that a total of 4,992 of these transcripts exhibited statistically significant up or down regulation in at least one of the four life-cycle stages with a median false discovery rate (FDR) of 0.06071% (3.03 spots) and a 90th percentile FDR of 0.40582% (20.26 spots). Whether one includes or excludes members of large gene families from the analysis, >50% of the genes with detectable signals were significantly regulated during the course of the T. cruzi life-cycle (Table 1). The significant regulation of thousands of T. cruzi genes is visually apparent in the heat map of significantly regulated genes (Figure 1). Note that the significantly regulated genes show consistency between biological replicates, particularly in stages where the extent of up or down regulation is the greatest. The expanded clusters in Figure 1 are highly enriched in trypomastigote-upregulated trans-sialidases (TS) and mucins.

Table 1 With or without the contribution of large gene family members, greater than 50% of the Trypanosoma cruzi genes detected on whole-genome, oligonucleotide microarrays were significantly stage-regulated at the RNA level.
Full size table
Figure 1
figure 1

Heat map of genes significantly regulated during the life-cycle of Trypanosoma cruzi. Ratios are log2 (stage/reference), thus yellow bars represent upregulation and blue bars represent downregulation. Two trypomastigote upregulated clusters are expanded. The spots annotated as 'not mappable to annotated ORF' were designed from earlier versions of the T. cruzi genome sequence and subsequently designated as 'obsolete' by TIGR. These spots no longer mapped within an annotated ORF in the final version of the genome sequence, e.g. they were in intergenic regions (such as untranslated regions) and/or were antisense (would hybridize with first strand cDNA from transcripts from the non-coding strand).

Full size image

Of the genes upregulated in two stages, 76% (1083/1423), were co-upregulated in stages occurring in the same host (mammalian hosts – amastigotes and trypomastigotes; insect hosts – epimastigotes and metacyclic trypomastigotes), whereas approximately 20% (282/1423) were co-upregulated in stages with similar biological functions. For example, genes co-upregulated in amastigotes and epimastigotes (dividing stages) were enriched in those involved in DNA repair, including proliferative cell nuclear antigen (PCNA). Likewise, genes co-upregulated in trypomastigotes and metacyclic trypomastigotes (non-dividing, infective stages) included never in mitosis (NIMA) related kinase and were enriched for trans-sialidases (TS), which are involved in invasion (reviewed in [30]).

The validity of T. cruzi microarray data is supported by qRT-PCR and by comparison to existing proteomic data

The accuracy of the microarray results was confirmed by quantitative reverse-transcriptase PCR (qRT-PCR) for a subset of the significantly regulated genes.

For 12 out of 12 genes showing significant up or downregulation by microarray analysis in trypomastigotes, qRT-PCR data agreed in direction of regulation (Figure 2A). Similar agreement was seen in the case of epimastigotes (16 out of 16) and amastigotes (13 out of 14) genes (Figure 2B and 2C).

Figure 2
figure 2

qRT-PCR confirmation of relative transcript abundances for microarray-identified, significantly regulated genes. Mean log2 ratios (stage/reference) for selected genes in selected stages determined by qRT-PCR and microarray analyses. Gene ID's are official gene symbols (prefix Tc00.1047053 is truncated) assigned to annotated genes in the reference CL Brener T. cruzi genome. Error bars are standard deviations for three biological replicates for both the microarray and qRT-PCR data. The qRT-PCR and microarray standard deviations reflect biological, not technical variation. A) trypomastigotes B) epimastigotes C) amastigotes.

Full size image

Further validation of the microarray results came from comparison to known protein expression profiles for selected genes. For example, proteomic analyses indicated most TS gene expression occurs in trypomastigotes and significantly fewer TS proteins are detected in amastigotes and metacyclic trypomastigotes (and none detected in epimastigotes) [8]. A similar pattern of expression was evident in the transcriptome data (Figure 3 and Additional file 8). Gene set enrichment (GSE) analysis [31, 32] of genes >2-fold upregulated in trypomastigotes and significantly regulated (by SAM analysis) also revealed enrichment of trans-sialidases (using the gene ontology (GO) molecular function term "exo-alpha-sialidase activity"; hypergeometric P-value of 4.1E-25; 1.8E-22 with Benjamini correction for FDR). Several TS mRNAs were detected in epimastigotes, in agreement with previous EST analyses [33]. Interestingly, of the 12 TS genes showing 2-fold or greater transcript upregulation in epimastigotes, 11 were also upregulated in metacyclic trypomastigote stages (Figure 4), consistent with the possibility they are translationally repressed in epimastigotes and subsequently translated in metacyclic trypomastigotes.

Figure 3
figure 3

Microarray-identified, significantly regulated Trypanosoma cruzi trans-sialidase (TS) genes. Shown are mean microarray log2 ratios (stage/reference) for TS significantly regulated in Trypanosoma cruzi amastigotes (AMA), trypomastigotes (TRYP), epimastigotes (EPI), and metacyclic trypomastigotes (META). The y-axis scale is provided as a reference for gene order for the genes in Additional file 8, which shows the mean log2 ratios for three biological replicates for each life cycle stage and their standard deviations. The genes are in the same order in each panel.

Full size image
Figure 4
figure 4

Microarray-identified, epimastigote upregulated Trypanosoma cruzi trans-sialidase (TS) genes. Mean microarray log2 ratios (stage/reference) in amastigotes (AMA), trypomastigotes (TRYP), epimastigotes (EPI), and metacyclic trypomastigotes (META) for significantly regulated TS that were at least two-fold upregulated in epimastigotes. Gene ID's are official gene symbols (prefix Tc00.1047053 is truncated) assigned to annotated genes in the reference CL Brener T. cruzi genome. Error bars are standard deviations for three biological replicates and reflect biological, not technical variation.

Full size image

Other gene groups showing concordance between the transcriptome and whole proteome analyses included genes for ribosomal proteins (downregulated in metacyclics) and genes in the histidine-to-glutamate pathway (upregulated in epimastigotes) [8] (Figure 5A and 5B and Additional file 9). We also observed the expected upregulation of mucin genes in trypomastigotes (reviewed in [34]; confirmed by hypergeometric P-value of 1.5E-37 for GSE) and the downregulation of flagellum-associated genes in amastigotes (which lack flagella; hypergeometric P-value of .00153 for GSE) (Figure 5C and 5D and Additional file 10).

Figure 5
figure 5

Transcript relative abundances for selected functional groups agree with published protein expression data. Shown are mean transcript log2 ratios (stage/reference) from T. cruzi oligonucleotide microarray analyses for A) ribosomal protein genes in metacyclic trypomastigotes, B) genes in the histidine-to-glutamate pathway in epimastigotes, C) mucin genes in trypomastigotes, and D) flagellum-associated genes in amastigotes. The ribosomal genes were selected based on their having the term 'ribosomal' in their annotation and their having either proteome or transcriptome data associated with them. Gene ID's in panels B and D are official gene symbols (prefix Tc00.1047053 is truncated) assigned to annotated genes in the reference CL Brener T. cruzi genome. In panels A and C the y-axis scale is provided as a reference for gene order for the genes in Additional files 9 and 10, respectively, which show the mean microarray ratios for three biological replicates for each life cycle stage and their standard deviations. Error bars in panels A, B, and D are standard deviations for three biological replicates and reflect biological, not technical variation.

Full size image

The T. cruzi microarray data provide expanded coverage of important functional groups

The concordance between the presently reported transcriptomic data and published proteomic data suggests that the mRNA relative abundances observed here will be useful indicators of the protein expression levels of genes for which protein expression data has not been reported, including many genes annotated as 'hypothetical.' Current proteomic methods often fail to detect or quantify proteins that are expressed at low levels, (e.g. protein kinases and phosphatases; reviewed in [35]) or proteins that have physical properties making them less amenable to proteomic analyses, (e.g. membrane proteins; reviewed in [36]). For example, this transcriptomic analysis detected 30 protein phosphatases and 128 protein kinases, which were significantly regulated at the transcript level (Additional files 1 and 2), and few of which have published protein expression data. Likewise, our transcriptome analysis detected 166 significantly regulated genes annotated as 'integral-to-membrane' (Additional file 3).

Members of many T. cruzi paralog clusters exhibit significant stage-regulated divergence in mRNA relative abundance patterns

Kinetoplastids appear to maintain multiple copies of genes for which high level expression is required [37, 38]. Hence we expected that genes in paralog clusters would have similar expression patterns and were surprised to find that over 10% (105/1023) of the T. cruzi paralog groups containing 3–15 members had divergent mRNA relative abundance patterns (Figure 6). One of these groups was the amastins, a paralog group previously reported to be largely amastigote-specific [37] but herein shown to include members with significant stage-specific upregulation in insect stages. The divergent expression patterns of the T. cruzi amastins were confirmed by qRT-PCR (Figure 6) and are also supported by the detection of amastin in the metacyclic trypomastigote proteome [8].

Figure 6
figure 6

Duplicate genes in Trypanosoma cruzi show divergent expression patterns. Microarray expression data for paralog groups containing 3–15 members were compared to identify groups showing divergent expression between duplicate genes. Shown are selected groups with representative divergent expression patterns. Gene ID's are official gene symbols (prefix Tc00.1047053 is truncated) assigned to annotated genes in the reference CL Brener T. cruzi genome. Expression data for the amastins were confirmed by qRT-PCR (top right panel). Error bars are standard deviations for three biological replicates for both the microarray and qRT-PCR data. The qRT-PCR and microarray standard deviations reflect biological, not technical variation.

Full size image

In addition to increasing transcript abundance, another presumed function of gene duplication is to allow for the development of new protein functions [38]. Such neofunctionalization may have occurred in the T. cruzi 60S ribosomal L18 protein paralog group (Figure 6). Tc00.1047053503395.40, which was upregulated in trypomastigotes, has a 130 amino acid N-terminal extension not found in any other publicly-available 60S ribosomal L18 sequence, suggesting the possibility that this paralog has a novel function in T. cruzi trypomastigotes.

Discussion

Transcriptomic analyses have not been as aggressively pursued in kinetoplastid organisms as they have been in other protozoan parasites, such as Plasmodium. This is likely due to the view that the predominantly post-transcriptional nature of gene expression regulation in kinetoplastids makes microarray studies in these organisms generally less informative than microarray studies in organisms for which transcription initiation plays a larger role in gene expression regulation. In support of this, the ranges of observed mRNA ratios between life-cycle stages in kinetoplastid organisms have been relatively narrow compared to the ranges observed in organisms with canonical RNA pol II promoters. For example, 1,000-fold induction in relative transcript abundance for CD70 was observed in a microarray study comparing Human T-lymphotropic virus type 1 (HTLV-1) carrying T-cell lines versus HTLV-1-negative T-cell lines [41], whereas in kinetoplastid microarray studies, fold inductions above ~8-fold are rarely observed [3, 42–46].

Further contributing to the debate over the utility of microarray studies in kinetoplastids has been the general observation that relatively few genes have exhibited significant stage regulation of mRNA relative abundances in previous microarray studies. For instance, Diehl et al., using arrays of 21,024 PCR-amplified genome shot-gun library inserts from T. brucei to compare relative transcript abundances between in vitro cultured human, long slender forms and procyclic parasites, observed that, although 75% of the array elements detected transcripts, only 2% displayed significant differences between the two life-cycle stages [39]. The observation that only 2% of the detected transcripts differed significantly in relative abundance between two very different life-cycle stages supported the concept that transcript abundances in T. brucei are largely constitutive, especially considering the two life-cycle stages were compared directly on the microarrays. This picture of predominantly constitutive genome expression at the level of transcript abundance in kinetoplastids was reinforced manifold by subsequent studies in T. brucei [40] and Leishmania spp. (reviewed in [41]).

In light of the findings from T. brucei and Leishmania spp. microarray studies as well as findings from our previous, limited microarray study in T. cruzi [3], we sought to determine if transcript levels are globally variable between life-cycle stages of T. cruzi using whole-genome oligonucleotide microarrays. When cDNAs from each life-cycle stage were co-hybridized with a reference cDNA sample comprised of all four life-cycle stages on oligonucleotide, whole genome microarrays, we observed that the relative transcript abundances for over 50% of the genes detected on the arrays were significantly regulated between the life-cycle stages.

The microarray data for selected genes were validated by qRT-PCR. There was generally good agreement between the two platforms in terms of direction of regulation, however there were some quantitative differences, especially in terms of the extent of up or downregulation estimated by these two techniques. These quantitative differences likely were the result of sequence-specific effects and differences in dynamic range for the two platforms [49, 50]. In addition, the RNA samples used for qRT-PCR were not the identical samples used for the microarrays.

Recent improvements in microarray design, such as the use of oligonucleotide probes designed for more uniform hybridization kinetics and with lower likelihood of cross-hybridization than amplicon-derived probes and the use of microarrays with greater genome coverage could in part account for why our study identified a higher percentage of significantly stage-regulated genes than previous microarray studies in other kinetoplastids [1–4, 39, 42]. However, even a recent study by Rochette et al. that used Leishmania whole genome oligonucleotide microarrays to directly compare amastigotes and procyclics reported that only 7% of the genes in Leishmania infantum and 9.3% of the genes in L. major were developmentally regulated [43]. It is also possible that the use of different statistical methods for analysis of microarray data my yield widely different estimates of the frequency of regulated genes. However, using less rigorous statistical criteria for determining significance, such as fold-change, results in higher false discovery rates [29] and thus would be predicted to overestimate the number of significantly regulated genes. Our reanalysis of the Papadopoulou data from L. infantum using our statistical methods found 715 genes (7% of the genes fulfilling our criteria for detection) to be significantly regulated between amastigotes and promastigotes with an FDR of 1.55% (an FDR nearly 4 times greater than what we report for T. cruzi). Thus, different statistical analysis methods cannot account for the differential estimates of the frequency of regulated genes in Leishmania and T. cruzi.

By incorporating all four life cycle stages of T. cruzi in our analysis, in contrast to the 2 stage comparisons done with other kinetoplastids (e.g. procyclic to amastigote [43] or procyclic to bloodstream forms [39]) we have increased the possibility of identifying a greater number of genes regulated in expression in at least one of the 4 life cycle stages. However, the contribution of increased number of stages studied to the overall number of significantly regulated genes was greatly mitigated by the fact that we used a reference design in which each life-cycle stage was compared to a mixture of all four life-cycle stages. This design minimized the number of microarrays required to compare the four life-cycle stages (each stage vs. reference = 24 hybridizations in the present study, each stage vs. each other stage = 36 hybridizations for equivalent replication), and also simplified the analysis, but had the disadvantage of buffering the ratios observed. By contrast, previous microarray studies in the kinetoplastids involved directly comparing one life-cycle stage with another on the microarrays, an experimental design which should maximize the number of significant ratios. Thus our estimate that >50% of genes in T. cruzi have significantly stage-regulated transcript levels is much more likely to be an underestimate than an overestimate.

The number of significantly stage-regulated transcripts reported here for T. cruzi is also likely to be underestimated because >2,000 of the oligonucleotide probes detected transcripts in two or fewer stages and were removed from further analysis for statistical reasons (significance determination for such genes would have been dubious, because SAM analysis of four groups (life-cycle stages) was unreliable for genes present in less than three groups). Also, almost certainly some of the genes that we reported as non-significantly regulated are significantly regulated in transitional stages of the life-cycle. For instance, Saxena et al., using microarrays of ~8300 PCR-amplified genome survey sequencing (GSS) clone inserts, identified 344 protein coding genes significantly regulated during axenic promastigote-to-amastigote differentiation in Leishmania donovani [52]. Of these, 136 genes, nearly 40% of the total number of differentially regulated genes identified in the study, displayed transient up or downregulation that would have been missed by only comparing fully differentiated parasites. Thus our analysis of only fully differentiated stages likely underestimated the percentage of genes we found significantly regulated at the transcript level during the T. cruzi life cycle, suggesting that time course studies looking at transitional stages in T. cruzi would be fruitful.

The disparity in the degree of transcript abundance regulation between T. cruzi and the other sequenced kinetoplastids suggests there may be differences in the number and kind of RNA binding proteins in their genomes, because almost all regulation of mRNA abundance in these organisms is at the level of mRNA stability. However, the genomes of T. cruzi, T. brucei, and L. major all have remarkably similar complements of RNA binding proteins (reviewed in [44]). Of the 77 proteins with RNA Recognition Motifs (RRMs) in T. cruzi, T. brucei, and L. major, only two were unique to T. cruzi, RBP4 (Tc00.1047053508901.20) and DRBD8 (Tc00.1047053503709.10, Tc00.1047053509581.50), both of which were upregulated in metacyclic trypomastigotes.

Present knowledge of the kinetoplastids does not elicit an obvious biological explanation for why gene expression regulation in T. cruz i might be so different from its closest relatives. Perhaps the most striking difference between T. cruzi and T. brucei and L. major is that only T. cruzi has both intracellular and extracellular life-cycle stages in the mammalian host. Amastigotes must replicate intracellularly and, although sequestered from the actions of antibodies and complement, are subjected to attack by CD8+ cytotoxic T lymphocytes (CTL) [45] and to the effects of host cytokines [46] and antimicrobials, including reactive oxygen species and nitric oxide [47]. Trypomastigotes are extracellular and thus must evade host defenses including complement, antibodies, and phagocytes. An enhanced capacity for regulating transcript abundances may have provided T. cruzi with the agility necessary to rapidly shift between such different environments.

The finding that T. cruzi divergently regulates mRNA relative abundances for members of paralog clusters was unexpected, given the post-transcriptional mode of gene expression regulation in this parasite. Divergent expression of duplicate genes is believed to occur by the accumulation of mutations in non-coding cis regulatory sequences, such as promoters [48]. Since kinetoplastids appear to be largely devoid of canonical promoters, it will be interesting to study the 3' untranslated regions of paralogs with divergent expression patterns to identify the sequences responsible for the differential mRNA abundances.

The differentially regulated T. cruzi duplicate genes identified here will be useful to study the retention and neofunctionalization of duplicate genes in an ancient eukaryote lacking transcriptional control of gene expression. In such an environment, the contributions of mRNA stability control and sub-cellular localization to the evolution of new genes can be studied in isolation from the effects of transcriptional control. Gene knockout studies are currently underway to determine if the array-identified, divergently-expressed paralogs, such as the trypomastigote upregulated 60S ribosomal L18 protein, have novel functions.

Conclusion

T. cruzi displays stage-regulated control of mRNA abundances for a significant proportion of its genes despite the fact that regulation of gene expression in the kinetoplastids is primarily post-transcriptional and despite the fact that previous genomic studies of the kinetoplastids have found that only a low proportion of their genomes are stage-regulated. Moreover, T. cruzi oligonucleotide microarray measurements largely agree with known protein expression data for key functional groups. Taken together, T. cruzi oligonucleotide microarray analysis is a useful screening tool to determine stage-regulated gene expression in this human pathogen.

Methods

Parasites

Brazil strain T. cruzi trypomastigotes were grown in monolayers of Vero cells (ATCC no. CCL-81) in RPMI supplemented with 5% horse serum as previously described [49]. Emergent trypomastigotes were harvested daily and examined by light microscopy to determine the percentages of amastigotes and trypomastigotes. Only preparations containing > 95% trypomastigotes were used in the subsequent studies. Amastigotes were prepared from axenically-induced trypomastigotes as described previously [50]. Briefly, emergent trypomastigote samples were centrifuged at 3,000 × g for 15 m at room temperature, brought to density of 5 × 106/ml in Protein-Free Hybridoma Medium (PFHM-II, Gibco, Bethesda, MD) at pH 5.0, and incubated at 37°C until greater than 95% of the parasites were fully converted to the amastigote stage as determined by microscopic examination. T. cruzi epimastigotes were grown in Liver Infusion Tryptose media (LIT) as previously described [51]. Cultures were harvested during mid-log phase by centrifugation at 3,000 × g for 10 m at room temperature. Metacyclic trypomastigotes were obtained from epimastigotes by axenic induction as previously described [52]. Briefly, log phase epimastigote cultures were centrifuged at 3,000 × g for 15 m at room temperature, brought to a density of 5 × 106/ml in Complete Grace's Insect Medium (Sigma no. G8142, Sigma, St. Louis, MO) supplemented with 10% fetal bovine serum, pH 6.6, and incubated at 29°C for 10–14 d. The percentages of metacyclics were determined by microscopic examination of parasites stained with Dif-Quick (Baxter Diagnostics, McGaw Park, IL).

Preparation of labeled cDNAs

Total RNA was isolated from parasites with the High Pure Total RNA Isolation Kit (Roche) per the manufacturer's instructions, including on-column DNase I digestion. Labeled first strand cDNA was synthesized from the total RNA samples using Superscript™ II reverse transcriptase (Life Technologies, Grand Island, NY) as previously described [3]. Briefly, 10 μg RNA were combined with 2 μg oligo-d(T) (Life Technologies), brought to 10 μl with water, denatured at 70°C for 10 m, and cooled on ice. Final 30 ul reactions were assembled with final concentrations of 1× First Strand Buffer (Life Technologies), 10 mM DTT (Life Technologies), 0.5 mM dATP, dCTP, and dGTP (Amersham Pharmacia Biotech, Piscataway, NJ), 0.1 mM dTTP (Amersham Pharmacia Biotech), 0.1 mM Cy3- or Cy5-dUTP (Amersham Pharmacia Biotech), and 13.3 U/μl Superscript II reverse transcriptase (Life Technologies). The reaction mixtures were incubated at 42°C for 2 h and stopped with the addition of 1.5 μL 20 mM EDTA. To degrade the RNA template, 1.5 μL 500 mM NaOH were added and the reactions were heated at 70°C for 10 m. The reactions were neutralized by adding 1.5 μL 500 mM HCl, and unincorporated fluorescent nucleotides were removed using GFX columns per the manufacturer's instructions (Amersham Pharmacia Biotech). The purified products were eluted using 50 μL TE, pH 8.0, and the labeled cDNA was completely dried in a SpeedVac (Savant Instruments, Holbrook, NY, USA) and resuspended in 10 μL water.

Microarrays

The T. cruzi microarrays were obtained from the Pathogen Functional Genomics Resource Center (PFGRC). The microarray description is available at http://pfgrc.tigr.org. The 12,288 unique array oligonucleotides were sense-strand 70-mers designed against open reading frames in the annotated CL Brener reference genome sequence and were printed in duplicate. The microarrays contained an additional 500 control oligonucleotides designed from Arabidopsis sequences, also printed in duplicate.

To assure accuracy in the spot-to-gene annotation, the sequences of the oligonucleotides used on the TIGR T. cruzi microarrays were remapped onto the T. cruzi genome and assigned to genes. The mapping was complicated by the fact that T. cruzi is diploid and the CL-Brener strain used for genome sequencing is a hybrid strain [53]. Thus, the two alleles for genes in the T. cruz i genome frequently differ by 1–2% in their coding sequences [54], and the terms 'allele' and 'gene' are conflated when describing the T. cruzi genome. Moreover, complete chromosomes were never assembled for the T. cruzi genome [54], and triploidy for some loci has been confirmed [55]. As a result, the total number of 'genes' identified by the oligonucleotides on the microarrays was greater than the number of unique oligonucleotides. Thus, oligonucleotides mapping with greater than 80% homology to three or fewer 'genes' were used in the subsequent analyses, and oligonucleotides that mapped to more than three genes were not included in the enumeration of significantly regulated genes. Additional file 4 shows the results of our remapping of the oligonucleotides to the T. cruzi annotated genome, the original TIGR mapping, and the results of the cross-hybridization prediction. Additional file 5 shows the representation of large gene families (cluster designations per [54]) on the microarrays. The mucin associated surface protein (MASP) family in particular was poorly represented on the microarrays.

Microarray prehybridization

The microarrays were prehybridized immediately preceding hybridization as described on the PFGRC website. Briefly, arrays are incubated in prehybridization solution (4× SSC, 40% formamide, 0.1% SDS, 0.5% BSA, and 25 mM Tris-HCl, pH 8.0) at 42°C for 1 h. Slides are then washed at room temperature in 0.2% SDS, followed by 3 washes with water, and dried by centrifugation.

Microarray hybridization

Six hybridizations were performed for each life-cycle stage. The hybridizations consisted of three dye-swap experiments from three independent samples (biological replicates). In each case, the experimental sample was from a single life-cycle stage and the control sample was an equal mixture of all four life-cycle stages. Hybridizations were performed as previously described [3]. Briefly, equal volumes of control and experimental labeled cDNAs were combined with mouse CoT1-DNA (Life Technologies), poly(A)-DNA (Amersham Pharmacia Biotech), and hybridization buffer to yield final concentrations of 4× SSC, 40% formamide, 0.1% SDS, 0.5 μg/μl CoT1-DNA, 0.5 μg/μl poly(A)-DNA, and 25 mM Tris-Cl, pH 8.0). Samples were heat denatured at 95°C for 3 m, centrifuged at 12,000 × g at room temperature for 1 m, applied to the prehybridized array in a MAUI mixer chamber (BioMicro Systems, Inc., Salt Lake City, Utah), placed in a MAUI hybridization apparatus and hybridized at 42°C for 16–20 h with constant mixing. Following hybridization, the array, with attached mixer, was removed from the MAUI apparatus, the mixer was removed from the array, and the array was washed once for 2 m at room temperature in 300 ml PBS with 0.05% SDS, followed by 3 washes for 2 m each at room temperature in 300 ml PBS, and one wash for 1 m at room temperature in 300 ml 0.2× PBS. The slide was then dried by centrifugation.

Scanning and data analysis

Immediately following hybridization and washing, arrays were scanned on a ScanArray 4000 (GSI Lumonics, Wilmington, MA) at the Integrated Biotech Laboratories at the University of Georgia. Microarrays were scanned at multiple laser and photomultiplier tube (PMT) settings in each channel to obtain Tagged Image File Format (TIFF) images of matching sensitivity in the two channels. Microarray scans were quantified using The Institute for Genomic Research (TIGR) SpotFinder module of TM4 http://www.tm4.org. Individual sub-grids within each scan-specific grid were aligned manually. Background correction was used to flag spots with signal intensities less than local background intensity plus one standard deviation. The Otsu algorithm [56] was used to identify spots. The signal intensity files generated in SpotFinder were imported into the TIGR Microarray Data Analysis System (MIDAS) module of TM4 http://www.tm4.org using default parameters: e.g. MIDAS filtered spots with signal intensity in either channel less than 1, spots with signal intensities less than 2 × local background in either channel, and spots with integrated signal intensities less than 10,000. Since spots typically contained approximately 100 pixels, integrated intensities of 10,000 corresponded to mean signal intensities of 100. Normalization was performed by locally weighted least squares regression (LOWESS) within each sub-grid using the default smoothing parameter of 0.33. Within-slide replicate spots analysis was used to convert the signal intensities of one replicate spot in a pair of replicate spots to the geometric mean of it and its replicate and to set the signal intensities for the other spot to zero, thus compressing the data. The within-slide replicate spots analysis also had the effect of setting the signal intensities to zero for both spots in a replicate pair to zero if the signal intensity in either channel for either spot was zero. The normalized signal intensity files thus generated in MIDAS were paired by dye-swap and checked for dye-swap consistency, again using the default parameters. Spots with greater than 2 standard deviations difference in log2 ratios between dye-swap replicates were assigned zero intensities. Spots that were within threshold consistency were averaged (geometric mean). Thus, each set of two input dye-swap files resulted in one output file. The above described operations, including setting minimum intensity cut-offs, setting signal intensities to zero for spots with signal intensities not significantly above background, averaging of within-slide replicate spots, and setting signal intensities to zero for spots with significant dye bias were standard data manipulations for array data QC and the utilization of technical replication (replicate spots and dye-swap hybridizations) to eliminate the influence of low-quality spots and/or dye bias from the downstream data analyses [57], thus leaving only biological variation to be evaluated, which was the intent of this study. Example data analysis reports for the MIDAS analyses are in Additional file 6. These analyses generated 12 result files, one from each biological replicate for each of the four life-cycle stages. Each row in the final result files corresponded to eight independent measurements (2 channels × 2 replicate spots × 2 dye-swap hybridizations). Using the TIGR MultiExperiment Viewer (MeV) software the result files were filtered to remove genes with zero intensity in more than 3 result files (percentage cut-off of 75%), leaving 10,256 spots. The genes and experiments were median centered and normalized using the default method for this software. To find spots with significant, repeatable, between-stage variation we analyzed the result files by multi-class SAM for 4 groups (amastigotes, trypomastigotes, epimastigotes, metacyclic trypomastigotes – 3 files each). The settings for the SAM analysis were: number of permutations = 1,000 (10 times the default value), select S0 using Tusher et al. method [29] (default), calculate q values? = yes (not default – slow), imputation engine = K nearest neighbors (default), number of neighbors = 10 (default), and construct hierarchical trees? = yes (not default). Significant spots were selected with a median false discovery rate (FDR) of 0.06071% and a 90th percentile FDR of 0.40582%. The microarray data were deposited in the Gene Expression Omnibus http://www.ncbi.nlm.nih.gov/geo/ under the accession GSE14641.

qRT-PCR

Genes from functional groups of interest were selected for qRT-PCR analysis in a given life-cycle stage based on the gene being significantly regulated in the microarray analysis, as determined by SAM.

Total RNA was prepared from 3 independent samples from each life-cycle stage as described above. RNA samples were treated with DNase I (Promega) at a final concentration of 1 U/μg RNA at 37°C for 45 m. RNA was purified from the DNase digestions using RNeasy RNA purification columns (Qiagen). First strand cDNA reactions from 5 μg total RNA were primed with a mixture of oligo d(T) and a cDNA primer (Tc-18S cDNA: AAGAAATATCGGTGAACTTTCG) specific to the 3' end of T. cruzi 18S rRNA. Briefly, 0.5 μg oligo d(T)V and 5 pmoles T. cruzi 18S rRNA primer were added to the RNA samples and the primer/template mixtures were denatured at 65°C for 5 m. The samples were cooled to 42°C and final 20 μl reactions were assembled containing 1× first strand buffer, 10 mM DTT, 0.5 mM dNTP, 40 U RNasin, and 200 U SuperScript II reverse transcriptase. The cDNA reactions were incubated at 42°C for 1.5 h. For each RNA sample, duplicate control cDNA reactions were prepared in which the reverse transcriptase was omitted. Following first strand synthesis, the reactions were heat inactivated at 70°C for 10 m, then treated with 2 U RNase H (Invitrogen) at 37°C for 45 m.

Quantifications of selected T. cruzi genes were performed for stage and reference cDNA samples in duplicate on an iCyler (Bio-Rad Laboratories, Hercules, CA) with an iQ5 Multicolor Real-Time PCR Detection System (Bio-Rad). The primer sequences used in the qRT-PCR analyses are available in Additional file 7. Reactions were prepared containing 5 pmoles forward and reverse primers, 1× iQ SYBR® Green Supermix (Bio-Rad), and 2 μl template DNA. Standard curves were prepared for each run using known quantities of T. cruzi genomic DNA (ten-fold dilutions beginning at 15 ng/μl) and primers for the gene being quantified. The raw quantifications were calculated using the iQ5 Optical Detection System software and normalized to the 18S rRNA values for each sample. The final stage/reference ratios were the averages, for each gene, of the 9 possible normalized stage/reference comparisons (3 stage samples × 3 reference samples).